_Z18populate_functionsv: 548| 1|const auto populate_functions() { 549| 1| using I = simdutf::implementation; 550| 1| using FuzzSignature = void (*)(std::span); 551| | 552| 1|#define ADD(lenfunc, conversionfunc) \ 553| 1| FuzzSignature { \ 554| 1| +[](std::span chardata) { \ 555| 1| const auto c = \ 556| 1| Conversion{ \ 559| 1| &I::lenfunc, &I::conversionfunc, \ 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ 562| 1| c.fuzz(chardata); \ 563| 1| } \ 564| 1| } 565| | 566| 1| return std::array{ 567| | // all these cases require valid input for invoking the convert function 568| | 569| | // see #493 570| | // IGNORE(latin1_length_from_utf16, convert_valid_utf16be_to_latin1), 571| 1| ADD(utf32_length_from_utf16be, convert_valid_utf16be_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 572| 1| ADD(utf8_length_from_utf16be, convert_valid_utf16be_to_utf8), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 573| | 574| | // see #493 575| | // IGNORE(latin1_length_from_utf16, convert_valid_utf16le_to_latin1), 576| 1| ADD(utf32_length_from_utf16le, convert_valid_utf16le_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 577| 1| ADD(utf8_length_from_utf16le, convert_valid_utf16le_to_utf8), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 578| | 579| | // see #493 580| | // IGNORE(latin1_length_from_utf32, convert_valid_utf32_to_latin1), 581| 1| ADD(utf16_length_from_utf32, convert_valid_utf32_to_utf16be), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 582| 1| ADD(utf16_length_from_utf32, convert_valid_utf32_to_utf16le), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 583| 1| ADD(utf8_length_from_utf32, convert_valid_utf32_to_utf8), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 584| | 585| | // see #493 586| | // IGNORE(latin1_length_from_utf8, convert_valid_utf8_to_latin1), 587| 1| ADD(utf16_length_from_utf8, convert_valid_utf8_to_utf16be), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 588| 1| ADD(utf16_length_from_utf8, convert_valid_utf8_to_utf16le), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 589| 1| ADD(utf32_length_from_utf8, convert_valid_utf8_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 590| | 591| | // all these cases operate on arbitrary data 592| 1| ADD(latin1_length_from_utf16, convert_utf16be_to_latin1), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 593| 1| ADD(utf32_length_from_utf16be, convert_utf16be_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 594| 1| ADD(utf8_length_from_utf16be, convert_utf16be_to_utf8), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 595| | 596| 1| ADD(latin1_length_from_utf16, convert_utf16le_to_latin1), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 597| 1| ADD(utf32_length_from_utf16le, convert_utf16le_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 598| 1| ADD(utf8_length_from_utf16le, convert_utf16le_to_utf8), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 599| | 600| 1| ADD(latin1_length_from_utf32, convert_utf32_to_latin1), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 601| 1| ADD(utf16_length_from_utf32, convert_utf32_to_utf16be), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 602| 1| ADD(utf16_length_from_utf32, convert_utf32_to_utf16le), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 603| 1| ADD(utf8_length_from_utf32, convert_utf32_to_utf8), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 604| | 605| 1| ADD(latin1_length_from_utf8, convert_utf8_to_latin1), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 606| 1| ADD(utf16_length_from_utf8, convert_utf8_to_utf16be), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 607| 1| ADD(utf16_length_from_utf8, convert_utf8_to_utf16le), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 608| 1| ADD(utf32_length_from_utf8, convert_utf8_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 609| | 610| | // all these cases operate on arbitrary data and use the _with_errors 611| | // variant 612| 1| ADD(latin1_length_from_utf16, convert_utf16be_to_latin1_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 613| 1| ADD(utf32_length_from_utf16be, convert_utf16be_to_utf32_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 614| 1| ADD(utf8_length_from_utf16be, convert_utf16be_to_utf8_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 615| | 616| 1| ADD(latin1_length_from_utf16, convert_utf16le_to_latin1_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 617| 1| ADD(utf32_length_from_utf16le, convert_utf16le_to_utf32_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 618| 1| ADD(utf8_length_from_utf16le, convert_utf16le_to_utf8_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 619| | 620| 1| ADD(latin1_length_from_utf32, convert_utf32_to_latin1_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 621| 1| ADD(utf16_length_from_utf32, convert_utf32_to_utf16be_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 622| 1| ADD(utf16_length_from_utf32, convert_utf32_to_utf16le_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 623| 1| ADD(utf8_length_from_utf32, convert_utf32_to_utf8_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 624| | 625| 1| ADD(latin1_length_from_utf8, convert_utf8_to_latin1_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 626| 1| ADD(utf16_length_from_utf8, convert_utf8_to_utf16be_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 627| 1| ADD(utf16_length_from_utf8, convert_utf8_to_utf16le_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 628| 1| ADD(utf32_length_from_utf8, convert_utf8_to_utf32_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 629| | 630| | // these are a bit special since all input is valid 631| 1| ADD(utf32_length_from_latin1, convert_latin1_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 632| 1| ADD(utf16_length_from_latin1, convert_latin1_to_utf16be), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 633| 1| ADD(utf16_length_from_latin1, convert_latin1_to_utf16le), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 634| 1| ADD(utf8_length_from_latin1, convert_latin1_to_utf8)}; ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 635| | 636| 1|#undef ADD 637| 1|} LLVMFuzzerTestOneInput: 639| 9.26k|extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size) { 640| 9.26k| static const auto fptrs = populate_functions(); 641| 9.26k| constexpr std::size_t Ncases = fptrs.size(); 642| | 643| | // pick one of the function pointers, based on the fuzz data 644| | // the first byte is which action to take. step forward 645| | // several bytes so the input is aligned. 646| 9.26k| if (size < 4) { ------------------ | Branch (646:7): [True: 3, False: 9.26k] ------------------ 647| 3| return 0; 648| 3| } 649| | 650| 9.26k| constexpr auto actionmask = std::bit_ceil(Ncases) - 1; 651| 9.26k| const auto action = data[0] & actionmask; 652| 9.26k| data += 4; 653| 9.26k| size -= 4; 654| | 655| 9.26k| if (action >= Ncases) { ------------------ | Branch (655:7): [True: 1, False: 9.26k] ------------------ 656| 1| return 0; 657| 1| } 658| | 659| 9.26k| if constexpr (use_separate_allocation) { 660| | // this is better at exercising null input and catch buffer underflows 661| 9.26k| const std::vector separate{data, data + size}; 662| 9.26k| fptrs[action](std::span(separate)); 663| | } else { 664| | std::span chardata{(const char*)data, size}; 665| | fptrs[action](chardata); 666| | } 667| | 668| 9.26k| return 0; 669| 9.26k|} conversion.cpp:_ZZ18populate_functionsvENK3$_0clENSt3__14spanIKcLm18446744073709551615EEE: 554| 87| +[](std::span chardata) { \ 555| 87| const auto c = \ 556| 87| Conversion{ \ 559| 87| &I::lenfunc, &I::conversionfunc, \ 560| 87| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 87|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 87| ::std::void_t(); \ | | 1180| 87| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 87| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 87| constexpr auto _size = _name.size(); \ | | 1183| 87| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 87| return _nameof; }() ------------------ 561| 87| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 87|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 87| ::std::void_t(); \ | | 1180| 87| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 87| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 87| constexpr auto _size = _name.size(); \ | | 1183| 87| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 87| return _nameof; }() ------------------ 562| 87| c.fuzz(chardata); \ 563| 87| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 227| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 227| const FromSpan from{reinterpret_cast(chardata.data()), 178| 227| chardata.size() / sizeof(FromType)}; 179| | 180| 227| static const bool do_print_testcase = 181| 227| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 227| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 227] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 227| do { 189| | // step 0 - is the input valid? 190| 227| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 227| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 227] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 227| From == UtfEncodings::UTF8) { 198| 227| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 227] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 227| } 201| | 202| | // step 2 - what is the required size of the output? 203| 227| const auto [output_length, length_agree] = 204| 227| calculate_length(from, inputisvalid); 205| 227| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 227] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 227| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 101, False: 126] | Branch (208:28): [True: 16, False: 85] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 16| return; 211| 16| } 212| | 213| | // step 3 - run the conversion 214| 211| const auto [written, outputs_agree] = 215| 211| do_conversion(from, output_length, inputisvalid); 216| 211| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 211] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 211| return; 221| 211| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 227| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 227| validation_result verify_valid_input(FromSpan src) const { 231| 227| validation_result ret{}; 232| | 233| 227| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 227| const auto implementations = get_supported_implementations(); 235| 227| std::vector results; 236| 227| results.reserve(implementations.size()); 237| | 238| 681| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 681, False: 227] ------------------ 239| 681| results.push_back( 240| 681| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 681| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 681| const bool validation2 = 245| 681| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 681| src.data(), src.size()); 247| 681| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 681] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 681| } 258| | 259| 227| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 227| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 227] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 227| } else { 273| 227| ret.implementations_agree = true; 274| 227| } 275| 227| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 227| return r.error == simdutf::SUCCESS; 277| 227| }); 278| 227| return ret; 279| 227| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 454| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 479| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 479| return r.error == simdutf::SUCCESS; 277| 479| }); _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 227| bool count_the_input(FromSpan src) const { 290| 227| const auto implementations = get_supported_implementations(); 291| 227| std::vector results; 292| 227| results.reserve(implementations.size()); 293| | 294| 681| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 681, False: 227] ------------------ 295| 681| std::size_t ret; 296| 681| if constexpr (From == UtfEncodings::UTF16BE) { 297| 681| ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 681| results.push_back(ret); 304| 681| } 305| 227| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 227| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 227] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 227| return true; 321| 227| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 454| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 227| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 227| length_result ret{}; 344| | 345| 227| const auto implementations = get_supported_implementations(); 346| 227| std::vector results; 347| 227| results.reserve(implementations.size()); 348| | 349| 681| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 681, False: 227] ------------------ 350| 681| const auto len = invoke_lengthcalc(impl, src); 351| 681| results.push_back(len); 352| 681| ret.length.push_back(len); 353| 681| } 354| | 355| 227| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 227| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 227] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 227| } else { 375| 227| ret.implementations_agree = true; 376| 227| } 377| 227| return ret; 378| 227| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 681| FromSpan src) const { 331| 681| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 681| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 454| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 211| const bool inputisvalid) const { 383| 211| conversion_result ret{}; 384| | 385| 211| const auto implementations = get_supported_implementations(); 386| | 387| 211| std::vector> results; 388| 211| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 211| std::vector> outputbuffers; 393| 211| outputbuffers.reserve(implementations.size()); 394| 844| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 633, False: 211] ------------------ 395| 633| auto impl = implementations[i]; 396| 633| const ToType canary1{42}; 397| 633| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 633| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 633| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 633| const auto success = [](const ConversionResult& r) -> bool { 402| 633| if constexpr (std::is_same_v) { 403| 633| return r != 0; 404| 633| } else { 405| 633| return r.error == simdutf::error_code::SUCCESS; 406| 633| } 407| 633| }(implret1); 408| 633| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 633| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 633| const ToType canary2{25}; 414| 633| const auto outputbuffer_first_run = outputbuffer; 415| 633| std::ranges::fill(outputbuffer, canary2); 416| 633| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 633| src.size(), outputbuffer.data()); 418| | 419| 633| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 633] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 633| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 378, False: 255] | Branch (423:29): [True: 366, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 366| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 366| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 366] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 366| } 440| 633| } 441| 633| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 366, False: 267] ------------------ 442| 633| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 211| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 85, False: 126] ------------------ 447| 255| for (auto& e : results) { ------------------ | Branch (447:20): [True: 255, False: 85] ------------------ 448| 255| e.outputhash.clear(); 449| 255| } 450| 85| } 451| | 452| 211| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 211| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 211] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 211| } else { 474| 211| ret.implementations_agree = true; 475| 211| } 476| 211| return ret; 477| 211| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 633| const auto success = [](const ConversionResult& r) -> bool { 402| 633| if constexpr (std::is_same_v) { 403| 633| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 633| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 422| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_1clENSt3__14spanIKcLm18446744073709551615EEE: 554| 248| +[](std::span chardata) { \ 555| 248| const auto c = \ 556| 248| Conversion{ \ 559| 248| &I::lenfunc, &I::conversionfunc, \ 560| 248| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 248|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 248| ::std::void_t(); \ | | 1180| 248| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 248| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 248| constexpr auto _size = _name.size(); \ | | 1183| 248| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 248| return _nameof; }() ------------------ 561| 248| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 248|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 248| ::std::void_t(); \ | | 1180| 248| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 248| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 248| constexpr auto _size = _name.size(); \ | | 1183| 248| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 248| return _nameof; }() ------------------ 562| 248| c.fuzz(chardata); \ 563| 248| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 421| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 421| const FromSpan from{reinterpret_cast(chardata.data()), 178| 421| chardata.size() / sizeof(FromType)}; 179| | 180| 421| static const bool do_print_testcase = 181| 421| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 421| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 421] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 421| do { 189| | // step 0 - is the input valid? 190| 421| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 421| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 421] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 421| From == UtfEncodings::UTF8) { 198| 421| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 421] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 421| } 201| | 202| | // step 2 - what is the required size of the output? 203| 421| const auto [output_length, length_agree] = 204| 421| calculate_length(from, inputisvalid); 205| 421| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 421] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 421| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 102, False: 319] | Branch (208:28): [True: 22, False: 80] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 22| return; 211| 22| } 212| | 213| | // step 3 - run the conversion 214| 399| const auto [written, outputs_agree] = 215| 399| do_conversion(from, output_length, inputisvalid); 216| 399| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 399] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 399| return; 221| 399| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 421| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 421| validation_result verify_valid_input(FromSpan src) const { 231| 421| validation_result ret{}; 232| | 233| 421| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 421| const auto implementations = get_supported_implementations(); 235| 421| std::vector results; 236| 421| results.reserve(implementations.size()); 237| | 238| 1.26k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.26k, False: 421] ------------------ 239| 1.26k| results.push_back( 240| 1.26k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.26k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.26k| const bool validation2 = 245| 1.26k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.26k| src.data(), src.size()); 247| 1.26k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.26k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.26k| } 258| | 259| 421| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 421| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 421] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 421| } else { 273| 421| ret.implementations_agree = true; 274| 421| } 275| 421| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 421| return r.error == simdutf::SUCCESS; 277| 421| }); 278| 421| return ret; 279| 421| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 842| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 1.05k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.05k| return r.error == simdutf::SUCCESS; 277| 1.05k| }); _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 421| bool count_the_input(FromSpan src) const { 290| 421| const auto implementations = get_supported_implementations(); 291| 421| std::vector results; 292| 421| results.reserve(implementations.size()); 293| | 294| 1.26k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.26k, False: 421] ------------------ 295| 1.26k| std::size_t ret; 296| 1.26k| if constexpr (From == UtfEncodings::UTF16BE) { 297| 1.26k| ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 1.26k| results.push_back(ret); 304| 1.26k| } 305| 421| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 421| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 421] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 421| return true; 321| 421| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 842| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 421| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 421| length_result ret{}; 344| | 345| 421| const auto implementations = get_supported_implementations(); 346| 421| std::vector results; 347| 421| results.reserve(implementations.size()); 348| | 349| 1.26k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.26k, False: 421] ------------------ 350| 1.26k| const auto len = invoke_lengthcalc(impl, src); 351| 1.26k| results.push_back(len); 352| 1.26k| ret.length.push_back(len); 353| 1.26k| } 354| | 355| 421| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 421| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 421] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 421| } else { 375| 421| ret.implementations_agree = true; 376| 421| } 377| 421| return ret; 378| 421| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.26k| FromSpan src) const { 331| 1.26k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.26k| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 842| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 399| const bool inputisvalid) const { 383| 399| conversion_result ret{}; 384| | 385| 399| const auto implementations = get_supported_implementations(); 386| | 387| 399| std::vector> results; 388| 399| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 399| std::vector> outputbuffers; 393| 399| outputbuffers.reserve(implementations.size()); 394| 1.59k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.19k, False: 399] ------------------ 395| 1.19k| auto impl = implementations[i]; 396| 1.19k| const ToType canary1{42}; 397| 1.19k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.19k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.19k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.19k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.19k| if constexpr (std::is_same_v) { 403| 1.19k| return r != 0; 404| 1.19k| } else { 405| 1.19k| return r.error == simdutf::error_code::SUCCESS; 406| 1.19k| } 407| 1.19k| }(implret1); 408| 1.19k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.19k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.19k| const ToType canary2{25}; 414| 1.19k| const auto outputbuffer_first_run = outputbuffer; 415| 1.19k| std::ranges::fill(outputbuffer, canary2); 416| 1.19k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.19k| src.size(), outputbuffer.data()); 418| | 419| 1.19k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.19k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.19k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 957, False: 240] | Branch (423:29): [True: 948, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 948| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 948| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 948] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 948| } 440| 1.19k| } 441| 1.19k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 948, False: 249] ------------------ 442| 1.19k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 399| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 80, False: 319] ------------------ 447| 240| for (auto& e : results) { ------------------ | Branch (447:20): [True: 240, False: 80] ------------------ 448| 240| e.outputhash.clear(); 449| 240| } 450| 80| } 451| | 452| 399| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 399| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 399] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 399| } else { 474| 399| ret.implementations_agree = true; 475| 399| } 476| 399| return ret; 477| 399| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.19k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.19k| if constexpr (std::is_same_v) { 403| 1.19k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.19k| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 798| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_2clENSt3__14spanIKcLm18446744073709551615EEE: 554| 70| +[](std::span chardata) { \ 555| 70| const auto c = \ 556| 70| Conversion{ \ 559| 70| &I::lenfunc, &I::conversionfunc, \ 560| 70| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 70|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 70| ::std::void_t(); \ | | 1180| 70| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 70| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 70| constexpr auto _size = _name.size(); \ | | 1183| 70| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 70| return _nameof; }() ------------------ 561| 70| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 70|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 70| ::std::void_t(); \ | | 1180| 70| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 70| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 70| constexpr auto _size = _name.size(); \ | | 1183| 70| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 70| return _nameof; }() ------------------ 562| 70| c.fuzz(chardata); \ 563| 70| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 233| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 233| const FromSpan from{reinterpret_cast(chardata.data()), 178| 233| chardata.size() / sizeof(FromType)}; 179| | 180| 233| static const bool do_print_testcase = 181| 233| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 233| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 233] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 233| do { 189| | // step 0 - is the input valid? 190| 233| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 233| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 233] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 233| From == UtfEncodings::UTF8) { 198| 233| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 233] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 233| } 201| | 202| | // step 2 - what is the required size of the output? 203| 233| const auto [output_length, length_agree] = 204| 233| calculate_length(from, inputisvalid); 205| 233| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 233] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 233| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 106, False: 127] | Branch (208:28): [True: 6, False: 100] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 6| return; 211| 6| } 212| | 213| | // step 3 - run the conversion 214| 227| const auto [written, outputs_agree] = 215| 227| do_conversion(from, output_length, inputisvalid); 216| 227| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 227] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 227| return; 221| 227| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 233| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 233| validation_result verify_valid_input(FromSpan src) const { 231| 233| validation_result ret{}; 232| | 233| 233| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 233| const auto implementations = get_supported_implementations(); 235| 233| std::vector results; 236| 233| results.reserve(implementations.size()); 237| | 238| 699| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 699, False: 233] ------------------ 239| 699| results.push_back( 240| 699| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 699| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 699| const bool validation2 = 245| 699| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 699| src.data(), src.size()); 247| 699| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 699] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 699| } 258| | 259| 233| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 233| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 233] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 233| } else { 273| 233| ret.implementations_agree = true; 274| 233| } 275| 233| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 233| return r.error == simdutf::SUCCESS; 277| 233| }); 278| 233| return ret; 279| 233| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 466| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 487| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 487| return r.error == simdutf::SUCCESS; 277| 487| }); _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 233| bool count_the_input(FromSpan src) const { 290| 233| const auto implementations = get_supported_implementations(); 291| 233| std::vector results; 292| 233| results.reserve(implementations.size()); 293| | 294| 699| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 699, False: 233] ------------------ 295| 699| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 699| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 699| ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 699| results.push_back(ret); 304| 699| } 305| 233| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 233| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 233] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 233| return true; 321| 233| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 466| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 233| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 233| length_result ret{}; 344| | 345| 233| const auto implementations = get_supported_implementations(); 346| 233| std::vector results; 347| 233| results.reserve(implementations.size()); 348| | 349| 699| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 699, False: 233] ------------------ 350| 699| const auto len = invoke_lengthcalc(impl, src); 351| 699| results.push_back(len); 352| 699| ret.length.push_back(len); 353| 699| } 354| | 355| 233| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 233| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 233] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 233| } else { 375| 233| ret.implementations_agree = true; 376| 233| } 377| 233| return ret; 378| 233| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 699| FromSpan src) const { 331| 699| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 699| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 466| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 227| const bool inputisvalid) const { 383| 227| conversion_result ret{}; 384| | 385| 227| const auto implementations = get_supported_implementations(); 386| | 387| 227| std::vector> results; 388| 227| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 227| std::vector> outputbuffers; 393| 227| outputbuffers.reserve(implementations.size()); 394| 908| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 681, False: 227] ------------------ 395| 681| auto impl = implementations[i]; 396| 681| const ToType canary1{42}; 397| 681| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 681| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 681| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 681| const auto success = [](const ConversionResult& r) -> bool { 402| 681| if constexpr (std::is_same_v) { 403| 681| return r != 0; 404| 681| } else { 405| 681| return r.error == simdutf::error_code::SUCCESS; 406| 681| } 407| 681| }(implret1); 408| 681| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 681| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 681| const ToType canary2{25}; 414| 681| const auto outputbuffer_first_run = outputbuffer; 415| 681| std::ranges::fill(outputbuffer, canary2); 416| 681| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 681| src.size(), outputbuffer.data()); 418| | 419| 681| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 681] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 681| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 381, False: 300] | Branch (423:29): [True: 369, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 369| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 369| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 369] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 369| } 440| 681| } 441| 681| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 369, False: 312] ------------------ 442| 681| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 227| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 100, False: 127] ------------------ 447| 300| for (auto& e : results) { ------------------ | Branch (447:20): [True: 300, False: 100] ------------------ 448| 300| e.outputhash.clear(); 449| 300| } 450| 100| } 451| | 452| 227| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 227| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 227] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 227| } else { 474| 227| ret.implementations_agree = true; 475| 227| } 476| 227| return ret; 477| 227| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 681| const auto success = [](const ConversionResult& r) -> bool { 402| 681| if constexpr (std::is_same_v) { 403| 681| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 681| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 454| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_3clENSt3__14spanIKcLm18446744073709551615EEE: 554| 215| +[](std::span chardata) { \ 555| 215| const auto c = \ 556| 215| Conversion{ \ 559| 215| &I::lenfunc, &I::conversionfunc, \ 560| 215| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 215|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 215| ::std::void_t(); \ | | 1180| 215| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 215| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 215| constexpr auto _size = _name.size(); \ | | 1183| 215| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 215| return _nameof; }() ------------------ 561| 215| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 215|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 215| ::std::void_t(); \ | | 1180| 215| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 215| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 215| constexpr auto _size = _name.size(); \ | | 1183| 215| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 215| return _nameof; }() ------------------ 562| 215| c.fuzz(chardata); \ 563| 215| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 433| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 433| const FromSpan from{reinterpret_cast(chardata.data()), 178| 433| chardata.size() / sizeof(FromType)}; 179| | 180| 433| static const bool do_print_testcase = 181| 433| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 433| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 433] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 433| do { 189| | // step 0 - is the input valid? 190| 433| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 433| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 433] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 433| From == UtfEncodings::UTF8) { 198| 433| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 433] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 433| } 201| | 202| | // step 2 - what is the required size of the output? 203| 433| const auto [output_length, length_agree] = 204| 433| calculate_length(from, inputisvalid); 205| 433| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 433] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 433| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 111, False: 322] | Branch (208:28): [True: 12, False: 99] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 12| return; 211| 12| } 212| | 213| | // step 3 - run the conversion 214| 421| const auto [written, outputs_agree] = 215| 421| do_conversion(from, output_length, inputisvalid); 216| 421| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 421] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 421| return; 221| 421| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 433| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 433| validation_result verify_valid_input(FromSpan src) const { 231| 433| validation_result ret{}; 232| | 233| 433| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 433| const auto implementations = get_supported_implementations(); 235| 433| std::vector results; 236| 433| results.reserve(implementations.size()); 237| | 238| 1.29k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.29k, False: 433] ------------------ 239| 1.29k| results.push_back( 240| 1.29k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.29k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.29k| const bool validation2 = 245| 1.29k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.29k| src.data(), src.size()); 247| 1.29k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.29k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.29k| } 258| | 259| 433| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 433| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 433] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 433| } else { 273| 433| ret.implementations_agree = true; 274| 433| } 275| 433| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 433| return r.error == simdutf::SUCCESS; 277| 433| }); 278| 433| return ret; 279| 433| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 866| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 1.07k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.07k| return r.error == simdutf::SUCCESS; 277| 1.07k| }); _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 433| bool count_the_input(FromSpan src) const { 290| 433| const auto implementations = get_supported_implementations(); 291| 433| std::vector results; 292| 433| results.reserve(implementations.size()); 293| | 294| 1.29k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.29k, False: 433] ------------------ 295| 1.29k| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 1.29k| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 1.29k| ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 1.29k| results.push_back(ret); 304| 1.29k| } 305| 433| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 433| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 433] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 433| return true; 321| 433| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 866| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 433| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 433| length_result ret{}; 344| | 345| 433| const auto implementations = get_supported_implementations(); 346| 433| std::vector results; 347| 433| results.reserve(implementations.size()); 348| | 349| 1.29k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.29k, False: 433] ------------------ 350| 1.29k| const auto len = invoke_lengthcalc(impl, src); 351| 1.29k| results.push_back(len); 352| 1.29k| ret.length.push_back(len); 353| 1.29k| } 354| | 355| 433| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 433| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 433] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 433| } else { 375| 433| ret.implementations_agree = true; 376| 433| } 377| 433| return ret; 378| 433| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.29k| FromSpan src) const { 331| 1.29k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.29k| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 866| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 421| const bool inputisvalid) const { 383| 421| conversion_result ret{}; 384| | 385| 421| const auto implementations = get_supported_implementations(); 386| | 387| 421| std::vector> results; 388| 421| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 421| std::vector> outputbuffers; 393| 421| outputbuffers.reserve(implementations.size()); 394| 1.68k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.26k, False: 421] ------------------ 395| 1.26k| auto impl = implementations[i]; 396| 1.26k| const ToType canary1{42}; 397| 1.26k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.26k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.26k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.26k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.26k| if constexpr (std::is_same_v) { 403| 1.26k| return r != 0; 404| 1.26k| } else { 405| 1.26k| return r.error == simdutf::error_code::SUCCESS; 406| 1.26k| } 407| 1.26k| }(implret1); 408| 1.26k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.26k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.26k| const ToType canary2{25}; 414| 1.26k| const auto outputbuffer_first_run = outputbuffer; 415| 1.26k| std::ranges::fill(outputbuffer, canary2); 416| 1.26k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.26k| src.size(), outputbuffer.data()); 418| | 419| 1.26k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.26k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.26k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 966, False: 297] | Branch (423:29): [True: 954, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 954| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 954| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 954] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 954| } 440| 1.26k| } 441| 1.26k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 954, False: 309] ------------------ 442| 1.26k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 421| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 99, False: 322] ------------------ 447| 297| for (auto& e : results) { ------------------ | Branch (447:20): [True: 297, False: 99] ------------------ 448| 297| e.outputhash.clear(); 449| 297| } 450| 99| } 451| | 452| 421| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 421| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 421] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 421| } else { 474| 421| ret.implementations_agree = true; 475| 421| } 476| 421| return ret; 477| 421| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.26k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.26k| if constexpr (std::is_same_v) { 403| 1.26k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.26k| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 842| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_4clENSt3__14spanIKcLm18446744073709551615EEE: 554| 69| +[](std::span chardata) { \ 555| 69| const auto c = \ 556| 69| Conversion{ \ 559| 69| &I::lenfunc, &I::conversionfunc, \ 560| 69| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 69|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 69| ::std::void_t(); \ | | 1180| 69| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 69| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 69| constexpr auto _size = _name.size(); \ | | 1183| 69| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 69| return _nameof; }() ------------------ 561| 69| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 69|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 69| ::std::void_t(); \ | | 1180| 69| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 69| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 69| constexpr auto _size = _name.size(); \ | | 1183| 69| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 69| return _nameof; }() ------------------ 562| 69| c.fuzz(chardata); \ 563| 69| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 339| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 339| const FromSpan from{reinterpret_cast(chardata.data()), 178| 339| chardata.size() / sizeof(FromType)}; 179| | 180| 339| static const bool do_print_testcase = 181| 339| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 339| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 339] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 339| do { 189| | // step 0 - is the input valid? 190| 339| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 339| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 339] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 339| const auto [output_length, length_agree] = 204| 339| calculate_length(from, inputisvalid); 205| 339| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 339] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 339| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 210, False: 129] | Branch (208:28): [True: 2, False: 208] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 2| return; 211| 2| } 212| | 213| | // step 3 - run the conversion 214| 337| const auto [written, outputs_agree] = 215| 337| do_conversion(from, output_length, inputisvalid); 216| 337| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 337] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 337| return; 221| 337| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 339| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 339| validation_result verify_valid_input(FromSpan src) const { 231| 339| validation_result ret{}; 232| | 233| 339| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 339| const auto implementations = get_supported_implementations(); 235| 339| std::vector results; 236| 339| results.reserve(implementations.size()); 237| | 238| 1.01k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.01k, False: 339] ------------------ 239| 1.01k| results.push_back( 240| 1.01k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.01k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.01k| const bool validation2 = 245| 1.01k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.01k| src.data(), src.size()); 247| 1.01k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.01k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.01k| } 258| | 259| 339| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 339| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 339] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 339| } else { 273| 339| ret.implementations_agree = true; 274| 339| } 275| 339| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 339| return r.error == simdutf::SUCCESS; 277| 339| }); 278| 339| return ret; 279| 339| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 678| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 597| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 597| return r.error == simdutf::SUCCESS; 277| 597| }); _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 339| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 339| length_result ret{}; 344| | 345| 339| const auto implementations = get_supported_implementations(); 346| 339| std::vector results; 347| 339| results.reserve(implementations.size()); 348| | 349| 1.01k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.01k, False: 339] ------------------ 350| 1.01k| const auto len = invoke_lengthcalc(impl, src); 351| 1.01k| results.push_back(len); 352| 1.01k| ret.length.push_back(len); 353| 1.01k| } 354| | 355| 339| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 339| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 339] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 339| } else { 375| 339| ret.implementations_agree = true; 376| 339| } 377| 339| return ret; 378| 339| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.01k| FromSpan src) const { 331| 1.01k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.01k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 678| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 337| const bool inputisvalid) const { 383| 337| conversion_result ret{}; 384| | 385| 337| const auto implementations = get_supported_implementations(); 386| | 387| 337| std::vector> results; 388| 337| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 337| std::vector> outputbuffers; 393| 337| outputbuffers.reserve(implementations.size()); 394| 1.34k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.01k, False: 337] ------------------ 395| 1.01k| auto impl = implementations[i]; 396| 1.01k| const ToType canary1{42}; 397| 1.01k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.01k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.01k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.01k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.01k| if constexpr (std::is_same_v) { 403| 1.01k| return r != 0; 404| 1.01k| } else { 405| 1.01k| return r.error == simdutf::error_code::SUCCESS; 406| 1.01k| } 407| 1.01k| }(implret1); 408| 1.01k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.01k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.01k| const ToType canary2{25}; 414| 1.01k| const auto outputbuffer_first_run = outputbuffer; 415| 1.01k| std::ranges::fill(outputbuffer, canary2); 416| 1.01k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.01k| src.size(), outputbuffer.data()); 418| | 419| 1.01k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.01k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.01k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 387, False: 624] | Branch (423:29): [True: 378, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 378| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 378| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 378] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 378| } 440| 1.01k| } 441| 1.01k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 378, False: 633] ------------------ 442| 1.01k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 337| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 208, False: 129] ------------------ 447| 624| for (auto& e : results) { ------------------ | Branch (447:20): [True: 624, False: 208] ------------------ 448| 624| e.outputhash.clear(); 449| 624| } 450| 208| } 451| | 452| 337| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 337| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 337] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 337| } else { 474| 337| ret.implementations_agree = true; 475| 337| } 476| 337| return ret; 477| 337| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.01k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.01k| if constexpr (std::is_same_v) { 403| 1.01k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.01k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 674| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_5clENSt3__14spanIKcLm18446744073709551615EEE: 554| 149| +[](std::span chardata) { \ 555| 149| const auto c = \ 556| 149| Conversion{ \ 559| 149| &I::lenfunc, &I::conversionfunc, \ 560| 149| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 149|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 149| ::std::void_t(); \ | | 1180| 149| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 149| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 149| constexpr auto _size = _name.size(); \ | | 1183| 149| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 149| return _nameof; }() ------------------ 561| 149| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 149|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 149| ::std::void_t(); \ | | 1180| 149| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 149| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 149| constexpr auto _size = _name.size(); \ | | 1183| 149| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 149| return _nameof; }() ------------------ 562| 149| c.fuzz(chardata); \ 563| 149| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 420| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 420| const FromSpan from{reinterpret_cast(chardata.data()), 178| 420| chardata.size() / sizeof(FromType)}; 179| | 180| 420| static const bool do_print_testcase = 181| 420| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 420| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 420] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 420| do { 189| | // step 0 - is the input valid? 190| 420| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 420| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 420] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 420| const auto [output_length, length_agree] = 204| 420| calculate_length(from, inputisvalid); 205| 420| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 420] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 420| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 204, False: 216] | Branch (208:28): [True: 9, False: 195] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 9| return; 211| 9| } 212| | 213| | // step 3 - run the conversion 214| 411| const auto [written, outputs_agree] = 215| 411| do_conversion(from, output_length, inputisvalid); 216| 411| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 411] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 411| return; 221| 411| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 420| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 420| validation_result verify_valid_input(FromSpan src) const { 231| 420| validation_result ret{}; 232| | 233| 420| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 420| const auto implementations = get_supported_implementations(); 235| 420| std::vector results; 236| 420| results.reserve(implementations.size()); 237| | 238| 1.26k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.26k, False: 420] ------------------ 239| 1.26k| results.push_back( 240| 1.26k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.26k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.26k| const bool validation2 = 245| 1.26k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.26k| src.data(), src.size()); 247| 1.26k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.26k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.26k| } 258| | 259| 420| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 420| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 420] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 420| } else { 273| 420| ret.implementations_agree = true; 274| 420| } 275| 420| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 420| return r.error == simdutf::SUCCESS; 277| 420| }); 278| 420| return ret; 279| 420| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 840| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 852| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 852| return r.error == simdutf::SUCCESS; 277| 852| }); _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 420| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 420| length_result ret{}; 344| | 345| 420| const auto implementations = get_supported_implementations(); 346| 420| std::vector results; 347| 420| results.reserve(implementations.size()); 348| | 349| 1.26k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.26k, False: 420] ------------------ 350| 1.26k| const auto len = invoke_lengthcalc(impl, src); 351| 1.26k| results.push_back(len); 352| 1.26k| ret.length.push_back(len); 353| 1.26k| } 354| | 355| 420| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 420| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 420] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 420| } else { 375| 420| ret.implementations_agree = true; 376| 420| } 377| 420| return ret; 378| 420| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.26k| FromSpan src) const { 331| 1.26k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.26k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 840| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 411| const bool inputisvalid) const { 383| 411| conversion_result ret{}; 384| | 385| 411| const auto implementations = get_supported_implementations(); 386| | 387| 411| std::vector> results; 388| 411| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 411| std::vector> outputbuffers; 393| 411| outputbuffers.reserve(implementations.size()); 394| 1.64k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.23k, False: 411] ------------------ 395| 1.23k| auto impl = implementations[i]; 396| 1.23k| const ToType canary1{42}; 397| 1.23k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.23k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.23k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.23k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.23k| if constexpr (std::is_same_v) { 403| 1.23k| return r != 0; 404| 1.23k| } else { 405| 1.23k| return r.error == simdutf::error_code::SUCCESS; 406| 1.23k| } 407| 1.23k| }(implret1); 408| 1.23k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.23k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.23k| const ToType canary2{25}; 414| 1.23k| const auto outputbuffer_first_run = outputbuffer; 415| 1.23k| std::ranges::fill(outputbuffer, canary2); 416| 1.23k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.23k| src.size(), outputbuffer.data()); 418| | 419| 1.23k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.23k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.23k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 648, False: 585] | Branch (423:29): [True: 636, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 636| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 636| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 636] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 636| } 440| 1.23k| } 441| 1.23k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 636, False: 597] ------------------ 442| 1.23k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 411| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 195, False: 216] ------------------ 447| 585| for (auto& e : results) { ------------------ | Branch (447:20): [True: 585, False: 195] ------------------ 448| 585| e.outputhash.clear(); 449| 585| } 450| 195| } 451| | 452| 411| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 411| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 411] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 411| } else { 474| 411| ret.implementations_agree = true; 475| 411| } 476| 411| return ret; 477| 411| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.23k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.23k| if constexpr (std::is_same_v) { 403| 1.23k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.23k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 822| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_6clENSt3__14spanIKcLm18446744073709551615EEE: 554| 120| +[](std::span chardata) { \ 555| 120| const auto c = \ 556| 120| Conversion{ \ 559| 120| &I::lenfunc, &I::conversionfunc, \ 560| 120| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 120|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 120| ::std::void_t(); \ | | 1180| 120| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 120| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 120| constexpr auto _size = _name.size(); \ | | 1183| 120| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 120| return _nameof; }() ------------------ 561| 120| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 120|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 120| ::std::void_t(); \ | | 1180| 120| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 120| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 120| constexpr auto _size = _name.size(); \ | | 1183| 120| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 120| return _nameof; }() ------------------ 562| 120| c.fuzz(chardata); \ 563| 120| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 444| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 444| const FromSpan from{reinterpret_cast(chardata.data()), 178| 444| chardata.size() / sizeof(FromType)}; 179| | 180| 444| static const bool do_print_testcase = 181| 444| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 444| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 444] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 444| do { 189| | // step 0 - is the input valid? 190| 444| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 444| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 444] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 444| const auto [output_length, length_agree] = 204| 444| calculate_length(from, inputisvalid); 205| 444| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 444] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 444| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 286, False: 158] | Branch (208:28): [True: 12, False: 274] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 12| return; 211| 12| } 212| | 213| | // step 3 - run the conversion 214| 432| const auto [written, outputs_agree] = 215| 432| do_conversion(from, output_length, inputisvalid); 216| 432| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 432] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 432| return; 221| 432| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 444| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 444| validation_result verify_valid_input(FromSpan src) const { 231| 444| validation_result ret{}; 232| | 233| 444| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 444| const auto implementations = get_supported_implementations(); 235| 444| std::vector results; 236| 444| results.reserve(implementations.size()); 237| | 238| 1.33k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.33k, False: 444] ------------------ 239| 1.33k| results.push_back( 240| 1.33k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.33k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.33k| const bool validation2 = 245| 1.33k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.33k| src.data(), src.size()); 247| 1.33k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.33k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.33k| } 258| | 259| 444| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 444| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 444] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 444| } else { 273| 444| ret.implementations_agree = true; 274| 444| } 275| 444| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 444| return r.error == simdutf::SUCCESS; 277| 444| }); 278| 444| return ret; 279| 444| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 888| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 760| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 760| return r.error == simdutf::SUCCESS; 277| 760| }); _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 444| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 444| length_result ret{}; 344| | 345| 444| const auto implementations = get_supported_implementations(); 346| 444| std::vector results; 347| 444| results.reserve(implementations.size()); 348| | 349| 1.33k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.33k, False: 444] ------------------ 350| 1.33k| const auto len = invoke_lengthcalc(impl, src); 351| 1.33k| results.push_back(len); 352| 1.33k| ret.length.push_back(len); 353| 1.33k| } 354| | 355| 444| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 444| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 444] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 444| } else { 375| 444| ret.implementations_agree = true; 376| 444| } 377| 444| return ret; 378| 444| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.33k| FromSpan src) const { 331| 1.33k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.33k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 888| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 432| const bool inputisvalid) const { 383| 432| conversion_result ret{}; 384| | 385| 432| const auto implementations = get_supported_implementations(); 386| | 387| 432| std::vector> results; 388| 432| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 432| std::vector> outputbuffers; 393| 432| outputbuffers.reserve(implementations.size()); 394| 1.72k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.29k, False: 432] ------------------ 395| 1.29k| auto impl = implementations[i]; 396| 1.29k| const ToType canary1{42}; 397| 1.29k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.29k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.29k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.29k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.29k| if constexpr (std::is_same_v) { 403| 1.29k| return r != 0; 404| 1.29k| } else { 405| 1.29k| return r.error == simdutf::error_code::SUCCESS; 406| 1.29k| } 407| 1.29k| }(implret1); 408| 1.29k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.29k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.29k| const ToType canary2{25}; 414| 1.29k| const auto outputbuffer_first_run = outputbuffer; 415| 1.29k| std::ranges::fill(outputbuffer, canary2); 416| 1.29k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.29k| src.size(), outputbuffer.data()); 418| | 419| 1.29k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.29k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.29k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 474, False: 822] | Branch (423:29): [True: 462, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 462| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 462| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 462] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 462| } 440| 1.29k| } 441| 1.29k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 462, False: 834] ------------------ 442| 1.29k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 432| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 274, False: 158] ------------------ 447| 822| for (auto& e : results) { ------------------ | Branch (447:20): [True: 822, False: 274] ------------------ 448| 822| e.outputhash.clear(); 449| 822| } 450| 274| } 451| | 452| 432| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 432| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 432] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 432| } else { 474| 432| ret.implementations_agree = true; 475| 432| } 476| 432| return ret; 477| 432| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.29k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.29k| if constexpr (std::is_same_v) { 403| 1.29k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.29k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 864| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_7clENSt3__14spanIKcLm18446744073709551615EEE: 554| 265| +[](std::span chardata) { \ 555| 265| const auto c = \ 556| 265| Conversion{ \ 559| 265| &I::lenfunc, &I::conversionfunc, \ 560| 265| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 265|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 265| ::std::void_t(); \ | | 1180| 265| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 265| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 265| constexpr auto _size = _name.size(); \ | | 1183| 265| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 265| return _nameof; }() ------------------ 561| 265| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 265|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 265| ::std::void_t(); \ | | 1180| 265| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 265| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 265| constexpr auto _size = _name.size(); \ | | 1183| 265| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 265| return _nameof; }() ------------------ 562| 265| c.fuzz(chardata); \ 563| 265| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 662| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 662| const FromSpan from{reinterpret_cast(chardata.data()), 178| 662| chardata.size() / sizeof(FromType)}; 179| | 180| 662| static const bool do_print_testcase = 181| 662| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 662| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 662] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 662| do { 189| | // step 0 - is the input valid? 190| 662| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 662| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 662] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 662| From == UtfEncodings::UTF8) { 198| 662| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 662] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 662| } 201| | 202| | // step 2 - what is the required size of the output? 203| 662| const auto [output_length, length_agree] = 204| 662| calculate_length(from, inputisvalid); 205| 662| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 662] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 662| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 289, False: 373] | Branch (208:28): [True: 10, False: 279] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 10| return; 211| 10| } 212| | 213| | // step 3 - run the conversion 214| 652| const auto [written, outputs_agree] = 215| 652| do_conversion(from, output_length, inputisvalid); 216| 652| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 652] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 652| return; 221| 652| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 662| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 662| validation_result verify_valid_input(FromSpan src) const { 231| 662| validation_result ret{}; 232| | 233| 662| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 662| const auto implementations = get_supported_implementations(); 235| 662| std::vector results; 236| 662| results.reserve(implementations.size()); 237| | 238| 1.98k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.98k, False: 662] ------------------ 239| 1.98k| results.push_back( 240| 1.98k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.98k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.98k| const bool validation2 = 245| 1.98k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.98k| src.data(), src.size()); 247| 1.98k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.98k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.98k| } 258| | 259| 662| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 662| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 662] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 662| } else { 273| 662| ret.implementations_agree = true; 274| 662| } 275| 662| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 662| return r.error == simdutf::SUCCESS; 277| 662| }); 278| 662| return ret; 279| 662| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 1.32k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 1.40k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.40k| return r.error == simdutf::SUCCESS; 277| 1.40k| }); _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 662| bool count_the_input(FromSpan src) const { 290| 662| const auto implementations = get_supported_implementations(); 291| 662| std::vector results; 292| 662| results.reserve(implementations.size()); 293| | 294| 1.98k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.98k, False: 662] ------------------ 295| 1.98k| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| 1.98k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.98k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.98k| } 303| 1.98k| results.push_back(ret); 304| 1.98k| } 305| 662| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 662| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 662] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 662| return true; 321| 662| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 1.32k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 662| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 662| length_result ret{}; 344| | 345| 662| const auto implementations = get_supported_implementations(); 346| 662| std::vector results; 347| 662| results.reserve(implementations.size()); 348| | 349| 1.98k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.98k, False: 662] ------------------ 350| 1.98k| const auto len = invoke_lengthcalc(impl, src); 351| 1.98k| results.push_back(len); 352| 1.98k| ret.length.push_back(len); 353| 1.98k| } 354| | 355| 662| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 662| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 662] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 662| } else { 375| 662| ret.implementations_agree = true; 376| 662| } 377| 662| return ret; 378| 662| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.98k| FromSpan src) const { 331| 1.98k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.98k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 1.32k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 652| const bool inputisvalid) const { 383| 652| conversion_result ret{}; 384| | 385| 652| const auto implementations = get_supported_implementations(); 386| | 387| 652| std::vector> results; 388| 652| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 652| std::vector> outputbuffers; 393| 652| outputbuffers.reserve(implementations.size()); 394| 2.60k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.95k, False: 652] ------------------ 395| 1.95k| auto impl = implementations[i]; 396| 1.95k| const ToType canary1{42}; 397| 1.95k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.95k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.95k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.95k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.95k| if constexpr (std::is_same_v) { 403| 1.95k| return r != 0; 404| 1.95k| } else { 405| 1.95k| return r.error == simdutf::error_code::SUCCESS; 406| 1.95k| } 407| 1.95k| }(implret1); 408| 1.95k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.95k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.95k| const ToType canary2{25}; 414| 1.95k| const auto outputbuffer_first_run = outputbuffer; 415| 1.95k| std::ranges::fill(outputbuffer, canary2); 416| 1.95k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.95k| src.size(), outputbuffer.data()); 418| | 419| 1.95k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.95k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.95k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 1.11k, False: 837] | Branch (423:29): [True: 1.11k, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 1.11k| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 1.11k| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 1.11k] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 1.11k| } 440| 1.95k| } 441| 1.95k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 1.11k, False: 843] ------------------ 442| 1.95k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 652| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 279, False: 373] ------------------ 447| 837| for (auto& e : results) { ------------------ | Branch (447:20): [True: 837, False: 279] ------------------ 448| 837| e.outputhash.clear(); 449| 837| } 450| 279| } 451| | 452| 652| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 652| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 652] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 652| } else { 474| 652| ret.implementations_agree = true; 475| 652| } 476| 652| return ret; 477| 652| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.95k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.95k| if constexpr (std::is_same_v) { 403| 1.95k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.95k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 1.30k| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_8clENSt3__14spanIKcLm18446744073709551615EEE: 554| 236| +[](std::span chardata) { \ 555| 236| const auto c = \ 556| 236| Conversion{ \ 559| 236| &I::lenfunc, &I::conversionfunc, \ 560| 236| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 236|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 236| ::std::void_t(); \ | | 1180| 236| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 236| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 236| constexpr auto _size = _name.size(); \ | | 1183| 236| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 236| return _nameof; }() ------------------ 561| 236| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 236|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 236| ::std::void_t(); \ | | 1180| 236| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 236| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 236| constexpr auto _size = _name.size(); \ | | 1183| 236| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 236| return _nameof; }() ------------------ 562| 236| c.fuzz(chardata); \ 563| 236| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 624| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 624| const FromSpan from{reinterpret_cast(chardata.data()), 178| 624| chardata.size() / sizeof(FromType)}; 179| | 180| 624| static const bool do_print_testcase = 181| 624| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 624| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 624] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 624| do { 189| | // step 0 - is the input valid? 190| 624| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 624| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 624] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 624| From == UtfEncodings::UTF8) { 198| 624| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 624] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 624| } 201| | 202| | // step 2 - what is the required size of the output? 203| 624| const auto [output_length, length_agree] = 204| 624| calculate_length(from, inputisvalid); 205| 624| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 624] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 624| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 277, False: 347] | Branch (208:28): [True: 11, False: 266] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 11| return; 211| 11| } 212| | 213| | // step 3 - run the conversion 214| 613| const auto [written, outputs_agree] = 215| 613| do_conversion(from, output_length, inputisvalid); 216| 613| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 613] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 613| return; 221| 613| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 624| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 624| validation_result verify_valid_input(FromSpan src) const { 231| 624| validation_result ret{}; 232| | 233| 624| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 624| const auto implementations = get_supported_implementations(); 235| 624| std::vector results; 236| 624| results.reserve(implementations.size()); 237| | 238| 1.87k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.87k, False: 624] ------------------ 239| 1.87k| results.push_back( 240| 1.87k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.87k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.87k| const bool validation2 = 245| 1.87k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.87k| src.data(), src.size()); 247| 1.87k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.87k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.87k| } 258| | 259| 624| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 624| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 624] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 624| } else { 273| 624| ret.implementations_agree = true; 274| 624| } 275| 624| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 624| return r.error == simdutf::SUCCESS; 277| 624| }); 278| 624| return ret; 279| 624| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 1.24k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 1.31k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.31k| return r.error == simdutf::SUCCESS; 277| 1.31k| }); _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 624| bool count_the_input(FromSpan src) const { 290| 624| const auto implementations = get_supported_implementations(); 291| 624| std::vector results; 292| 624| results.reserve(implementations.size()); 293| | 294| 1.87k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.87k, False: 624] ------------------ 295| 1.87k| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| 1.87k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.87k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.87k| } 303| 1.87k| results.push_back(ret); 304| 1.87k| } 305| 624| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 624| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 624] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 624| return true; 321| 624| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 1.24k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 624| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 624| length_result ret{}; 344| | 345| 624| const auto implementations = get_supported_implementations(); 346| 624| std::vector results; 347| 624| results.reserve(implementations.size()); 348| | 349| 1.87k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.87k, False: 624] ------------------ 350| 1.87k| const auto len = invoke_lengthcalc(impl, src); 351| 1.87k| results.push_back(len); 352| 1.87k| ret.length.push_back(len); 353| 1.87k| } 354| | 355| 624| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 624| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 624] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 624| } else { 375| 624| ret.implementations_agree = true; 376| 624| } 377| 624| return ret; 378| 624| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.87k| FromSpan src) const { 331| 1.87k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.87k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 1.24k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 613| const bool inputisvalid) const { 383| 613| conversion_result ret{}; 384| | 385| 613| const auto implementations = get_supported_implementations(); 386| | 387| 613| std::vector> results; 388| 613| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 613| std::vector> outputbuffers; 393| 613| outputbuffers.reserve(implementations.size()); 394| 2.45k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.83k, False: 613] ------------------ 395| 1.83k| auto impl = implementations[i]; 396| 1.83k| const ToType canary1{42}; 397| 1.83k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.83k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.83k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.83k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.83k| if constexpr (std::is_same_v) { 403| 1.83k| return r != 0; 404| 1.83k| } else { 405| 1.83k| return r.error == simdutf::error_code::SUCCESS; 406| 1.83k| } 407| 1.83k| }(implret1); 408| 1.83k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.83k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.83k| const ToType canary2{25}; 414| 1.83k| const auto outputbuffer_first_run = outputbuffer; 415| 1.83k| std::ranges::fill(outputbuffer, canary2); 416| 1.83k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.83k| src.size(), outputbuffer.data()); 418| | 419| 1.83k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.83k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.83k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 1.04k, False: 798] | Branch (423:29): [True: 1.03k, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 1.03k| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 1.03k| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 1.03k] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 1.03k| } 440| 1.83k| } 441| 1.83k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 1.03k, False: 804] ------------------ 442| 1.83k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 613| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 266, False: 347] ------------------ 447| 798| for (auto& e : results) { ------------------ | Branch (447:20): [True: 798, False: 266] ------------------ 448| 798| e.outputhash.clear(); 449| 798| } 450| 266| } 451| | 452| 613| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 613| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 613] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 613| } else { 474| 613| ret.implementations_agree = true; 475| 613| } 476| 613| return ret; 477| 613| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.83k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.83k| if constexpr (std::is_same_v) { 403| 1.83k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.83k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 1.22k| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_9clENSt3__14spanIKcLm18446744073709551615EEE: 554| 242| +[](std::span chardata) { \ 555| 242| const auto c = \ 556| 242| Conversion{ \ 559| 242| &I::lenfunc, &I::conversionfunc, \ 560| 242| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 242|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 242| ::std::void_t(); \ | | 1180| 242| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 242| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 242| constexpr auto _size = _name.size(); \ | | 1183| 242| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 242| return _nameof; }() ------------------ 561| 242| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 242|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 242| ::std::void_t(); \ | | 1180| 242| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 242| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 242| constexpr auto _size = _name.size(); \ | | 1183| 242| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 242| return _nameof; }() ------------------ 562| 242| c.fuzz(chardata); \ 563| 242| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 621| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 621| const FromSpan from{reinterpret_cast(chardata.data()), 178| 621| chardata.size() / sizeof(FromType)}; 179| | 180| 621| static const bool do_print_testcase = 181| 621| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 621| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 621] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 621| do { 189| | // step 0 - is the input valid? 190| 621| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 621| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 621] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 621| From == UtfEncodings::UTF8) { 198| 621| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 621] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 621| } 201| | 202| | // step 2 - what is the required size of the output? 203| 621| const auto [output_length, length_agree] = 204| 621| calculate_length(from, inputisvalid); 205| 621| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 621] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 621| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 257, False: 364] | Branch (208:28): [True: 10, False: 247] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 10| return; 211| 10| } 212| | 213| | // step 3 - run the conversion 214| 611| const auto [written, outputs_agree] = 215| 611| do_conversion(from, output_length, inputisvalid); 216| 611| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 611] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 611| return; 221| 611| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 621| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 621| validation_result verify_valid_input(FromSpan src) const { 231| 621| validation_result ret{}; 232| | 233| 621| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 621| const auto implementations = get_supported_implementations(); 235| 621| std::vector results; 236| 621| results.reserve(implementations.size()); 237| | 238| 1.86k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.86k, False: 621] ------------------ 239| 1.86k| results.push_back( 240| 1.86k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.86k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.86k| const bool validation2 = 245| 1.86k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.86k| src.data(), src.size()); 247| 1.86k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.86k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.86k| } 258| | 259| 621| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 621| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 621] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 621| } else { 273| 621| ret.implementations_agree = true; 274| 621| } 275| 621| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 621| return r.error == simdutf::SUCCESS; 277| 621| }); 278| 621| return ret; 279| 621| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 1.24k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 1.34k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.34k| return r.error == simdutf::SUCCESS; 277| 1.34k| }); _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 621| bool count_the_input(FromSpan src) const { 290| 621| const auto implementations = get_supported_implementations(); 291| 621| std::vector results; 292| 621| results.reserve(implementations.size()); 293| | 294| 1.86k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.86k, False: 621] ------------------ 295| 1.86k| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| 1.86k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.86k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.86k| } 303| 1.86k| results.push_back(ret); 304| 1.86k| } 305| 621| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 621| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 621] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 621| return true; 321| 621| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 1.24k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 621| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 621| length_result ret{}; 344| | 345| 621| const auto implementations = get_supported_implementations(); 346| 621| std::vector results; 347| 621| results.reserve(implementations.size()); 348| | 349| 1.86k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.86k, False: 621] ------------------ 350| 1.86k| const auto len = invoke_lengthcalc(impl, src); 351| 1.86k| results.push_back(len); 352| 1.86k| ret.length.push_back(len); 353| 1.86k| } 354| | 355| 621| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 621| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 621] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 621| } else { 375| 621| ret.implementations_agree = true; 376| 621| } 377| 621| return ret; 378| 621| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.86k| FromSpan src) const { 331| 1.86k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.86k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 1.24k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 611| const bool inputisvalid) const { 383| 611| conversion_result ret{}; 384| | 385| 611| const auto implementations = get_supported_implementations(); 386| | 387| 611| std::vector> results; 388| 611| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 611| std::vector> outputbuffers; 393| 611| outputbuffers.reserve(implementations.size()); 394| 2.44k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.83k, False: 611] ------------------ 395| 1.83k| auto impl = implementations[i]; 396| 1.83k| const ToType canary1{42}; 397| 1.83k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.83k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.83k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.83k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.83k| if constexpr (std::is_same_v) { 403| 1.83k| return r != 0; 404| 1.83k| } else { 405| 1.83k| return r.error == simdutf::error_code::SUCCESS; 406| 1.83k| } 407| 1.83k| }(implret1); 408| 1.83k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.83k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.83k| const ToType canary2{25}; 414| 1.83k| const auto outputbuffer_first_run = outputbuffer; 415| 1.83k| std::ranges::fill(outputbuffer, canary2); 416| 1.83k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.83k| src.size(), outputbuffer.data()); 418| | 419| 1.83k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.83k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.83k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 1.09k, False: 741] | Branch (423:29): [True: 1.08k, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 1.08k| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 1.08k| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 1.08k] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 1.08k| } 440| 1.83k| } 441| 1.83k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 1.08k, False: 747] ------------------ 442| 1.83k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 611| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 247, False: 364] ------------------ 447| 741| for (auto& e : results) { ------------------ | Branch (447:20): [True: 741, False: 247] ------------------ 448| 741| e.outputhash.clear(); 449| 741| } 450| 247| } 451| | 452| 611| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 611| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 611] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 611| } else { 474| 611| ret.implementations_agree = true; 475| 611| } 476| 611| return ret; 477| 611| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.83k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.83k| if constexpr (std::is_same_v) { 403| 1.83k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.83k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 1.22k| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_10clENSt3__14spanIKcLm18446744073709551615EEE: 554| 57| +[](std::span chardata) { \ 555| 57| const auto c = \ 556| 57| Conversion{ \ 559| 57| &I::lenfunc, &I::conversionfunc, \ 560| 57| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 57|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 57| ::std::void_t(); \ | | 1180| 57| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 57| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 57| constexpr auto _size = _name.size(); \ | | 1183| 57| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 57| return _nameof; }() ------------------ 561| 57| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 57|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 57| ::std::void_t(); \ | | 1180| 57| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 57| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 57| constexpr auto _size = _name.size(); \ | | 1183| 57| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 57| return _nameof; }() ------------------ 562| 57| c.fuzz(chardata); \ 563| 57| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 57| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 57| const FromSpan from{reinterpret_cast(chardata.data()), 178| 57| chardata.size() / sizeof(FromType)}; 179| | 180| 57| static const bool do_print_testcase = 181| 57| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 57| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 57] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 57| do { 189| | // step 0 - is the input valid? 190| 57| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 57| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 57] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 57| From == UtfEncodings::UTF8) { 198| 57| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 57] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 57| } 201| | 202| | // step 2 - what is the required size of the output? 203| 57| const auto [output_length, length_agree] = 204| 57| calculate_length(from, inputisvalid); 205| 57| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 57] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 57| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 11, False: 46] | Branch (208:28): [True: 0, False: 11] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 57| const auto [written, outputs_agree] = 215| 57| do_conversion(from, output_length, inputisvalid); 216| 57| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 57] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 57| return; 221| 57| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 57| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 230| 57| validation_result verify_valid_input(FromSpan src) const { 231| 57| validation_result ret{}; 232| | 233| 57| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 57| const auto implementations = get_supported_implementations(); 235| 57| std::vector results; 236| 57| results.reserve(implementations.size()); 237| | 238| 171| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 171, False: 57] ------------------ 239| 171| results.push_back( 240| 171| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 171| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 171| const bool validation2 = 245| 171| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 171| src.data(), src.size()); 247| 171| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 171] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 171| } 258| | 259| 57| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 57| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 57] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 57| } else { 273| 57| ret.implementations_agree = true; 274| 57| } 275| 57| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 57| return r.error == simdutf::SUCCESS; 277| 57| }); 278| 57| return ret; 279| 57| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 114| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 149| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 149| return r.error == simdutf::SUCCESS; 277| 149| }); _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEE: 289| 57| bool count_the_input(FromSpan src) const { 290| 57| const auto implementations = get_supported_implementations(); 291| 57| std::vector results; 292| 57| results.reserve(implementations.size()); 293| | 294| 171| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 171, False: 57] ------------------ 295| 171| std::size_t ret; 296| 171| if constexpr (From == UtfEncodings::UTF16BE) { 297| 171| ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 171| results.push_back(ret); 304| 171| } 305| 57| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 57| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 57] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 57| return true; 321| 57| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 114| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 57| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 57| length_result ret{}; 344| | 345| 57| const auto implementations = get_supported_implementations(); 346| 57| std::vector results; 347| 57| results.reserve(implementations.size()); 348| | 349| 171| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 171, False: 57] ------------------ 350| 171| const auto len = invoke_lengthcalc(impl, src); 351| 171| results.push_back(len); 352| 171| ret.length.push_back(len); 353| 171| } 354| | 355| 57| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 57| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 57] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 57| } else { 375| 57| ret.implementations_agree = true; 376| 57| } 377| 57| return ret; 378| 57| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 171| FromSpan src) const { 339| 171| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 171| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 114| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 57| const bool inputisvalid) const { 383| 57| conversion_result ret{}; 384| | 385| 57| const auto implementations = get_supported_implementations(); 386| | 387| 57| std::vector> results; 388| 57| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 57| std::vector> outputbuffers; 393| 57| outputbuffers.reserve(implementations.size()); 394| 228| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 171, False: 57] ------------------ 395| 171| auto impl = implementations[i]; 396| 171| const ToType canary1{42}; 397| 171| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 171| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 171| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 171| const auto success = [](const ConversionResult& r) -> bool { 402| 171| if constexpr (std::is_same_v) { 403| 171| return r != 0; 404| 171| } else { 405| 171| return r.error == simdutf::error_code::SUCCESS; 406| 171| } 407| 171| }(implret1); 408| 171| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 171| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 171| const ToType canary2{25}; 414| 171| const auto outputbuffer_first_run = outputbuffer; 415| 171| std::ranges::fill(outputbuffer, canary2); 416| 171| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 171| src.size(), outputbuffer.data()); 418| | 419| 171| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 171] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 171| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 138, False: 33] | Branch (423:29): [True: 69, False: 69] ------------------ 424| | // only care about the output if the input is valid 425| 69| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 69| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 69] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 69| } 440| 171| } 441| 171| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 69, False: 102] ------------------ 442| 171| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 57| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 11, False: 46] ------------------ 447| 33| for (auto& e : results) { ------------------ | Branch (447:20): [True: 33, False: 11] ------------------ 448| 33| e.outputhash.clear(); 449| 33| } 450| 11| } 451| | 452| 57| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 57| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 57] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 57| } else { 474| 57| ret.implementations_agree = true; 475| 57| } 476| 57| return ret; 477| 57| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 171| const auto success = [](const ConversionResult& r) -> bool { 402| 171| if constexpr (std::is_same_v) { 403| 171| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 171| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 114| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_11clENSt3__14spanIKcLm18446744073709551615EEE: 554| 140| +[](std::span chardata) { \ 555| 140| const auto c = \ 556| 140| Conversion{ \ 559| 140| &I::lenfunc, &I::conversionfunc, \ 560| 140| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 140|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 140| ::std::void_t(); \ | | 1180| 140| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 140| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 140| constexpr auto _size = _name.size(); \ | | 1183| 140| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 140| return _nameof; }() ------------------ 561| 140| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 140|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 140| ::std::void_t(); \ | | 1180| 140| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 140| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 140| constexpr auto _size = _name.size(); \ | | 1183| 140| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 140| return _nameof; }() ------------------ 562| 140| c.fuzz(chardata); \ 563| 140| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_12clENSt3__14spanIKcLm18446744073709551615EEE: 554| 173| +[](std::span chardata) { \ 555| 173| const auto c = \ 556| 173| Conversion{ \ 559| 173| &I::lenfunc, &I::conversionfunc, \ 560| 173| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 173|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 173| ::std::void_t(); \ | | 1180| 173| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 173| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 173| constexpr auto _size = _name.size(); \ | | 1183| 173| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 173| return _nameof; }() ------------------ 561| 173| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 173|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 173| ::std::void_t(); \ | | 1180| 173| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 173| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 173| constexpr auto _size = _name.size(); \ | | 1183| 173| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 173| return _nameof; }() ------------------ 562| 173| c.fuzz(chardata); \ 563| 173| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_13clENSt3__14spanIKcLm18446744073709551615EEE: 554| 62| +[](std::span chardata) { \ 555| 62| const auto c = \ 556| 62| Conversion{ \ 559| 62| &I::lenfunc, &I::conversionfunc, \ 560| 62| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 62|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 62| ::std::void_t(); \ | | 1180| 62| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 62| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 62| constexpr auto _size = _name.size(); \ | | 1183| 62| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 62| return _nameof; }() ------------------ 561| 62| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 62|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 62| ::std::void_t(); \ | | 1180| 62| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 62| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 62| constexpr auto _size = _name.size(); \ | | 1183| 62| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 62| return _nameof; }() ------------------ 562| 62| c.fuzz(chardata); \ 563| 62| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 62| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 62| const FromSpan from{reinterpret_cast(chardata.data()), 178| 62| chardata.size() / sizeof(FromType)}; 179| | 180| 62| static const bool do_print_testcase = 181| 62| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 62| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 62] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 62| do { 189| | // step 0 - is the input valid? 190| 62| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 62| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 62] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 62| From == UtfEncodings::UTF8) { 198| 62| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 62] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 62| } 201| | 202| | // step 2 - what is the required size of the output? 203| 62| const auto [output_length, length_agree] = 204| 62| calculate_length(from, inputisvalid); 205| 62| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 62] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 62| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 16, False: 46] | Branch (208:28): [True: 0, False: 16] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 62| const auto [written, outputs_agree] = 215| 62| do_conversion(from, output_length, inputisvalid); 216| 62| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 62] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 62| return; 221| 62| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 62| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 230| 62| validation_result verify_valid_input(FromSpan src) const { 231| 62| validation_result ret{}; 232| | 233| 62| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 62| const auto implementations = get_supported_implementations(); 235| 62| std::vector results; 236| 62| results.reserve(implementations.size()); 237| | 238| 186| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 186, False: 62] ------------------ 239| 186| results.push_back( 240| 186| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 186| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 186| const bool validation2 = 245| 186| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 186| src.data(), src.size()); 247| 186| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 186] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 186| } 258| | 259| 62| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 62| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 62] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 62| } else { 273| 62| ret.implementations_agree = true; 274| 62| } 275| 62| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 62| return r.error == simdutf::SUCCESS; 277| 62| }); 278| 62| return ret; 279| 62| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 124| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 154| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 154| return r.error == simdutf::SUCCESS; 277| 154| }); _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEE: 289| 62| bool count_the_input(FromSpan src) const { 290| 62| const auto implementations = get_supported_implementations(); 291| 62| std::vector results; 292| 62| results.reserve(implementations.size()); 293| | 294| 186| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 186, False: 62] ------------------ 295| 186| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 186| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 186| ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 186| results.push_back(ret); 304| 186| } 305| 62| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 62| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 62] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 62| return true; 321| 62| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 124| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 62| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 62| length_result ret{}; 344| | 345| 62| const auto implementations = get_supported_implementations(); 346| 62| std::vector results; 347| 62| results.reserve(implementations.size()); 348| | 349| 186| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 186, False: 62] ------------------ 350| 186| const auto len = invoke_lengthcalc(impl, src); 351| 186| results.push_back(len); 352| 186| ret.length.push_back(len); 353| 186| } 354| | 355| 62| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 62| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 62] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 62| } else { 375| 62| ret.implementations_agree = true; 376| 62| } 377| 62| return ret; 378| 62| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 186| FromSpan src) const { 339| 186| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 186| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 124| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 62| const bool inputisvalid) const { 383| 62| conversion_result ret{}; 384| | 385| 62| const auto implementations = get_supported_implementations(); 386| | 387| 62| std::vector> results; 388| 62| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 62| std::vector> outputbuffers; 393| 62| outputbuffers.reserve(implementations.size()); 394| 248| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 186, False: 62] ------------------ 395| 186| auto impl = implementations[i]; 396| 186| const ToType canary1{42}; 397| 186| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 186| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 186| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 186| const auto success = [](const ConversionResult& r) -> bool { 402| 186| if constexpr (std::is_same_v) { 403| 186| return r != 0; 404| 186| } else { 405| 186| return r.error == simdutf::error_code::SUCCESS; 406| 186| } 407| 186| }(implret1); 408| 186| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 186| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 186| const ToType canary2{25}; 414| 186| const auto outputbuffer_first_run = outputbuffer; 415| 186| std::ranges::fill(outputbuffer, canary2); 416| 186| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 186| src.size(), outputbuffer.data()); 418| | 419| 186| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 186] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 186| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 138, False: 48] | Branch (423:29): [True: 60, False: 78] ------------------ 424| | // only care about the output if the input is valid 425| 60| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 60| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 60] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 60| } 440| 186| } 441| 186| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 60, False: 126] ------------------ 442| 186| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 62| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 16, False: 46] ------------------ 447| 48| for (auto& e : results) { ------------------ | Branch (447:20): [True: 48, False: 16] ------------------ 448| 48| e.outputhash.clear(); 449| 48| } 450| 16| } 451| | 452| 62| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 62| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 62] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 62| } else { 474| 62| ret.implementations_agree = true; 475| 62| } 476| 62| return ret; 477| 62| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 186| const auto success = [](const ConversionResult& r) -> bool { 402| 186| if constexpr (std::is_same_v) { 403| 186| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 186| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 124| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_14clENSt3__14spanIKcLm18446744073709551615EEE: 554| 163| +[](std::span chardata) { \ 555| 163| const auto c = \ 556| 163| Conversion{ \ 559| 163| &I::lenfunc, &I::conversionfunc, \ 560| 163| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 163|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 163| ::std::void_t(); \ | | 1180| 163| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 163| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 163| constexpr auto _size = _name.size(); \ | | 1183| 163| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 163| return _nameof; }() ------------------ 561| 163| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 163|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 163| ::std::void_t(); \ | | 1180| 163| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 163| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 163| constexpr auto _size = _name.size(); \ | | 1183| 163| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 163| return _nameof; }() ------------------ 562| 163| c.fuzz(chardata); \ 563| 163| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_15clENSt3__14spanIKcLm18446744073709551615EEE: 554| 218| +[](std::span chardata) { \ 555| 218| const auto c = \ 556| 218| Conversion{ \ 559| 218| &I::lenfunc, &I::conversionfunc, \ 560| 218| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 218|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 218| ::std::void_t(); \ | | 1180| 218| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 218| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 218| constexpr auto _size = _name.size(); \ | | 1183| 218| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 218| return _nameof; }() ------------------ 561| 218| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 218|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 218| ::std::void_t(); \ | | 1180| 218| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 218| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 218| constexpr auto _size = _name.size(); \ | | 1183| 218| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 218| return _nameof; }() ------------------ 562| 218| c.fuzz(chardata); \ 563| 218| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_16clENSt3__14spanIKcLm18446744073709551615EEE: 554| 114| +[](std::span chardata) { \ 555| 114| const auto c = \ 556| 114| Conversion{ \ 559| 114| &I::lenfunc, &I::conversionfunc, \ 560| 114| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 114|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 114| ::std::void_t(); \ | | 1180| 114| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 114| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 114| constexpr auto _size = _name.size(); \ | | 1183| 114| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 114| return _nameof; }() ------------------ 561| 114| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 114|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 114| ::std::void_t(); \ | | 1180| 114| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 114| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 114| constexpr auto _size = _name.size(); \ | | 1183| 114| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 114| return _nameof; }() ------------------ 562| 114| c.fuzz(chardata); \ 563| 114| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 114| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 114| const FromSpan from{reinterpret_cast(chardata.data()), 178| 114| chardata.size() / sizeof(FromType)}; 179| | 180| 114| static const bool do_print_testcase = 181| 114| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 114| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 114] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 114| do { 189| | // step 0 - is the input valid? 190| 114| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 114| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 114] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 114| const auto [output_length, length_agree] = 204| 114| calculate_length(from, inputisvalid); 205| 114| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 114] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 114| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 81, False: 33] | Branch (208:28): [True: 0, False: 81] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 114| const auto [written, outputs_agree] = 215| 114| do_conversion(from, output_length, inputisvalid); 216| 114| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 114] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 114| return; 221| 114| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 114| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 230| 114| validation_result verify_valid_input(FromSpan src) const { 231| 114| validation_result ret{}; 232| | 233| 114| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 114| const auto implementations = get_supported_implementations(); 235| 114| std::vector results; 236| 114| results.reserve(implementations.size()); 237| | 238| 342| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 342, False: 114] ------------------ 239| 342| results.push_back( 240| 342| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 342| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 342| const bool validation2 = 245| 342| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 342| src.data(), src.size()); 247| 342| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 342] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 342| } 258| | 259| 114| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 114| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 114] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 114| } else { 273| 114| ret.implementations_agree = true; 274| 114| } 275| 114| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 114| return r.error == simdutf::SUCCESS; 277| 114| }); 278| 114| return ret; 279| 114| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 228| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 180| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 180| return r.error == simdutf::SUCCESS; 277| 180| }); _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 114| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 114| length_result ret{}; 344| | 345| 114| const auto implementations = get_supported_implementations(); 346| 114| std::vector results; 347| 114| results.reserve(implementations.size()); 348| | 349| 342| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 342, False: 114] ------------------ 350| 342| const auto len = invoke_lengthcalc(impl, src); 351| 342| results.push_back(len); 352| 342| ret.length.push_back(len); 353| 342| } 354| | 355| 114| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 114| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 114] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 114| } else { 375| 114| ret.implementations_agree = true; 376| 114| } 377| 114| return ret; 378| 114| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 342| FromSpan src) const { 339| 342| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 342| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 228| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 114| const bool inputisvalid) const { 383| 114| conversion_result ret{}; 384| | 385| 114| const auto implementations = get_supported_implementations(); 386| | 387| 114| std::vector> results; 388| 114| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 114| std::vector> outputbuffers; 393| 114| outputbuffers.reserve(implementations.size()); 394| 456| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 342, False: 114] ------------------ 395| 342| auto impl = implementations[i]; 396| 342| const ToType canary1{42}; 397| 342| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 342| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 342| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 342| const auto success = [](const ConversionResult& r) -> bool { 402| 342| if constexpr (std::is_same_v) { 403| 342| return r != 0; 404| 342| } else { 405| 342| return r.error == simdutf::error_code::SUCCESS; 406| 342| } 407| 342| }(implret1); 408| 342| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 342| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 342| const ToType canary2{25}; 414| 342| const auto outputbuffer_first_run = outputbuffer; 415| 342| std::ranges::fill(outputbuffer, canary2); 416| 342| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 342| src.size(), outputbuffer.data()); 418| | 419| 342| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 342] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 342| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 99, False: 243] | Branch (423:29): [True: 78, False: 21] ------------------ 424| | // only care about the output if the input is valid 425| 78| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 78| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 78] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 78| } 440| 342| } 441| 342| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 78, False: 264] ------------------ 442| 342| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 114| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 81, False: 33] ------------------ 447| 243| for (auto& e : results) { ------------------ | Branch (447:20): [True: 243, False: 81] ------------------ 448| 243| e.outputhash.clear(); 449| 243| } 450| 81| } 451| | 452| 114| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 114| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 114] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 114| } else { 474| 114| ret.implementations_agree = true; 475| 114| } 476| 114| return ret; 477| 114| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 342| const auto success = [](const ConversionResult& r) -> bool { 402| 342| if constexpr (std::is_same_v) { 403| 342| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 342| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 228| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_17clENSt3__14spanIKcLm18446744073709551615EEE: 554| 270| +[](std::span chardata) { \ 555| 270| const auto c = \ 556| 270| Conversion{ \ 559| 270| &I::lenfunc, &I::conversionfunc, \ 560| 270| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 270|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 270| ::std::void_t(); \ | | 1180| 270| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 270| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 270| constexpr auto _size = _name.size(); \ | | 1183| 270| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 270| return _nameof; }() ------------------ 561| 270| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 270|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 270| ::std::void_t(); \ | | 1180| 270| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 270| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 270| constexpr auto _size = _name.size(); \ | | 1183| 270| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 270| return _nameof; }() ------------------ 562| 270| c.fuzz(chardata); \ 563| 270| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_18clENSt3__14spanIKcLm18446744073709551615EEE: 554| 271| +[](std::span chardata) { \ 555| 271| const auto c = \ 556| 271| Conversion{ \ 559| 271| &I::lenfunc, &I::conversionfunc, \ 560| 271| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 271|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 271| ::std::void_t(); \ | | 1180| 271| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 271| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 271| constexpr auto _size = _name.size(); \ | | 1183| 271| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 271| return _nameof; }() ------------------ 561| 271| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 271|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 271| ::std::void_t(); \ | | 1180| 271| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 271| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 271| constexpr auto _size = _name.size(); \ | | 1183| 271| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 271| return _nameof; }() ------------------ 562| 271| c.fuzz(chardata); \ 563| 271| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_19clENSt3__14spanIKcLm18446744073709551615EEE: 554| 324| +[](std::span chardata) { \ 555| 324| const auto c = \ 556| 324| Conversion{ \ 559| 324| &I::lenfunc, &I::conversionfunc, \ 560| 324| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 324|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 324| ::std::void_t(); \ | | 1180| 324| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 324| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 324| constexpr auto _size = _name.size(); \ | | 1183| 324| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 324| return _nameof; }() ------------------ 561| 324| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 324|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 324| ::std::void_t(); \ | | 1180| 324| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 324| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 324| constexpr auto _size = _name.size(); \ | | 1183| 324| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 324| return _nameof; }() ------------------ 562| 324| c.fuzz(chardata); \ 563| 324| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_20clENSt3__14spanIKcLm18446744073709551615EEE: 554| 293| +[](std::span chardata) { \ 555| 293| const auto c = \ 556| 293| Conversion{ \ 559| 293| &I::lenfunc, &I::conversionfunc, \ 560| 293| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 293|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 293| ::std::void_t(); \ | | 1180| 293| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 293| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 293| constexpr auto _size = _name.size(); \ | | 1183| 293| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 293| return _nameof; }() ------------------ 561| 293| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 293|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 293| ::std::void_t(); \ | | 1180| 293| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 293| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 293| constexpr auto _size = _name.size(); \ | | 1183| 293| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 293| return _nameof; }() ------------------ 562| 293| c.fuzz(chardata); \ 563| 293| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 293| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 293| const FromSpan from{reinterpret_cast(chardata.data()), 178| 293| chardata.size() / sizeof(FromType)}; 179| | 180| 293| static const bool do_print_testcase = 181| 293| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 293| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 293] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 293| do { 189| | // step 0 - is the input valid? 190| 293| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 293| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 293] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 293| From == UtfEncodings::UTF8) { 198| 293| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 293] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 293| } 201| | 202| | // step 2 - what is the required size of the output? 203| 293| const auto [output_length, length_agree] = 204| 293| calculate_length(from, inputisvalid); 205| 293| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 293] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 293| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 199, False: 94] | Branch (208:28): [True: 0, False: 199] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 293| const auto [written, outputs_agree] = 215| 293| do_conversion(from, output_length, inputisvalid); 216| 293| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 293] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 293| return; 221| 293| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 293| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 293| validation_result verify_valid_input(FromSpan src) const { 231| 293| validation_result ret{}; 232| | 233| 293| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 293| const auto implementations = get_supported_implementations(); 235| 293| std::vector results; 236| 293| results.reserve(implementations.size()); 237| | 238| 879| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 879, False: 293] ------------------ 239| 879| results.push_back( 240| 879| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 879| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 879| const bool validation2 = 245| 879| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 879| src.data(), src.size()); 247| 879| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 879] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 879| } 258| | 259| 293| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 293| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 293] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 293| } else { 273| 293| ret.implementations_agree = true; 274| 293| } 275| 293| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 293| return r.error == simdutf::SUCCESS; 277| 293| }); 278| 293| return ret; 279| 293| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 586| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 481| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 481| return r.error == simdutf::SUCCESS; 277| 481| }); _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 293| bool count_the_input(FromSpan src) const { 290| 293| const auto implementations = get_supported_implementations(); 291| 293| std::vector results; 292| 293| results.reserve(implementations.size()); 293| | 294| 879| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 879, False: 293] ------------------ 295| 879| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| 879| } else if constexpr (From == UtfEncodings::UTF8) { 301| 879| ret = impl->count_utf8(src.data(), src.size()); 302| 879| } 303| 879| results.push_back(ret); 304| 879| } 305| 293| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 293| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 293] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 293| return true; 321| 293| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 586| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 293| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 293| length_result ret{}; 344| | 345| 293| const auto implementations = get_supported_implementations(); 346| 293| std::vector results; 347| 293| results.reserve(implementations.size()); 348| | 349| 879| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 879, False: 293] ------------------ 350| 879| const auto len = invoke_lengthcalc(impl, src); 351| 879| results.push_back(len); 352| 879| ret.length.push_back(len); 353| 879| } 354| | 355| 293| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 293| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 293] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 293| } else { 375| 293| ret.implementations_agree = true; 376| 293| } 377| 293| return ret; 378| 293| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 879| FromSpan src) const { 331| 879| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 879| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 586| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 293| const bool inputisvalid) const { 383| 293| conversion_result ret{}; 384| | 385| 293| const auto implementations = get_supported_implementations(); 386| | 387| 293| std::vector> results; 388| 293| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 293| std::vector> outputbuffers; 393| 293| outputbuffers.reserve(implementations.size()); 394| 1.17k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 879, False: 293] ------------------ 395| 879| auto impl = implementations[i]; 396| 879| const ToType canary1{42}; 397| 879| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 879| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 879| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 879| const auto success = [](const ConversionResult& r) -> bool { 402| 879| if constexpr (std::is_same_v) { 403| 879| return r != 0; 404| 879| } else { 405| 879| return r.error == simdutf::error_code::SUCCESS; 406| 879| } 407| 879| }(implret1); 408| 879| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 879| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 879| const ToType canary2{25}; 414| 879| const auto outputbuffer_first_run = outputbuffer; 415| 879| std::ranges::fill(outputbuffer, canary2); 416| 879| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 879| src.size(), outputbuffer.data()); 418| | 419| 879| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 879] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 879| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 282, False: 597] | Branch (423:29): [True: 252, False: 30] ------------------ 424| | // only care about the output if the input is valid 425| 252| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 252| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 252] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 252| } 440| 879| } 441| 879| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 252, False: 627] ------------------ 442| 879| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 293| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 199, False: 94] ------------------ 447| 597| for (auto& e : results) { ------------------ | Branch (447:20): [True: 597, False: 199] ------------------ 448| 597| e.outputhash.clear(); 449| 597| } 450| 199| } 451| | 452| 293| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 293| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 293] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 293| } else { 474| 293| ret.implementations_agree = true; 475| 293| } 476| 293| return ret; 477| 293| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 879| const auto success = [](const ConversionResult& r) -> bool { 402| 879| if constexpr (std::is_same_v) { 403| 879| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 879| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 586| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_21clENSt3__14spanIKcLm18446744073709551615EEE: 554| 397| +[](std::span chardata) { \ 555| 397| const auto c = \ 556| 397| Conversion{ \ 559| 397| &I::lenfunc, &I::conversionfunc, \ 560| 397| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 397|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 397| ::std::void_t(); \ | | 1180| 397| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 397| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 397| constexpr auto _size = _name.size(); \ | | 1183| 397| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 397| return _nameof; }() ------------------ 561| 397| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 397|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 397| ::std::void_t(); \ | | 1180| 397| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 397| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 397| constexpr auto _size = _name.size(); \ | | 1183| 397| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 397| return _nameof; }() ------------------ 562| 397| c.fuzz(chardata); \ 563| 397| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_22clENSt3__14spanIKcLm18446744073709551615EEE: 554| 388| +[](std::span chardata) { \ 555| 388| const auto c = \ 556| 388| Conversion{ \ 559| 388| &I::lenfunc, &I::conversionfunc, \ 560| 388| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 388|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 388| ::std::void_t(); \ | | 1180| 388| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 388| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 388| constexpr auto _size = _name.size(); \ | | 1183| 388| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 388| return _nameof; }() ------------------ 561| 388| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 388|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 388| ::std::void_t(); \ | | 1180| 388| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 388| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 388| constexpr auto _size = _name.size(); \ | | 1183| 388| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 388| return _nameof; }() ------------------ 562| 388| c.fuzz(chardata); \ 563| 388| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_23clENSt3__14spanIKcLm18446744073709551615EEE: 554| 379| +[](std::span chardata) { \ 555| 379| const auto c = \ 556| 379| Conversion{ \ 559| 379| &I::lenfunc, &I::conversionfunc, \ 560| 379| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 379|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 379| ::std::void_t(); \ | | 1180| 379| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 379| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 379| constexpr auto _size = _name.size(); \ | | 1183| 379| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 379| return _nameof; }() ------------------ 561| 379| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 379|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 379| ::std::void_t(); \ | | 1180| 379| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 379| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 379| constexpr auto _size = _name.size(); \ | | 1183| 379| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 379| return _nameof; }() ------------------ 562| 379| c.fuzz(chardata); \ 563| 379| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_24clENSt3__14spanIKcLm18446744073709551615EEE: 554| 133| +[](std::span chardata) { \ 555| 133| const auto c = \ 556| 133| Conversion{ \ 559| 133| &I::lenfunc, &I::conversionfunc, \ 560| 133| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 133|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 133| ::std::void_t(); \ | | 1180| 133| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 133| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 133| constexpr auto _size = _name.size(); \ | | 1183| 133| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 133| return _nameof; }() ------------------ 561| 133| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 133|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 133| ::std::void_t(); \ | | 1180| 133| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 133| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 133| constexpr auto _size = _name.size(); \ | | 1183| 133| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 133| return _nameof; }() ------------------ 562| 133| c.fuzz(chardata); \ 563| 133| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 133| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 133| const FromSpan from{reinterpret_cast(chardata.data()), 178| 133| chardata.size() / sizeof(FromType)}; 179| | 180| 133| static const bool do_print_testcase = 181| 133| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 133| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 133] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 133| do { 189| | // step 0 - is the input valid? 190| 133| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 133| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 133] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 133| From == UtfEncodings::UTF8) { 198| 133| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 133] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 133| } 201| | 202| | // step 2 - what is the required size of the output? 203| 133| const auto [output_length, length_agree] = 204| 133| calculate_length(from, inputisvalid); 205| 133| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 133] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 133| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 31, False: 102] | Branch (208:28): [True: 0, False: 31] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 133| const auto [written, outputs_agree] = 215| 133| do_conversion(from, output_length, inputisvalid); 216| 133| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 133] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 133| return; 221| 133| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 133| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEE: 230| 133| validation_result verify_valid_input(FromSpan src) const { 231| 133| validation_result ret{}; 232| | 233| 133| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 133| const auto implementations = get_supported_implementations(); 235| 133| std::vector results; 236| 133| results.reserve(implementations.size()); 237| | 238| 399| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 399, False: 133] ------------------ 239| 399| results.push_back( 240| 399| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 399| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 399| const bool validation2 = 245| 399| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 399| src.data(), src.size()); 247| 399| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 399] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 399| } 258| | 259| 133| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 133| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 133] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 133| } else { 273| 133| ret.implementations_agree = true; 274| 133| } 275| 133| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 133| return r.error == simdutf::SUCCESS; 277| 133| }); 278| 133| return ret; 279| 133| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS5_S5_EEDaSJ_SM_: 259| 266| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKS5_E_clESI_: 275| 337| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 337| return r.error == simdutf::SUCCESS; 277| 337| }); _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEE: 289| 133| bool count_the_input(FromSpan src) const { 290| 133| const auto implementations = get_supported_implementations(); 291| 133| std::vector results; 292| 133| results.reserve(implementations.size()); 293| | 294| 399| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 399, False: 133] ------------------ 295| 399| std::size_t ret; 296| 399| if constexpr (From == UtfEncodings::UTF16BE) { 297| 399| ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 399| results.push_back(ret); 304| 399| } 305| 133| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 133| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 133] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 133| return true; 321| 133| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 266| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEb: 342| 133| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 133| length_result ret{}; 344| | 345| 133| const auto implementations = get_supported_implementations(); 346| 133| std::vector results; 347| 133| results.reserve(implementations.size()); 348| | 349| 399| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 399, False: 133] ------------------ 350| 399| const auto len = invoke_lengthcalc(impl, src); 351| 399| results.push_back(len); 352| 399| ret.length.push_back(len); 353| 399| } 354| | 355| 133| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 133| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 133] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 133| } else { 375| 133| ret.implementations_agree = true; 376| 133| } 377| 133| return ret; 378| 133| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSF_NSt3__14spanIS6_Lm18446744073709551615EEE: 338| 399| FromSpan src) const { 339| 399| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 399| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 266| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 133| const bool inputisvalid) const { 383| 133| conversion_result ret{}; 384| | 385| 133| const auto implementations = get_supported_implementations(); 386| | 387| 133| std::vector> results; 388| 133| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 133| std::vector> outputbuffers; 393| 133| outputbuffers.reserve(implementations.size()); 394| 532| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 399, False: 133] ------------------ 395| 399| auto impl = implementations[i]; 396| 399| const ToType canary1{42}; 397| 399| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 399| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 399| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 399| const auto success = [](const ConversionResult& r) -> bool { 402| 399| if constexpr (std::is_same_v) { 403| 399| return r != 0; 404| 399| } else { 405| 399| return r.error == simdutf::error_code::SUCCESS; 406| 399| } 407| 399| }(implret1); 408| 399| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 399| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 399| const ToType canary2{25}; 414| 399| const auto outputbuffer_first_run = outputbuffer; 415| 399| std::ranges::fill(outputbuffer, canary2); 416| 399| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 399| src.size(), outputbuffer.data()); 418| | 419| 399| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 399] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 399| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 306, False: 93] | Branch (423:29): [True: 138, False: 168] ------------------ 424| | // only care about the output if the input is valid 425| 138| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 138| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 138] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 138| } 440| 399| } 441| 399| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 138, False: 261] ------------------ 442| 399| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 133| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 31, False: 102] ------------------ 447| 93| for (auto& e : results) { ------------------ | Branch (447:20): [True: 93, False: 31] ------------------ 448| 93| e.outputhash.clear(); 449| 93| } 450| 31| } 451| | 452| 133| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 133| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 133] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 133| } else { 474| 133| ret.implementations_agree = true; 475| 133| } 476| 133| return ret; 477| 133| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS5_E_clESM_: 401| 399| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 399| } else { 405| 399| return r.error == simdutf::error_code::SUCCESS; 406| 399| } 407| 399| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS5_ESU_EEDaSN_SQ_: 452| 266| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_25clENSt3__14spanIKcLm18446744073709551615EEE: 554| 173| +[](std::span chardata) { \ 555| 173| const auto c = \ 556| 173| Conversion{ \ 559| 173| &I::lenfunc, &I::conversionfunc, \ 560| 173| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 173|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 173| ::std::void_t(); \ | | 1180| 173| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 173| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 173| constexpr auto _size = _name.size(); \ | | 1183| 173| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 173| return _nameof; }() ------------------ 561| 173| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 173|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 173| ::std::void_t(); \ | | 1180| 173| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 173| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 173| constexpr auto _size = _name.size(); \ | | 1183| 173| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 173| return _nameof; }() ------------------ 562| 173| c.fuzz(chardata); \ 563| 173| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 173| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 173| const FromSpan from{reinterpret_cast(chardata.data()), 178| 173| chardata.size() / sizeof(FromType)}; 179| | 180| 173| static const bool do_print_testcase = 181| 173| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 173| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 173] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 173| do { 189| | // step 0 - is the input valid? 190| 173| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 173| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 173] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 173| From == UtfEncodings::UTF8) { 198| 173| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 173] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 173| } 201| | 202| | // step 2 - what is the required size of the output? 203| 173| const auto [output_length, length_agree] = 204| 173| calculate_length(from, inputisvalid); 205| 173| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 173] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 173| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 92, False: 81] | Branch (208:28): [True: 0, False: 92] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 173| const auto [written, outputs_agree] = 215| 173| do_conversion(from, output_length, inputisvalid); 216| 173| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 173] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 173| return; 221| 173| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 173| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 173| validation_result verify_valid_input(FromSpan src) const { 231| 173| validation_result ret{}; 232| | 233| 173| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 173| const auto implementations = get_supported_implementations(); 235| 173| std::vector results; 236| 173| results.reserve(implementations.size()); 237| | 238| 519| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 519, False: 173] ------------------ 239| 519| results.push_back( 240| 519| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 519| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 519| const bool validation2 = 245| 519| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 519| src.data(), src.size()); 247| 519| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 519] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 519| } 258| | 259| 173| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 173| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 173] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 173| } else { 273| 173| ret.implementations_agree = true; 274| 173| } 275| 173| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 173| return r.error == simdutf::SUCCESS; 277| 173| }); 278| 173| return ret; 279| 173| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 346| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 335| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 335| return r.error == simdutf::SUCCESS; 277| 335| }); _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 173| bool count_the_input(FromSpan src) const { 290| 173| const auto implementations = get_supported_implementations(); 291| 173| std::vector results; 292| 173| results.reserve(implementations.size()); 293| | 294| 519| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 519, False: 173] ------------------ 295| 519| std::size_t ret; 296| 519| if constexpr (From == UtfEncodings::UTF16BE) { 297| 519| ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 519| results.push_back(ret); 304| 519| } 305| 173| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 173| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 173] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 173| return true; 321| 173| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 346| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 173| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 173| length_result ret{}; 344| | 345| 173| const auto implementations = get_supported_implementations(); 346| 173| std::vector results; 347| 173| results.reserve(implementations.size()); 348| | 349| 519| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 519, False: 173] ------------------ 350| 519| const auto len = invoke_lengthcalc(impl, src); 351| 519| results.push_back(len); 352| 519| ret.length.push_back(len); 353| 519| } 354| | 355| 173| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 173| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 173] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 173| } else { 375| 173| ret.implementations_agree = true; 376| 173| } 377| 173| return ret; 378| 173| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 519| FromSpan src) const { 331| 519| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 519| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 346| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 173| const bool inputisvalid) const { 383| 173| conversion_result ret{}; 384| | 385| 173| const auto implementations = get_supported_implementations(); 386| | 387| 173| std::vector> results; 388| 173| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 173| std::vector> outputbuffers; 393| 173| outputbuffers.reserve(implementations.size()); 394| 692| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 519, False: 173] ------------------ 395| 519| auto impl = implementations[i]; 396| 519| const ToType canary1{42}; 397| 519| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 519| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 519| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 519| const auto success = [](const ConversionResult& r) -> bool { 402| 519| if constexpr (std::is_same_v) { 403| 519| return r != 0; 404| 519| } else { 405| 519| return r.error == simdutf::error_code::SUCCESS; 406| 519| } 407| 519| }(implret1); 408| 519| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 519| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 519| const ToType canary2{25}; 414| 519| const auto outputbuffer_first_run = outputbuffer; 415| 519| std::ranges::fill(outputbuffer, canary2); 416| 519| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 519| src.size(), outputbuffer.data()); 418| | 419| 519| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 519] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 519| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 243, False: 276] | Branch (423:29): [True: 243, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 243| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 243| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 243] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 243| } 440| 519| } 441| 519| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 243, False: 276] ------------------ 442| 519| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 173| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 92, False: 81] ------------------ 447| 276| for (auto& e : results) { ------------------ | Branch (447:20): [True: 276, False: 92] ------------------ 448| 276| e.outputhash.clear(); 449| 276| } 450| 92| } 451| | 452| 173| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 173| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 173] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 173| } else { 474| 173| ret.implementations_agree = true; 475| 173| } 476| 173| return ret; 477| 173| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 519| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 519| } else { 405| 519| return r.error == simdutf::error_code::SUCCESS; 406| 519| } 407| 519| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 346| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_26clENSt3__14spanIKcLm18446744073709551615EEE: 554| 327| +[](std::span chardata) { \ 555| 327| const auto c = \ 556| 327| Conversion{ \ 559| 327| &I::lenfunc, &I::conversionfunc, \ 560| 327| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 327|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 327| ::std::void_t(); \ | | 1180| 327| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 327| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 327| constexpr auto _size = _name.size(); \ | | 1183| 327| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 327| return _nameof; }() ------------------ 561| 327| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 327|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 327| ::std::void_t(); \ | | 1180| 327| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 327| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 327| constexpr auto _size = _name.size(); \ | | 1183| 327| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 327| return _nameof; }() ------------------ 562| 327| c.fuzz(chardata); \ 563| 327| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 327| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 327| const FromSpan from{reinterpret_cast(chardata.data()), 178| 327| chardata.size() / sizeof(FromType)}; 179| | 180| 327| static const bool do_print_testcase = 181| 327| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 327| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 327] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 327| do { 189| | // step 0 - is the input valid? 190| 327| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 327| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 327] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 327| From == UtfEncodings::UTF8) { 198| 327| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 327] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 327| } 201| | 202| | // step 2 - what is the required size of the output? 203| 327| const auto [output_length, length_agree] = 204| 327| calculate_length(from, inputisvalid); 205| 327| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 327] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 327| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 136, False: 191] | Branch (208:28): [True: 0, False: 136] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 327| const auto [written, outputs_agree] = 215| 327| do_conversion(from, output_length, inputisvalid); 216| 327| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 327] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 327| return; 221| 327| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 327| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 327| validation_result verify_valid_input(FromSpan src) const { 231| 327| validation_result ret{}; 232| | 233| 327| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 327| const auto implementations = get_supported_implementations(); 235| 327| std::vector results; 236| 327| results.reserve(implementations.size()); 237| | 238| 981| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 981, False: 327] ------------------ 239| 981| results.push_back( 240| 981| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 981| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 981| const bool validation2 = 245| 981| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 981| src.data(), src.size()); 247| 981| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 981] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 981| } 258| | 259| 327| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 327| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 327] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 327| } else { 273| 327| ret.implementations_agree = true; 274| 327| } 275| 327| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 327| return r.error == simdutf::SUCCESS; 277| 327| }); 278| 327| return ret; 279| 327| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 654| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 709| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 709| return r.error == simdutf::SUCCESS; 277| 709| }); _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 327| bool count_the_input(FromSpan src) const { 290| 327| const auto implementations = get_supported_implementations(); 291| 327| std::vector results; 292| 327| results.reserve(implementations.size()); 293| | 294| 981| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 981, False: 327] ------------------ 295| 981| std::size_t ret; 296| 981| if constexpr (From == UtfEncodings::UTF16BE) { 297| 981| ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 981| results.push_back(ret); 304| 981| } 305| 327| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 327| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 327] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 327| return true; 321| 327| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 654| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 327| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 327| length_result ret{}; 344| | 345| 327| const auto implementations = get_supported_implementations(); 346| 327| std::vector results; 347| 327| results.reserve(implementations.size()); 348| | 349| 981| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 981, False: 327] ------------------ 350| 981| const auto len = invoke_lengthcalc(impl, src); 351| 981| results.push_back(len); 352| 981| ret.length.push_back(len); 353| 981| } 354| | 355| 327| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 327| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 327] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 327| } else { 375| 327| ret.implementations_agree = true; 376| 327| } 377| 327| return ret; 378| 327| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 981| FromSpan src) const { 331| 981| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 981| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 654| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 327| const bool inputisvalid) const { 383| 327| conversion_result ret{}; 384| | 385| 327| const auto implementations = get_supported_implementations(); 386| | 387| 327| std::vector> results; 388| 327| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 327| std::vector> outputbuffers; 393| 327| outputbuffers.reserve(implementations.size()); 394| 1.30k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 981, False: 327] ------------------ 395| 981| auto impl = implementations[i]; 396| 981| const ToType canary1{42}; 397| 981| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 981| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 981| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 981| const auto success = [](const ConversionResult& r) -> bool { 402| 981| if constexpr (std::is_same_v) { 403| 981| return r != 0; 404| 981| } else { 405| 981| return r.error == simdutf::error_code::SUCCESS; 406| 981| } 407| 981| }(implret1); 408| 981| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 981| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 981| const ToType canary2{25}; 414| 981| const auto outputbuffer_first_run = outputbuffer; 415| 981| std::ranges::fill(outputbuffer, canary2); 416| 981| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 981| src.size(), outputbuffer.data()); 418| | 419| 981| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 981] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 981| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 573, False: 408] | Branch (423:29): [True: 573, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 573| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 573| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 573] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 573| } 440| 981| } 441| 981| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 573, False: 408] ------------------ 442| 981| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 327| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 136, False: 191] ------------------ 447| 408| for (auto& e : results) { ------------------ | Branch (447:20): [True: 408, False: 136] ------------------ 448| 408| e.outputhash.clear(); 449| 408| } 450| 136| } 451| | 452| 327| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 327| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 327] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 327| } else { 474| 327| ret.implementations_agree = true; 475| 327| } 476| 327| return ret; 477| 327| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 981| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 981| } else { 405| 981| return r.error == simdutf::error_code::SUCCESS; 406| 981| } 407| 981| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 654| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_27clENSt3__14spanIKcLm18446744073709551615EEE: 554| 139| +[](std::span chardata) { \ 555| 139| const auto c = \ 556| 139| Conversion{ \ 559| 139| &I::lenfunc, &I::conversionfunc, \ 560| 139| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 139|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 139| ::std::void_t(); \ | | 1180| 139| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 139| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 139| constexpr auto _size = _name.size(); \ | | 1183| 139| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 139| return _nameof; }() ------------------ 561| 139| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 139|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 139| ::std::void_t(); \ | | 1180| 139| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 139| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 139| constexpr auto _size = _name.size(); \ | | 1183| 139| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 139| return _nameof; }() ------------------ 562| 139| c.fuzz(chardata); \ 563| 139| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 139| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 139| const FromSpan from{reinterpret_cast(chardata.data()), 178| 139| chardata.size() / sizeof(FromType)}; 179| | 180| 139| static const bool do_print_testcase = 181| 139| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 139| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 139] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 139| do { 189| | // step 0 - is the input valid? 190| 139| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 139| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 139] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 139| From == UtfEncodings::UTF8) { 198| 139| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 139] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 139| } 201| | 202| | // step 2 - what is the required size of the output? 203| 139| const auto [output_length, length_agree] = 204| 139| calculate_length(from, inputisvalid); 205| 139| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 139] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 139| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 24, False: 115] | Branch (208:28): [True: 0, False: 24] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 139| const auto [written, outputs_agree] = 215| 139| do_conversion(from, output_length, inputisvalid); 216| 139| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 139] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 139| return; 221| 139| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 139| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEE: 230| 139| validation_result verify_valid_input(FromSpan src) const { 231| 139| validation_result ret{}; 232| | 233| 139| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 139| const auto implementations = get_supported_implementations(); 235| 139| std::vector results; 236| 139| results.reserve(implementations.size()); 237| | 238| 417| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 417, False: 139] ------------------ 239| 417| results.push_back( 240| 417| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 417| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 417| const bool validation2 = 245| 417| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 417| src.data(), src.size()); 247| 417| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 417] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 417| } 258| | 259| 139| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 139| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 139] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 139| } else { 273| 139| ret.implementations_agree = true; 274| 139| } 275| 139| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 139| return r.error == simdutf::SUCCESS; 277| 139| }); 278| 139| return ret; 279| 139| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS5_S5_EEDaSJ_SM_: 259| 278| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKS5_E_clESI_: 275| 369| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 369| return r.error == simdutf::SUCCESS; 277| 369| }); _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEE: 289| 139| bool count_the_input(FromSpan src) const { 290| 139| const auto implementations = get_supported_implementations(); 291| 139| std::vector results; 292| 139| results.reserve(implementations.size()); 293| | 294| 417| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 417, False: 139] ------------------ 295| 417| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 417| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 417| ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 417| results.push_back(ret); 304| 417| } 305| 139| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 139| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 139] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 139| return true; 321| 139| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 278| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEb: 342| 139| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 139| length_result ret{}; 344| | 345| 139| const auto implementations = get_supported_implementations(); 346| 139| std::vector results; 347| 139| results.reserve(implementations.size()); 348| | 349| 417| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 417, False: 139] ------------------ 350| 417| const auto len = invoke_lengthcalc(impl, src); 351| 417| results.push_back(len); 352| 417| ret.length.push_back(len); 353| 417| } 354| | 355| 139| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 139| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 139] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 139| } else { 375| 139| ret.implementations_agree = true; 376| 139| } 377| 139| return ret; 378| 139| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSF_NSt3__14spanIS6_Lm18446744073709551615EEE: 338| 417| FromSpan src) const { 339| 417| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 417| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 278| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 139| const bool inputisvalid) const { 383| 139| conversion_result ret{}; 384| | 385| 139| const auto implementations = get_supported_implementations(); 386| | 387| 139| std::vector> results; 388| 139| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 139| std::vector> outputbuffers; 393| 139| outputbuffers.reserve(implementations.size()); 394| 556| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 417, False: 139] ------------------ 395| 417| auto impl = implementations[i]; 396| 417| const ToType canary1{42}; 397| 417| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 417| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 417| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 417| const auto success = [](const ConversionResult& r) -> bool { 402| 417| if constexpr (std::is_same_v) { 403| 417| return r != 0; 404| 417| } else { 405| 417| return r.error == simdutf::error_code::SUCCESS; 406| 417| } 407| 417| }(implret1); 408| 417| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 417| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 417| const ToType canary2{25}; 414| 417| const auto outputbuffer_first_run = outputbuffer; 415| 417| std::ranges::fill(outputbuffer, canary2); 416| 417| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 417| src.size(), outputbuffer.data()); 418| | 419| 417| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 417] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 417| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 345, False: 72] | Branch (423:29): [True: 141, False: 204] ------------------ 424| | // only care about the output if the input is valid 425| 141| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 141| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 141] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 141| } 440| 417| } 441| 417| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 141, False: 276] ------------------ 442| 417| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 139| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 24, False: 115] ------------------ 447| 72| for (auto& e : results) { ------------------ | Branch (447:20): [True: 72, False: 24] ------------------ 448| 72| e.outputhash.clear(); 449| 72| } 450| 24| } 451| | 452| 139| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 139| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 139] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 139| } else { 474| 139| ret.implementations_agree = true; 475| 139| } 476| 139| return ret; 477| 139| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS5_E_clESM_: 401| 417| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 417| } else { 405| 417| return r.error == simdutf::error_code::SUCCESS; 406| 417| } 407| 417| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS5_ESU_EEDaSN_SQ_: 452| 278| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_28clENSt3__14spanIKcLm18446744073709551615EEE: 554| 167| +[](std::span chardata) { \ 555| 167| const auto c = \ 556| 167| Conversion{ \ 559| 167| &I::lenfunc, &I::conversionfunc, \ 560| 167| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 167|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 167| ::std::void_t(); \ | | 1180| 167| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 167| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 167| constexpr auto _size = _name.size(); \ | | 1183| 167| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 167| return _nameof; }() ------------------ 561| 167| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 167|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 167| ::std::void_t(); \ | | 1180| 167| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 167| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 167| constexpr auto _size = _name.size(); \ | | 1183| 167| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 167| return _nameof; }() ------------------ 562| 167| c.fuzz(chardata); \ 563| 167| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 167| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 167| const FromSpan from{reinterpret_cast(chardata.data()), 178| 167| chardata.size() / sizeof(FromType)}; 179| | 180| 167| static const bool do_print_testcase = 181| 167| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 167| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 167] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 167| do { 189| | // step 0 - is the input valid? 190| 167| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 167| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 167] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 167| From == UtfEncodings::UTF8) { 198| 167| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 167] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 167| } 201| | 202| | // step 2 - what is the required size of the output? 203| 167| const auto [output_length, length_agree] = 204| 167| calculate_length(from, inputisvalid); 205| 167| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 167] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 167| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 90, False: 77] | Branch (208:28): [True: 0, False: 90] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 167| const auto [written, outputs_agree] = 215| 167| do_conversion(from, output_length, inputisvalid); 216| 167| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 167] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 167| return; 221| 167| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 167| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 167| validation_result verify_valid_input(FromSpan src) const { 231| 167| validation_result ret{}; 232| | 233| 167| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 167| const auto implementations = get_supported_implementations(); 235| 167| std::vector results; 236| 167| results.reserve(implementations.size()); 237| | 238| 501| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 501, False: 167] ------------------ 239| 501| results.push_back( 240| 501| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 501| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 501| const bool validation2 = 245| 501| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 501| src.data(), src.size()); 247| 501| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 501] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 501| } 258| | 259| 167| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 167| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 167] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 167| } else { 273| 167| ret.implementations_agree = true; 274| 167| } 275| 167| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 167| return r.error == simdutf::SUCCESS; 277| 167| }); 278| 167| return ret; 279| 167| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 334| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 321| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 321| return r.error == simdutf::SUCCESS; 277| 321| }); _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 167| bool count_the_input(FromSpan src) const { 290| 167| const auto implementations = get_supported_implementations(); 291| 167| std::vector results; 292| 167| results.reserve(implementations.size()); 293| | 294| 501| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 501, False: 167] ------------------ 295| 501| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 501| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 501| ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 501| results.push_back(ret); 304| 501| } 305| 167| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 167| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 167] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 167| return true; 321| 167| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 334| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 167| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 167| length_result ret{}; 344| | 345| 167| const auto implementations = get_supported_implementations(); 346| 167| std::vector results; 347| 167| results.reserve(implementations.size()); 348| | 349| 501| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 501, False: 167] ------------------ 350| 501| const auto len = invoke_lengthcalc(impl, src); 351| 501| results.push_back(len); 352| 501| ret.length.push_back(len); 353| 501| } 354| | 355| 167| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 167| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 167] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 167| } else { 375| 167| ret.implementations_agree = true; 376| 167| } 377| 167| return ret; 378| 167| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 501| FromSpan src) const { 331| 501| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 501| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 334| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 167| const bool inputisvalid) const { 383| 167| conversion_result ret{}; 384| | 385| 167| const auto implementations = get_supported_implementations(); 386| | 387| 167| std::vector> results; 388| 167| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 167| std::vector> outputbuffers; 393| 167| outputbuffers.reserve(implementations.size()); 394| 668| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 501, False: 167] ------------------ 395| 501| auto impl = implementations[i]; 396| 501| const ToType canary1{42}; 397| 501| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 501| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 501| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 501| const auto success = [](const ConversionResult& r) -> bool { 402| 501| if constexpr (std::is_same_v) { 403| 501| return r != 0; 404| 501| } else { 405| 501| return r.error == simdutf::error_code::SUCCESS; 406| 501| } 407| 501| }(implret1); 408| 501| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 501| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 501| const ToType canary2{25}; 414| 501| const auto outputbuffer_first_run = outputbuffer; 415| 501| std::ranges::fill(outputbuffer, canary2); 416| 501| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 501| src.size(), outputbuffer.data()); 418| | 419| 501| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 501] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 501| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 231, False: 270] | Branch (423:29): [True: 231, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 231| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 231| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 231] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 231| } 440| 501| } 441| 501| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 231, False: 270] ------------------ 442| 501| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 167| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 90, False: 77] ------------------ 447| 270| for (auto& e : results) { ------------------ | Branch (447:20): [True: 270, False: 90] ------------------ 448| 270| e.outputhash.clear(); 449| 270| } 450| 90| } 451| | 452| 167| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 167| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 167] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 167| } else { 474| 167| ret.implementations_agree = true; 475| 167| } 476| 167| return ret; 477| 167| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 501| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 501| } else { 405| 501| return r.error == simdutf::error_code::SUCCESS; 406| 501| } 407| 501| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 334| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_29clENSt3__14spanIKcLm18446744073709551615EEE: 554| 321| +[](std::span chardata) { \ 555| 321| const auto c = \ 556| 321| Conversion{ \ 559| 321| &I::lenfunc, &I::conversionfunc, \ 560| 321| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 321|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 321| ::std::void_t(); \ | | 1180| 321| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 321| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 321| constexpr auto _size = _name.size(); \ | | 1183| 321| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 321| return _nameof; }() ------------------ 561| 321| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 321|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 321| ::std::void_t(); \ | | 1180| 321| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 321| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 321| constexpr auto _size = _name.size(); \ | | 1183| 321| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 321| return _nameof; }() ------------------ 562| 321| c.fuzz(chardata); \ 563| 321| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 321| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 321| const FromSpan from{reinterpret_cast(chardata.data()), 178| 321| chardata.size() / sizeof(FromType)}; 179| | 180| 321| static const bool do_print_testcase = 181| 321| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 321| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 321] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 321| do { 189| | // step 0 - is the input valid? 190| 321| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 321| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 321] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 321| From == UtfEncodings::UTF8) { 198| 321| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 321] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 321| } 201| | 202| | // step 2 - what is the required size of the output? 203| 321| const auto [output_length, length_agree] = 204| 321| calculate_length(from, inputisvalid); 205| 321| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 321] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 321| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 120, False: 201] | Branch (208:28): [True: 0, False: 120] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 321| const auto [written, outputs_agree] = 215| 321| do_conversion(from, output_length, inputisvalid); 216| 321| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 321] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 321| return; 221| 321| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 321| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 321| validation_result verify_valid_input(FromSpan src) const { 231| 321| validation_result ret{}; 232| | 233| 321| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 321| const auto implementations = get_supported_implementations(); 235| 321| std::vector results; 236| 321| results.reserve(implementations.size()); 237| | 238| 963| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 963, False: 321] ------------------ 239| 963| results.push_back( 240| 963| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 963| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 963| const bool validation2 = 245| 963| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 963| src.data(), src.size()); 247| 963| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 963] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 963| } 258| | 259| 321| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 321| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 321] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 321| } else { 273| 321| ret.implementations_agree = true; 274| 321| } 275| 321| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 321| return r.error == simdutf::SUCCESS; 277| 321| }); 278| 321| return ret; 279| 321| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 642| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 723| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 723| return r.error == simdutf::SUCCESS; 277| 723| }); _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 321| bool count_the_input(FromSpan src) const { 290| 321| const auto implementations = get_supported_implementations(); 291| 321| std::vector results; 292| 321| results.reserve(implementations.size()); 293| | 294| 963| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 963, False: 321] ------------------ 295| 963| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 963| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 963| ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 963| results.push_back(ret); 304| 963| } 305| 321| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 321| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 321] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 321| return true; 321| 321| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 642| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 321| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 321| length_result ret{}; 344| | 345| 321| const auto implementations = get_supported_implementations(); 346| 321| std::vector results; 347| 321| results.reserve(implementations.size()); 348| | 349| 963| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 963, False: 321] ------------------ 350| 963| const auto len = invoke_lengthcalc(impl, src); 351| 963| results.push_back(len); 352| 963| ret.length.push_back(len); 353| 963| } 354| | 355| 321| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 321| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 321] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 321| } else { 375| 321| ret.implementations_agree = true; 376| 321| } 377| 321| return ret; 378| 321| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 963| FromSpan src) const { 331| 963| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 963| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 642| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 321| const bool inputisvalid) const { 383| 321| conversion_result ret{}; 384| | 385| 321| const auto implementations = get_supported_implementations(); 386| | 387| 321| std::vector> results; 388| 321| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 321| std::vector> outputbuffers; 393| 321| outputbuffers.reserve(implementations.size()); 394| 1.28k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 963, False: 321] ------------------ 395| 963| auto impl = implementations[i]; 396| 963| const ToType canary1{42}; 397| 963| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 963| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 963| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 963| const auto success = [](const ConversionResult& r) -> bool { 402| 963| if constexpr (std::is_same_v) { 403| 963| return r != 0; 404| 963| } else { 405| 963| return r.error == simdutf::error_code::SUCCESS; 406| 963| } 407| 963| }(implret1); 408| 963| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 963| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 963| const ToType canary2{25}; 414| 963| const auto outputbuffer_first_run = outputbuffer; 415| 963| std::ranges::fill(outputbuffer, canary2); 416| 963| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 963| src.size(), outputbuffer.data()); 418| | 419| 963| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 963] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 963| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 603, False: 360] | Branch (423:29): [True: 603, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 603| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 603| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 603] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 603| } 440| 963| } 441| 963| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 603, False: 360] ------------------ 442| 963| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 321| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 120, False: 201] ------------------ 447| 360| for (auto& e : results) { ------------------ | Branch (447:20): [True: 360, False: 120] ------------------ 448| 360| e.outputhash.clear(); 449| 360| } 450| 120| } 451| | 452| 321| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 321| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 321] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 321| } else { 474| 321| ret.implementations_agree = true; 475| 321| } 476| 321| return ret; 477| 321| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 963| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 963| } else { 405| 963| return r.error == simdutf::error_code::SUCCESS; 406| 963| } 407| 963| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 642| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_30clENSt3__14spanIKcLm18446744073709551615EEE: 554| 218| +[](std::span chardata) { \ 555| 218| const auto c = \ 556| 218| Conversion{ \ 559| 218| &I::lenfunc, &I::conversionfunc, \ 560| 218| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 218|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 218| ::std::void_t(); \ | | 1180| 218| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 218| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 218| constexpr auto _size = _name.size(); \ | | 1183| 218| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 218| return _nameof; }() ------------------ 561| 218| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 218|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 218| ::std::void_t(); \ | | 1180| 218| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 218| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 218| constexpr auto _size = _name.size(); \ | | 1183| 218| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 218| return _nameof; }() ------------------ 562| 218| c.fuzz(chardata); \ 563| 218| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 218| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 218| const FromSpan from{reinterpret_cast(chardata.data()), 178| 218| chardata.size() / sizeof(FromType)}; 179| | 180| 218| static const bool do_print_testcase = 181| 218| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 218| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 218] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 218| do { 189| | // step 0 - is the input valid? 190| 218| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 218| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 218] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 218| const auto [output_length, length_agree] = 204| 218| calculate_length(from, inputisvalid); 205| 218| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 218] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 218| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 189, False: 29] | Branch (208:28): [True: 0, False: 189] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 218| const auto [written, outputs_agree] = 215| 218| do_conversion(from, output_length, inputisvalid); 216| 218| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 218] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 218| return; 221| 218| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 218| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEE: 230| 218| validation_result verify_valid_input(FromSpan src) const { 231| 218| validation_result ret{}; 232| | 233| 218| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 218| const auto implementations = get_supported_implementations(); 235| 218| std::vector results; 236| 218| results.reserve(implementations.size()); 237| | 238| 654| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 654, False: 218] ------------------ 239| 654| results.push_back( 240| 654| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 654| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 654| const bool validation2 = 245| 654| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 654| src.data(), src.size()); 247| 654| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 654] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 654| } 258| | 259| 218| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 218| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 218] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 218| } else { 273| 218| ret.implementations_agree = true; 274| 218| } 275| 218| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 218| return r.error == simdutf::SUCCESS; 277| 218| }); 278| 218| return ret; 279| 218| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS5_S5_EEDaSJ_SM_: 259| 436| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKS5_E_clESI_: 275| 276| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 276| return r.error == simdutf::SUCCESS; 277| 276| }); _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEb: 342| 218| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 218| length_result ret{}; 344| | 345| 218| const auto implementations = get_supported_implementations(); 346| 218| std::vector results; 347| 218| results.reserve(implementations.size()); 348| | 349| 654| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 654, False: 218] ------------------ 350| 654| const auto len = invoke_lengthcalc(impl, src); 351| 654| results.push_back(len); 352| 654| ret.length.push_back(len); 353| 654| } 354| | 355| 218| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 218| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 218] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 218| } else { 375| 218| ret.implementations_agree = true; 376| 218| } 377| 218| return ret; 378| 218| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSF_NSt3__14spanIS6_Lm18446744073709551615EEE: 338| 654| FromSpan src) const { 339| 654| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 654| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 436| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 218| const bool inputisvalid) const { 383| 218| conversion_result ret{}; 384| | 385| 218| const auto implementations = get_supported_implementations(); 386| | 387| 218| std::vector> results; 388| 218| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 218| std::vector> outputbuffers; 393| 218| outputbuffers.reserve(implementations.size()); 394| 872| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 654, False: 218] ------------------ 395| 654| auto impl = implementations[i]; 396| 654| const ToType canary1{42}; 397| 654| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 654| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 654| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 654| const auto success = [](const ConversionResult& r) -> bool { 402| 654| if constexpr (std::is_same_v) { 403| 654| return r != 0; 404| 654| } else { 405| 654| return r.error == simdutf::error_code::SUCCESS; 406| 654| } 407| 654| }(implret1); 408| 654| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 654| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 654| const ToType canary2{25}; 414| 654| const auto outputbuffer_first_run = outputbuffer; 415| 654| std::ranges::fill(outputbuffer, canary2); 416| 654| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 654| src.size(), outputbuffer.data()); 418| | 419| 654| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 654] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 654| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 87, False: 567] | Branch (423:29): [True: 63, False: 24] ------------------ 424| | // only care about the output if the input is valid 425| 63| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 63| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 63] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 63| } 440| 654| } 441| 654| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 63, False: 591] ------------------ 442| 654| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 218| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 189, False: 29] ------------------ 447| 567| for (auto& e : results) { ------------------ | Branch (447:20): [True: 567, False: 189] ------------------ 448| 567| e.outputhash.clear(); 449| 567| } 450| 189| } 451| | 452| 218| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 218| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 218] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 218| } else { 474| 218| ret.implementations_agree = true; 475| 218| } 476| 218| return ret; 477| 218| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS5_E_clESM_: 401| 654| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 654| } else { 405| 654| return r.error == simdutf::error_code::SUCCESS; 406| 654| } 407| 654| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS5_ESU_EEDaSN_SQ_: 452| 436| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_31clENSt3__14spanIKcLm18446744073709551615EEE: 554| 347| +[](std::span chardata) { \ 555| 347| const auto c = \ 556| 347| Conversion{ \ 559| 347| &I::lenfunc, &I::conversionfunc, \ 560| 347| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 347|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 347| ::std::void_t(); \ | | 1180| 347| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 347| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 347| constexpr auto _size = _name.size(); \ | | 1183| 347| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 347| return _nameof; }() ------------------ 561| 347| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 347|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 347| ::std::void_t(); \ | | 1180| 347| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 347| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 347| constexpr auto _size = _name.size(); \ | | 1183| 347| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 347| return _nameof; }() ------------------ 562| 347| c.fuzz(chardata); \ 563| 347| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 347| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 347| const FromSpan from{reinterpret_cast(chardata.data()), 178| 347| chardata.size() / sizeof(FromType)}; 179| | 180| 347| static const bool do_print_testcase = 181| 347| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 347| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 347] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 347| do { 189| | // step 0 - is the input valid? 190| 347| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 347| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 347] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 347| const auto [output_length, length_agree] = 204| 347| calculate_length(from, inputisvalid); 205| 347| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 347] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 347| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 256, False: 91] | Branch (208:28): [True: 0, False: 256] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 347| const auto [written, outputs_agree] = 215| 347| do_conversion(from, output_length, inputisvalid); 216| 347| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 347] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 347| return; 221| 347| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 347| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 347| validation_result verify_valid_input(FromSpan src) const { 231| 347| validation_result ret{}; 232| | 233| 347| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 347| const auto implementations = get_supported_implementations(); 235| 347| std::vector results; 236| 347| results.reserve(implementations.size()); 237| | 238| 1.04k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.04k, False: 347] ------------------ 239| 1.04k| results.push_back( 240| 1.04k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.04k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.04k| const bool validation2 = 245| 1.04k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.04k| src.data(), src.size()); 247| 1.04k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.04k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.04k| } 258| | 259| 347| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 347| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 347] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 347| } else { 273| 347| ret.implementations_agree = true; 274| 347| } 275| 347| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 347| return r.error == simdutf::SUCCESS; 277| 347| }); 278| 347| return ret; 279| 347| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 694| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 529| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 529| return r.error == simdutf::SUCCESS; 277| 529| }); _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 347| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 347| length_result ret{}; 344| | 345| 347| const auto implementations = get_supported_implementations(); 346| 347| std::vector results; 347| 347| results.reserve(implementations.size()); 348| | 349| 1.04k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.04k, False: 347] ------------------ 350| 1.04k| const auto len = invoke_lengthcalc(impl, src); 351| 1.04k| results.push_back(len); 352| 1.04k| ret.length.push_back(len); 353| 1.04k| } 354| | 355| 347| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 347| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 347] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 347| } else { 375| 347| ret.implementations_agree = true; 376| 347| } 377| 347| return ret; 378| 347| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.04k| FromSpan src) const { 331| 1.04k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.04k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 694| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 347| const bool inputisvalid) const { 383| 347| conversion_result ret{}; 384| | 385| 347| const auto implementations = get_supported_implementations(); 386| | 387| 347| std::vector> results; 388| 347| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 347| std::vector> outputbuffers; 393| 347| outputbuffers.reserve(implementations.size()); 394| 1.38k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.04k, False: 347] ------------------ 395| 1.04k| auto impl = implementations[i]; 396| 1.04k| const ToType canary1{42}; 397| 1.04k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.04k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.04k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.04k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.04k| if constexpr (std::is_same_v) { 403| 1.04k| return r != 0; 404| 1.04k| } else { 405| 1.04k| return r.error == simdutf::error_code::SUCCESS; 406| 1.04k| } 407| 1.04k| }(implret1); 408| 1.04k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.04k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.04k| const ToType canary2{25}; 414| 1.04k| const auto outputbuffer_first_run = outputbuffer; 415| 1.04k| std::ranges::fill(outputbuffer, canary2); 416| 1.04k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.04k| src.size(), outputbuffer.data()); 418| | 419| 1.04k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.04k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.04k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 273, False: 768] | Branch (423:29): [True: 273, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 273| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 273| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 273] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 273| } 440| 1.04k| } 441| 1.04k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 273, False: 768] ------------------ 442| 1.04k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 347| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 256, False: 91] ------------------ 447| 768| for (auto& e : results) { ------------------ | Branch (447:20): [True: 768, False: 256] ------------------ 448| 768| e.outputhash.clear(); 449| 768| } 450| 256| } 451| | 452| 347| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 347| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 347] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 347| } else { 474| 347| ret.implementations_agree = true; 475| 347| } 476| 347| return ret; 477| 347| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.04k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.04k| } else { 405| 1.04k| return r.error == simdutf::error_code::SUCCESS; 406| 1.04k| } 407| 1.04k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 694| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_32clENSt3__14spanIKcLm18446744073709551615EEE: 554| 337| +[](std::span chardata) { \ 555| 337| const auto c = \ 556| 337| Conversion{ \ 559| 337| &I::lenfunc, &I::conversionfunc, \ 560| 337| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 337|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 337| ::std::void_t(); \ | | 1180| 337| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 337| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 337| constexpr auto _size = _name.size(); \ | | 1183| 337| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 337| return _nameof; }() ------------------ 561| 337| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 337|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 337| ::std::void_t(); \ | | 1180| 337| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 337| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 337| constexpr auto _size = _name.size(); \ | | 1183| 337| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 337| return _nameof; }() ------------------ 562| 337| c.fuzz(chardata); \ 563| 337| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 337| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 337| const FromSpan from{reinterpret_cast(chardata.data()), 178| 337| chardata.size() / sizeof(FromType)}; 179| | 180| 337| static const bool do_print_testcase = 181| 337| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 337| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 337] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 337| do { 189| | // step 0 - is the input valid? 190| 337| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 337| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 337] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 337| const auto [output_length, length_agree] = 204| 337| calculate_length(from, inputisvalid); 205| 337| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 337] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 337| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 252, False: 85] | Branch (208:28): [True: 0, False: 252] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 337| const auto [written, outputs_agree] = 215| 337| do_conversion(from, output_length, inputisvalid); 216| 337| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 337] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 337| return; 221| 337| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 337| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 337| validation_result verify_valid_input(FromSpan src) const { 231| 337| validation_result ret{}; 232| | 233| 337| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 337| const auto implementations = get_supported_implementations(); 235| 337| std::vector results; 236| 337| results.reserve(implementations.size()); 237| | 238| 1.01k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.01k, False: 337] ------------------ 239| 1.01k| results.push_back( 240| 1.01k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.01k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.01k| const bool validation2 = 245| 1.01k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.01k| src.data(), src.size()); 247| 1.01k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.01k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.01k| } 258| | 259| 337| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 337| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 337] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 337| } else { 273| 337| ret.implementations_agree = true; 274| 337| } 275| 337| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 337| return r.error == simdutf::SUCCESS; 277| 337| }); 278| 337| return ret; 279| 337| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 674| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 507| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 507| return r.error == simdutf::SUCCESS; 277| 507| }); _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 337| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 337| length_result ret{}; 344| | 345| 337| const auto implementations = get_supported_implementations(); 346| 337| std::vector results; 347| 337| results.reserve(implementations.size()); 348| | 349| 1.01k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.01k, False: 337] ------------------ 350| 1.01k| const auto len = invoke_lengthcalc(impl, src); 351| 1.01k| results.push_back(len); 352| 1.01k| ret.length.push_back(len); 353| 1.01k| } 354| | 355| 337| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 337| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 337] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 337| } else { 375| 337| ret.implementations_agree = true; 376| 337| } 377| 337| return ret; 378| 337| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.01k| FromSpan src) const { 331| 1.01k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.01k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 674| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 337| const bool inputisvalid) const { 383| 337| conversion_result ret{}; 384| | 385| 337| const auto implementations = get_supported_implementations(); 386| | 387| 337| std::vector> results; 388| 337| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 337| std::vector> outputbuffers; 393| 337| outputbuffers.reserve(implementations.size()); 394| 1.34k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.01k, False: 337] ------------------ 395| 1.01k| auto impl = implementations[i]; 396| 1.01k| const ToType canary1{42}; 397| 1.01k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.01k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.01k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.01k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.01k| if constexpr (std::is_same_v) { 403| 1.01k| return r != 0; 404| 1.01k| } else { 405| 1.01k| return r.error == simdutf::error_code::SUCCESS; 406| 1.01k| } 407| 1.01k| }(implret1); 408| 1.01k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.01k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.01k| const ToType canary2{25}; 414| 1.01k| const auto outputbuffer_first_run = outputbuffer; 415| 1.01k| std::ranges::fill(outputbuffer, canary2); 416| 1.01k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.01k| src.size(), outputbuffer.data()); 418| | 419| 1.01k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.01k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.01k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 255, False: 756] | Branch (423:29): [True: 255, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 255| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 255| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 255] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 255| } 440| 1.01k| } 441| 1.01k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 255, False: 756] ------------------ 442| 1.01k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 337| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 252, False: 85] ------------------ 447| 756| for (auto& e : results) { ------------------ | Branch (447:20): [True: 756, False: 252] ------------------ 448| 756| e.outputhash.clear(); 449| 756| } 450| 252| } 451| | 452| 337| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 337| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 337] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 337| } else { 474| 337| ret.implementations_agree = true; 475| 337| } 476| 337| return ret; 477| 337| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.01k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.01k| } else { 405| 1.01k| return r.error == simdutf::error_code::SUCCESS; 406| 1.01k| } 407| 1.01k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 674| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_33clENSt3__14spanIKcLm18446744073709551615EEE: 554| 416| +[](std::span chardata) { \ 555| 416| const auto c = \ 556| 416| Conversion{ \ 559| 416| &I::lenfunc, &I::conversionfunc, \ 560| 416| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 416|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 416| ::std::void_t(); \ | | 1180| 416| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 416| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 416| constexpr auto _size = _name.size(); \ | | 1183| 416| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 416| return _nameof; }() ------------------ 561| 416| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 416|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 416| ::std::void_t(); \ | | 1180| 416| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 416| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 416| constexpr auto _size = _name.size(); \ | | 1183| 416| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 416| return _nameof; }() ------------------ 562| 416| c.fuzz(chardata); \ 563| 416| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 416| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 416| const FromSpan from{reinterpret_cast(chardata.data()), 178| 416| chardata.size() / sizeof(FromType)}; 179| | 180| 416| static const bool do_print_testcase = 181| 416| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 416| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 416] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 416| do { 189| | // step 0 - is the input valid? 190| 416| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 416| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 416] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 416| const auto [output_length, length_agree] = 204| 416| calculate_length(from, inputisvalid); 205| 416| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 416] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 416| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 309, False: 107] | Branch (208:28): [True: 0, False: 309] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 416| const auto [written, outputs_agree] = 215| 416| do_conversion(from, output_length, inputisvalid); 216| 416| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 416] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 416| return; 221| 416| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 416| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 416| validation_result verify_valid_input(FromSpan src) const { 231| 416| validation_result ret{}; 232| | 233| 416| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 416| const auto implementations = get_supported_implementations(); 235| 416| std::vector results; 236| 416| results.reserve(implementations.size()); 237| | 238| 1.24k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.24k, False: 416] ------------------ 239| 1.24k| results.push_back( 240| 1.24k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.24k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.24k| const bool validation2 = 245| 1.24k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.24k| src.data(), src.size()); 247| 1.24k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.24k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.24k| } 258| | 259| 416| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 416| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 416] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 416| } else { 273| 416| ret.implementations_agree = true; 274| 416| } 275| 416| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 416| return r.error == simdutf::SUCCESS; 277| 416| }); 278| 416| return ret; 279| 416| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 832| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 630| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 630| return r.error == simdutf::SUCCESS; 277| 630| }); _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 416| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 416| length_result ret{}; 344| | 345| 416| const auto implementations = get_supported_implementations(); 346| 416| std::vector results; 347| 416| results.reserve(implementations.size()); 348| | 349| 1.24k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.24k, False: 416] ------------------ 350| 1.24k| const auto len = invoke_lengthcalc(impl, src); 351| 1.24k| results.push_back(len); 352| 1.24k| ret.length.push_back(len); 353| 1.24k| } 354| | 355| 416| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 416| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 416] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 416| } else { 375| 416| ret.implementations_agree = true; 376| 416| } 377| 416| return ret; 378| 416| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.24k| FromSpan src) const { 331| 1.24k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.24k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 832| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 416| const bool inputisvalid) const { 383| 416| conversion_result ret{}; 384| | 385| 416| const auto implementations = get_supported_implementations(); 386| | 387| 416| std::vector> results; 388| 416| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 416| std::vector> outputbuffers; 393| 416| outputbuffers.reserve(implementations.size()); 394| 1.66k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.24k, False: 416] ------------------ 395| 1.24k| auto impl = implementations[i]; 396| 1.24k| const ToType canary1{42}; 397| 1.24k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.24k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.24k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.24k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.24k| if constexpr (std::is_same_v) { 403| 1.24k| return r != 0; 404| 1.24k| } else { 405| 1.24k| return r.error == simdutf::error_code::SUCCESS; 406| 1.24k| } 407| 1.24k| }(implret1); 408| 1.24k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.24k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.24k| const ToType canary2{25}; 414| 1.24k| const auto outputbuffer_first_run = outputbuffer; 415| 1.24k| std::ranges::fill(outputbuffer, canary2); 416| 1.24k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.24k| src.size(), outputbuffer.data()); 418| | 419| 1.24k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.24k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.24k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 321, False: 927] | Branch (423:29): [True: 321, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 321| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 321| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 321] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 321| } 440| 1.24k| } 441| 1.24k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 321, False: 927] ------------------ 442| 1.24k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 416| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 309, False: 107] ------------------ 447| 927| for (auto& e : results) { ------------------ | Branch (447:20): [True: 927, False: 309] ------------------ 448| 927| e.outputhash.clear(); 449| 927| } 450| 309| } 451| | 452| 416| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 416| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 416] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 416| } else { 474| 416| ret.implementations_agree = true; 475| 416| } 476| 416| return ret; 477| 416| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.24k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.24k| } else { 405| 1.24k| return r.error == simdutf::error_code::SUCCESS; 406| 1.24k| } 407| 1.24k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 832| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_34clENSt3__14spanIKcLm18446744073709551615EEE: 554| 256| +[](std::span chardata) { \ 555| 256| const auto c = \ 556| 256| Conversion{ \ 559| 256| &I::lenfunc, &I::conversionfunc, \ 560| 256| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 256|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 256| ::std::void_t(); \ | | 1180| 256| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 256| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 256| constexpr auto _size = _name.size(); \ | | 1183| 256| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 256| return _nameof; }() ------------------ 561| 256| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 256|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 256| ::std::void_t(); \ | | 1180| 256| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 256| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 256| constexpr auto _size = _name.size(); \ | | 1183| 256| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 256| return _nameof; }() ------------------ 562| 256| c.fuzz(chardata); \ 563| 256| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 256| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 256| const FromSpan from{reinterpret_cast(chardata.data()), 178| 256| chardata.size() / sizeof(FromType)}; 179| | 180| 256| static const bool do_print_testcase = 181| 256| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 256| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 256] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 256| do { 189| | // step 0 - is the input valid? 190| 256| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 256| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 256] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 256| From == UtfEncodings::UTF8) { 198| 256| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 256] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 256| } 201| | 202| | // step 2 - what is the required size of the output? 203| 256| const auto [output_length, length_agree] = 204| 256| calculate_length(from, inputisvalid); 205| 256| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 256] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 256| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 139, False: 117] | Branch (208:28): [True: 0, False: 139] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 256| const auto [written, outputs_agree] = 215| 256| do_conversion(from, output_length, inputisvalid); 216| 256| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 256] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 256| return; 221| 256| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 256| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 256| validation_result verify_valid_input(FromSpan src) const { 231| 256| validation_result ret{}; 232| | 233| 256| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 256| const auto implementations = get_supported_implementations(); 235| 256| std::vector results; 236| 256| results.reserve(implementations.size()); 237| | 238| 768| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 768, False: 256] ------------------ 239| 768| results.push_back( 240| 768| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 768| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 768| const bool validation2 = 245| 768| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 768| src.data(), src.size()); 247| 768| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 768] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 768| } 258| | 259| 256| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 256| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 256] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 256| } else { 273| 256| ret.implementations_agree = true; 274| 256| } 275| 256| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 256| return r.error == simdutf::SUCCESS; 277| 256| }); 278| 256| return ret; 279| 256| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 512| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 490| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 490| return r.error == simdutf::SUCCESS; 277| 490| }); _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 256| bool count_the_input(FromSpan src) const { 290| 256| const auto implementations = get_supported_implementations(); 291| 256| std::vector results; 292| 256| results.reserve(implementations.size()); 293| | 294| 768| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 768, False: 256] ------------------ 295| 768| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| 768| } else if constexpr (From == UtfEncodings::UTF8) { 301| 768| ret = impl->count_utf8(src.data(), src.size()); 302| 768| } 303| 768| results.push_back(ret); 304| 768| } 305| 256| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 256| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 256] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 256| return true; 321| 256| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 512| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 256| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 256| length_result ret{}; 344| | 345| 256| const auto implementations = get_supported_implementations(); 346| 256| std::vector results; 347| 256| results.reserve(implementations.size()); 348| | 349| 768| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 768, False: 256] ------------------ 350| 768| const auto len = invoke_lengthcalc(impl, src); 351| 768| results.push_back(len); 352| 768| ret.length.push_back(len); 353| 768| } 354| | 355| 256| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 256| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 256] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 256| } else { 375| 256| ret.implementations_agree = true; 376| 256| } 377| 256| return ret; 378| 256| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 768| FromSpan src) const { 331| 768| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 768| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 512| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 256| const bool inputisvalid) const { 383| 256| conversion_result ret{}; 384| | 385| 256| const auto implementations = get_supported_implementations(); 386| | 387| 256| std::vector> results; 388| 256| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 256| std::vector> outputbuffers; 393| 256| outputbuffers.reserve(implementations.size()); 394| 1.02k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 768, False: 256] ------------------ 395| 768| auto impl = implementations[i]; 396| 768| const ToType canary1{42}; 397| 768| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 768| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 768| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 768| const auto success = [](const ConversionResult& r) -> bool { 402| 768| if constexpr (std::is_same_v) { 403| 768| return r != 0; 404| 768| } else { 405| 768| return r.error == simdutf::error_code::SUCCESS; 406| 768| } 407| 768| }(implret1); 408| 768| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 768| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 768| const ToType canary2{25}; 414| 768| const auto outputbuffer_first_run = outputbuffer; 415| 768| std::ranges::fill(outputbuffer, canary2); 416| 768| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 768| src.size(), outputbuffer.data()); 418| | 419| 768| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 768] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 768| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 351, False: 417] | Branch (423:29): [True: 339, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 339| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 339| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 339] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 339| } 440| 768| } 441| 768| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 339, False: 429] ------------------ 442| 768| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 256| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 139, False: 117] ------------------ 447| 417| for (auto& e : results) { ------------------ | Branch (447:20): [True: 417, False: 139] ------------------ 448| 417| e.outputhash.clear(); 449| 417| } 450| 139| } 451| | 452| 256| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 256| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 256] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 256| } else { 474| 256| ret.implementations_agree = true; 475| 256| } 476| 256| return ret; 477| 256| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 768| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 768| } else { 405| 768| return r.error == simdutf::error_code::SUCCESS; 406| 768| } 407| 768| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 512| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_35clENSt3__14spanIKcLm18446744073709551615EEE: 554| 353| +[](std::span chardata) { \ 555| 353| const auto c = \ 556| 353| Conversion{ \ 559| 353| &I::lenfunc, &I::conversionfunc, \ 560| 353| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 353|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 353| ::std::void_t(); \ | | 1180| 353| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 353| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 353| constexpr auto _size = _name.size(); \ | | 1183| 353| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 353| return _nameof; }() ------------------ 561| 353| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 353|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 353| ::std::void_t(); \ | | 1180| 353| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 353| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 353| constexpr auto _size = _name.size(); \ | | 1183| 353| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 353| return _nameof; }() ------------------ 562| 353| c.fuzz(chardata); \ 563| 353| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 353| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 353| const FromSpan from{reinterpret_cast(chardata.data()), 178| 353| chardata.size() / sizeof(FromType)}; 179| | 180| 353| static const bool do_print_testcase = 181| 353| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 353| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 353] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 353| do { 189| | // step 0 - is the input valid? 190| 353| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 353| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 353] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 353| From == UtfEncodings::UTF8) { 198| 353| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 353] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 353| } 201| | 202| | // step 2 - what is the required size of the output? 203| 353| const auto [output_length, length_agree] = 204| 353| calculate_length(from, inputisvalid); 205| 353| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 353] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 353| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 197, False: 156] | Branch (208:28): [True: 0, False: 197] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 353| const auto [written, outputs_agree] = 215| 353| do_conversion(from, output_length, inputisvalid); 216| 353| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 353] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 353| return; 221| 353| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 353| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 353| validation_result verify_valid_input(FromSpan src) const { 231| 353| validation_result ret{}; 232| | 233| 353| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 353| const auto implementations = get_supported_implementations(); 235| 353| std::vector results; 236| 353| results.reserve(implementations.size()); 237| | 238| 1.05k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.05k, False: 353] ------------------ 239| 1.05k| results.push_back( 240| 1.05k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.05k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.05k| const bool validation2 = 245| 1.05k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.05k| src.data(), src.size()); 247| 1.05k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.05k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.05k| } 258| | 259| 353| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 353| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 353] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 353| } else { 273| 353| ret.implementations_agree = true; 274| 353| } 275| 353| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 353| return r.error == simdutf::SUCCESS; 277| 353| }); 278| 353| return ret; 279| 353| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 706| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 665| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 665| return r.error == simdutf::SUCCESS; 277| 665| }); _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 353| bool count_the_input(FromSpan src) const { 290| 353| const auto implementations = get_supported_implementations(); 291| 353| std::vector results; 292| 353| results.reserve(implementations.size()); 293| | 294| 1.05k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.05k, False: 353] ------------------ 295| 1.05k| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| 1.05k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.05k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.05k| } 303| 1.05k| results.push_back(ret); 304| 1.05k| } 305| 353| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 353| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 353] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 353| return true; 321| 353| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 706| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 353| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 353| length_result ret{}; 344| | 345| 353| const auto implementations = get_supported_implementations(); 346| 353| std::vector results; 347| 353| results.reserve(implementations.size()); 348| | 349| 1.05k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.05k, False: 353] ------------------ 350| 1.05k| const auto len = invoke_lengthcalc(impl, src); 351| 1.05k| results.push_back(len); 352| 1.05k| ret.length.push_back(len); 353| 1.05k| } 354| | 355| 353| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 353| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 353] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 353| } else { 375| 353| ret.implementations_agree = true; 376| 353| } 377| 353| return ret; 378| 353| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.05k| FromSpan src) const { 331| 1.05k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.05k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 706| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 353| const bool inputisvalid) const { 383| 353| conversion_result ret{}; 384| | 385| 353| const auto implementations = get_supported_implementations(); 386| | 387| 353| std::vector> results; 388| 353| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 353| std::vector> outputbuffers; 393| 353| outputbuffers.reserve(implementations.size()); 394| 1.41k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.05k, False: 353] ------------------ 395| 1.05k| auto impl = implementations[i]; 396| 1.05k| const ToType canary1{42}; 397| 1.05k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.05k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.05k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.05k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.05k| if constexpr (std::is_same_v) { 403| 1.05k| return r != 0; 404| 1.05k| } else { 405| 1.05k| return r.error == simdutf::error_code::SUCCESS; 406| 1.05k| } 407| 1.05k| }(implret1); 408| 1.05k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.05k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.05k| const ToType canary2{25}; 414| 1.05k| const auto outputbuffer_first_run = outputbuffer; 415| 1.05k| std::ranges::fill(outputbuffer, canary2); 416| 1.05k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.05k| src.size(), outputbuffer.data()); 418| | 419| 1.05k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.05k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.05k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 468, False: 591] | Branch (423:29): [True: 468, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 468| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 468| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 468] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 468| } 440| 1.05k| } 441| 1.05k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 468, False: 591] ------------------ 442| 1.05k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 353| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 197, False: 156] ------------------ 447| 591| for (auto& e : results) { ------------------ | Branch (447:20): [True: 591, False: 197] ------------------ 448| 591| e.outputhash.clear(); 449| 591| } 450| 197| } 451| | 452| 353| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 353| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 353] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 353| } else { 474| 353| ret.implementations_agree = true; 475| 353| } 476| 353| return ret; 477| 353| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.05k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.05k| } else { 405| 1.05k| return r.error == simdutf::error_code::SUCCESS; 406| 1.05k| } 407| 1.05k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 706| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_36clENSt3__14spanIKcLm18446744073709551615EEE: 554| 305| +[](std::span chardata) { \ 555| 305| const auto c = \ 556| 305| Conversion{ \ 559| 305| &I::lenfunc, &I::conversionfunc, \ 560| 305| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 305|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 305| ::std::void_t(); \ | | 1180| 305| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 305| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 305| constexpr auto _size = _name.size(); \ | | 1183| 305| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 305| return _nameof; }() ------------------ 561| 305| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 305|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 305| ::std::void_t(); \ | | 1180| 305| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 305| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 305| constexpr auto _size = _name.size(); \ | | 1183| 305| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 305| return _nameof; }() ------------------ 562| 305| c.fuzz(chardata); \ 563| 305| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 305| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 305| const FromSpan from{reinterpret_cast(chardata.data()), 178| 305| chardata.size() / sizeof(FromType)}; 179| | 180| 305| static const bool do_print_testcase = 181| 305| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 305| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 305] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 305| do { 189| | // step 0 - is the input valid? 190| 305| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 305| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 305] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 305| From == UtfEncodings::UTF8) { 198| 305| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 305] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 305| } 201| | 202| | // step 2 - what is the required size of the output? 203| 305| const auto [output_length, length_agree] = 204| 305| calculate_length(from, inputisvalid); 205| 305| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 305] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 305| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 184, False: 121] | Branch (208:28): [True: 0, False: 184] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 305| const auto [written, outputs_agree] = 215| 305| do_conversion(from, output_length, inputisvalid); 216| 305| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 305] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 305| return; 221| 305| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 305| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 305| validation_result verify_valid_input(FromSpan src) const { 231| 305| validation_result ret{}; 232| | 233| 305| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 305| const auto implementations = get_supported_implementations(); 235| 305| std::vector results; 236| 305| results.reserve(implementations.size()); 237| | 238| 915| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 915, False: 305] ------------------ 239| 915| results.push_back( 240| 915| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 915| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 915| const bool validation2 = 245| 915| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 915| src.data(), src.size()); 247| 915| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 915] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 915| } 258| | 259| 305| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 305| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 305] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 305| } else { 273| 305| ret.implementations_agree = true; 274| 305| } 275| 305| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 305| return r.error == simdutf::SUCCESS; 277| 305| }); 278| 305| return ret; 279| 305| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 610| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 547| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 547| return r.error == simdutf::SUCCESS; 277| 547| }); _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 305| bool count_the_input(FromSpan src) const { 290| 305| const auto implementations = get_supported_implementations(); 291| 305| std::vector results; 292| 305| results.reserve(implementations.size()); 293| | 294| 915| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 915, False: 305] ------------------ 295| 915| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| 915| } else if constexpr (From == UtfEncodings::UTF8) { 301| 915| ret = impl->count_utf8(src.data(), src.size()); 302| 915| } 303| 915| results.push_back(ret); 304| 915| } 305| 305| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 305| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 305] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 305| return true; 321| 305| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 610| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 305| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 305| length_result ret{}; 344| | 345| 305| const auto implementations = get_supported_implementations(); 346| 305| std::vector results; 347| 305| results.reserve(implementations.size()); 348| | 349| 915| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 915, False: 305] ------------------ 350| 915| const auto len = invoke_lengthcalc(impl, src); 351| 915| results.push_back(len); 352| 915| ret.length.push_back(len); 353| 915| } 354| | 355| 305| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 305| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 305] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 305| } else { 375| 305| ret.implementations_agree = true; 376| 305| } 377| 305| return ret; 378| 305| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 915| FromSpan src) const { 331| 915| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 915| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 610| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 305| const bool inputisvalid) const { 383| 305| conversion_result ret{}; 384| | 385| 305| const auto implementations = get_supported_implementations(); 386| | 387| 305| std::vector> results; 388| 305| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 305| std::vector> outputbuffers; 393| 305| outputbuffers.reserve(implementations.size()); 394| 1.22k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 915, False: 305] ------------------ 395| 915| auto impl = implementations[i]; 396| 915| const ToType canary1{42}; 397| 915| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 915| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 915| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 915| const auto success = [](const ConversionResult& r) -> bool { 402| 915| if constexpr (std::is_same_v) { 403| 915| return r != 0; 404| 915| } else { 405| 915| return r.error == simdutf::error_code::SUCCESS; 406| 915| } 407| 915| }(implret1); 408| 915| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 915| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 915| const ToType canary2{25}; 414| 915| const auto outputbuffer_first_run = outputbuffer; 415| 915| std::ranges::fill(outputbuffer, canary2); 416| 915| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 915| src.size(), outputbuffer.data()); 418| | 419| 915| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 915] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 915| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 363, False: 552] | Branch (423:29): [True: 363, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 363| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 363| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 363] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 363| } 440| 915| } 441| 915| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 363, False: 552] ------------------ 442| 915| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 305| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 184, False: 121] ------------------ 447| 552| for (auto& e : results) { ------------------ | Branch (447:20): [True: 552, False: 184] ------------------ 448| 552| e.outputhash.clear(); 449| 552| } 450| 184| } 451| | 452| 305| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 305| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 305] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 305| } else { 474| 305| ret.implementations_agree = true; 475| 305| } 476| 305| return ret; 477| 305| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 915| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 915| } else { 405| 915| return r.error == simdutf::error_code::SUCCESS; 406| 915| } 407| 915| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 610| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_37clENSt3__14spanIKcLm18446744073709551615EEE: 554| 373| +[](std::span chardata) { \ 555| 373| const auto c = \ 556| 373| Conversion{ \ 559| 373| &I::lenfunc, &I::conversionfunc, \ 560| 373| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 373|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 373| ::std::void_t(); \ | | 1180| 373| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 373| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 373| constexpr auto _size = _name.size(); \ | | 1183| 373| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 373| return _nameof; }() ------------------ 561| 373| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 373|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 373| ::std::void_t(); \ | | 1180| 373| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 373| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 373| constexpr auto _size = _name.size(); \ | | 1183| 373| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 373| return _nameof; }() ------------------ 562| 373| c.fuzz(chardata); \ 563| 373| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 373| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 373| const FromSpan from{reinterpret_cast(chardata.data()), 178| 373| chardata.size() / sizeof(FromType)}; 179| | 180| 373| static const bool do_print_testcase = 181| 373| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 373| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 373] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 373| do { 189| | // step 0 - is the input valid? 190| 373| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 373| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 373] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 373| From == UtfEncodings::UTF8) { 198| 373| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 373] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 373| } 201| | 202| | // step 2 - what is the required size of the output? 203| 373| const auto [output_length, length_agree] = 204| 373| calculate_length(from, inputisvalid); 205| 373| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 373] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 373| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 217, False: 156] | Branch (208:28): [True: 0, False: 217] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 373| const auto [written, outputs_agree] = 215| 373| do_conversion(from, output_length, inputisvalid); 216| 373| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 373] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 373| return; 221| 373| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 373| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 373| validation_result verify_valid_input(FromSpan src) const { 231| 373| validation_result ret{}; 232| | 233| 373| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 373| const auto implementations = get_supported_implementations(); 235| 373| std::vector results; 236| 373| results.reserve(implementations.size()); 237| | 238| 1.11k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.11k, False: 373] ------------------ 239| 1.11k| results.push_back( 240| 1.11k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.11k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.11k| const bool validation2 = 245| 1.11k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.11k| src.data(), src.size()); 247| 1.11k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.11k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.11k| } 258| | 259| 373| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 373| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 373] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 373| } else { 273| 373| ret.implementations_agree = true; 274| 373| } 275| 373| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 373| return r.error == simdutf::SUCCESS; 277| 373| }); 278| 373| return ret; 279| 373| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 746| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 685| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 685| return r.error == simdutf::SUCCESS; 277| 685| }); _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 373| bool count_the_input(FromSpan src) const { 290| 373| const auto implementations = get_supported_implementations(); 291| 373| std::vector results; 292| 373| results.reserve(implementations.size()); 293| | 294| 1.11k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.11k, False: 373] ------------------ 295| 1.11k| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| 1.11k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.11k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.11k| } 303| 1.11k| results.push_back(ret); 304| 1.11k| } 305| 373| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 373| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 373] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 373| return true; 321| 373| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 746| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 373| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 373| length_result ret{}; 344| | 345| 373| const auto implementations = get_supported_implementations(); 346| 373| std::vector results; 347| 373| results.reserve(implementations.size()); 348| | 349| 1.11k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.11k, False: 373] ------------------ 350| 1.11k| const auto len = invoke_lengthcalc(impl, src); 351| 1.11k| results.push_back(len); 352| 1.11k| ret.length.push_back(len); 353| 1.11k| } 354| | 355| 373| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 373| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 373] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 373| } else { 375| 373| ret.implementations_agree = true; 376| 373| } 377| 373| return ret; 378| 373| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.11k| FromSpan src) const { 331| 1.11k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.11k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 746| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 373| const bool inputisvalid) const { 383| 373| conversion_result ret{}; 384| | 385| 373| const auto implementations = get_supported_implementations(); 386| | 387| 373| std::vector> results; 388| 373| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 373| std::vector> outputbuffers; 393| 373| outputbuffers.reserve(implementations.size()); 394| 1.49k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.11k, False: 373] ------------------ 395| 1.11k| auto impl = implementations[i]; 396| 1.11k| const ToType canary1{42}; 397| 1.11k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.11k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.11k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.11k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.11k| if constexpr (std::is_same_v) { 403| 1.11k| return r != 0; 404| 1.11k| } else { 405| 1.11k| return r.error == simdutf::error_code::SUCCESS; 406| 1.11k| } 407| 1.11k| }(implret1); 408| 1.11k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.11k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.11k| const ToType canary2{25}; 414| 1.11k| const auto outputbuffer_first_run = outputbuffer; 415| 1.11k| std::ranges::fill(outputbuffer, canary2); 416| 1.11k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.11k| src.size(), outputbuffer.data()); 418| | 419| 1.11k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.11k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.11k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 468, False: 651] | Branch (423:29): [True: 468, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 468| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 468| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 468] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 468| } 440| 1.11k| } 441| 1.11k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 468, False: 651] ------------------ 442| 1.11k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 373| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 217, False: 156] ------------------ 447| 651| for (auto& e : results) { ------------------ | Branch (447:20): [True: 651, False: 217] ------------------ 448| 651| e.outputhash.clear(); 449| 651| } 450| 217| } 451| | 452| 373| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 373| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 373] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 373| } else { 474| 373| ret.implementations_agree = true; 475| 373| } 476| 373| return ret; 477| 373| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.11k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.11k| } else { 405| 1.11k| return r.error == simdutf::error_code::SUCCESS; 406| 1.11k| } 407| 1.11k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 746| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_38clENSt3__14spanIKcLm18446744073709551615EEE: 554| 62| +[](std::span chardata) { \ 555| 62| const auto c = \ 556| 62| Conversion{ \ 559| 62| &I::lenfunc, &I::conversionfunc, \ 560| 62| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 62|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 62| ::std::void_t(); \ | | 1180| 62| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 62| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 62| constexpr auto _size = _name.size(); \ | | 1183| 62| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 62| return _nameof; }() ------------------ 561| 62| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 62|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 62| ::std::void_t(); \ | | 1180| 62| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 62| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 62| constexpr auto _size = _name.size(); \ | | 1183| 62| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 62| return _nameof; }() ------------------ 562| 62| c.fuzz(chardata); \ 563| 62| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE4fuzzENSt3__14spanIS5_Lm18446744073709551615EEE: 175| 62| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 62| const FromSpan from{reinterpret_cast(chardata.data()), 178| 62| chardata.size() / sizeof(FromType)}; 179| | 180| 62| static const bool do_print_testcase = 181| 62| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 62| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 62] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 62| do { 189| | // step 0 - is the input valid? 190| 62| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 62| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 62] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 62| const auto [output_length, length_agree] = 204| 62| calculate_length(from, inputisvalid); 205| 62| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 62] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 62| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 62] | Branch (208:28): [True: 0, False: 0] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 62| const auto [written, outputs_agree] = 215| 62| do_conversion(from, output_length, inputisvalid); 216| 62| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 62] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 62| return; 221| 62| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 62| } _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 283| 62| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 62| return validation_result{.valid = true, .implementations_agree = true}; 287| 62| } _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 62| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 62| length_result ret{}; 344| | 345| 62| const auto implementations = get_supported_implementations(); 346| 62| std::vector results; 347| 62| results.reserve(implementations.size()); 348| | 349| 186| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 186, False: 62] ------------------ 350| 186| const auto len = invoke_lengthcalc(impl, src); 351| 186| results.push_back(len); 352| 186| ret.length.push_back(len); 353| 186| } 354| | 355| 62| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 62| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 62] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 62| } else { 375| 62| ret.implementations_agree = true; 376| 62| } 377| 62| return ret; 378| 62| } _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 186| FromSpan src) const { 339| 186| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 186| } _ZZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 124| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 62| const bool inputisvalid) const { 383| 62| conversion_result ret{}; 384| | 385| 62| const auto implementations = get_supported_implementations(); 386| | 387| 62| std::vector> results; 388| 62| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 62| std::vector> outputbuffers; 393| 62| outputbuffers.reserve(implementations.size()); 394| 248| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 186, False: 62] ------------------ 395| 186| auto impl = implementations[i]; 396| 186| const ToType canary1{42}; 397| 186| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 186| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 186| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 186| const auto success = [](const ConversionResult& r) -> bool { 402| 186| if constexpr (std::is_same_v) { 403| 186| return r != 0; 404| 186| } else { 405| 186| return r.error == simdutf::error_code::SUCCESS; 406| 186| } 407| 186| }(implret1); 408| 186| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 186| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 186| const ToType canary2{25}; 414| 186| const auto outputbuffer_first_run = outputbuffer; 415| 186| std::ranges::fill(outputbuffer, canary2); 416| 186| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 186| src.size(), outputbuffer.data()); 418| | 419| 186| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 186] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 186| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 186, False: 0] | Branch (423:29): [True: 180, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 180| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 180| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 180] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 180| } 440| 186| } 441| 186| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 180, False: 6] ------------------ 442| 186| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 62| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 62] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 62| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 62| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 62] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 62| } else { 474| 62| ret.implementations_agree = true; 475| 62| } 476| 62| return ret; 477| 62| } _ZZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 186| const auto success = [](const ConversionResult& r) -> bool { 402| 186| if constexpr (std::is_same_v) { 403| 186| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 186| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 124| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_39clENSt3__14spanIKcLm18446744073709551615EEE: 554| 48| +[](std::span chardata) { \ 555| 48| const auto c = \ 556| 48| Conversion{ \ 559| 48| &I::lenfunc, &I::conversionfunc, \ 560| 48| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 48|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 48| ::std::void_t(); \ | | 1180| 48| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 48| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 48| constexpr auto _size = _name.size(); \ | | 1183| 48| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 48| return _nameof; }() ------------------ 561| 48| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 48|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 48| ::std::void_t(); \ | | 1180| 48| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 48| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 48| constexpr auto _size = _name.size(); \ | | 1183| 48| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 48| return _nameof; }() ------------------ 562| 48| c.fuzz(chardata); \ 563| 48| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE4fuzzENSt3__14spanIS5_Lm18446744073709551615EEE: 175| 48| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 48| const FromSpan from{reinterpret_cast(chardata.data()), 178| 48| chardata.size() / sizeof(FromType)}; 179| | 180| 48| static const bool do_print_testcase = 181| 48| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 48| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 48] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 48| do { 189| | // step 0 - is the input valid? 190| 48| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 48| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 48] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 48| const auto [output_length, length_agree] = 204| 48| calculate_length(from, inputisvalid); 205| 48| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 48] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 48| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 48] | Branch (208:28): [True: 0, False: 0] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 48| const auto [written, outputs_agree] = 215| 48| do_conversion(from, output_length, inputisvalid); 216| 48| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 48] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 48| return; 221| 48| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 48| } _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 283| 48| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 48| return validation_result{.valid = true, .implementations_agree = true}; 287| 48| } _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 48| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 48| length_result ret{}; 344| | 345| 48| const auto implementations = get_supported_implementations(); 346| 48| std::vector results; 347| 48| results.reserve(implementations.size()); 348| | 349| 144| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 144, False: 48] ------------------ 350| 144| const auto len = invoke_lengthcalc(impl, src); 351| 144| results.push_back(len); 352| 144| ret.length.push_back(len); 353| 144| } 354| | 355| 48| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 48| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 48] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 48| } else { 375| 48| ret.implementations_agree = true; 376| 48| } 377| 48| return ret; 378| 48| } _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 144| FromSpan src) const { 339| 144| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 144| } _ZZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 96| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 48| const bool inputisvalid) const { 383| 48| conversion_result ret{}; 384| | 385| 48| const auto implementations = get_supported_implementations(); 386| | 387| 48| std::vector> results; 388| 48| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 48| std::vector> outputbuffers; 393| 48| outputbuffers.reserve(implementations.size()); 394| 192| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 144, False: 48] ------------------ 395| 144| auto impl = implementations[i]; 396| 144| const ToType canary1{42}; 397| 144| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 144| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 144| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 144| const auto success = [](const ConversionResult& r) -> bool { 402| 144| if constexpr (std::is_same_v) { 403| 144| return r != 0; 404| 144| } else { 405| 144| return r.error == simdutf::error_code::SUCCESS; 406| 144| } 407| 144| }(implret1); 408| 144| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 144| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 144| const ToType canary2{25}; 414| 144| const auto outputbuffer_first_run = outputbuffer; 415| 144| std::ranges::fill(outputbuffer, canary2); 416| 144| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 144| src.size(), outputbuffer.data()); 418| | 419| 144| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 144] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 144| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 144, False: 0] | Branch (423:29): [True: 138, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 138| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 138| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 138] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 138| } 440| 144| } 441| 144| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 138, False: 6] ------------------ 442| 144| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 48| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 48] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 48| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 48| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 48] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 48| } else { 474| 48| ret.implementations_agree = true; 475| 48| } 476| 48| return ret; 477| 48| } _ZZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 144| const auto success = [](const ConversionResult& r) -> bool { 402| 144| if constexpr (std::is_same_v) { 403| 144| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 144| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 96| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_40clENSt3__14spanIKcLm18446744073709551615EEE: 554| 38| +[](std::span chardata) { \ 555| 38| const auto c = \ 556| 38| Conversion{ \ 559| 38| &I::lenfunc, &I::conversionfunc, \ 560| 38| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 38|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 38| ::std::void_t(); \ | | 1180| 38| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 38| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 38| constexpr auto _size = _name.size(); \ | | 1183| 38| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 38| return _nameof; }() ------------------ 561| 38| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 38|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 38| ::std::void_t(); \ | | 1180| 38| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 38| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 38| constexpr auto _size = _name.size(); \ | | 1183| 38| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 38| return _nameof; }() ------------------ 562| 38| c.fuzz(chardata); \ 563| 38| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE4fuzzENSt3__14spanIS5_Lm18446744073709551615EEE: 175| 38| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 38| const FromSpan from{reinterpret_cast(chardata.data()), 178| 38| chardata.size() / sizeof(FromType)}; 179| | 180| 38| static const bool do_print_testcase = 181| 38| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 38| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 38] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 38| do { 189| | // step 0 - is the input valid? 190| 38| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 38| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 38] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 38| const auto [output_length, length_agree] = 204| 38| calculate_length(from, inputisvalid); 205| 38| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 38] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 38| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 38] | Branch (208:28): [True: 0, False: 0] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 38| const auto [written, outputs_agree] = 215| 38| do_conversion(from, output_length, inputisvalid); 216| 38| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 38] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 38| return; 221| 38| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 38| } _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 283| 38| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 38| return validation_result{.valid = true, .implementations_agree = true}; 287| 38| } _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 38| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 38| length_result ret{}; 344| | 345| 38| const auto implementations = get_supported_implementations(); 346| 38| std::vector results; 347| 38| results.reserve(implementations.size()); 348| | 349| 114| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 114, False: 38] ------------------ 350| 114| const auto len = invoke_lengthcalc(impl, src); 351| 114| results.push_back(len); 352| 114| ret.length.push_back(len); 353| 114| } 354| | 355| 38| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 38| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 38] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 38| } else { 375| 38| ret.implementations_agree = true; 376| 38| } 377| 38| return ret; 378| 38| } _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 114| FromSpan src) const { 339| 114| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 114| } _ZZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 76| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 38| const bool inputisvalid) const { 383| 38| conversion_result ret{}; 384| | 385| 38| const auto implementations = get_supported_implementations(); 386| | 387| 38| std::vector> results; 388| 38| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 38| std::vector> outputbuffers; 393| 38| outputbuffers.reserve(implementations.size()); 394| 152| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 114, False: 38] ------------------ 395| 114| auto impl = implementations[i]; 396| 114| const ToType canary1{42}; 397| 114| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 114| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 114| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 114| const auto success = [](const ConversionResult& r) -> bool { 402| 114| if constexpr (std::is_same_v) { 403| 114| return r != 0; 404| 114| } else { 405| 114| return r.error == simdutf::error_code::SUCCESS; 406| 114| } 407| 114| }(implret1); 408| 114| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 114| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 114| const ToType canary2{25}; 414| 114| const auto outputbuffer_first_run = outputbuffer; 415| 114| std::ranges::fill(outputbuffer, canary2); 416| 114| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 114| src.size(), outputbuffer.data()); 418| | 419| 114| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 114] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 114| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 114, False: 0] | Branch (423:29): [True: 108, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 108| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 108| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 108] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 108| } 440| 114| } 441| 114| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 108, False: 6] ------------------ 442| 114| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 38| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 38] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 38| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 38| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 38] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 38| } else { 474| 38| ret.implementations_agree = true; 475| 38| } 476| 38| return ret; 477| 38| } _ZZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 114| const auto success = [](const ConversionResult& r) -> bool { 402| 114| if constexpr (std::is_same_v) { 403| 114| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 114| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 76| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_41clENSt3__14spanIKcLm18446744073709551615EEE: 554| 302| +[](std::span chardata) { \ 555| 302| const auto c = \ 556| 302| Conversion{ \ 559| 302| &I::lenfunc, &I::conversionfunc, \ 560| 302| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 302|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 302| ::std::void_t(); \ | | 1180| 302| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 302| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 302| constexpr auto _size = _name.size(); \ | | 1183| 302| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 302| return _nameof; }() ------------------ 561| 302| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 302|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 302| ::std::void_t(); \ | | 1180| 302| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 302| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 302| constexpr auto _size = _name.size(); \ | | 1183| 302| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 302| return _nameof; }() ------------------ 562| 302| c.fuzz(chardata); \ 563| 302| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 302| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 302| const FromSpan from{reinterpret_cast(chardata.data()), 178| 302| chardata.size() / sizeof(FromType)}; 179| | 180| 302| static const bool do_print_testcase = 181| 302| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 302| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 302] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 302| do { 189| | // step 0 - is the input valid? 190| 302| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 302| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 302] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 302| const auto [output_length, length_agree] = 204| 302| calculate_length(from, inputisvalid); 205| 302| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 302] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 302| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 302] | Branch (208:28): [True: 0, False: 0] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 302| const auto [written, outputs_agree] = 215| 302| do_conversion(from, output_length, inputisvalid); 216| 302| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 302] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 302| return; 221| 302| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 302| } _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 283| 302| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 302| return validation_result{.valid = true, .implementations_agree = true}; 287| 302| } _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 302| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 302| length_result ret{}; 344| | 345| 302| const auto implementations = get_supported_implementations(); 346| 302| std::vector results; 347| 302| results.reserve(implementations.size()); 348| | 349| 906| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 906, False: 302] ------------------ 350| 906| const auto len = invoke_lengthcalc(impl, src); 351| 906| results.push_back(len); 352| 906| ret.length.push_back(len); 353| 906| } 354| | 355| 302| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 302| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 302] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 302| } else { 375| 302| ret.implementations_agree = true; 376| 302| } 377| 302| return ret; 378| 302| } _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 906| FromSpan src) const { 331| 906| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 906| } _ZZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 604| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 302| const bool inputisvalid) const { 383| 302| conversion_result ret{}; 384| | 385| 302| const auto implementations = get_supported_implementations(); 386| | 387| 302| std::vector> results; 388| 302| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 302| std::vector> outputbuffers; 393| 302| outputbuffers.reserve(implementations.size()); 394| 1.20k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 906, False: 302] ------------------ 395| 906| auto impl = implementations[i]; 396| 906| const ToType canary1{42}; 397| 906| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 906| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 906| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 906| const auto success = [](const ConversionResult& r) -> bool { 402| 906| if constexpr (std::is_same_v) { 403| 906| return r != 0; 404| 906| } else { 405| 906| return r.error == simdutf::error_code::SUCCESS; 406| 906| } 407| 906| }(implret1); 408| 906| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 906| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 906| const ToType canary2{25}; 414| 906| const auto outputbuffer_first_run = outputbuffer; 415| 906| std::ranges::fill(outputbuffer, canary2); 416| 906| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 906| src.size(), outputbuffer.data()); 418| | 419| 906| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 906] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 906| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 906, False: 0] | Branch (423:29): [True: 900, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 900| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 900| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 900] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 900| } 440| 906| } 441| 906| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 900, False: 6] ------------------ 442| 906| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 302| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 302] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 302| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 302| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 302] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 302| } else { 474| 302| ret.implementations_agree = true; 475| 302| } 476| 302| return ret; 477| 302| } _ZZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 906| const auto success = [](const ConversionResult& r) -> bool { 402| 906| if constexpr (std::is_same_v) { 403| 906| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 906| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 604| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZN10FNV1A_hash14fnv1ahash_implENSt3__14spanIKcLm18446744073709551615EEE: 75| 17.5k| static constexpr std::uint64_t fnv1ahash_impl(std::span bytes) { 76| 17.5k| auto hash = offset; 77| | 78| 640M| for (auto byte : bytes) { ------------------ | Branch (78:20): [True: 640M, False: 17.5k] ------------------ 79| 640M| hash ^= static_cast(byte); 80| 640M| hash *= prime; 81| 640M| } 82| | 83| 17.5k| return hash; 84| 17.5k| } _ZN10FNV1A_hash14fnv1ahash_implIKDsLm18446744073709551615EQaaaa11is_hashableIT_Entsr3stdE9is_same_vIS2_cEntsr3stdE9is_same_vIS2_hEEEmNSt3__14spanIS2_XT0_EEE: 89| 15.0k| static constexpr std::uint64_t fnv1ahash_impl(std::span data) { 90| 15.0k| return fnv1ahash_impl({reinterpret_cast(data.data()), 91| 15.0k| data.size_bytes()}); 92| 15.0k| } _ZN10FNV1A_hash14fnv1ahash_implENSt3__14spanIKhLm18446744073709551615EEE: 65| 23.5k| fnv1ahash_impl(std::span bytes) { 66| 23.5k| auto hash = offset; 67| | 68| 2.26G| for (std::uint64_t byte : bytes) { ------------------ | Branch (68:29): [True: 2.26G, False: 23.5k] ------------------ 69| 2.26G| hash ^= byte; 70| 2.26G| hash *= prime; 71| 2.26G| } 72| | 73| 23.5k| return hash; 74| 23.5k| } _Z29get_supported_implementationsv: 26| 33.4k|get_supported_implementations() { 27| 33.4k| static const auto impl = []() -> auto { 28| 33.4k| std::vector ret; 29| 33.4k| for (auto e : simdutf::get_available_implementations()) { 30| 33.4k| std::cerr << "implementation " << e->name() << " is available? " 31| 33.4k| << e->supported_by_runtime_system() << '\n'; 32| 33.4k| if (e->supported_by_runtime_system()) { 33| 33.4k| ret.push_back(e); 34| 33.4k| } 35| 33.4k| } 36| 33.4k| return ret; 37| 33.4k| }(); 38| 33.4k| return {impl.data(), impl.size()}; 39| 33.4k|} _ZZ29get_supported_implementationsvENKUlvE_clEv: 27| 1| static const auto impl = []() -> auto { 28| 1| std::vector ret; 29| 4| for (auto e : simdutf::get_available_implementations()) { ------------------ | Branch (29:17): [True: 4, False: 1] ------------------ 30| 4| std::cerr << "implementation " << e->name() << " is available? " 31| 4| << e->supported_by_runtime_system() << '\n'; 32| 4| if (e->supported_by_runtime_system()) { ------------------ | Branch (32:11): [True: 3, False: 1] ------------------ 33| 3| ret.push_back(e); 34| 3| } 35| 4| } 36| 1| return ret; 37| 1| }(); _ZneRKN7simdutf6resultES2_: 42| 29.2k|inline bool operator!=(const simdutf::result& a, const simdutf::result& b) { 43| 29.2k| return a.count != b.count || a.error != b.error; ------------------ | Branch (43:10): [True: 0, False: 29.2k] | Branch (43:32): [True: 0, False: 29.2k] ------------------ 44| 29.2k|} _ZN10FNV1A_hash6as_strIJNSt3__16vectorIDiNS1_9allocatorIDiEEEEEEENS1_12basic_stringIcNS1_11char_traitsIcEENS3_IcEEEEDpRKT_: 100| 8.41k| template static std::string as_str(const Data&... data) { 101| 8.41k| static_assert(sizeof...(Data) > 0, "must hash with at least one argument"); 102| 8.41k| std::uint64_t h; 103| | if constexpr (sizeof...(Data) > 1) { 104| | const std::array hashes{fnv1ahash_impl(data)...}; 105| | const auto s = std::span(hashes); 106| | h = fnv1ahash_impl(s); 107| 8.41k| } else { 108| 8.41k| h = fnv1ahash_impl(data...); 109| 8.41k| } 110| 8.41k| constexpr std::size_t expected_chars = 16; 111| 8.41k| std::string ret(expected_chars, '0'); 112| 8.41k| auto c = std::to_chars(ret.data(), ret.data() + ret.size(), h, 16); 113| 8.41k| assert(c.ec == std::errc{}); ------------------ | Branch (113:5): [True: 8.41k, False: 0] ------------------ 114| 8.41k| auto nwritten = c.ptr - ret.data(); 115| 8.41k| assert(nwritten <= expected_chars); ------------------ | Branch (115:5): [True: 8.41k, False: 0] ------------------ 116| 8.41k| std::rotate(ret.data(), c.ptr, ret.data() + expected_chars); 117| 8.41k| return ret; 118| 8.41k| } _ZN10FNV1A_hash14fnv1ahash_implIDiQ11is_hashableIT_EEEmRKNSt3__16vectorIS1_NS2_9allocatorIS1_EEEE: 96| 8.41k| static constexpr std::uint64_t fnv1ahash_impl(const std::vector& data) { 97| 8.41k| return fnv1ahash_impl(std::span(data)); 98| 8.41k| } _ZN10FNV1A_hash14fnv1ahash_implIKDiLm18446744073709551615EQaaaa11is_hashableIT_Entsr3stdE9is_same_vIS2_cEntsr3stdE9is_same_vIS2_hEEEmNSt3__14spanIS2_XT0_EEE: 89| 8.41k| static constexpr std::uint64_t fnv1ahash_impl(std::span data) { 90| 8.41k| return fnv1ahash_impl({reinterpret_cast(data.data()), 91| 8.41k| data.size_bytes()}); 92| 8.41k| } _ZN10FNV1A_hash6as_strIJNSt3__16vectorIcNS1_9allocatorIcEEEEEEENS1_12basic_stringIcNS1_11char_traitsIcEES4_EEDpRKT_: 100| 17.5k| template static std::string as_str(const Data&... data) { 101| 17.5k| static_assert(sizeof...(Data) > 0, "must hash with at least one argument"); 102| 17.5k| std::uint64_t h; 103| | if constexpr (sizeof...(Data) > 1) { 104| | const std::array hashes{fnv1ahash_impl(data)...}; 105| | const auto s = std::span(hashes); 106| | h = fnv1ahash_impl(s); 107| 17.5k| } else { 108| 17.5k| h = fnv1ahash_impl(data...); 109| 17.5k| } 110| 17.5k| constexpr std::size_t expected_chars = 16; 111| 17.5k| std::string ret(expected_chars, '0'); 112| 17.5k| auto c = std::to_chars(ret.data(), ret.data() + ret.size(), h, 16); 113| 17.5k| assert(c.ec == std::errc{}); ------------------ | Branch (113:5): [True: 17.5k, False: 0] ------------------ 114| 17.5k| auto nwritten = c.ptr - ret.data(); 115| 17.5k| assert(nwritten <= expected_chars); ------------------ | Branch (115:5): [True: 17.5k, False: 0] ------------------ 116| 17.5k| std::rotate(ret.data(), c.ptr, ret.data() + expected_chars); 117| 17.5k| return ret; 118| 17.5k| } _ZN10FNV1A_hash14fnv1ahash_implIcQ11is_hashableIT_EEEmRKNSt3__16vectorIS1_NS2_9allocatorIS1_EEEE: 96| 17.5k| static constexpr std::uint64_t fnv1ahash_impl(const std::vector& data) { 97| 17.5k| return fnv1ahash_impl(std::span(data)); 98| 17.5k| } _ZN10FNV1A_hash6as_strIJNSt3__16vectorIDsNS1_9allocatorIDsEEEEEEENS1_12basic_stringIcNS1_11char_traitsIcEENS3_IcEEEEDpRKT_: 100| 15.0k| template static std::string as_str(const Data&... data) { 101| 15.0k| static_assert(sizeof...(Data) > 0, "must hash with at least one argument"); 102| 15.0k| std::uint64_t h; 103| | if constexpr (sizeof...(Data) > 1) { 104| | const std::array hashes{fnv1ahash_impl(data)...}; 105| | const auto s = std::span(hashes); 106| | h = fnv1ahash_impl(s); 107| 15.0k| } else { 108| 15.0k| h = fnv1ahash_impl(data...); 109| 15.0k| } 110| 15.0k| constexpr std::size_t expected_chars = 16; 111| 15.0k| std::string ret(expected_chars, '0'); 112| 15.0k| auto c = std::to_chars(ret.data(), ret.data() + ret.size(), h, 16); 113| 15.0k| assert(c.ec == std::errc{}); ------------------ | Branch (113:5): [True: 15.0k, False: 0] ------------------ 114| 15.0k| auto nwritten = c.ptr - ret.data(); 115| 15.0k| assert(nwritten <= expected_chars); ------------------ | Branch (115:5): [True: 15.0k, False: 0] ------------------ 116| 15.0k| std::rotate(ret.data(), c.ptr, ret.data() + expected_chars); 117| 15.0k| return ret; 118| 15.0k| } _ZN10FNV1A_hash14fnv1ahash_implIDsQ11is_hashableIT_EEEmRKNSt3__16vectorIS1_NS2_9allocatorIS1_EEEE: 96| 15.0k| static constexpr std::uint64_t fnv1ahash_impl(const std::vector& data) { 97| 15.0k| return fnv1ahash_impl(std::span(data)); 98| 15.0k| } _ZeqRKN7simdutf6resultES2_: 45| 7.73k|inline bool operator==(const simdutf::result& a, const simdutf::result& b) { 46| 7.73k| return a.count == b.count && a.error == b.error; ------------------ | Branch (46:10): [True: 7.73k, False: 0] | Branch (46:32): [True: 7.73k, False: 0] ------------------ 47| 7.73k|} _ZNK23ValidationFunctionTraitIL12UtfEncodings0EE24ValidationWithErrorsNameMUlvE_clEv: 1178| 2|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 2| ::std::void_t(); \ 1180| 2| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 2| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 2| constexpr auto _size = _name.size(); \ 1183| 2| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 2| return _nameof; }() _ZNK6nameof7cstringILt28EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 4| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt28EE4dataEv: 233| 4| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt28EE4sizeEv: 235| 4| [[nodiscard]] constexpr size_type size() const noexcept { return N; } _ZNK23ValidationFunctionTraitIL12UtfEncodings1EE24ValidationWithErrorsNameMUlvE_clEv: 1178| 2|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 2| ::std::void_t(); \ 1180| 2| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 2| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 2| constexpr auto _size = _name.size(); \ 1183| 2| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 2| return _nameof; }() _ZNK23ValidationFunctionTraitIL12UtfEncodings3EE24ValidationWithErrorsNameMUlvE_clEv: 1178| 2|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 2| ::std::void_t(); \ 1180| 2| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 2| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 2| constexpr auto _size = _name.size(); \ 1183| 2| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 2| return _nameof; }() _ZNK6nameof7cstringILt26EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 2| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt26EE4dataEv: 233| 2| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt26EE4sizeEv: 235| 2| [[nodiscard]] constexpr size_type size() const noexcept { return N; } _ZNK23ValidationFunctionTraitIL12UtfEncodings2EE24ValidationWithErrorsNameMUlvE_clEv: 1178| 2|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 2| ::std::void_t(); \ 1180| 2| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 2| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 2| constexpr auto _size = _name.size(); \ 1183| 2| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 2| return _nameof; }() _ZNK6nameof7cstringILt25EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 1.00k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt25EE4dataEv: 233| 1.00k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt25EE4sizeEv: 235| 1.00k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_0clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 87|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 87| ::std::void_t(); \ 1180| 87| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 87| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 87| constexpr auto _size = _name.size(); \ 1183| 87| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 87| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_0clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 87|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 87| ::std::void_t(); \ 1180| 87| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 87| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 87| constexpr auto _size = _name.size(); \ 1183| 87| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 87| return _nameof; }() _ZNK6nameof7cstringILt30EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 375| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt30EE4dataEv: 233| 375| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt30EE4sizeEv: 235| 375| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_1clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 248|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 248| ::std::void_t(); \ 1180| 248| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 248| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 248| constexpr auto _size = _name.size(); \ 1183| 248| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 248| return _nameof; }() _ZNK6nameof7cstringILt24EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 3.21k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt24EE4dataEv: 233| 3.21k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt24EE4sizeEv: 235| 3.21k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_1clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 248|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 248| ::std::void_t(); \ 1180| 248| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 248| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 248| constexpr auto _size = _name.size(); \ 1183| 248| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 248| return _nameof; }() _ZNK6nameof7cstringILt29EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 964| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt29EE4dataEv: 233| 964| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt29EE4sizeEv: 235| 964| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_2clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 70|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 70| ::std::void_t(); \ 1180| 70| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 70| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 70| constexpr auto _size = _name.size(); \ 1183| 70| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 70| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_2clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 70|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 70| ::std::void_t(); \ 1180| 70| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 70| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 70| constexpr auto _size = _name.size(); \ 1183| 70| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 70| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_3clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 215|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 215| ::std::void_t(); \ 1180| 215| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 215| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 215| constexpr auto _size = _name.size(); \ 1183| 215| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 215| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_3clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 215|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 215| ::std::void_t(); \ 1180| 215| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 215| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 215| constexpr auto _size = _name.size(); \ 1183| 215| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 215| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_4clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 69|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 69| ::std::void_t(); \ 1180| 69| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 69| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 69| constexpr auto _size = _name.size(); \ 1183| 69| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 69| return _nameof; }() _ZNK6nameof7cstringILt23EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 3.64k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt23EE4dataEv: 233| 3.64k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt23EE4sizeEv: 235| 3.64k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_4clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 69|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 69| ::std::void_t(); \ 1180| 69| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 69| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 69| constexpr auto _size = _name.size(); \ 1183| 69| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 69| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_5clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 149|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 149| ::std::void_t(); \ 1180| 149| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 149| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 149| constexpr auto _size = _name.size(); \ 1183| 149| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 149| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_5clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 149|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 149| ::std::void_t(); \ 1180| 149| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 149| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 149| constexpr auto _size = _name.size(); \ 1183| 149| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 149| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_6clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 120|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 120| ::std::void_t(); \ 1180| 120| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 120| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 120| constexpr auto _size = _name.size(); \ 1183| 120| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 120| return _nameof; }() _ZNK6nameof7cstringILt22EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 4.39k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt22EE4dataEv: 233| 4.39k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt22EE4sizeEv: 235| 4.39k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_6clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 120|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 120| ::std::void_t(); \ 1180| 120| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 120| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 120| constexpr auto _size = _name.size(); \ 1183| 120| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 120| return _nameof; }() _ZNK6nameof7cstringILt27EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 362| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt27EE4dataEv: 233| 362| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt27EE4sizeEv: 235| 362| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_7clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 265|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 265| ::std::void_t(); \ 1180| 265| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 265| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 265| constexpr auto _size = _name.size(); \ 1183| 265| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 265| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_7clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 265|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 265| ::std::void_t(); \ 1180| 265| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 265| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 265| constexpr auto _size = _name.size(); \ 1183| 265| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 265| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_8clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 236|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 236| ::std::void_t(); \ 1180| 236| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 236| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 236| constexpr auto _size = _name.size(); \ 1183| 236| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 236| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_8clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 236|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 236| ::std::void_t(); \ 1180| 236| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 236| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 236| constexpr auto _size = _name.size(); \ 1183| 236| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 236| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_9clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 242|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 242| ::std::void_t(); \ 1180| 242| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 242| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 242| constexpr auto _size = _name.size(); \ 1183| 242| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 242| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_9clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 242|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 242| ::std::void_t(); \ 1180| 242| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 242| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 242| constexpr auto _size = _name.size(); \ 1183| 242| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 242| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_10clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 57|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 57| ::std::void_t(); \ 1180| 57| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 57| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 57| constexpr auto _size = _name.size(); \ 1183| 57| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 57| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_10clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 57|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 57| ::std::void_t(); \ 1180| 57| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 57| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 57| constexpr auto _size = _name.size(); \ 1183| 57| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 57| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_11clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 140|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 140| ::std::void_t(); \ 1180| 140| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 140| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 140| constexpr auto _size = _name.size(); \ 1183| 140| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 140| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_11clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 140|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 140| ::std::void_t(); \ 1180| 140| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 140| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 140| constexpr auto _size = _name.size(); \ 1183| 140| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 140| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_12clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 173|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 173| ::std::void_t(); \ 1180| 173| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 173| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 173| constexpr auto _size = _name.size(); \ 1183| 173| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 173| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_12clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 173|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 173| ::std::void_t(); \ 1180| 173| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 173| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 173| constexpr auto _size = _name.size(); \ 1183| 173| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 173| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_13clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 62|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 62| ::std::void_t(); \ 1180| 62| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 62| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 62| constexpr auto _size = _name.size(); \ 1183| 62| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 62| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_13clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 62|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 62| ::std::void_t(); \ 1180| 62| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 62| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 62| constexpr auto _size = _name.size(); \ 1183| 62| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 62| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_14clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 163|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 163| ::std::void_t(); \ 1180| 163| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 163| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 163| constexpr auto _size = _name.size(); \ 1183| 163| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 163| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_14clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 163|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 163| ::std::void_t(); \ 1180| 163| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 163| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 163| constexpr auto _size = _name.size(); \ 1183| 163| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 163| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_15clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 218|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 218| ::std::void_t(); \ 1180| 218| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 218| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 218| constexpr auto _size = _name.size(); \ 1183| 218| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 218| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_15clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 218|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 218| ::std::void_t(); \ 1180| 218| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 218| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 218| constexpr auto _size = _name.size(); \ 1183| 218| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 218| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_16clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 114|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 114| ::std::void_t(); \ 1180| 114| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 114| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 114| constexpr auto _size = _name.size(); \ 1183| 114| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 114| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_16clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 114|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 114| ::std::void_t(); \ 1180| 114| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 114| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 114| constexpr auto _size = _name.size(); \ 1183| 114| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 114| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_17clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 270|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 270| ::std::void_t(); \ 1180| 270| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 270| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 270| constexpr auto _size = _name.size(); \ 1183| 270| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 270| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_17clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 270|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 270| ::std::void_t(); \ 1180| 270| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 270| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 270| constexpr auto _size = _name.size(); \ 1183| 270| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 270| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_18clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 271|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 271| ::std::void_t(); \ 1180| 271| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 271| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 271| constexpr auto _size = _name.size(); \ 1183| 271| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 271| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_18clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 271|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 271| ::std::void_t(); \ 1180| 271| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 271| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 271| constexpr auto _size = _name.size(); \ 1183| 271| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 271| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_19clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 324|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 324| ::std::void_t(); \ 1180| 324| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 324| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 324| constexpr auto _size = _name.size(); \ 1183| 324| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 324| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_19clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 324|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 324| ::std::void_t(); \ 1180| 324| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 324| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 324| constexpr auto _size = _name.size(); \ 1183| 324| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 324| return _nameof; }() _ZNK6nameof7cstringILt21EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 703| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt21EE4dataEv: 233| 703| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt21EE4sizeEv: 235| 703| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_20clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 293|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 293| ::std::void_t(); \ 1180| 293| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 293| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 293| constexpr auto _size = _name.size(); \ 1183| 293| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 293| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_20clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 293|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 293| ::std::void_t(); \ 1180| 293| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 293| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 293| constexpr auto _size = _name.size(); \ 1183| 293| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 293| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_21clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 397|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 397| ::std::void_t(); \ 1180| 397| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 397| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 397| constexpr auto _size = _name.size(); \ 1183| 397| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 397| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_21clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 397|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 397| ::std::void_t(); \ 1180| 397| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 397| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 397| constexpr auto _size = _name.size(); \ 1183| 397| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 397| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_22clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 388|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 388| ::std::void_t(); \ 1180| 388| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 388| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 388| constexpr auto _size = _name.size(); \ 1183| 388| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 388| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_22clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 388|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 388| ::std::void_t(); \ 1180| 388| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 388| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 388| constexpr auto _size = _name.size(); \ 1183| 388| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 388| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_23clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 379|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 379| ::std::void_t(); \ 1180| 379| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 379| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 379| constexpr auto _size = _name.size(); \ 1183| 379| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 379| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_23clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 379|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 379| ::std::void_t(); \ 1180| 379| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 379| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 379| constexpr auto _size = _name.size(); \ 1183| 379| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 379| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_24clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 133|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 133| ::std::void_t(); \ 1180| 133| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 133| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 133| constexpr auto _size = _name.size(); \ 1183| 133| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 133| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_24clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 133|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 133| ::std::void_t(); \ 1180| 133| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 133| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 133| constexpr auto _size = _name.size(); \ 1183| 133| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 133| return _nameof; }() _ZNK6nameof7cstringILt37EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 272| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt37EE4dataEv: 233| 272| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt37EE4sizeEv: 235| 272| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_25clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 173|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 173| ::std::void_t(); \ 1180| 173| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 173| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 173| constexpr auto _size = _name.size(); \ 1183| 173| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 173| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_25clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 173|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 173| ::std::void_t(); \ 1180| 173| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 173| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 173| constexpr auto _size = _name.size(); \ 1183| 173| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 173| return _nameof; }() _ZNK6nameof7cstringILt36EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 1.02k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt36EE4dataEv: 233| 1.02k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt36EE4sizeEv: 235| 1.02k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_26clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 327|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 327| ::std::void_t(); \ 1180| 327| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 327| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 327| constexpr auto _size = _name.size(); \ 1183| 327| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 327| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_26clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 327|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 327| ::std::void_t(); \ 1180| 327| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 327| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 327| constexpr auto _size = _name.size(); \ 1183| 327| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 327| return _nameof; }() _ZNK6nameof7cstringILt35EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 1.52k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt35EE4dataEv: 233| 1.52k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt35EE4sizeEv: 235| 1.52k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_27clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 139|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 139| ::std::void_t(); \ 1180| 139| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 139| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 139| constexpr auto _size = _name.size(); \ 1183| 139| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 139| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_27clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 139|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 139| ::std::void_t(); \ 1180| 139| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 139| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 139| constexpr auto _size = _name.size(); \ 1183| 139| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 139| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_28clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 167|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 167| ::std::void_t(); \ 1180| 167| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 167| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 167| constexpr auto _size = _name.size(); \ 1183| 167| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 167| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_28clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 167|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 167| ::std::void_t(); \ 1180| 167| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 167| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 167| constexpr auto _size = _name.size(); \ 1183| 167| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 167| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_29clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 321|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 321| ::std::void_t(); \ 1180| 321| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 321| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 321| constexpr auto _size = _name.size(); \ 1183| 321| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 321| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_29clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 321|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 321| ::std::void_t(); \ 1180| 321| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 321| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 321| constexpr auto _size = _name.size(); \ 1183| 321| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 321| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_30clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 218|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 218| ::std::void_t(); \ 1180| 218| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 218| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 218| constexpr auto _size = _name.size(); \ 1183| 218| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 218| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_30clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 218|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 218| ::std::void_t(); \ 1180| 218| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 218| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 218| constexpr auto _size = _name.size(); \ 1183| 218| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 218| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_31clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 347|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 347| ::std::void_t(); \ 1180| 347| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 347| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 347| constexpr auto _size = _name.size(); \ 1183| 347| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 347| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_31clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 347|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 347| ::std::void_t(); \ 1180| 347| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 347| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 347| constexpr auto _size = _name.size(); \ 1183| 347| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 347| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_32clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 337|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 337| ::std::void_t(); \ 1180| 337| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 337| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 337| constexpr auto _size = _name.size(); \ 1183| 337| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 337| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_32clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 337|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 337| ::std::void_t(); \ 1180| 337| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 337| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 337| constexpr auto _size = _name.size(); \ 1183| 337| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 337| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_33clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 416|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 416| ::std::void_t(); \ 1180| 416| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 416| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 416| constexpr auto _size = _name.size(); \ 1183| 416| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 416| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_33clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 416|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 416| ::std::void_t(); \ 1180| 416| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 416| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 416| constexpr auto _size = _name.size(); \ 1183| 416| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 416| return _nameof; }() _ZNK6nameof7cstringILt33EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 789| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt33EE4dataEv: 233| 789| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt33EE4sizeEv: 235| 789| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_34clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 256|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 256| ::std::void_t(); \ 1180| 256| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 256| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 256| constexpr auto _size = _name.size(); \ 1183| 256| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 256| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_34clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 256|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 256| ::std::void_t(); \ 1180| 256| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 256| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 256| constexpr auto _size = _name.size(); \ 1183| 256| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 256| return _nameof; }() _ZNK6nameof7cstringILt34EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 256| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt34EE4dataEv: 233| 256| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt34EE4sizeEv: 235| 256| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_35clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 353|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 353| ::std::void_t(); \ 1180| 353| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 353| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 353| constexpr auto _size = _name.size(); \ 1183| 353| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 353| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_35clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 353|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 353| ::std::void_t(); \ 1180| 353| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 353| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 353| constexpr auto _size = _name.size(); \ 1183| 353| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 353| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_36clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 305|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 305| ::std::void_t(); \ 1180| 305| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 305| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 305| constexpr auto _size = _name.size(); \ 1183| 305| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 305| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_36clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 305|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 305| ::std::void_t(); \ 1180| 305| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 305| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 305| constexpr auto _size = _name.size(); \ 1183| 305| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 305| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_37clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 373|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 373| ::std::void_t(); \ 1180| 373| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 373| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 373| constexpr auto _size = _name.size(); \ 1183| 373| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 373| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_37clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 373|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 373| ::std::void_t(); \ 1180| 373| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 373| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 373| constexpr auto _size = _name.size(); \ 1183| 373| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 373| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_38clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 62|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 62| ::std::void_t(); \ 1180| 62| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 62| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 62| constexpr auto _size = _name.size(); \ 1183| 62| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 62| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_38clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 62|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 62| ::std::void_t(); \ 1180| 62| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 62| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 62| constexpr auto _size = _name.size(); \ 1183| 62| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 62| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_39clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 48|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 48| ::std::void_t(); \ 1180| 48| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 48| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 48| constexpr auto _size = _name.size(); \ 1183| 48| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 48| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_39clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 48|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 48| ::std::void_t(); \ 1180| 48| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 48| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 48| constexpr auto _size = _name.size(); \ 1183| 48| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 48| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_40clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 38|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 38| ::std::void_t(); \ 1180| 38| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 38| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 38| constexpr auto _size = _name.size(); \ 1183| 38| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 38| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_40clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 38|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 38| ::std::void_t(); \ 1180| 38| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 38| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 38| constexpr auto _size = _name.size(); \ 1183| 38| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 38| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_41clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 302|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 302| ::std::void_t(); \ 1180| 302| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 302| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 302| constexpr auto _size = _name.size(); \ 1183| 302| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 302| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_41clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 302|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 302| ::std::void_t(); \ 1180| 302| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 302| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 302| constexpr auto _size = _name.size(); \ 1183| 302| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 302| return _nameof; }() _ZN7simdutf6resultC2ENS_10error_codeEm: 87| 76.0k| : error{err}, count{pos} {} _ZNK7simdutf6result6is_errEv: 93| 5.32k| simdutf_really_inline simdutf_constexpr23 bool is_err() const noexcept { 94| 5.32k| return error != error_code::SUCCESS; 95| 5.32k| } _ZN7simdutf11full_resultC2ENS_10error_codeEmm: 111| 3.23k| : error{err}, input_count{pos_in}, output_count{pos_out} {} _ZNK7simdutf11full_resultcvNS_6resultEEv: 117| 3.23k| simdutf_really_inline simdutf_constexpr23 operator result() const noexcept { 118| 3.23k| if (error == error_code::SUCCESS) { ------------------ | Branch (118:9): [True: 2.33k, False: 894] ------------------ 119| 2.33k| return result{error, output_count}; 120| 2.33k| } else { 121| 894| return result{error, input_count}; 122| 894| } 123| 3.23k| } _ZNK7simdutf14implementation4nameEv: 5106| 4| virtual std::string_view name() const noexcept { return _name; } _ZNK7simdutf14implementation25required_instruction_setsEv: 5157| 8| virtual uint32_t required_instruction_sets() const { 5158| 8| return _required_instruction_sets; 5159| 8| } _ZNK7simdutf14implementation24utf16_length_from_latin1Em: 6347| 258| utf16_length_from_latin1(size_t length) const noexcept { 6348| 258| return length; 6349| 258| } _ZNK7simdutf14implementation24latin1_length_from_utf32Em: 6535| 996| latin1_length_from_utf32(size_t length) const noexcept { 6536| 996| return length; 6537| 996| } _ZNK7simdutf14implementation24latin1_length_from_utf16Em: 6572| 1.17k| latin1_length_from_utf16(size_t length) const noexcept { 6573| 1.17k| return length; 6574| 1.17k| } _ZNK7simdutf14implementation24utf32_length_from_latin1Em: 6605| 186| utf32_length_from_latin1(size_t length) const noexcept { 6606| 186| return length; 6607| 186| } _ZN7simdutf14implementationC2EPKcS2_j: 7043| 4| : _name(name), _description(description), 7044| 4| _required_instruction_sets(required_instruction_sets) {} _ZN7simdutf8internal29available_implementation_listC2Ev: 7075| 1| simdutf_really_inline available_implementation_list() {} simdutf.cpp:_ZN7simdutf8internalL30detect_supported_architecturesEv: 236| 8|static inline uint32_t detect_supported_architectures() { 237| 8| uint32_t eax; 238| 8| uint32_t ebx = 0; 239| 8| uint32_t ecx = 0; 240| 8| uint32_t edx = 0; 241| 8| uint32_t host_isa = 0x0; 242| | 243| | // EBX for EAX=0x1 244| 8| eax = 0x1; 245| 8| cpuid(&eax, &ebx, &ecx, &edx); 246| | 247| 8| if (ecx & cpuid_bit::sse42) { ------------------ | Branch (247:7): [True: 8, False: 0] ------------------ 248| 8| host_isa |= instruction_set::SSE42; 249| 8| } 250| | 251| 8| if (ecx & cpuid_bit::pclmulqdq) { ------------------ | Branch (251:7): [True: 8, False: 0] ------------------ 252| 8| host_isa |= instruction_set::PCLMULQDQ; 253| 8| } 254| | 255| 8| if ((ecx & cpuid_bit::osxsave) != cpuid_bit::osxsave) { ------------------ | Branch (255:7): [True: 0, False: 8] ------------------ 256| 0| return host_isa; 257| 0| } 258| | 259| | // xgetbv for checking if the OS saves registers 260| 8| uint64_t xcr0 = xgetbv(); 261| | 262| 8| if ((xcr0 & cpuid_bit::xcr0_bit::avx256_saved) == 0) { ------------------ | Branch (262:7): [True: 0, False: 8] ------------------ 263| 0| return host_isa; 264| 0| } 265| | // ECX for EAX=0x7 266| 8| eax = 0x7; 267| 8| ecx = 0x0; // Sub-leaf = 0 268| 8| cpuid(&eax, &ebx, &ecx, &edx); 269| 8| if (ebx & cpuid_bit::ebx::avx2) { ------------------ | Branch (269:7): [True: 8, False: 0] ------------------ 270| 8| host_isa |= instruction_set::AVX2; 271| 8| } 272| 8| if (ebx & cpuid_bit::ebx::bmi1) { ------------------ | Branch (272:7): [True: 8, False: 0] ------------------ 273| 8| host_isa |= instruction_set::BMI1; 274| 8| } 275| 8| if (ebx & cpuid_bit::ebx::bmi2) { ------------------ | Branch (275:7): [True: 8, False: 0] ------------------ 276| 8| host_isa |= instruction_set::BMI2; 277| 8| } 278| 8| if (!((xcr0 & cpuid_bit::xcr0_bit::avx512_saved) == ------------------ | Branch (278:7): [True: 8, False: 0] ------------------ 279| 8| cpuid_bit::xcr0_bit::avx512_saved)) { 280| 8| return host_isa; 281| 8| } 282| 0| if (ebx & cpuid_bit::ebx::avx512f) { ------------------ | Branch (282:7): [True: 0, False: 0] ------------------ 283| 0| host_isa |= instruction_set::AVX512F; 284| 0| } 285| 0| if (ebx & cpuid_bit::ebx::avx512bw) { ------------------ | Branch (285:7): [True: 0, False: 0] ------------------ 286| 0| host_isa |= instruction_set::AVX512BW; 287| 0| } 288| 0| if (ebx & cpuid_bit::ebx::avx512cd) { ------------------ | Branch (288:7): [True: 0, False: 0] ------------------ 289| 0| host_isa |= instruction_set::AVX512CD; 290| 0| } 291| 0| if (ebx & cpuid_bit::ebx::avx512dq) { ------------------ | Branch (291:7): [True: 0, False: 0] ------------------ 292| 0| host_isa |= instruction_set::AVX512DQ; 293| 0| } 294| 0| if (ebx & cpuid_bit::ebx::avx512vl) { ------------------ | Branch (294:7): [True: 0, False: 0] ------------------ 295| 0| host_isa |= instruction_set::AVX512VL; 296| 0| } 297| 0| if (ecx & cpuid_bit::ecx::avx512vbmi2) { ------------------ | Branch (297:7): [True: 0, False: 0] ------------------ 298| 0| host_isa |= instruction_set::AVX512VBMI2; 299| 0| } 300| 0| if (ecx & cpuid_bit::ecx::avx512vpopcnt) { ------------------ | Branch (300:7): [True: 0, False: 0] ------------------ 301| 0| host_isa |= instruction_set::AVX512VPOPCNTDQ; 302| 0| } 303| 0| return host_isa; 304| 8|} simdutf.cpp:_ZN7simdutf8internalL5cpuidEPjS1_S1_S1_: 202| 16| uint32_t *edx) { 203| | #if defined(_MSC_VER) 204| | int cpu_info[4]; 205| | __cpuidex(cpu_info, *eax, *ecx); 206| | *eax = cpu_info[0]; 207| | *ebx = cpu_info[1]; 208| | *ecx = cpu_info[2]; 209| | *edx = cpu_info[3]; 210| | #elif (defined(HAVE_GCC_GET_CPUID) && defined(USE_GCC_GET_CPUID)) || \ 211| | defined(__FILC__) 212| | uint32_t level = *eax; 213| | __get_cpuid(level, eax, ebx, ecx, edx); 214| | #else 215| 16| uint32_t a = *eax, b, c = *ecx, d; 216| 16| asm volatile("cpuid\n\t" : "+a"(a), "=b"(b), "+c"(c), "=d"(d)); 217| 16| *eax = a; 218| 16| *ebx = b; 219| 16| *ecx = c; 220| 16| *edx = d; 221| 16| #endif 222| 16|} simdutf.cpp:_ZN7simdutf8internalL6xgetbvEv: 224| 8|static inline uint64_t xgetbv() { 225| | #if defined(_MSC_VER) 226| | return _xgetbv(0); 227| | #elif defined(__FILC__) 228| | return zxgetbv(); 229| | #else 230| 8| uint32_t xcr0_lo, xcr0_hi; 231| 8| asm volatile("xgetbv\n\t" : "=a"(xcr0_lo), "=d"(xcr0_hi) : "c"(0)); 232| 8| return xcr0_lo | ((uint64_t)xcr0_hi << 32); 233| 8| #endif 234| 8|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_16latin123utf8_length_from_latin1EPKcm: 10| 604| size_t len) { 11| 604| const uint8_t *c = reinterpret_cast(buf); 12| 604| size_t answer = 0; 13| 4.71k| for (size_t i = 0; i < len; i++) { ------------------ | Branch (13:22): [True: 4.11k, False: 604] ------------------ 14| 4.11k| if ((c[i] >> 7)) { ------------------ | Branch (14:9): [True: 1.16k, False: 2.94k] ------------------ 15| 1.16k| answer++; 16| 1.16k| } 17| 4.11k| } 18| 604| return answer + len; 19| 604|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115latin1_to_utf167convertILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 14| 192| char16_t *utf16_output) { 15| 192| size_t pos = 0; 16| 192| char16_t *start{utf16_output}; 17| | 18| 1.49M| while (pos < len) { ------------------ | Branch (18:10): [True: 1.49M, False: 192] ------------------ 19| 1.49M| uint16_t word = 20| 1.49M| uint8_t(data[pos]); // extend Latin-1 char to 16-bit Unicode code point 21| 1.49M| *utf16_output++ = 22| 1.49M| char16_t(match_system(big_endian) ? word : u16_swap_bytes(word)); ------------------ | Branch (22:18): [True: 1.49M, Folded] ------------------ 23| 1.49M| pos++; 24| 1.49M| } 25| | 26| 192| return utf16_output - start; 27| 192|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115latin1_to_utf167convertILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 14| 252| char16_t *utf16_output) { 15| 252| size_t pos = 0; 16| 252| char16_t *start{utf16_output}; 17| | 18| 3.04M| while (pos < len) { ------------------ | Branch (18:10): [True: 3.04M, False: 252] ------------------ 19| 3.04M| uint16_t word = 20| 3.04M| uint8_t(data[pos]); // extend Latin-1 char to 16-bit Unicode code point 21| 3.04M| *utf16_output++ = 22| 3.04M| char16_t(match_system(big_endian) ? word : u16_swap_bytes(word)); ------------------ | Branch (22:18): [Folded, False: 3.04M] ------------------ 23| 3.04M| pos++; 24| 3.04M| } 25| | 26| 252| return utf16_output - start; 27| 252|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115latin1_to_utf327convertIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mPDi: 14| 304| char32_t *utf32_output) { 15| 304| char32_t *start{utf32_output}; 16| 6.93M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (16:22): [True: 6.93M, False: 304] ------------------ 17| 6.93M| *utf32_output++ = uint8_t(data[i]); 18| 6.93M| } 19| 304| return utf32_output - start; 20| 304|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf87convertEPKcmPc: 66| 1.71k| char *utf8_output) { 67| 1.71k| return convert(reinterpret_cast(buf), len, 68| 1.71k| utf8_output); 69| 1.71k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf87convertIPKhPcQaasr7simdutf6detailE22indexes_into_byte_likeIT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEEmS7_mS8_: 17| 1.71k| OutputPtr utf8_output) { 18| | // const unsigned char *data = reinterpret_cast(buf); 19| 1.71k| size_t pos = 0; 20| 1.71k| size_t utf8_pos = 0; 21| | 22| 8.58M| while (pos < len) { ------------------ | Branch (22:10): [True: 8.57M, False: 1.71k] ------------------ 23| |#if SIMDUTF_CPLUSPLUS23 24| | if !consteval 25| |#endif 26| 8.57M| { 27| | // try to convert the next block of 16 ASCII bytes 28| 8.57M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (28:11): [True: 8.56M, False: 15.3k] ------------------ 29| | // they are ascii 30| 8.56M| uint64_t v1; 31| 8.56M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 32| 8.56M| uint64_t v2; 33| 8.56M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 34| 8.56M| uint64_t v{v1 | 35| 8.56M| v2}; // We are only interested in these bits: 1000 1000 1000 36| | // 1000, so it makes sense to concatenate everything 37| 8.56M| if ((v & 0x8080808080808080) == ------------------ | Branch (37:13): [True: 3.00M, False: 5.55M] ------------------ 38| 8.56M| 0) { // if NONE of these are set, e.g. all of them are zero, then 39| | // everything is ASCII 40| 3.00M| size_t final_pos = pos + 16; 41| 51.1M| while (pos < final_pos) { ------------------ | Branch (41:18): [True: 48.1M, False: 3.00M] ------------------ 42| 48.1M| utf8_output[utf8_pos++] = char(data[pos]); 43| 48.1M| pos++; 44| 48.1M| } 45| 3.00M| continue; 46| 3.00M| } 47| 8.56M| } // if (pos + 16 <= len) 48| 8.57M| } // !consteval scope 49| | 50| 5.57M| unsigned char byte = data[pos]; 51| 5.57M| if ((byte & 0x80) == 0) { // if ASCII ------------------ | Branch (51:9): [True: 2.42M, False: 3.14M] ------------------ 52| | // will generate one UTF-8 bytes 53| 2.42M| utf8_output[utf8_pos++] = char(byte); 54| 2.42M| pos++; 55| 3.14M| } else { 56| | // will generate two UTF-8 bytes 57| 3.14M| utf8_output[utf8_pos++] = char((byte >> 6) | 0b11000000); 58| 3.14M| utf8_output[utf8_pos++] = char((byte & 0b111111) | 0b10000000); 59| 3.14M| pos++; 60| 3.14M| } 61| 5.57M| } // while 62| 1.71k| return utf8_pos; 63| 1.71k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf823utf8_length_from_latin1IPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_m: 152| 302|utf8_length_from_latin1(InputPtr input, size_t length) noexcept { 153| 302| size_t answer = length; 154| 302| size_t i = 0; 155| | 156| |#if SIMDUTF_CPLUSPLUS23 157| | if !consteval 158| |#endif 159| 302| { 160| 302| auto pop = [](uint64_t v) { 161| 302| return (size_t)(((v >> 7) & UINT64_C(0x0101010101010101)) * 162| 302| UINT64_C(0x0101010101010101) >> 163| 302| 56); 164| 302| }; 165| 839k| for (; i + 32 <= length; i += 32) { ------------------ | Branch (165:12): [True: 838k, False: 302] ------------------ 166| 838k| uint64_t v; 167| 838k| memcpy(&v, input + i, 8); 168| 838k| answer += pop(v); 169| 838k| memcpy(&v, input + i + 8, sizeof(v)); 170| 838k| answer += pop(v); 171| 838k| memcpy(&v, input + i + 16, sizeof(v)); 172| 838k| answer += pop(v); 173| 838k| memcpy(&v, input + i + 24, sizeof(v)); 174| 838k| answer += pop(v); 175| 838k| } 176| 592| for (; i + 8 <= length; i += 8) { ------------------ | Branch (176:12): [True: 290, False: 302] ------------------ 177| 290| uint64_t v; 178| 290| memcpy(&v, input + i, sizeof(v)); 179| 290| answer += pop(v); 180| 290| } 181| 302| } // !consteval scope 182| 814| for (; i + 1 <= length; i += 1) { ------------------ | Branch (182:10): [True: 512, False: 302] ------------------ 183| 512| answer += static_cast(input[i]) >> 7; 184| 512| } 185| 302| return answer; 186| 302|} simdutf.cpp:_ZZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf823utf8_length_from_latin1IPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mENKUlmE_clEm: 160| 3.35M| auto pop = [](uint64_t v) { 161| 3.35M| return (size_t)(((v >> 7) & UINT64_C(0x0101010101010101)) * 162| | UINT64_C(0x0101010101010101) >> 163| 3.35M| 56); 164| 3.35M| }; _ZN7simdutf6scalar14u16_swap_bytesEt: 8| 207M|u16_swap_bytes(const uint16_t word) { 9| 207M| return uint16_t((word >> 8) | (word << 8)); 10| 207M|} _ZN7simdutf6scalar5utf1614swap_if_neededILNS_10endiannessE1EEEtt: 27| 91.0M|template constexpr uint16_t swap_if_needed(uint16_t c) { 28| 91.0M| return !match_system(big_endian) ? scalar::u16_swap_bytes(c) : c; ------------------ | Branch (28:10): [True: 91.0M, Folded] ------------------ 29| 91.0M|} _ZN7simdutf6scalar5utf1614swap_if_neededILNS_10endiannessE0EEEtt: 27| 95.2M|template constexpr uint16_t swap_if_needed(uint16_t c) { 28| 95.2M| return !match_system(big_endian) ? scalar::u16_swap_bytes(c) : c; ------------------ | Branch (28:10): [Folded, False: 95.2M] ------------------ 29| 95.2M|} _ZN7simdutf6scalar5utf168validateILNS_10endiannessE0EEEbPKDsm: 22| 3.80k|validate(const char16_t *data, size_t len) noexcept { 23| 3.80k| uint64_t pos = 0; 24| 20.3M| while (pos < len) { ------------------ | Branch (24:10): [True: 20.3M, False: 2.63k] ------------------ 25| 20.3M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 26| 20.3M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (26:9): [True: 16.6k, False: 20.3M] ------------------ 27| 16.6k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 204, False: 16.4k] ------------------ 28| 204| return false; 29| 204| } 30| 16.4k| char16_t diff = char16_t(word - 0xD800); 31| 16.4k| if (diff > 0x3FF) { ------------------ | Branch (31:11): [True: 356, False: 16.0k] ------------------ 32| 356| return false; 33| 356| } 34| 16.0k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (34:28): [Folded, False: 16.0k] ------------------ 35| 16.0k| ? u16_swap_bytes(data[pos + 1]) 36| 16.0k| : data[pos + 1]; 37| 16.0k| char16_t diff2 = char16_t(next_word - 0xDC00); 38| 16.0k| if (diff2 > 0x3FF) { ------------------ | Branch (38:11): [True: 614, False: 15.4k] ------------------ 39| 614| return false; 40| 614| } 41| 15.4k| pos += 2; 42| 20.3M| } else { 43| 20.3M| pos++; 44| 20.3M| } 45| 20.3M| } 46| 2.63k| return true; 47| 3.80k|} _ZN7simdutf6scalar5utf168validateILNS_10endiannessE1EEEbPKDsm: 22| 3.72k|validate(const char16_t *data, size_t len) noexcept { 23| 3.72k| uint64_t pos = 0; 24| 20.9M| while (pos < len) { ------------------ | Branch (24:10): [True: 20.9M, False: 2.56k] ------------------ 25| 20.9M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 26| 20.9M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (26:9): [True: 16.2k, False: 20.9M] ------------------ 27| 16.2k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 195, False: 16.0k] ------------------ 28| 195| return false; 29| 195| } 30| 16.0k| char16_t diff = char16_t(word - 0xD800); 31| 16.0k| if (diff > 0x3FF) { ------------------ | Branch (31:11): [True: 417, False: 15.6k] ------------------ 32| 417| return false; 33| 417| } 34| 15.6k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (34:28): [True: 15.6k, Folded] ------------------ 35| 15.6k| ? u16_swap_bytes(data[pos + 1]) 36| 15.6k| : data[pos + 1]; 37| 15.6k| char16_t diff2 = char16_t(next_word - 0xDC00); 38| 15.6k| if (diff2 > 0x3FF) { ------------------ | Branch (38:11): [True: 550, False: 15.1k] ------------------ 39| 550| return false; 40| 550| } 41| 15.1k| pos += 2; 42| 20.9M| } else { 43| 20.9M| pos++; 44| 20.9M| } 45| 20.9M| } 46| 2.56k| return true; 47| 3.72k|} _ZN7simdutf6scalar5utf1620validate_with_errorsILNS_10endiannessE0EEENS_6resultEPKDsm: 51| 4.03k|validate_with_errors(const char16_t *data, size_t len) noexcept { 52| 4.03k| size_t pos = 0; 53| 20.3M| while (pos < len) { ------------------ | Branch (53:10): [True: 20.3M, False: 2.63k] ------------------ 54| 20.3M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 55| 20.3M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (55:9): [True: 16.9k, False: 20.3M] ------------------ 56| 16.9k| if (pos + 1 >= len) { ------------------ | Branch (56:11): [True: 204, False: 16.7k] ------------------ 57| 204| return result(error_code::SURROGATE, pos); 58| 204| } 59| 16.7k| char16_t diff = char16_t(word - 0xD800); 60| 16.7k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 474, False: 16.2k] ------------------ 61| 474| return result(error_code::SURROGATE, pos); 62| 474| } 63| 16.2k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (63:28): [Folded, False: 16.2k] ------------------ 64| 16.2k| ? u16_swap_bytes(data[pos + 1]) 65| 16.2k| : data[pos + 1]; 66| 16.2k| char16_t diff2 = uint16_t(next_word - 0xDC00); 67| 16.2k| if (diff2 > 0x3FF) { ------------------ | Branch (67:11): [True: 723, False: 15.5k] ------------------ 68| 723| return result(error_code::SURROGATE, pos); 69| 723| } 70| 15.5k| pos += 2; 71| 20.3M| } else { 72| 20.3M| pos++; 73| 20.3M| } 74| 20.3M| } 75| 2.63k| return result(error_code::SUCCESS, pos); 76| 4.03k|} _ZN7simdutf6scalar5utf1620validate_with_errorsILNS_10endiannessE1EEENS_6resultEPKDsm: 51| 3.98k|validate_with_errors(const char16_t *data, size_t len) noexcept { 52| 3.98k| size_t pos = 0; 53| 20.9M| while (pos < len) { ------------------ | Branch (53:10): [True: 20.9M, False: 2.56k] ------------------ 54| 20.9M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 55| 20.9M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (55:9): [True: 16.5k, False: 20.9M] ------------------ 56| 16.5k| if (pos + 1 >= len) { ------------------ | Branch (56:11): [True: 195, False: 16.3k] ------------------ 57| 195| return result(error_code::SURROGATE, pos); 58| 195| } 59| 16.3k| char16_t diff = char16_t(word - 0xD800); 60| 16.3k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 540, False: 15.8k] ------------------ 61| 540| return result(error_code::SURROGATE, pos); 62| 540| } 63| 15.8k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (63:28): [True: 15.8k, Folded] ------------------ 64| 15.8k| ? u16_swap_bytes(data[pos + 1]) 65| 15.8k| : data[pos + 1]; 66| 15.8k| char16_t diff2 = uint16_t(next_word - 0xDC00); 67| 15.8k| if (diff2 > 0x3FF) { ------------------ | Branch (67:11): [True: 684, False: 15.1k] ------------------ 68| 684| return result(error_code::SURROGATE, pos); 69| 684| } 70| 15.1k| pos += 2; 71| 20.9M| } else { 72| 20.9M| pos++; 73| 20.9M| } 74| 20.9M| } 75| 2.56k| return result(error_code::SUCCESS, pos); 76| 3.98k|} _ZN7simdutf6scalar5utf1617count_code_pointsILNS_10endiannessE0EEEmPKDsm: 79| 4.86k|simdutf_constexpr23 size_t count_code_points(const char16_t *p, size_t len) { 80| | // We are not BOM aware. 81| 4.86k| size_t counter{0}; 82| 29.5M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (82:22): [True: 29.5M, False: 4.86k] ------------------ 83| 29.5M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 84| 29.5M| counter += ((word & 0xFC00) != 0xDC00); 85| 29.5M| } 86| 4.86k| return counter; 87| 4.86k|} _ZN7simdutf6scalar5utf1617count_code_pointsILNS_10endiannessE1EEEmPKDsm: 79| 4.81k|simdutf_constexpr23 size_t count_code_points(const char16_t *p, size_t len) { 80| | // We are not BOM aware. 81| 4.81k| size_t counter{0}; 82| 27.5M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (82:22): [True: 27.5M, False: 4.81k] ------------------ 83| 27.5M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 84| 27.5M| counter += ((word & 0xFC00) != 0xDC00); 85| 27.5M| } 86| 4.81k| return counter; 87| 4.81k|} _ZN7simdutf6scalar5utf1622utf8_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 91| 2.26k| size_t len) { 92| | // We are not BOM aware. 93| 2.26k| size_t counter{0}; 94| 16.9M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (94:22): [True: 16.9M, False: 2.26k] ------------------ 95| 16.9M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 96| 16.9M| counter++; // ASCII 97| 16.9M| counter += static_cast( 98| 16.9M| word > 99| 16.9M| 0x7F); // non-ASCII is at least 2 bytes, surrogates are 2*2 == 4 bytes 100| 16.9M| counter += static_cast((word > 0x7FF && word <= 0xD7FF) || ------------------ | Branch (100:37): [True: 3.63M, False: 13.2M] | Branch (100:53): [True: 3.35M, False: 283k] ------------------ 101| 13.5M| (word >= 0xE000)); // three-byte ------------------ | Branch (101:36): [True: 224k, False: 13.3M] ------------------ 102| 16.9M| } 103| 2.26k| return counter; 104| 2.26k|} _ZN7simdutf6scalar5utf1622utf8_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 91| 2.24k| size_t len) { 92| | // We are not BOM aware. 93| 2.24k| size_t counter{0}; 94| 15.1M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (94:22): [True: 15.1M, False: 2.24k] ------------------ 95| 15.1M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 96| 15.1M| counter++; // ASCII 97| 15.1M| counter += static_cast( 98| 15.1M| word > 99| 15.1M| 0x7F); // non-ASCII is at least 2 bytes, surrogates are 2*2 == 4 bytes 100| 15.1M| counter += static_cast((word > 0x7FF && word <= 0xD7FF) || ------------------ | Branch (100:37): [True: 3.49M, False: 11.7M] | Branch (100:53): [True: 3.34M, False: 150k] ------------------ 101| 11.8M| (word >= 0xE000)); // three-byte ------------------ | Branch (101:36): [True: 127k, False: 11.7M] ------------------ 102| 15.1M| } 103| 2.24k| return counter; 104| 2.24k|} _ZN7simdutf6scalar5utf1623utf32_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 108| 400| size_t len) { 109| | // We are not BOM aware. 110| 400| size_t counter{0}; 111| 7.40M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (111:22): [True: 7.40M, False: 400] ------------------ 112| 7.40M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 113| 7.40M| counter += ((word & 0xFC00) != 0xDC00); 114| 7.40M| } 115| 400| return counter; 116| 400|} _ZN7simdutf6scalar5utf1623utf32_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 108| 400| size_t len) { 109| | // We are not BOM aware. 110| 400| size_t counter{0}; 111| 5.82M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (111:22): [True: 5.82M, False: 400] ------------------ 112| 5.82M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 113| 5.82M| counter += ((word & 0xFC00) != 0xDC00); 114| 5.82M| } 115| 400| return counter; 116| 400|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin17convertILNS_10endiannessE0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 272| OutputPtr latin_output) { 18| 272| if (len == 0) { ------------------ | Branch (18:7): [True: 6, False: 266] ------------------ 19| 6| return 0; 20| 6| } 21| 266| size_t pos = 0; 22| 266| const auto latin_output_start = latin_output; 23| 266| uint16_t word = 0; 24| 266| uint16_t too_large = 0; 25| | 26| 567k| while (pos < len) { ------------------ | Branch (26:10): [True: 567k, False: 266] ------------------ 27| 567k| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (27:12): [Folded, False: 567k] ------------------ 28| 567k| too_large |= word; 29| 567k| *latin_output++ = char(word & 0xFF); 30| 567k| pos++; 31| 567k| } 32| 266| if ((too_large & 0xFF00) != 0) { ------------------ | Branch (32:7): [True: 174, False: 92] ------------------ 33| 174| return 0; 34| 174| } 35| | 36| 92| return latin_output - latin_output_start; 37| 266|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin17convertILNS_10endiannessE1EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 244| OutputPtr latin_output) { 18| 244| if (len == 0) { ------------------ | Branch (18:7): [True: 6, False: 238] ------------------ 19| 6| return 0; 20| 6| } 21| 238| size_t pos = 0; 22| 238| const auto latin_output_start = latin_output; 23| 238| uint16_t word = 0; 24| 238| uint16_t too_large = 0; 25| | 26| 1.73M| while (pos < len) { ------------------ | Branch (26:10): [True: 1.73M, False: 238] ------------------ 27| 1.73M| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (27:12): [True: 1.73M, Folded] ------------------ 28| 1.73M| too_large |= word; 29| 1.73M| *latin_output++ = char(word & 0xFF); 30| 1.73M| pos++; 31| 1.73M| } 32| 238| if ((too_large & 0xFF00) != 0) { ------------------ | Branch (32:7): [True: 124, False: 114] ------------------ 33| 124| return 0; 34| 124| } 35| | 36| 114| return latin_output - latin_output_start; 37| 238|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin119convert_with_errorsILNS_10endiannessE0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEENS_6resultES8_mS9_: 45| 466| OutputPtr latin_output) { 46| 466| if (len == 0) { ------------------ | Branch (46:7): [True: 4, False: 462] ------------------ 47| 4| return result(error_code::SUCCESS, 0); 48| 4| } 49| 462| size_t pos = 0; 50| 462| auto start = latin_output; 51| 462| uint16_t word; 52| | 53| 412k| while (pos < len) { ------------------ | Branch (53:10): [True: 412k, False: 184] ------------------ 54| |#if SIMDUTF_CPLUSPLUS23 55| | if !consteval 56| |#endif 57| 412k| { 58| 412k| if (pos + 16 <= len) { // if it is safe to read 32 more bytes, check that ------------------ | Branch (58:11): [True: 410k, False: 1.67k] ------------------ 59| | // they are Latin1 60| 410k| uint64_t v1, v2, v3, v4; 61| 410k| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 62| 410k| ::memcpy(&v2, data + pos + 4, sizeof(uint64_t)); 63| 410k| ::memcpy(&v3, data + pos + 8, sizeof(uint64_t)); 64| 410k| ::memcpy(&v4, data + pos + 12, sizeof(uint64_t)); 65| | 66| | if constexpr (!match_system(big_endian)) { 67| | v1 = (v1 >> 8) | (v1 << (64 - 8)); 68| | } 69| | if constexpr (!match_system(big_endian)) { 70| | v2 = (v2 >> 8) | (v2 << (64 - 8)); 71| | } 72| | if constexpr (!match_system(big_endian)) { 73| | v3 = (v3 >> 8) | (v3 << (64 - 8)); 74| | } 75| | if constexpr (!match_system(big_endian)) { 76| | v4 = (v4 >> 8) | (v4 << (64 - 8)); 77| | } 78| | 79| 410k| if (((v1 | v2 | v3 | v4) & 0xFF00FF00FF00FF00) == 0) { ------------------ | Branch (79:13): [True: 410k, False: 260] ------------------ 80| 410k| size_t final_pos = pos + 16; 81| 6.97M| while (pos < final_pos) { ------------------ | Branch (81:18): [True: 6.56M, False: 410k] ------------------ 82| 6.56M| *latin_output++ = !match_system(big_endian) ------------------ | Branch (82:31): [Folded, False: 6.56M] ------------------ 83| 6.56M| ? char(u16_swap_bytes(data[pos])) 84| 6.56M| : char(data[pos]); 85| 6.56M| pos++; 86| 6.56M| } 87| 410k| continue; 88| 410k| } 89| 410k| } 90| 412k| } 91| | 92| 1.93k| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (92:12): [Folded, False: 1.93k] ------------------ 93| 1.93k| if ((word & 0xFF00) == 0) { ------------------ | Branch (93:9): [True: 1.65k, False: 278] ------------------ 94| 1.65k| *latin_output++ = char(word & 0xFF); 95| 1.65k| pos++; 96| 1.65k| } else { 97| 278| return result(error_code::TOO_LARGE, pos); 98| 278| } 99| 1.93k| } 100| 184| return result(error_code::SUCCESS, latin_output - start); 101| 462|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin119convert_with_errorsILNS_10endiannessE1EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEENS_6resultES8_mS9_: 45| 464| OutputPtr latin_output) { 46| 464| if (len == 0) { ------------------ | Branch (46:7): [True: 4, False: 460] ------------------ 47| 4| return result(error_code::SUCCESS, 0); 48| 4| } 49| 460| size_t pos = 0; 50| 460| auto start = latin_output; 51| 460| uint16_t word; 52| | 53| 524k| while (pos < len) { ------------------ | Branch (53:10): [True: 523k, False: 194] ------------------ 54| |#if SIMDUTF_CPLUSPLUS23 55| | if !consteval 56| |#endif 57| 523k| { 58| 523k| if (pos + 16 <= len) { // if it is safe to read 32 more bytes, check that ------------------ | Branch (58:11): [True: 522k, False: 1.55k] ------------------ 59| | // they are Latin1 60| 522k| uint64_t v1, v2, v3, v4; 61| 522k| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 62| 522k| ::memcpy(&v2, data + pos + 4, sizeof(uint64_t)); 63| 522k| ::memcpy(&v3, data + pos + 8, sizeof(uint64_t)); 64| 522k| ::memcpy(&v4, data + pos + 12, sizeof(uint64_t)); 65| | 66| 522k| if constexpr (!match_system(big_endian)) { 67| 522k| v1 = (v1 >> 8) | (v1 << (64 - 8)); 68| 522k| } 69| 522k| if constexpr (!match_system(big_endian)) { 70| 522k| v2 = (v2 >> 8) | (v2 << (64 - 8)); 71| 522k| } 72| 522k| if constexpr (!match_system(big_endian)) { 73| 522k| v3 = (v3 >> 8) | (v3 << (64 - 8)); 74| 522k| } 75| 522k| if constexpr (!match_system(big_endian)) { 76| 522k| v4 = (v4 >> 8) | (v4 << (64 - 8)); 77| 522k| } 78| | 79| 522k| if (((v1 | v2 | v3 | v4) & 0xFF00FF00FF00FF00) == 0) { ------------------ | Branch (79:13): [True: 522k, False: 326] ------------------ 80| 522k| size_t final_pos = pos + 16; 81| 8.87M| while (pos < final_pos) { ------------------ | Branch (81:18): [True: 8.35M, False: 522k] ------------------ 82| 8.35M| *latin_output++ = !match_system(big_endian) ------------------ | Branch (82:31): [True: 8.35M, Folded] ------------------ 83| 8.35M| ? char(u16_swap_bytes(data[pos])) 84| 8.35M| : char(data[pos]); 85| 8.35M| pos++; 86| 8.35M| } 87| 522k| continue; 88| 522k| } 89| 522k| } 90| 523k| } 91| | 92| 1.87k| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (92:12): [True: 1.87k, Folded] ------------------ 93| 1.87k| if ((word & 0xFF00) == 0) { ------------------ | Branch (93:9): [True: 1.61k, False: 266] ------------------ 94| 1.61k| *latin_output++ = char(word & 0xFF); 95| 1.61k| pos++; 96| 1.61k| } else { 97| 266| return result(error_code::TOO_LARGE, pos); 98| 266| } 99| 1.87k| } 100| 194| return result(error_code::SUCCESS, latin_output - start); 101| 460|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf327convertILNS_10endiannessE0EEEmPKDsmPDi: 11| 840| char32_t *utf32_output) { 12| 840| size_t pos = 0; 13| 840| char32_t *start{utf32_output}; 14| 2.51M| while (pos < len) { ------------------ | Branch (14:10): [True: 2.51M, False: 494] ------------------ 15| 2.51M| uint16_t word = 16| 2.51M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [Folded, False: 2.51M] ------------------ 17| 2.51M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 2.51M, False: 3.66k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 2.51M| *utf32_output++ = char32_t(word); 20| 2.51M| pos++; 21| 2.51M| } else { 22| | // must be a surrogate pair 23| 3.66k| uint16_t diff = uint16_t(word - 0xD800); 24| 3.66k| if (diff > 0x3FF) { ------------------ | Branch (24:11): [True: 126, False: 3.54k] ------------------ 25| 126| return 0; 26| 126| } 27| 3.54k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 72, False: 3.47k] ------------------ 28| 72| return 0; 29| 72| } // minimal bound checking 30| 3.47k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (30:28): [Folded, False: 3.47k] ------------------ 31| 3.47k| ? u16_swap_bytes(data[pos + 1]) 32| 3.47k| : data[pos + 1]; 33| 3.47k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 34| 3.47k| if (diff2 > 0x3FF) { ------------------ | Branch (34:11): [True: 148, False: 3.32k] ------------------ 35| 148| return 0; 36| 148| } 37| 3.32k| uint32_t value = (diff << 10) + diff2 + 0x10000; 38| 3.32k| *utf32_output++ = char32_t(value); 39| 3.32k| pos += 2; 40| 3.32k| } 41| 2.51M| } 42| 494| return utf32_output - start; 43| 840|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf327convertILNS_10endiannessE1EEEmPKDsmPDi: 11| 820| char32_t *utf32_output) { 12| 820| size_t pos = 0; 13| 820| char32_t *start{utf32_output}; 14| 1.59M| while (pos < len) { ------------------ | Branch (14:10): [True: 1.59M, False: 500] ------------------ 15| 1.59M| uint16_t word = 16| 1.59M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [True: 1.59M, Folded] ------------------ 17| 1.59M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 1.59M, False: 1.82k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 1.59M| *utf32_output++ = char32_t(word); 20| 1.59M| pos++; 21| 1.59M| } else { 22| | // must be a surrogate pair 23| 1.82k| uint16_t diff = uint16_t(word - 0xD800); 24| 1.82k| if (diff > 0x3FF) { ------------------ | Branch (24:11): [True: 86, False: 1.74k] ------------------ 25| 86| return 0; 26| 86| } 27| 1.74k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 114, False: 1.62k] ------------------ 28| 114| return 0; 29| 114| } // minimal bound checking 30| 1.62k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (30:28): [True: 1.62k, Folded] ------------------ 31| 1.62k| ? u16_swap_bytes(data[pos + 1]) 32| 1.62k| : data[pos + 1]; 33| 1.62k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 34| 1.62k| if (diff2 > 0x3FF) { ------------------ | Branch (34:11): [True: 120, False: 1.50k] ------------------ 35| 120| return 0; 36| 120| } 37| 1.50k| uint32_t value = (diff << 10) + diff2 + 0x10000; 38| 1.50k| *utf32_output++ = char32_t(value); 39| 1.50k| pos += 2; 40| 1.50k| } 41| 1.59M| } 42| 500| return utf32_output - start; 43| 820|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3219convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKDsmPDi: 47| 690| char32_t *utf32_output) { 48| 690| size_t pos = 0; 49| 690| char32_t *start{utf32_output}; 50| 2.54M| while (pos < len) { ------------------ | Branch (50:10): [True: 2.54M, False: 378] ------------------ 51| 2.54M| uint16_t word = 52| 2.54M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (52:9): [Folded, False: 2.54M] ------------------ 53| 2.54M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (53:9): [True: 2.53M, False: 9.83k] ------------------ 54| | // No surrogate pair, extend 16-bit word to 32-bit word 55| 2.53M| *utf32_output++ = char32_t(word); 56| 2.53M| pos++; 57| 2.53M| } else { 58| | // must be a surrogate pair 59| 9.83k| uint16_t diff = uint16_t(word - 0xD800); 60| 9.83k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 78, False: 9.75k] ------------------ 61| 78| return result(error_code::SURROGATE, pos); 62| 78| } 63| 9.75k| if (pos + 1 >= len) { ------------------ | Branch (63:11): [True: 84, False: 9.66k] ------------------ 64| 84| return result(error_code::SURROGATE, pos); 65| 84| } // minimal bound checking 66| 9.66k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (66:28): [Folded, False: 9.66k] ------------------ 67| 9.66k| ? u16_swap_bytes(data[pos + 1]) 68| 9.66k| : data[pos + 1]; 69| 9.66k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 70| 9.66k| if (diff2 > 0x3FF) { ------------------ | Branch (70:11): [True: 150, False: 9.51k] ------------------ 71| 150| return result(error_code::SURROGATE, pos); 72| 150| } 73| 9.51k| uint32_t value = (diff << 10) + diff2 + 0x10000; 74| 9.51k| *utf32_output++ = char32_t(value); 75| 9.51k| pos += 2; 76| 9.51k| } 77| 2.54M| } 78| 378| return result(error_code::SUCCESS, utf32_output - start); 79| 690|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3219convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKDsmPDi: 47| 758| char32_t *utf32_output) { 48| 758| size_t pos = 0; 49| 758| char32_t *start{utf32_output}; 50| 2.40M| while (pos < len) { ------------------ | Branch (50:10): [True: 2.40M, False: 414] ------------------ 51| 2.40M| uint16_t word = 52| 2.40M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (52:9): [True: 2.40M, Folded] ------------------ 53| 2.40M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (53:9): [True: 2.39M, False: 10.3k] ------------------ 54| | // No surrogate pair, extend 16-bit word to 32-bit word 55| 2.39M| *utf32_output++ = char32_t(word); 56| 2.39M| pos++; 57| 2.39M| } else { 58| | // must be a surrogate pair 59| 10.3k| uint16_t diff = uint16_t(word - 0xD800); 60| 10.3k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 116, False: 10.2k] ------------------ 61| 116| return result(error_code::SURROGATE, pos); 62| 116| } 63| 10.2k| if (pos + 1 >= len) { ------------------ | Branch (63:11): [True: 78, False: 10.1k] ------------------ 64| 78| return result(error_code::SURROGATE, pos); 65| 78| } // minimal bound checking 66| 10.1k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (66:28): [True: 10.1k, Folded] ------------------ 67| 10.1k| ? u16_swap_bytes(data[pos + 1]) 68| 10.1k| : data[pos + 1]; 69| 10.1k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 70| 10.1k| if (diff2 > 0x3FF) { ------------------ | Branch (70:11): [True: 150, False: 9.99k] ------------------ 71| 150| return result(error_code::SURROGATE, pos); 72| 150| } 73| 9.99k| uint32_t value = (diff << 10) + diff2 + 0x10000; 74| 9.99k| *utf32_output++ = char32_t(value); 75| 9.99k| pos += 2; 76| 9.99k| } 77| 2.40M| } 78| 414| return result(error_code::SUCCESS, utf32_output - start); 79| 758|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3213convert_validILNS_10endiannessE0EEEmPKDsmPDi: 11| 128| char32_t *utf32_output) { 12| 128| size_t pos = 0; 13| 128| char32_t *start{utf32_output}; 14| 1.85M| while (pos < len) { ------------------ | Branch (14:10): [True: 1.85M, False: 128] ------------------ 15| 1.85M| uint16_t word = 16| 1.85M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [Folded, False: 1.85M] ------------------ 17| 1.85M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 1.85M, False: 3.45k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 1.85M| *utf32_output++ = char32_t(word); 20| 1.85M| pos++; 21| 1.85M| } else { 22| | // must be a surrogate pair 23| 3.45k| uint16_t diff = uint16_t(word - 0xD800); 24| 3.45k| if (pos + 1 >= len) { ------------------ | Branch (24:11): [True: 0, False: 3.45k] ------------------ 25| 0| return 0; 26| 0| } // minimal bound checking 27| 3.45k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (27:28): [Folded, False: 3.45k] ------------------ 28| 3.45k| ? u16_swap_bytes(data[pos + 1]) 29| 3.45k| : data[pos + 1]; 30| 3.45k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 31| 3.45k| uint32_t value = (diff << 10) + diff2 + 0x10000; 32| 3.45k| *utf32_output++ = char32_t(value); 33| 3.45k| pos += 2; 34| 3.45k| } 35| 1.85M| } 36| 128| return utf32_output - start; 37| 128|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3213convert_validILNS_10endiannessE1EEEmPKDsmPDi: 11| 142| char32_t *utf32_output) { 12| 142| size_t pos = 0; 13| 142| char32_t *start{utf32_output}; 14| 4.30M| while (pos < len) { ------------------ | Branch (14:10): [True: 4.30M, False: 142] ------------------ 15| 4.30M| uint16_t word = 16| 4.30M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [True: 4.30M, Folded] ------------------ 17| 4.30M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 4.30M, False: 944] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 4.30M| *utf32_output++ = char32_t(word); 20| 4.30M| pos++; 21| 4.30M| } else { 22| | // must be a surrogate pair 23| 944| uint16_t diff = uint16_t(word - 0xD800); 24| 944| if (pos + 1 >= len) { ------------------ | Branch (24:11): [True: 0, False: 944] ------------------ 25| 0| return 0; 26| 0| } // minimal bound checking 27| 944| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (27:28): [True: 944, Folded] ------------------ 28| 944| ? u16_swap_bytes(data[pos + 1]) 29| 944| : data[pos + 1]; 30| 944| uint16_t diff2 = uint16_t(next_word - 0xDC00); 31| 944| uint32_t value = (diff << 10) + diff2 + 0x10000; 32| 944| *utf32_output++ = char32_t(value); 33| 944| pos += 2; 34| 944| } 35| 4.30M| } 36| 142| return utf32_output - start; 37| 142|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf87convertILNS_10endiannessE0EPKDsPcQsr7simdutf6detailE18indexes_into_utf16IT0_EEEmS8_mT1_: 17| 1.88k| OutputPtr utf8_output) { 18| 1.88k| size_t pos = 0; 19| 1.88k| const auto start = utf8_output; 20| 3.43M| while (pos < len) { ------------------ | Branch (20:10): [True: 3.43M, False: 1.51k] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 3.43M| { 25| | // try to convert the next block of 8 bytes 26| 3.43M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 3.43M, False: 3.43k] ------------------ 27| | // they are ascii 28| 3.43M| uint64_t v; 29| 3.43M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| | if constexpr (!match_system(big_endian)) { 31| | v = (v >> 8) | (v << (64 - 8)); 32| | } 33| 3.43M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.54M, False: 1.88M] ------------------ 34| 1.54M| size_t final_pos = pos + 4; 35| 7.74M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 6.19M, False: 1.54M] ------------------ 36| 6.19M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [Folded, False: 6.19M] ------------------ 37| 6.19M| ? char(u16_swap_bytes(data[pos])) 38| 6.19M| : char(data[pos]); 39| 6.19M| pos++; 40| 6.19M| } 41| 1.54M| continue; 42| 1.54M| } 43| 3.43M| } 44| 3.43M| } 45| 1.88M| uint16_t word = 46| 1.88M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (46:9): [Folded, False: 1.88M] ------------------ 47| 1.88M| if ((word & 0xFF80) == 0) { ------------------ | Branch (47:9): [True: 162k, False: 1.72M] ------------------ 48| | // will generate one UTF-8 bytes 49| 162k| *utf8_output++ = char(word); 50| 162k| pos++; 51| 1.72M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (51:16): [True: 1.41M, False: 308k] ------------------ 52| | // will generate two UTF-8 bytes 53| | // we have 0b110XXXXX 0b10XXXXXX 54| 1.41M| *utf8_output++ = char((word >> 6) | 0b11000000); 55| 1.41M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 56| 1.41M| pos++; 57| 1.41M| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (57:16): [True: 302k, False: 6.16k] ------------------ 58| | // will generate three UTF-8 bytes 59| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 60| 302k| *utf8_output++ = char((word >> 12) | 0b11100000); 61| 302k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 62| 302k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 63| 302k| pos++; 64| 302k| } else { 65| | // must be a surrogate pair 66| 6.16k| if (pos + 1 >= len) { ------------------ | Branch (66:11): [True: 66, False: 6.10k] ------------------ 67| 66| return 0; 68| 66| } 69| 6.10k| uint16_t diff = uint16_t(word - 0xD800); 70| 6.10k| if (diff > 0x3FF) { ------------------ | Branch (70:11): [True: 104, False: 5.99k] ------------------ 71| 104| return 0; 72| 104| } 73| 5.99k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (73:28): [Folded, False: 5.99k] ------------------ 74| 5.99k| ? u16_swap_bytes(data[pos + 1]) 75| 5.99k| : data[pos + 1]; 76| 5.99k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 77| 5.99k| if (diff2 > 0x3FF) { ------------------ | Branch (77:11): [True: 200, False: 5.79k] ------------------ 78| 200| return 0; 79| 200| } 80| 5.79k| uint32_t value = (diff << 10) + diff2 + 0x10000; 81| | // will generate four UTF-8 bytes 82| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 83| 5.79k| *utf8_output++ = char((value >> 18) | 0b11110000); 84| 5.79k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 85| 5.79k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 86| 5.79k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 87| 5.79k| pos += 2; 88| 5.79k| } 89| 1.88M| } 90| 1.51k| return utf8_output - start; 91| 1.88k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf87convertILNS_10endiannessE1EPKDsPcQsr7simdutf6detailE18indexes_into_utf16IT0_EEEmS8_mT1_: 17| 1.73k| OutputPtr utf8_output) { 18| 1.73k| size_t pos = 0; 19| 1.73k| const auto start = utf8_output; 20| 3.01M| while (pos < len) { ------------------ | Branch (20:10): [True: 3.01M, False: 1.45k] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 3.01M| { 25| | // try to convert the next block of 8 bytes 26| 3.01M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 3.00M, False: 3.20k] ------------------ 27| | // they are ascii 28| 3.00M| uint64_t v; 29| 3.00M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| 3.00M| if constexpr (!match_system(big_endian)) { 31| 3.00M| v = (v >> 8) | (v << (64 - 8)); 32| 3.00M| } 33| 3.00M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.10M, False: 1.90M] ------------------ 34| 1.10M| size_t final_pos = pos + 4; 35| 5.50M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 4.40M, False: 1.10M] ------------------ 36| 4.40M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [True: 4.40M, Folded] ------------------ 37| 4.40M| ? char(u16_swap_bytes(data[pos])) 38| 4.40M| : char(data[pos]); 39| 4.40M| pos++; 40| 4.40M| } 41| 1.10M| continue; 42| 1.10M| } 43| 3.00M| } 44| 3.01M| } 45| 1.91M| uint16_t word = 46| 1.91M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (46:9): [True: 1.91M, Folded] ------------------ 47| 1.91M| if ((word & 0xFF80) == 0) { ------------------ | Branch (47:9): [True: 80.0k, False: 1.83M] ------------------ 48| | // will generate one UTF-8 bytes 49| 80.0k| *utf8_output++ = char(word); 50| 80.0k| pos++; 51| 1.83M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (51:16): [True: 28.8k, False: 1.80M] ------------------ 52| | // will generate two UTF-8 bytes 53| | // we have 0b110XXXXX 0b10XXXXXX 54| 28.8k| *utf8_output++ = char((word >> 6) | 0b11000000); 55| 28.8k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 56| 28.8k| pos++; 57| 1.80M| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (57:16): [True: 1.79M, False: 3.05k] ------------------ 58| | // will generate three UTF-8 bytes 59| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 60| 1.79M| *utf8_output++ = char((word >> 12) | 0b11100000); 61| 1.79M| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 62| 1.79M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 63| 1.79M| pos++; 64| 1.79M| } else { 65| | // must be a surrogate pair 66| 3.05k| if (pos + 1 >= len) { ------------------ | Branch (66:11): [True: 30, False: 3.02k] ------------------ 67| 30| return 0; 68| 30| } 69| 3.02k| uint16_t diff = uint16_t(word - 0xD800); 70| 3.02k| if (diff > 0x3FF) { ------------------ | Branch (70:11): [True: 126, False: 2.89k] ------------------ 71| 126| return 0; 72| 126| } 73| 2.89k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (73:28): [True: 2.89k, Folded] ------------------ 74| 2.89k| ? u16_swap_bytes(data[pos + 1]) 75| 2.89k| : data[pos + 1]; 76| 2.89k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 77| 2.89k| if (diff2 > 0x3FF) { ------------------ | Branch (77:11): [True: 128, False: 2.77k] ------------------ 78| 128| return 0; 79| 128| } 80| 2.77k| uint32_t value = (diff << 10) + diff2 + 0x10000; 81| | // will generate four UTF-8 bytes 82| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 83| 2.77k| *utf8_output++ = char((value >> 18) | 0b11110000); 84| 2.77k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 85| 2.77k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 86| 2.77k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 87| 2.77k| pos += 2; 88| 2.77k| } 89| 1.91M| } 90| 1.45k| return utf8_output - start; 91| 1.73k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf819convert_with_errorsILNS_10endiannessE0ELb0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT1_Esr7simdutf6detailE26index_assignable_from_charIT2_EEENS_11full_resultES8_mS9_m: 101| 1.61k| size_t utf8_len = 0) { 102| 1.61k| if (check_output && utf8_len == 0) { ------------------ | Branch (102:7): [Folded, False: 1.61k] | Branch (102:23): [True: 0, False: 0] ------------------ 103| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, 0, 0); 104| 0| } 105| | 106| 1.61k| size_t pos = 0; 107| 1.61k| auto start = utf8_output; 108| 1.61k| auto end = utf8_output + utf8_len; 109| | 110| 6.34M| while (pos < len) { ------------------ | Branch (110:10): [True: 6.34M, False: 1.19k] ------------------ 111| |#if SIMDUTF_CPLUSPLUS23 112| | if !consteval 113| |#endif 114| 6.34M| { 115| | // try to convert the next block of 8 bytes 116| 6.34M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (116:11): [True: 6.34M, False: 2.70k] ------------------ 117| | // they are ascii 118| 6.34M| uint64_t v; 119| 6.34M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 120| | if constexpr (!match_system(big_endian)) 121| | v = (v >> 8) | (v << (64 - 8)); 122| 6.34M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (122:13): [True: 1.33M, False: 5.00M] ------------------ 123| 1.33M| size_t final_pos = pos + 4; 124| 6.67M| while (pos < final_pos) { ------------------ | Branch (124:18): [True: 5.34M, False: 1.33M] ------------------ 125| 5.34M| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (125:17): [Folded, False: 5.34M] | Branch (125:33): [True: 0, False: 0] ------------------ 126| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, pos, 127| 0| utf8_output - start); 128| 0| } 129| 5.34M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (129:30): [Folded, False: 5.34M] ------------------ 130| 5.34M| ? char(u16_swap_bytes(data[pos])) 131| 5.34M| : char(data[pos]); 132| 5.34M| pos++; 133| 5.34M| } 134| 1.33M| continue; 135| 1.33M| } 136| 6.34M| } 137| 6.34M| } 138| | 139| 5.00M| uint16_t word = 140| 5.00M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (140:9): [Folded, False: 5.00M] ------------------ 141| 5.00M| if ((word & 0xFF80) == 0) { ------------------ | Branch (141:9): [True: 208k, False: 4.80M] ------------------ 142| | // will generate one UTF-8 bytes 143| 208k| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (143:11): [Folded, False: 208k] | Branch (143:27): [True: 0, False: 0] ------------------ 144| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, pos, 145| 0| utf8_output - start); 146| 0| } 147| 208k| *utf8_output++ = char(word); 148| 208k| pos++; 149| 4.80M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (149:16): [True: 967k, False: 3.83M] ------------------ 150| | // will generate two UTF-8 bytes 151| | // we have 0b110XXXXX 0b10XXXXXX 152| 967k| if (check_output && size_t(end - utf8_output) < 2) { ------------------ | Branch (152:11): [Folded, False: 967k] | Branch (152:27): [True: 0, False: 0] ------------------ 153| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, pos, 154| 0| utf8_output - start); 155| 0| } 156| 967k| *utf8_output++ = char((word >> 6) | 0b11000000); 157| 967k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 158| 967k| pos++; 159| | 160| 3.83M| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (160:16): [True: 3.83M, False: 3.71k] ------------------ 161| | // will generate three UTF-8 bytes 162| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 163| 3.83M| if (check_output && size_t(end - utf8_output) < 3) { ------------------ | Branch (163:11): [Folded, False: 3.83M] | Branch (163:27): [True: 0, False: 0] ------------------ 164| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, pos, 165| 0| utf8_output - start); 166| 0| } 167| 3.83M| *utf8_output++ = char((word >> 12) | 0b11100000); 168| 3.83M| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 169| 3.83M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 170| 3.83M| pos++; 171| 3.83M| } else { 172| | 173| 3.71k| if (check_output && size_t(end - utf8_output) < 4) { ------------------ | Branch (173:11): [Folded, False: 3.71k] | Branch (173:27): [True: 0, False: 0] ------------------ 174| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, pos, 175| 0| utf8_output - start); 176| 0| } 177| | // must be a surrogate pair 178| 3.71k| if (pos + 1 >= len) { ------------------ | Branch (178:11): [True: 36, False: 3.67k] ------------------ 179| 36| return full_result(error_code::SURROGATE, pos, utf8_output - start); 180| 36| } 181| 3.67k| uint16_t diff = uint16_t(word - 0xD800); 182| 3.67k| if (diff > 0x3FF) { ------------------ | Branch (182:11): [True: 126, False: 3.55k] ------------------ 183| 126| return full_result(error_code::SURROGATE, pos, utf8_output - start); 184| 126| } 185| 3.55k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (185:28): [Folded, False: 3.55k] ------------------ 186| 3.55k| ? u16_swap_bytes(data[pos + 1]) 187| 3.55k| : data[pos + 1]; 188| 3.55k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 189| 3.55k| if (diff2 > 0x3FF) { ------------------ | Branch (189:11): [True: 254, False: 3.29k] ------------------ 190| 254| return full_result(error_code::SURROGATE, pos, utf8_output - start); 191| 254| } 192| 3.29k| uint32_t value = (diff << 10) + diff2 + 0x10000; 193| | // will generate four UTF-8 bytes 194| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 195| 3.29k| *utf8_output++ = char((value >> 18) | 0b11110000); 196| 3.29k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 197| 3.29k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 198| 3.29k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 199| 3.29k| pos += 2; 200| 3.29k| } 201| 5.00M| } 202| 1.19k| return full_result(error_code::SUCCESS, pos, utf8_output - start); 203| 1.61k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf819convert_with_errorsILNS_10endiannessE1ELb0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT1_Esr7simdutf6detailE26index_assignable_from_charIT2_EEENS_11full_resultES8_mS9_m: 101| 1.61k| size_t utf8_len = 0) { 102| 1.61k| if (check_output && utf8_len == 0) { ------------------ | Branch (102:7): [Folded, False: 1.61k] | Branch (102:23): [True: 0, False: 0] ------------------ 103| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, 0, 0); 104| 0| } 105| | 106| 1.61k| size_t pos = 0; 107| 1.61k| auto start = utf8_output; 108| 1.61k| auto end = utf8_output + utf8_len; 109| | 110| 2.77M| while (pos < len) { ------------------ | Branch (110:10): [True: 2.77M, False: 1.13k] ------------------ 111| |#if SIMDUTF_CPLUSPLUS23 112| | if !consteval 113| |#endif 114| 2.77M| { 115| | // try to convert the next block of 8 bytes 116| 2.77M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (116:11): [True: 2.77M, False: 2.46k] ------------------ 117| | // they are ascii 118| 2.77M| uint64_t v; 119| 2.77M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 120| | if constexpr (!match_system(big_endian)) 121| 2.77M| v = (v >> 8) | (v << (64 - 8)); 122| 2.77M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (122:13): [True: 1.66M, False: 1.10M] ------------------ 123| 1.66M| size_t final_pos = pos + 4; 124| 8.34M| while (pos < final_pos) { ------------------ | Branch (124:18): [True: 6.67M, False: 1.66M] ------------------ 125| 6.67M| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (125:17): [Folded, False: 6.67M] | Branch (125:33): [True: 0, False: 0] ------------------ 126| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, pos, 127| 0| utf8_output - start); 128| 0| } 129| 6.67M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (129:30): [True: 6.67M, Folded] ------------------ 130| 6.67M| ? char(u16_swap_bytes(data[pos])) 131| 6.67M| : char(data[pos]); 132| 6.67M| pos++; 133| 6.67M| } 134| 1.66M| continue; 135| 1.66M| } 136| 2.77M| } 137| 2.77M| } 138| | 139| 1.10M| uint16_t word = 140| 1.10M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (140:9): [True: 1.10M, Folded] ------------------ 141| 1.10M| if ((word & 0xFF80) == 0) { ------------------ | Branch (141:9): [True: 289k, False: 816k] ------------------ 142| | // will generate one UTF-8 bytes 143| 289k| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (143:11): [Folded, False: 289k] | Branch (143:27): [True: 0, False: 0] ------------------ 144| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, pos, 145| 0| utf8_output - start); 146| 0| } 147| 289k| *utf8_output++ = char(word); 148| 289k| pos++; 149| 816k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (149:16): [True: 127k, False: 689k] ------------------ 150| | // will generate two UTF-8 bytes 151| | // we have 0b110XXXXX 0b10XXXXXX 152| 127k| if (check_output && size_t(end - utf8_output) < 2) { ------------------ | Branch (152:11): [Folded, False: 127k] | Branch (152:27): [True: 0, False: 0] ------------------ 153| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, pos, 154| 0| utf8_output - start); 155| 0| } 156| 127k| *utf8_output++ = char((word >> 6) | 0b11000000); 157| 127k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 158| 127k| pos++; 159| | 160| 689k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (160:16): [True: 679k, False: 9.39k] ------------------ 161| | // will generate three UTF-8 bytes 162| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 163| 679k| if (check_output && size_t(end - utf8_output) < 3) { ------------------ | Branch (163:11): [Folded, False: 679k] | Branch (163:27): [True: 0, False: 0] ------------------ 164| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, pos, 165| 0| utf8_output - start); 166| 0| } 167| 679k| *utf8_output++ = char((word >> 12) | 0b11100000); 168| 679k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 169| 679k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 170| 679k| pos++; 171| 679k| } else { 172| | 173| 9.39k| if (check_output && size_t(end - utf8_output) < 4) { ------------------ | Branch (173:11): [Folded, False: 9.39k] | Branch (173:27): [True: 0, False: 0] ------------------ 174| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, pos, 175| 0| utf8_output - start); 176| 0| } 177| | // must be a surrogate pair 178| 9.39k| if (pos + 1 >= len) { ------------------ | Branch (178:11): [True: 48, False: 9.34k] ------------------ 179| 48| return full_result(error_code::SURROGATE, pos, utf8_output - start); 180| 48| } 181| 9.34k| uint16_t diff = uint16_t(word - 0xD800); 182| 9.34k| if (diff > 0x3FF) { ------------------ | Branch (182:11): [True: 176, False: 9.16k] ------------------ 183| 176| return full_result(error_code::SURROGATE, pos, utf8_output - start); 184| 176| } 185| 9.16k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (185:28): [True: 9.16k, Folded] ------------------ 186| 9.16k| ? u16_swap_bytes(data[pos + 1]) 187| 9.16k| : data[pos + 1]; 188| 9.16k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 189| 9.16k| if (diff2 > 0x3FF) { ------------------ | Branch (189:11): [True: 254, False: 8.91k] ------------------ 190| 254| return full_result(error_code::SURROGATE, pos, utf8_output - start); 191| 254| } 192| 8.91k| uint32_t value = (diff << 10) + diff2 + 0x10000; 193| | // will generate four UTF-8 bytes 194| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 195| 8.91k| *utf8_output++ = char((value >> 18) | 0b11110000); 196| 8.91k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 197| 8.91k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 198| 8.91k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 199| 8.91k| pos += 2; 200| 8.91k| } 201| 1.10M| } 202| 1.13k| return full_result(error_code::SUCCESS, pos, utf8_output - start); 203| 1.61k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf813convert_validILNS_10endiannessE0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 406| OutputPtr utf8_output) { 18| 406| size_t pos = 0; 19| 406| auto start = utf8_output; 20| 2.03M| while (pos < len) { ------------------ | Branch (20:10): [True: 2.03M, False: 406] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 2.03M| { 25| | // try to convert the next block of 4 ASCII characters 26| 2.03M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 2.03M, False: 842] ------------------ 27| | // they are ascii 28| 2.03M| uint64_t v; 29| 2.03M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| | if constexpr (!match_system(big_endian)) { 31| | v = (v >> 8) | (v << (64 - 8)); 32| | } 33| 2.03M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.54M, False: 489k] ------------------ 34| 1.54M| size_t final_pos = pos + 4; 35| 7.74M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 6.19M, False: 1.54M] ------------------ 36| 6.19M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [Folded, False: 6.19M] ------------------ 37| 6.19M| ? char(u16_swap_bytes(data[pos])) 38| 6.19M| : char(data[pos]); 39| 6.19M| pos++; 40| 6.19M| } 41| 1.54M| continue; 42| 1.54M| } 43| 2.03M| } 44| 2.03M| } 45| | 46| 490k| uint16_t word = 47| 490k| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (47:9): [Folded, False: 490k] ------------------ 48| 490k| if ((word & 0xFF80) == 0) { ------------------ | Branch (48:9): [True: 261k, False: 228k] ------------------ 49| | // will generate one UTF-8 bytes 50| 261k| *utf8_output++ = char(word); 51| 261k| pos++; 52| 261k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (52:16): [True: 77.4k, False: 151k] ------------------ 53| | // will generate two UTF-8 bytes 54| | // we have 0b110XXXXX 0b10XXXXXX 55| 77.4k| *utf8_output++ = char((word >> 6) | 0b11000000); 56| 77.4k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 57| 77.4k| pos++; 58| 151k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (58:16): [True: 148k, False: 2.60k] ------------------ 59| | // will generate three UTF-8 bytes 60| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 61| 148k| *utf8_output++ = char((word >> 12) | 0b11100000); 62| 148k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 63| 148k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 64| 148k| pos++; 65| 148k| } else { 66| | // must be a surrogate pair 67| 2.60k| uint16_t diff = uint16_t(word - 0xD800); 68| 2.60k| if (pos + 1 >= len) { ------------------ | Branch (68:11): [True: 0, False: 2.60k] ------------------ 69| 0| return 0; 70| 0| } // minimal bound checking 71| 2.60k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (71:28): [Folded, False: 2.60k] ------------------ 72| 2.60k| ? u16_swap_bytes(data[pos + 1]) 73| 2.60k| : data[pos + 1]; 74| 2.60k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 75| 2.60k| uint32_t value = (diff << 10) + diff2 + 0x10000; 76| | // will generate four UTF-8 bytes 77| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 78| 2.60k| *utf8_output++ = char((value >> 18) | 0b11110000); 79| 2.60k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 80| 2.60k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 81| 2.60k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 82| 2.60k| pos += 2; 83| 2.60k| } 84| 490k| } 85| 406| return utf8_output - start; 86| 406|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf813convert_validILNS_10endiannessE1EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 452| OutputPtr utf8_output) { 18| 452| size_t pos = 0; 19| 452| auto start = utf8_output; 20| 3.13M| while (pos < len) { ------------------ | Branch (20:10): [True: 3.13M, False: 452] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 3.13M| { 25| | // try to convert the next block of 4 ASCII characters 26| 3.13M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 3.13M, False: 940] ------------------ 27| | // they are ascii 28| 3.13M| uint64_t v; 29| 3.13M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| 3.13M| if constexpr (!match_system(big_endian)) { 31| 3.13M| v = (v >> 8) | (v << (64 - 8)); 32| 3.13M| } 33| 3.13M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.95M, False: 1.17M] ------------------ 34| 1.95M| size_t final_pos = pos + 4; 35| 9.77M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 7.82M, False: 1.95M] ------------------ 36| 7.82M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [True: 7.82M, Folded] ------------------ 37| 7.82M| ? char(u16_swap_bytes(data[pos])) 38| 7.82M| : char(data[pos]); 39| 7.82M| pos++; 40| 7.82M| } 41| 1.95M| continue; 42| 1.95M| } 43| 3.13M| } 44| 3.13M| } 45| | 46| 1.17M| uint16_t word = 47| 1.17M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (47:9): [True: 1.17M, Folded] ------------------ 48| 1.17M| if ((word & 0xFF80) == 0) { ------------------ | Branch (48:9): [True: 319k, False: 860k] ------------------ 49| | // will generate one UTF-8 bytes 50| 319k| *utf8_output++ = char(word); 51| 319k| pos++; 52| 860k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (52:16): [True: 98.9k, False: 761k] ------------------ 53| | // will generate two UTF-8 bytes 54| | // we have 0b110XXXXX 0b10XXXXXX 55| 98.9k| *utf8_output++ = char((word >> 6) | 0b11000000); 56| 98.9k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 57| 98.9k| pos++; 58| 761k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (58:16): [True: 759k, False: 2.80k] ------------------ 59| | // will generate three UTF-8 bytes 60| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 61| 759k| *utf8_output++ = char((word >> 12) | 0b11100000); 62| 759k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 63| 759k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 64| 759k| pos++; 65| 759k| } else { 66| | // must be a surrogate pair 67| 2.80k| uint16_t diff = uint16_t(word - 0xD800); 68| 2.80k| if (pos + 1 >= len) { ------------------ | Branch (68:11): [True: 0, False: 2.80k] ------------------ 69| 0| return 0; 70| 0| } // minimal bound checking 71| 2.80k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (71:28): [True: 2.80k, Folded] ------------------ 72| 2.80k| ? u16_swap_bytes(data[pos + 1]) 73| 2.80k| : data[pos + 1]; 74| 2.80k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 75| 2.80k| uint32_t value = (diff << 10) + diff2 + 0x10000; 76| | // will generate four UTF-8 bytes 77| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 78| 2.80k| *utf8_output++ = char((value >> 18) | 0b11110000); 79| 2.80k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 80| 2.80k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 81| 2.80k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 82| 2.80k| pos += 2; 83| 2.80k| } 84| 1.17M| } 85| 452| return utf8_output - start; 86| 452|} _ZN7simdutf6scalar5utf328validateEPKDim: 25| 5.67k| size_t len) noexcept { 26| 5.67k| return validate(reinterpret_cast(buf), len); 27| 5.67k|} _ZN7simdutf6scalar5utf3220validate_with_errorsEPKDim: 49| 7.86k|validate_with_errors(const char32_t *buf, size_t len) noexcept { 50| 7.86k| return validate_with_errors(reinterpret_cast(buf), len); 51| 7.86k|} _ZN7simdutf6scalar5utf3222utf8_length_from_utf32EPKDim: 54| 2.58k| size_t len) { 55| | // We are not BOM aware. 56| 2.58k| size_t counter{0}; 57| 2.79M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (57:22): [True: 2.79M, False: 2.58k] ------------------ 58| | // credit: @ttsugriy for the vectorizable approach 59| 2.79M| counter++; // ASCII 60| 2.79M| counter += static_cast(p[i] > 0x7F); // two-byte 61| 2.79M| counter += static_cast(p[i] > 0x7FF); // three-byte 62| 2.79M| counter += static_cast(p[i] > 0xFFFF); // four-bytes 63| 2.79M| } 64| 2.58k| return counter; 65| 2.58k|} _ZN7simdutf6scalar5utf3223utf16_length_from_utf32EPKDim: 68| 4.32k|utf16_length_from_utf32(const char32_t *p, size_t len) { 69| | // We are not BOM aware. 70| 4.32k| size_t counter{0}; 71| 4.66M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (71:22): [True: 4.65M, False: 4.32k] ------------------ 72| 4.65M| counter++; // non-surrogate word 73| 4.65M| counter += static_cast(p[i] > 0xFFFF); // surrogate pair 74| 4.65M| } 75| 4.32k| return counter; 76| 4.32k|} _ZN7simdutf6scalar5utf328validateIPKjQsr7simdutf6detailE19indexes_into_uint32IT_EEEbS5_m: 13| 5.67k| size_t len) noexcept { 14| 5.67k| uint64_t pos = 0; 15| 1.99M| for (; pos < len; pos++) { ------------------ | Branch (15:10): [True: 1.99M, False: 2.50k] ------------------ 16| 1.99M| uint32_t word = data[pos]; 17| 1.99M| if (word > 0x10FFFF || (word >= 0xD800 && word <= 0xDFFF)) { ------------------ | Branch (17:9): [True: 2.74k, False: 1.99M] | Branch (17:29): [True: 40.8k, False: 1.95M] | Branch (17:47): [True: 435, False: 40.4k] ------------------ 18| 3.17k| return false; 19| 3.17k| } 20| 1.99M| } 21| 2.50k| return true; 22| 5.67k|} _ZN7simdutf6scalar5utf3220validate_with_errorsIPKjQsr7simdutf6detailE19indexes_into_uint32IT_EEENS_6resultES5_m: 34| 7.86k|validate_with_errors(InputPtr data, size_t len) noexcept { 35| 7.86k| size_t pos = 0; 36| 2.00M| for (; pos < len; pos++) { ------------------ | Branch (36:10): [True: 2.00M, False: 2.50k] ------------------ 37| 2.00M| uint32_t word = data[pos]; 38| 2.00M| if (word > 0x10FFFF) { ------------------ | Branch (38:9): [True: 4.52k, False: 1.99M] ------------------ 39| 4.52k| return result(error_code::TOO_LARGE, pos); 40| 4.52k| } 41| 1.99M| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (41:9): [True: 42.2k, False: 1.95M] | Branch (41:27): [True: 834, False: 41.4k] ------------------ 42| 834| return result(error_code::SURROGATE, pos); 43| 834| } 44| 1.99M| } 45| 2.50k| return result(error_code::SUCCESS, pos); 46| 7.86k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf32_to_latin17convertEPKDimPc: 12| 544| char *latin1_output) { 13| 544| char *start = latin1_output; 14| 544| uint32_t utf32_char; 15| 544| size_t pos = 0; 16| 544| uint32_t too_large = 0; 17| | 18| 1.35M| while (pos < len) { ------------------ | Branch (18:10): [True: 1.35M, False: 544] ------------------ 19| 1.35M| utf32_char = (uint32_t)data[pos]; 20| 1.35M| too_large |= utf32_char; 21| 1.35M| *latin1_output++ = (char)(utf32_char & 0xFF); 22| 1.35M| pos++; 23| 1.35M| } 24| 544| if ((too_large & 0xFFFFFF00) != 0) { ------------------ | Branch (24:7): [True: 404, False: 140] ------------------ 25| 404| return 0; 26| 404| } 27| 140| return latin1_output - start; 28| 544|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf32_to_latin119convert_with_errorsEPKDimPc: 32| 1.27k| char *latin1_output) { 33| 1.27k| char *start{latin1_output}; 34| 1.27k| size_t pos = 0; 35| 268k| while (pos < len) { ------------------ | Branch (35:10): [True: 268k, False: 88] ------------------ 36| |#if SIMDUTF_CPLUSPLUS23 37| | if !consteval 38| |#endif 39| 268k| { 40| 268k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (40:11): [True: 267k, False: 294] ------------------ 41| | // they are Latin1 42| 267k| uint64_t v; 43| 267k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 44| 267k| if ((v & 0xFFFFFF00FFFFFF00) == 0) { ------------------ | Branch (44:13): [True: 266k, False: 1.09k] ------------------ 45| 266k| *latin1_output++ = char(data[pos]); 46| 266k| *latin1_output++ = char(data[pos + 1]); 47| 266k| pos += 2; 48| 266k| continue; 49| 266k| } 50| 267k| } 51| 268k| } 52| | 53| 1.38k| uint32_t utf32_char = data[pos]; 54| 1.38k| if ((utf32_char & 0xFFFFFF00) == ------------------ | Branch (54:9): [True: 206, False: 1.18k] ------------------ 55| 1.38k| 0) { // Check if the character can be represented in Latin-1 56| 206| *latin1_output++ = (char)(utf32_char & 0xFF); 57| 206| pos++; 58| 1.18k| } else { 59| 1.18k| return result(error_code::TOO_LARGE, pos); 60| 1.18k| }; 61| 206| } 62| 88| return result(error_code::SUCCESS, latin1_output - start); 63| 1.27k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf167convertILNS_10endiannessE0EEEmPKDimPDs: 11| 1.91k| char16_t *utf16_output) { 12| 1.91k| size_t pos = 0; 13| 1.91k| char16_t *start{utf16_output}; 14| 670k| while (pos < len) { ------------------ | Branch (14:10): [True: 669k, False: 1.00k] ------------------ 15| 669k| uint32_t word = data[pos]; 16| 669k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 665k, False: 3.73k] ------------------ 17| 665k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (17:11): [True: 1.80k, False: 663k] | Branch (17:29): [True: 218, False: 1.58k] ------------------ 18| 218| return 0; 19| 218| } 20| | // will not generate a surrogate pair 21| 665k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (21:25): [Folded, False: 665k] ------------------ 22| 665k| ? char16_t(u16_swap_bytes(uint16_t(word))) 23| 665k| : char16_t(word); 24| 665k| } else { 25| | // will generate a surrogate pair 26| 3.73k| if (word > 0x10FFFF) { ------------------ | Branch (26:11): [True: 694, False: 3.03k] ------------------ 27| 694| return 0; 28| 694| } 29| 3.03k| word -= 0x10000; 30| 3.03k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 31| 3.03k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 32| | if constexpr (!match_system(big_endian)) { 33| | high_surrogate = u16_swap_bytes(high_surrogate); 34| | low_surrogate = u16_swap_bytes(low_surrogate); 35| | } 36| 3.03k| *utf16_output++ = char16_t(high_surrogate); 37| 3.03k| *utf16_output++ = char16_t(low_surrogate); 38| 3.03k| } 39| 668k| pos++; 40| 668k| } 41| 1.00k| return utf16_output - start; 42| 1.91k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf167convertILNS_10endiannessE1EEEmPKDimPDs: 11| 1.56k| char16_t *utf16_output) { 12| 1.56k| size_t pos = 0; 13| 1.56k| char16_t *start{utf16_output}; 14| 110k| while (pos < len) { ------------------ | Branch (14:10): [True: 109k, False: 628] ------------------ 15| 109k| uint32_t word = data[pos]; 16| 109k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 106k, False: 2.88k] ------------------ 17| 106k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (17:11): [True: 1.10k, False: 105k] | Branch (17:29): [True: 166, False: 940] ------------------ 18| 166| return 0; 19| 166| } 20| | // will not generate a surrogate pair 21| 106k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (21:25): [True: 106k, Folded] ------------------ 22| 106k| ? char16_t(u16_swap_bytes(uint16_t(word))) 23| 106k| : char16_t(word); 24| 106k| } else { 25| | // will generate a surrogate pair 26| 2.88k| if (word > 0x10FFFF) { ------------------ | Branch (26:11): [True: 772, False: 2.10k] ------------------ 27| 772| return 0; 28| 772| } 29| 2.10k| word -= 0x10000; 30| 2.10k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 31| 2.10k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 32| 2.10k| if constexpr (!match_system(big_endian)) { 33| 2.10k| high_surrogate = u16_swap_bytes(high_surrogate); 34| 2.10k| low_surrogate = u16_swap_bytes(low_surrogate); 35| 2.10k| } 36| 2.10k| *utf16_output++ = char16_t(high_surrogate); 37| 2.10k| *utf16_output++ = char16_t(low_surrogate); 38| 2.10k| } 39| 108k| pos++; 40| 108k| } 41| 628| return utf16_output - start; 42| 1.56k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1619convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKDimPDs: 46| 1.99k| char16_t *utf16_output) { 47| 1.99k| size_t pos = 0; 48| 1.99k| char16_t *start{utf16_output}; 49| 547k| while (pos < len) { ------------------ | Branch (49:10): [True: 546k, False: 484] ------------------ 50| 546k| uint32_t word = data[pos]; 51| 546k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (51:9): [True: 539k, False: 7.21k] ------------------ 52| 539k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (52:11): [True: 1.52k, False: 538k] | Branch (52:29): [True: 240, False: 1.28k] ------------------ 53| 240| return result(error_code::SURROGATE, pos); 54| 240| } 55| | // will not generate a surrogate pair 56| 539k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (56:25): [Folded, False: 539k] ------------------ 57| 539k| ? char16_t(u16_swap_bytes(uint16_t(word))) 58| 539k| : char16_t(word); 59| 539k| } else { 60| | // will generate a surrogate pair 61| 7.21k| if (word > 0x10FFFF) { ------------------ | Branch (61:11): [True: 1.27k, False: 5.94k] ------------------ 62| 1.27k| return result(error_code::TOO_LARGE, pos); 63| 1.27k| } 64| 5.94k| word -= 0x10000; 65| 5.94k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 66| 5.94k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 67| | if constexpr (!match_system(big_endian)) { 68| | high_surrogate = u16_swap_bytes(high_surrogate); 69| | low_surrogate = u16_swap_bytes(low_surrogate); 70| | } 71| 5.94k| *utf16_output++ = char16_t(high_surrogate); 72| 5.94k| *utf16_output++ = char16_t(low_surrogate); 73| 5.94k| } 74| 545k| pos++; 75| 545k| } 76| 484| return result(error_code::SUCCESS, utf16_output - start); 77| 1.99k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1619convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKDimPDs: 46| 2.05k| char16_t *utf16_output) { 47| 2.05k| size_t pos = 0; 48| 2.05k| char16_t *start{utf16_output}; 49| 37.3k| while (pos < len) { ------------------ | Branch (49:10): [True: 36.8k, False: 518] ------------------ 50| 36.8k| uint32_t word = data[pos]; 51| 36.8k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (51:9): [True: 31.9k, False: 4.92k] ------------------ 52| 31.9k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (52:11): [True: 1.83k, False: 30.0k] | Branch (52:29): [True: 240, False: 1.59k] ------------------ 53| 240| return result(error_code::SURROGATE, pos); 54| 240| } 55| | // will not generate a surrogate pair 56| 31.6k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (56:25): [True: 31.6k, Folded] ------------------ 57| 31.6k| ? char16_t(u16_swap_bytes(uint16_t(word))) 58| 31.6k| : char16_t(word); 59| 31.6k| } else { 60| | // will generate a surrogate pair 61| 4.92k| if (word > 0x10FFFF) { ------------------ | Branch (61:11): [True: 1.29k, False: 3.63k] ------------------ 62| 1.29k| return result(error_code::TOO_LARGE, pos); 63| 1.29k| } 64| 3.63k| word -= 0x10000; 65| 3.63k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 66| 3.63k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 67| 3.63k| if constexpr (!match_system(big_endian)) { 68| 3.63k| high_surrogate = u16_swap_bytes(high_surrogate); 69| 3.63k| low_surrogate = u16_swap_bytes(low_surrogate); 70| 3.63k| } 71| 3.63k| *utf16_output++ = char16_t(high_surrogate); 72| 3.63k| *utf16_output++ = char16_t(low_surrogate); 73| 3.63k| } 74| 35.3k| pos++; 75| 35.3k| } 76| 518| return result(error_code::SUCCESS, utf16_output - start); 77| 2.05k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1613convert_validILNS_10endiannessE0EEEmPKDimPDs: 11| 280| char16_t *utf16_output) { 12| 280| size_t pos = 0; 13| 280| char16_t *start{utf16_output}; 14| 547k| while (pos < len) { ------------------ | Branch (14:10): [True: 546k, False: 280] ------------------ 15| 546k| uint32_t word = data[pos]; 16| 546k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 526k, False: 20.2k] ------------------ 17| | // will not generate a surrogate pair 18| 526k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (18:25): [Folded, False: 526k] ------------------ 19| 526k| ? char16_t(u16_swap_bytes(uint16_t(word))) 20| 526k| : char16_t(word); 21| 526k| pos++; 22| 526k| } else { 23| | // will generate a surrogate pair 24| 20.2k| word -= 0x10000; 25| 20.2k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 26| 20.2k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 27| | if constexpr (!match_system(big_endian)) { 28| | high_surrogate = u16_swap_bytes(high_surrogate); 29| | low_surrogate = u16_swap_bytes(low_surrogate); 30| | } 31| 20.2k| *utf16_output++ = char16_t(high_surrogate); 32| 20.2k| *utf16_output++ = char16_t(low_surrogate); 33| 20.2k| pos++; 34| 20.2k| } 35| 546k| } 36| 280| return utf16_output - start; 37| 280|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1613convert_validILNS_10endiannessE1EEEmPKDimPDs: 11| 134| char16_t *utf16_output) { 12| 134| size_t pos = 0; 13| 134| char16_t *start{utf16_output}; 14| 32.5k| while (pos < len) { ------------------ | Branch (14:10): [True: 32.3k, False: 134] ------------------ 15| 32.3k| uint32_t word = data[pos]; 16| 32.3k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 11.6k, False: 20.7k] ------------------ 17| | // will not generate a surrogate pair 18| 11.6k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (18:25): [True: 11.6k, Folded] ------------------ 19| 11.6k| ? char16_t(u16_swap_bytes(uint16_t(word))) 20| 11.6k| : char16_t(word); 21| 11.6k| pos++; 22| 20.7k| } else { 23| | // will generate a surrogate pair 24| 20.7k| word -= 0x10000; 25| 20.7k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 26| 20.7k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 27| 20.7k| if constexpr (!match_system(big_endian)) { 28| 20.7k| high_surrogate = u16_swap_bytes(high_surrogate); 29| 20.7k| low_surrogate = u16_swap_bytes(low_surrogate); 30| 20.7k| } 31| 20.7k| *utf16_output++ = char16_t(high_surrogate); 32| 20.7k| *utf16_output++ = char16_t(low_surrogate); 33| 20.7k| pos++; 34| 20.7k| } 35| 32.3k| } 36| 134| return utf16_output - start; 37| 134|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf32_to_utf87convertIPKDiPcQaasr7simdutf6detailE18indexes_into_utf32IT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEEmS7_mS8_: 17| 1.81k| OutputPtr utf8_output) { 18| 1.81k| size_t pos = 0; 19| 1.81k| auto start = utf8_output; 20| 263k| while (pos < len) { ------------------ | Branch (20:10): [True: 263k, False: 716] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 263k| { // try to convert the next block of 2 ASCII characters 25| 263k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (25:11): [True: 262k, False: 952] ------------------ 26| | // they are ascii 27| 262k| uint64_t v; 28| 262k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 29| 262k| if ((v & 0xFFFFFF80FFFFFF80) == 0) { ------------------ | Branch (29:13): [True: 242k, False: 19.7k] ------------------ 30| 242k| *utf8_output++ = char(data[pos]); 31| 242k| *utf8_output++ = char(data[pos + 1]); 32| 242k| pos += 2; 33| 242k| continue; 34| 242k| } 35| 262k| } 36| 263k| } 37| | 38| 20.6k| uint32_t word = data[pos]; 39| 20.6k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (39:9): [True: 3.31k, False: 17.3k] ------------------ 40| | // will generate one UTF-8 bytes 41| 3.31k| *utf8_output++ = char(word); 42| 3.31k| pos++; 43| 17.3k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (43:16): [True: 3.64k, False: 13.6k] ------------------ 44| | // will generate two UTF-8 bytes 45| | // we have 0b110XXXXX 0b10XXXXXX 46| 3.64k| *utf8_output++ = char((word >> 6) | 0b11000000); 47| 3.64k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 48| 3.64k| pos++; 49| 13.6k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (49:16): [True: 8.71k, False: 4.98k] ------------------ 50| | // will generate three UTF-8 bytes 51| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 52| 8.71k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (52:11): [True: 6.29k, False: 2.42k] | Branch (52:29): [True: 150, False: 6.14k] ------------------ 53| 150| return 0; 54| 150| } 55| 8.56k| *utf8_output++ = char((word >> 12) | 0b11100000); 56| 8.56k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 57| 8.56k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 58| 8.56k| pos++; 59| 8.56k| } else { 60| | // will generate four UTF-8 bytes 61| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 62| 4.98k| if (word > 0x10FFFF) { ------------------ | Branch (62:11): [True: 946, False: 4.04k] ------------------ 63| 946| return 0; 64| 946| } 65| 4.04k| *utf8_output++ = char((word >> 18) | 0b11110000); 66| 4.04k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 67| 4.04k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 68| 4.04k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 69| 4.04k| pos++; 70| 4.04k| } 71| 20.6k| } 72| 716| return utf8_output - start; 73| 1.81k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf32_to_utf819convert_with_errorsIPKDiPcQaasr7simdutf6detailE18indexes_into_utf32IT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEENS_6resultES7_mS8_: 81| 2.48k| OutputPtr utf8_output) { 82| 2.48k| size_t pos = 0; 83| 2.48k| auto start = utf8_output; 84| 211k| while (pos < len) { ------------------ | Branch (84:10): [True: 210k, False: 634] ------------------ 85| |#if SIMDUTF_CPLUSPLUS23 86| | if !consteval 87| |#endif 88| 210k| { // try to convert the next block of 2 ASCII characters 89| 210k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (89:11): [True: 209k, False: 1.01k] ------------------ 90| | // they are ascii 91| 209k| uint64_t v; 92| 209k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 93| 209k| if ((v & 0xFFFFFF80FFFFFF80) == 0) { ------------------ | Branch (93:13): [True: 172k, False: 37.0k] ------------------ 94| 172k| *utf8_output++ = char(data[pos]); 95| 172k| *utf8_output++ = char(data[pos + 1]); 96| 172k| pos += 2; 97| 172k| continue; 98| 172k| } 99| 209k| } 100| 210k| } 101| | 102| 38.0k| uint32_t word = data[pos]; 103| 38.0k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (103:9): [True: 4.44k, False: 33.5k] ------------------ 104| | // will generate one UTF-8 bytes 105| 4.44k| *utf8_output++ = char(word); 106| 4.44k| pos++; 107| 33.5k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (107:16): [True: 16.7k, False: 16.8k] ------------------ 108| | // will generate two UTF-8 bytes 109| | // we have 0b110XXXXX 0b10XXXXXX 110| 16.7k| *utf8_output++ = char((word >> 6) | 0b11000000); 111| 16.7k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 112| 16.7k| pos++; 113| 16.8k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (113:16): [True: 10.9k, False: 5.89k] ------------------ 114| | // will generate three UTF-8 bytes 115| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 116| 10.9k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (116:11): [True: 7.67k, False: 3.28k] | Branch (116:29): [True: 318, False: 7.36k] ------------------ 117| 318| return result(error_code::SURROGATE, pos); 118| 318| } 119| 10.6k| *utf8_output++ = char((word >> 12) | 0b11100000); 120| 10.6k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 121| 10.6k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 122| 10.6k| pos++; 123| 10.6k| } else { 124| | // will generate four UTF-8 bytes 125| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 126| 5.89k| if (word > 0x10FFFF) { ------------------ | Branch (126:11): [True: 1.53k, False: 4.35k] ------------------ 127| 1.53k| return result(error_code::TOO_LARGE, pos); 128| 1.53k| } 129| 4.35k| *utf8_output++ = char((word >> 18) | 0b11110000); 130| 4.35k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 131| 4.35k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 132| 4.35k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 133| 4.35k| pos++; 134| 4.35k| } 135| 38.0k| } 136| 634| return result(error_code::SUCCESS, utf8_output - start); 137| 2.48k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf32_to_utf813convert_validIPKDiPcQaasr7simdutf6detailE18indexes_into_utf32IT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEEmS7_mS8_: 17| 216| OutputPtr utf8_output) { 18| 216| size_t pos = 0; 19| 216| auto start = utf8_output; 20| 64.0k| while (pos < len) { ------------------ | Branch (20:10): [True: 63.7k, False: 216] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 63.7k| { // try to convert the next block of 2 ASCII characters 25| 63.7k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (25:11): [True: 63.6k, False: 188] ------------------ 26| | // they are ascii 27| 63.6k| uint64_t v; 28| 63.6k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 29| 63.6k| if ((v & 0xFFFFFF80FFFFFF80) == 0) { ------------------ | Branch (29:13): [True: 50.1k, False: 13.4k] ------------------ 30| 50.1k| *utf8_output++ = char(data[pos]); 31| 50.1k| *utf8_output++ = char(data[pos + 1]); 32| 50.1k| pos += 2; 33| 50.1k| continue; 34| 50.1k| } 35| 63.6k| } 36| 63.7k| } 37| | 38| 13.6k| uint32_t word = data[pos]; 39| 13.6k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (39:9): [True: 934, False: 12.7k] ------------------ 40| | // will generate one UTF-8 bytes 41| 934| *utf8_output++ = char(word); 42| 934| pos++; 43| 12.7k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (43:16): [True: 10.4k, False: 2.25k] ------------------ 44| | // will generate two UTF-8 bytes 45| | // we have 0b110XXXXX 0b10XXXXXX 46| 10.4k| *utf8_output++ = char((word >> 6) | 0b11000000); 47| 10.4k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 48| 10.4k| pos++; 49| 10.4k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (49:16): [True: 996, False: 1.25k] ------------------ 50| | // will generate three UTF-8 bytes 51| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 52| 996| *utf8_output++ = char((word >> 12) | 0b11100000); 53| 996| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 54| 996| *utf8_output++ = char((word & 0b111111) | 0b10000000); 55| 996| pos++; 56| 1.25k| } else { 57| | // will generate four UTF-8 bytes 58| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 59| 1.25k| *utf8_output++ = char((word >> 18) | 0b11110000); 60| 1.25k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 61| 1.25k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 62| 1.25k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 63| 1.25k| pos++; 64| 1.25k| } 65| 13.6k| } 66| 216| return utf8_output - start; 67| 216|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf88validateEPKcm: 113| 3.48k| size_t len) noexcept { 114| 3.48k| return validate(reinterpret_cast(buf), len); 115| 3.48k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf88validateIPKhEEbT_m: 14| 3.48k| size_t len) noexcept { 15| 3.48k| static_assert( 16| 3.48k| std::is_same::type, uint8_t>::value, 17| 3.48k| "dereferencing the data pointer must result in a uint8_t"); 18| 3.48k| uint64_t pos = 0; 19| 3.48k| uint32_t code_point = 0; 20| 8.88M| while (pos < len) { ------------------ | Branch (20:10): [True: 8.88M, False: 619] ------------------ 21| 8.88M| uint64_t next_pos; 22| |#if SIMDUTF_CPLUSPLUS23 23| | if !consteval 24| |#endif 25| 8.88M| { // check if the next 16 bytes are ascii. 26| 8.88M| next_pos = pos + 16; 27| 8.88M| if (next_pos <= len) { // if it is safe to read 16 more bytes, check ------------------ | Branch (27:11): [True: 8.88M, False: 3.40k] ------------------ 28| | // that they are ascii 29| 8.88M| uint64_t v1{}; 30| 8.88M| std::memcpy(&v1, data + pos, sizeof(uint64_t)); 31| 8.88M| uint64_t v2{}; 32| 8.88M| std::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 33| 8.88M| uint64_t v{v1 | v2}; 34| 8.88M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (34:13): [True: 7.67M, False: 1.20M] ------------------ 35| 7.67M| pos = next_pos; 36| 7.67M| continue; 37| 7.67M| } 38| 8.88M| } 39| 8.88M| } 40| | 41| 1.21M| unsigned char byte = data[pos]; 42| | 43| 6.16M| while (byte < 0b10000000) { ------------------ | Branch (43:12): [True: 4.95M, False: 1.21M] ------------------ 44| 4.95M| if (++pos == len) { ------------------ | Branch (44:11): [True: 1.10k, False: 4.95M] ------------------ 45| 1.10k| return true; 46| 1.10k| } 47| 4.95M| byte = data[pos]; 48| 4.95M| } 49| | 50| 1.21M| if ((byte & 0b11100000) == 0b11000000) { ------------------ | Branch (50:9): [True: 804k, False: 405k] ------------------ 51| 804k| next_pos = pos + 2; 52| 804k| if (next_pos > len) { ------------------ | Branch (52:11): [True: 65, False: 804k] ------------------ 53| 65| return false; 54| 65| } 55| 804k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (55:11): [True: 239, False: 803k] ------------------ 56| 239| return false; 57| 239| } 58| | // range check 59| 803k| code_point = (byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 60| 803k| if (code_point < 0x80) { ------------------ | Branch (60:11): [True: 35, False: 803k] ------------------ 61| 35| return false; 62| 35| } 63| 803k| } else if ((byte & 0b11110000) == 0b11100000) { ------------------ | Branch (63:16): [True: 336k, False: 69.5k] ------------------ 64| 336k| next_pos = pos + 3; 65| 336k| if (next_pos > len) { ------------------ | Branch (65:11): [True: 57, False: 336k] ------------------ 66| 57| return false; 67| 57| } 68| 336k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (68:11): [True: 133, False: 336k] ------------------ 69| 133| return false; 70| 133| } 71| 336k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (71:11): [True: 42, False: 336k] ------------------ 72| 42| return false; 73| 42| } 74| | // range check 75| 336k| code_point = (byte & 0b00001111) << 12 | 76| 336k| (data[pos + 1] & 0b00111111) << 6 | 77| 336k| (data[pos + 2] & 0b00111111); 78| 336k| if ((code_point < 0x800) || ------------------ | Branch (78:11): [True: 40, False: 335k] ------------------ 79| 335k| (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (79:12): [True: 7.39k, False: 328k] | Branch (79:35): [True: 23, False: 7.37k] ------------------ 80| 63| return false; 81| 63| } 82| 336k| } else if ((byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (82:16): [True: 68.7k, False: 836] ------------------ 83| 68.7k| next_pos = pos + 4; 84| 68.7k| if (next_pos > len) { ------------------ | Branch (84:11): [True: 43, False: 68.6k] ------------------ 85| 43| return false; 86| 43| } 87| 68.6k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (87:11): [True: 83, False: 68.5k] ------------------ 88| 83| return false; 89| 83| } 90| 68.5k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (90:11): [True: 42, False: 68.5k] ------------------ 91| 42| return false; 92| 42| } 93| 68.5k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (93:11): [True: 43, False: 68.5k] ------------------ 94| 43| return false; 95| 43| } 96| | // range check 97| 68.5k| code_point = 98| 68.5k| (byte & 0b00000111) << 18 | (data[pos + 1] & 0b00111111) << 12 | 99| 68.5k| (data[pos + 2] & 0b00111111) << 6 | (data[pos + 3] & 0b00111111); 100| 68.5k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (100:11): [True: 42, False: 68.4k] | Branch (100:35): [True: 36, False: 68.4k] ------------------ 101| 78| return false; 102| 78| } 103| 68.5k| } else { 104| | // we may have a continuation 105| 836| return false; 106| 836| } 107| 1.20M| pos = next_pos; 108| 1.20M| } 109| 619| return true; 110| 3.48k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf820validate_with_errorsEPKcm: 221| 6.73k|validate_with_errors(const char *buf, size_t len) noexcept { 222| 6.73k| return validate_with_errors(reinterpret_cast(buf), len); 223| 6.73k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf820validate_with_errorsIPKhEENS_6resultET_m: 119| 6.73k|validate_with_errors(BytePtr data, size_t len) noexcept { 120| 6.73k| static_assert( 121| 6.73k| std::is_same::type, uint8_t>::value, 122| 6.73k| "dereferencing the data pointer must result in a uint8_t"); 123| 6.73k| size_t pos = 0; 124| 6.73k| uint32_t code_point = 0; 125| 8.89M| while (pos < len) { ------------------ | Branch (125:10): [True: 8.89M, False: 619] ------------------ 126| | // check of the next 16 bytes are ascii. 127| 8.89M| size_t next_pos = pos + 16; 128| 8.89M| if (next_pos <= ------------------ | Branch (128:9): [True: 8.88M, False: 5.22k] ------------------ 129| 8.89M| len) { // if it is safe to read 16 more bytes, check that they are ascii 130| 8.88M| uint64_t v1; 131| 8.88M| std::memcpy(&v1, data + pos, sizeof(uint64_t)); 132| 8.88M| uint64_t v2; 133| 8.88M| std::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 134| 8.88M| uint64_t v{v1 | v2}; 135| 8.88M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (135:11): [True: 7.67M, False: 1.21M] ------------------ 136| 7.67M| pos = next_pos; 137| 7.67M| continue; 138| 7.67M| } 139| 8.88M| } 140| 1.21M| unsigned char byte = data[pos]; 141| | 142| 6.19M| while (byte < 0b10000000) { ------------------ | Branch (142:12): [True: 4.97M, False: 1.21M] ------------------ 143| 4.97M| if (++pos == len) { ------------------ | Branch (143:11): [True: 1.10k, False: 4.97M] ------------------ 144| 1.10k| return result(error_code::SUCCESS, len); 145| 1.10k| } 146| 4.97M| byte = data[pos]; 147| 4.97M| } 148| | 149| 1.21M| if ((byte & 0b11100000) == 0b11000000) { ------------------ | Branch (149:9): [True: 806k, False: 409k] ------------------ 150| 806k| next_pos = pos + 2; 151| 806k| if (next_pos > len) { ------------------ | Branch (151:11): [True: 195, False: 806k] ------------------ 152| 195| return result(error_code::TOO_SHORT, pos); 153| 195| } 154| 806k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (154:11): [True: 717, False: 805k] ------------------ 155| 717| return result(error_code::TOO_SHORT, pos); 156| 717| } 157| | // range check 158| 805k| code_point = (byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 159| 805k| if (code_point < 0x80) { ------------------ | Branch (159:11): [True: 105, False: 805k] ------------------ 160| 105| return result(error_code::OVERLONG, pos); 161| 105| } 162| 805k| } else if ((byte & 0b11110000) == 0b11100000) { ------------------ | Branch (162:16): [True: 337k, False: 71.9k] ------------------ 163| 337k| next_pos = pos + 3; 164| 337k| if (next_pos > len) { ------------------ | Branch (164:11): [True: 171, False: 337k] ------------------ 165| 171| return result(error_code::TOO_SHORT, pos); 166| 171| } 167| 337k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (167:11): [True: 399, False: 337k] ------------------ 168| 399| return result(error_code::TOO_SHORT, pos); 169| 399| } 170| 337k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (170:11): [True: 126, False: 337k] ------------------ 171| 126| return result(error_code::TOO_SHORT, pos); 172| 126| } 173| | // range check 174| 337k| code_point = (byte & 0b00001111) << 12 | 175| 337k| (data[pos + 1] & 0b00111111) << 6 | 176| 337k| (data[pos + 2] & 0b00111111); 177| 337k| if (code_point < 0x800) { ------------------ | Branch (177:11): [True: 120, False: 337k] ------------------ 178| 120| return result(error_code::OVERLONG, pos); 179| 120| } 180| 337k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (180:11): [True: 8.01k, False: 329k] | Branch (180:34): [True: 69, False: 7.94k] ------------------ 181| 69| return result(error_code::SURROGATE, pos); 182| 69| } 183| 337k| } else if ((byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (183:16): [True: 69.7k, False: 2.23k] ------------------ 184| 69.7k| next_pos = pos + 4; 185| 69.7k| if (next_pos > len) { ------------------ | Branch (185:11): [True: 129, False: 69.6k] ------------------ 186| 129| return result(error_code::TOO_SHORT, pos); 187| 129| } 188| 69.6k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (188:11): [True: 249, False: 69.3k] ------------------ 189| 249| return result(error_code::TOO_SHORT, pos); 190| 249| } 191| 69.3k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (191:11): [True: 126, False: 69.2k] ------------------ 192| 126| return result(error_code::TOO_SHORT, pos); 193| 126| } 194| 69.2k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (194:11): [True: 129, False: 69.1k] ------------------ 195| 129| return result(error_code::TOO_SHORT, pos); 196| 129| } 197| | // range check 198| 69.1k| code_point = 199| 69.1k| (byte & 0b00000111) << 18 | (data[pos + 1] & 0b00111111) << 12 | 200| 69.1k| (data[pos + 2] & 0b00111111) << 6 | (data[pos + 3] & 0b00111111); 201| 69.1k| if (code_point <= 0xffff) { ------------------ | Branch (201:11): [True: 126, False: 68.9k] ------------------ 202| 126| return result(error_code::OVERLONG, pos); 203| 126| } 204| 68.9k| if (0x10ffff < code_point) { ------------------ | Branch (204:11): [True: 108, False: 68.8k] ------------------ 205| 108| return result(error_code::TOO_LARGE, pos); 206| 108| } 207| 68.9k| } else { 208| | // we either have too many continuation bytes or an invalid leading byte 209| 2.23k| if ((byte & 0b11000000) == 0b10000000) { ------------------ | Branch (209:11): [True: 1.17k, False: 1.06k] ------------------ 210| 1.17k| return result(error_code::TOO_LONG, pos); 211| 1.17k| } else { 212| 1.06k| return result(error_code::HEADER_BITS, pos); 213| 1.06k| } 214| 2.23k| } 215| 1.21M| pos = next_pos; 216| 1.21M| } 217| 619| return result(error_code::SUCCESS, len); 218| 6.73k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf831rewind_and_validate_with_errorsEPKcS4_m: 232| 3.51k| const char *start, const char *buf, size_t len) noexcept { 233| | // First check that we start with a leading byte 234| 3.51k| if ((*start & 0b11000000) == 0b10000000) { ------------------ | Branch (234:7): [True: 270, False: 3.24k] ------------------ 235| 270| return result(error_code::TOO_LONG, 0); 236| 270| } 237| 3.24k| size_t extra_len{0}; 238| | // A leading byte cannot be further than 4 bytes away 239| 3.44k| for (int i = 0; i < 5; i++) { ------------------ | Branch (239:19): [True: 3.44k, False: 0] ------------------ 240| 3.44k| unsigned char byte = *buf; 241| 3.44k| if ((byte & 0b11000000) != 0b10000000) { ------------------ | Branch (241:9): [True: 3.24k, False: 192] ------------------ 242| 3.24k| break; 243| 3.24k| } else { 244| 192| buf--; 245| 192| extra_len++; 246| 192| } 247| 3.44k| } 248| | 249| 3.24k| result res = validate_with_errors(buf, len + extra_len); 250| 3.24k| res.count -= extra_len; // Might underflow 251| 3.24k| return res; 252| 3.51k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf817count_code_pointsIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_m: 258| 15.0k|simdutf_constexpr23 size_t count_code_points(InputPtr data, size_t len) { 259| 15.0k| size_t counter{0}; 260| 264M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (260:22): [True: 264M, False: 15.0k] ------------------ 261| | // -65 is 0b10111111, anything larger in two-complement's should start a new 262| | // code point. 263| 264M| if (int8_t(data[i]) > -65) { ------------------ | Branch (263:9): [True: 261M, False: 3.13M] ------------------ 264| 261M| counter++; 265| 261M| } 266| 264M| } 267| 15.0k| return counter; 268| 15.0k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf822utf16_length_from_utf8IPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_m: 274| 5.83k|simdutf_constexpr23 size_t utf16_length_from_utf8(InputPtr data, size_t len) { 275| 5.83k| size_t counter{0}; 276| 97.4M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (276:22): [True: 97.4M, False: 5.83k] ------------------ 277| 97.4M| if (int8_t(data[i]) > -65) { ------------------ | Branch (277:9): [True: 95.7M, False: 1.65M] ------------------ 278| 95.7M| counter++; 279| 95.7M| } 280| 97.4M| if (uint8_t(data[i]) >= 240) { ------------------ | Branch (280:9): [True: 137k, False: 97.3M] ------------------ 281| 137k| counter++; 282| 137k| } 283| 97.4M| } 284| 5.83k| return counter; 285| 5.83k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf8_to_latin130rewind_and_convert_with_errorsEmPKcmPc: 176| 1.01k| char *latin1_output) { 177| 1.01k| size_t extra_len{0}; 178| | // We potentially need to go back in time and find a leading byte. 179| | // In theory '3' would be sufficient, but sometimes the error can go back 180| | // quite far. 181| 1.01k| size_t how_far_back = prior_bytes; 182| | // size_t how_far_back = 3; // 3 bytes in the past + current position 183| | // if(how_far_back >= prior_bytes) { how_far_back = prior_bytes; } 184| 1.01k| bool found_leading_bytes{false}; 185| | // important: it is i <= how_far_back and not 'i < how_far_back'. 186| 1.10k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (186:22): [True: 1.02k, False: 88] ------------------ 187| 1.02k| unsigned char byte = buf[-static_cast(i)]; 188| 1.02k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 189| 1.02k| if (found_leading_bytes) { ------------------ | Branch (189:9): [True: 928, False: 92] ------------------ 190| 928| if (i > 0 && byte < 128) { ------------------ | Branch (190:11): [True: 4, False: 924] | Branch (190:20): [True: 4, False: 0] ------------------ 191| | // If we had to go back and the leading byte is ascii 192| | // then we can stop right away. 193| 4| return result(error_code::TOO_LONG, 0 - i + 1); 194| 4| } 195| 924| buf -= i; 196| 924| extra_len = i; 197| 924| break; 198| 928| } 199| 1.02k| } 200| | // 201| | // It is possible for this function to return a negative count in its result. 202| | // C++ Standard Section 18.1 defines size_t is in which is described 203| | // in C Standard as . C Standard Section 4.1.5 defines size_t as an 204| | // unsigned integral type of the result of the sizeof operator 205| | // 206| | // An unsigned type will simply wrap round arithmetically (well defined). 207| | // 208| 1.01k| if (!found_leading_bytes) { ------------------ | Branch (208:7): [True: 88, False: 924] ------------------ 209| | // If how_far_back == 3, we may have four consecutive continuation bytes!!! 210| | // [....] [continuation] [continuation] [continuation] | [buf is 211| | // continuation] Or we possibly have a stream that does not start with a 212| | // leading byte. 213| 88| return result(error_code::TOO_LONG, 0 - how_far_back); 214| 88| } 215| 924| result res = convert_with_errors(buf, len + extra_len, latin1_output); 216| 924| if (res.error) { ------------------ | Branch (216:7): [True: 480, False: 444] ------------------ 217| 480| res.count -= extra_len; 218| 480| } 219| 924| return res; 220| 1.01k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf8_to_latin119convert_with_errorsIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEENS_6resultES6_mPc: 92| 1.43k| char *latin_output) { 93| 1.43k| size_t pos = 0; 94| 1.43k| char *start{latin_output}; 95| | 96| 1.44M| while (pos < len) { ------------------ | Branch (96:10): [True: 1.44M, False: 670] ------------------ 97| |#if SIMDUTF_CPLUSPLUS23 98| | if !consteval 99| |#endif 100| 1.44M| { 101| | // try to convert the next block of 16 ASCII bytes 102| 1.44M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (102:11): [True: 1.44M, False: 6.37k] ------------------ 103| | // they are ascii 104| 1.44M| uint64_t v1; 105| 1.44M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 106| 1.44M| uint64_t v2; 107| 1.44M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 108| 1.44M| uint64_t v{v1 | v2}; // We are only interested in these bits: 1000 1000 109| | // 1000 1000...etc 110| 1.44M| if ((v & 0x8080808080808080) == ------------------ | Branch (110:13): [True: 1.22M, False: 210k] ------------------ 111| 1.44M| 0) { // if NONE of these are set, e.g. all of them are zero, then 112| | // everything is ASCII 113| 1.22M| size_t final_pos = pos + 16; 114| 20.8M| while (pos < final_pos) { ------------------ | Branch (114:18): [True: 19.6M, False: 1.22M] ------------------ 115| 19.6M| *latin_output++ = char(data[pos]); 116| 19.6M| pos++; 117| 19.6M| } 118| 1.22M| continue; 119| 1.22M| } 120| 1.44M| } 121| 1.44M| } 122| | // suppose it is not an all ASCII byte sequence 123| 217k| uint8_t leading_byte = data[pos]; // leading byte 124| 217k| if (leading_byte < 0b10000000) { ------------------ | Branch (124:9): [True: 189k, False: 27.8k] ------------------ 125| | // converting one ASCII byte !!! 126| 189k| *latin_output++ = char(leading_byte); 127| 189k| pos++; 128| 189k| } else if ((leading_byte & 0b11100000) == ------------------ | Branch (128:16): [True: 27.3k, False: 496] ------------------ 129| 27.8k| 0b11000000) { // the first three bits indicate: 130| | // We have a two-byte UTF-8 131| 27.3k| if (pos + 1 >= len) { ------------------ | Branch (131:11): [True: 30, False: 27.3k] ------------------ 132| 30| return result(error_code::TOO_SHORT, pos); 133| 30| } // minimal bound checking 134| 27.3k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (134:11): [True: 180, False: 27.1k] ------------------ 135| 180| return result(error_code::TOO_SHORT, pos); 136| 180| } // checks if the next byte is a valid continuation byte in UTF-8. A 137| | // valid continuation byte starts with 10. 138| | // range check - 139| 27.1k| uint32_t code_point = 140| 27.1k| (leading_byte & 0b00011111) << 6 | 141| 27.1k| (data[pos + 1] & 142| 27.1k| 0b00111111); // assembles the Unicode code point from the two bytes. 143| | // It does this by discarding the leading 110 and 10 144| | // bits from the two bytes, shifting the remaining bits 145| | // of the first byte, and then combining the results 146| | // with a bitwise OR operation. 147| 27.1k| if (code_point < 0x80) { ------------------ | Branch (147:11): [True: 12, False: 27.1k] ------------------ 148| 12| return result(error_code::OVERLONG, pos); 149| 12| } 150| 27.1k| if (0xFF < code_point) { ------------------ | Branch (150:11): [True: 48, False: 27.0k] ------------------ 151| 48| return result(error_code::TOO_LARGE, pos); 152| 48| } // We only care about the range 129-255 which is Non-ASCII latin1 153| | // characters 154| 27.0k| *latin_output++ = char(code_point); 155| 27.0k| pos += 2; 156| 27.0k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (156:16): [True: 60, False: 436] ------------------ 157| | // We have a three-byte UTF-8 158| 60| return result(error_code::TOO_LARGE, pos); 159| 436| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (159:16): [True: 54, False: 382] ------------------ 160| | // we have a 4-byte UTF-8 word. 161| 54| return result(error_code::TOO_LARGE, pos); 162| 382| } else { 163| | // we either have too many continuation bytes or an invalid leading byte 164| 382| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (164:11): [True: 196, False: 186] ------------------ 165| 196| return result(error_code::TOO_LONG, pos); 166| 196| } 167| | 168| 186| return result(error_code::HEADER_BITS, pos); 169| 382| } 170| 217k| } 171| 670| return result(error_code::SUCCESS, latin_output - start); 172| 1.43k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf8_to_latin17convertIPKcPcQaasr7simdutf6detailE22indexes_into_byte_likeIT_Esr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mS8_: 17| 1.22k| OutputPtr latin_output) { 18| 1.22k| size_t pos = 0; 19| 1.22k| auto start = latin_output; 20| | 21| 1.49M| while (pos < len) { ------------------ | Branch (21:10): [True: 1.49M, False: 508] ------------------ 22| |#if SIMDUTF_CPLUSPLUS23 23| | if !consteval 24| |#endif 25| 1.49M| { 26| | // try to convert the next block of 16 ASCII bytes 27| 1.49M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (27:11): [True: 1.49M, False: 3.89k] ------------------ 28| | // they are ascii 29| 1.49M| uint64_t v1; 30| 1.49M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 31| 1.49M| uint64_t v2; 32| 1.49M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 33| 1.49M| uint64_t v{v1 | v2}; // We are only interested in these bits: 1000 1000 34| | // 1000 1000 .... etc 35| 1.49M| if ((v & 0x8080808080808080) == ------------------ | Branch (35:13): [True: 1.35M, False: 134k] ------------------ 36| 1.49M| 0) { // if NONE of these are set, e.g. all of them are zero, then 37| | // everything is ASCII 38| 1.35M| size_t final_pos = pos + 16; 39| 23.1M| while (pos < final_pos) { ------------------ | Branch (39:18): [True: 21.7M, False: 1.35M] ------------------ 40| 21.7M| *latin_output++ = char(data[pos]); 41| 21.7M| pos++; 42| 21.7M| } 43| 1.35M| continue; 44| 1.35M| } 45| 1.49M| } 46| 1.49M| } 47| | 48| | // suppose it is not an all ASCII byte sequence 49| 138k| uint8_t leading_byte = data[pos]; // leading byte 50| 138k| if (leading_byte < 0b10000000) { ------------------ | Branch (50:9): [True: 121k, False: 16.6k] ------------------ 51| | // converting one ASCII byte !!! 52| 121k| *latin_output++ = char(leading_byte); 53| 121k| pos++; 54| 121k| } else if ((leading_byte & 0b11100000) == ------------------ | Branch (54:16): [True: 16.1k, False: 504] ------------------ 55| 16.6k| 0b11000000) { // the first three bits indicate: 56| | // We have a two-byte UTF-8 57| 16.1k| if (pos + 1 >= len) { ------------------ | Branch (57:11): [True: 36, False: 16.0k] ------------------ 58| 36| return 0; 59| 36| } // minimal bound checking 60| 16.0k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (60:11): [True: 96, False: 15.9k] ------------------ 61| 96| return 0; 62| 96| } // checks if the next byte is a valid continuation byte in UTF-8. A 63| | // valid continuation byte starts with 10. 64| | // range check - 65| 15.9k| uint32_t code_point = 66| 15.9k| (leading_byte & 0b00011111) << 6 | 67| 15.9k| (data[pos + 1] & 68| 15.9k| 0b00111111); // assembles the Unicode code point from the two bytes. 69| | // It does this by discarding the leading 110 and 10 70| | // bits from the two bytes, shifting the remaining bits 71| | // of the first byte, and then combining the results 72| | // with a bitwise OR operation. 73| 15.9k| if (code_point < 0x80 || 0xFF < code_point) { ------------------ | Branch (73:11): [True: 16, False: 15.9k] | Branch (73:32): [True: 66, False: 15.8k] ------------------ 74| 82| return 0; // We only care about the range 129-255 which is Non-ASCII 75| | // latin1 characters. A code_point beneath 0x80 is invalid as 76| | // it is already covered by bytes whose leading bit is zero. 77| 82| } 78| 15.8k| *latin_output++ = char(code_point); 79| 15.8k| pos += 2; 80| 15.8k| } else { 81| 504| return 0; 82| 504| } 83| 138k| } 84| 508| return latin_output - start; 85| 1.22k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf167convertILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.66k| char16_t *utf16_output) { 17| 1.66k| size_t pos = 0; 18| 1.66k| char16_t *start{utf16_output}; 19| 1.47M| while (pos < len) { ------------------ | Branch (19:10): [True: 1.47M, False: 728] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| | // try to convert the next block of 16 ASCII bytes 24| 1.47M| { 25| 1.47M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (25:11): [True: 1.47M, False: 4.96k] ------------------ 26| | // they are ascii 27| 1.47M| uint64_t v1; 28| 1.47M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 29| 1.47M| uint64_t v2; 30| 1.47M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 31| 1.47M| uint64_t v{v1 | v2}; 32| 1.47M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (32:13): [True: 917k, False: 556k] ------------------ 33| 917k| size_t final_pos = pos + 16; 34| 15.5M| while (pos < final_pos) { ------------------ | Branch (34:18): [True: 14.6M, False: 917k] ------------------ 35| 14.6M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (35:31): [Folded, False: 14.6M] ------------------ 36| 14.6M| ? char16_t(u16_swap_bytes(data[pos])) 37| 14.6M| : char16_t(data[pos]); 38| 14.6M| pos++; 39| 14.6M| } 40| 917k| continue; 41| 917k| } 42| 1.47M| } 43| 1.47M| } 44| | 45| 561k| uint8_t leading_byte = data[pos]; // leading byte 46| 561k| if (leading_byte < 0b10000000) { ------------------ | Branch (46:9): [True: 444k, False: 116k] ------------------ 47| | // converting one ASCII byte !!! 48| 444k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (48:25): [Folded, False: 444k] ------------------ 49| 444k| ? char16_t(u16_swap_bytes(leading_byte)) 50| 444k| : char16_t(leading_byte); 51| 444k| pos++; 52| 444k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (52:16): [True: 76.0k, False: 40.7k] ------------------ 53| | // We have a two-byte UTF-8, it should become 54| | // a single UTF-16 word. 55| 76.0k| if (pos + 1 >= len) { ------------------ | Branch (55:11): [True: 18, False: 76.0k] ------------------ 56| 18| return 0; 57| 18| } // minimal bound checking 58| 76.0k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (58:11): [True: 98, False: 75.9k] ------------------ 59| 98| return 0; 60| 98| } 61| | // range check 62| 75.9k| uint32_t code_point = 63| 75.9k| (leading_byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 64| 75.9k| if (code_point < 0x80) { ------------------ | Branch (64:11): [True: 20, False: 75.9k] ------------------ 65| 20| return 0; 66| 20| } 67| | if constexpr (!match_system(big_endian)) { 68| | code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 69| | } 70| 75.9k| *utf16_output++ = char16_t(code_point); 71| 75.9k| pos += 2; 72| 75.9k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (72:16): [True: 33.8k, False: 6.92k] ------------------ 73| | // We have a three-byte UTF-8, it should become 74| | // a single UTF-16 word. 75| 33.8k| if (pos + 2 >= len) { ------------------ | Branch (75:11): [True: 60, False: 33.7k] ------------------ 76| 60| return 0; 77| 60| } // minimal bound checking 78| | 79| 33.7k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (79:11): [True: 54, False: 33.7k] ------------------ 80| 54| return 0; 81| 54| } 82| 33.7k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (82:11): [True: 40, False: 33.6k] ------------------ 83| 40| return 0; 84| 40| } 85| | // range check 86| 33.6k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 87| 33.6k| (data[pos + 1] & 0b00111111) << 6 | 88| 33.6k| (data[pos + 2] & 0b00111111); 89| 33.6k| if (code_point < 0x800 || (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (89:11): [True: 46, False: 33.6k] | Branch (89:34): [True: 772, False: 32.8k] | Branch (89:57): [True: 18, False: 754] ------------------ 90| 64| return 0; 91| 64| } 92| | if constexpr (!match_system(big_endian)) { 93| | code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 94| | } 95| 33.6k| *utf16_output++ = char16_t(code_point); 96| 33.6k| pos += 3; 97| 33.6k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (97:16): [True: 6.51k, False: 408] ------------------ 98| | // we have a 4-byte UTF-8 word. 99| 6.51k| if (pos + 3 >= len) { ------------------ | Branch (99:11): [True: 24, False: 6.49k] ------------------ 100| 24| return 0; 101| 24| } // minimal bound checking 102| 6.49k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (102:11): [True: 34, False: 6.45k] ------------------ 103| 34| return 0; 104| 34| } 105| 6.45k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (105:11): [True: 10, False: 6.44k] ------------------ 106| 10| return 0; 107| 10| } 108| 6.44k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (108:11): [True: 40, False: 6.40k] ------------------ 109| 40| return 0; 110| 40| } 111| | 112| | // range check 113| 6.40k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 114| 6.40k| (data[pos + 1] & 0b00111111) << 12 | 115| 6.40k| (data[pos + 2] & 0b00111111) << 6 | 116| 6.40k| (data[pos + 3] & 0b00111111); 117| 6.40k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (117:11): [True: 36, False: 6.37k] | Branch (117:35): [True: 34, False: 6.33k] ------------------ 118| 70| return 0; 119| 70| } 120| 6.33k| code_point -= 0x10000; 121| 6.33k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 122| 6.33k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 123| | if constexpr (!match_system(big_endian)) { 124| | high_surrogate = u16_swap_bytes(high_surrogate); 125| | low_surrogate = u16_swap_bytes(low_surrogate); 126| | } 127| 6.33k| *utf16_output++ = char16_t(high_surrogate); 128| 6.33k| *utf16_output++ = char16_t(low_surrogate); 129| 6.33k| pos += 4; 130| 6.33k| } else { 131| 408| return 0; 132| 408| } 133| 561k| } 134| 728| return utf16_output - start; 135| 1.66k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf167convertILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.67k| char16_t *utf16_output) { 17| 1.67k| size_t pos = 0; 18| 1.67k| char16_t *start{utf16_output}; 19| 2.30M| while (pos < len) { ------------------ | Branch (19:10): [True: 2.30M, False: 704] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| | // try to convert the next block of 16 ASCII bytes 24| 2.30M| { 25| 2.30M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (25:11): [True: 2.29M, False: 5.03k] ------------------ 26| | // they are ascii 27| 2.29M| uint64_t v1; 28| 2.29M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 29| 2.29M| uint64_t v2; 30| 2.29M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 31| 2.29M| uint64_t v{v1 | v2}; 32| 2.29M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (32:13): [True: 1.37M, False: 917k] ------------------ 33| 1.37M| size_t final_pos = pos + 16; 34| 23.4M| while (pos < final_pos) { ------------------ | Branch (34:18): [True: 22.0M, False: 1.37M] ------------------ 35| 22.0M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (35:31): [True: 22.0M, Folded] ------------------ 36| 22.0M| ? char16_t(u16_swap_bytes(data[pos])) 37| 22.0M| : char16_t(data[pos]); 38| 22.0M| pos++; 39| 22.0M| } 40| 1.37M| continue; 41| 1.37M| } 42| 2.29M| } 43| 2.30M| } 44| | 45| 922k| uint8_t leading_byte = data[pos]; // leading byte 46| 922k| if (leading_byte < 0b10000000) { ------------------ | Branch (46:9): [True: 751k, False: 171k] ------------------ 47| | // converting one ASCII byte !!! 48| 751k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (48:25): [True: 751k, Folded] ------------------ 49| 751k| ? char16_t(u16_swap_bytes(leading_byte)) 50| 751k| : char16_t(leading_byte); 51| 751k| pos++; 52| 751k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (52:16): [True: 107k, False: 63.4k] ------------------ 53| | // We have a two-byte UTF-8, it should become 54| | // a single UTF-16 word. 55| 107k| if (pos + 1 >= len) { ------------------ | Branch (55:11): [True: 78, False: 107k] ------------------ 56| 78| return 0; 57| 78| } // minimal bound checking 58| 107k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (58:11): [True: 80, False: 107k] ------------------ 59| 80| return 0; 60| 80| } 61| | // range check 62| 107k| uint32_t code_point = 63| 107k| (leading_byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 64| 107k| if (code_point < 0x80) { ------------------ | Branch (64:11): [True: 24, False: 107k] ------------------ 65| 24| return 0; 66| 24| } 67| 107k| if constexpr (!match_system(big_endian)) { 68| 107k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 69| 107k| } 70| 107k| *utf16_output++ = char16_t(code_point); 71| 107k| pos += 2; 72| 107k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (72:16): [True: 52.2k, False: 11.2k] ------------------ 73| | // We have a three-byte UTF-8, it should become 74| | // a single UTF-16 word. 75| 52.2k| if (pos + 2 >= len) { ------------------ | Branch (75:11): [True: 24, False: 52.1k] ------------------ 76| 24| return 0; 77| 24| } // minimal bound checking 78| | 79| 52.1k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (79:11): [True: 64, False: 52.1k] ------------------ 80| 64| return 0; 81| 64| } 82| 52.1k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (82:11): [True: 32, False: 52.0k] ------------------ 83| 32| return 0; 84| 32| } 85| | // range check 86| 52.0k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 87| 52.0k| (data[pos + 1] & 0b00111111) << 6 | 88| 52.0k| (data[pos + 2] & 0b00111111); 89| 52.0k| if (code_point < 0x800 || (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (89:11): [True: 32, False: 52.0k] | Branch (89:34): [True: 706, False: 51.3k] | Branch (89:57): [True: 14, False: 692] ------------------ 90| 46| return 0; 91| 46| } 92| 52.0k| if constexpr (!match_system(big_endian)) { 93| 52.0k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 94| 52.0k| } 95| 52.0k| *utf16_output++ = char16_t(code_point); 96| 52.0k| pos += 3; 97| 52.0k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (97:16): [True: 10.8k, False: 394] ------------------ 98| | // we have a 4-byte UTF-8 word. 99| 10.8k| if (pos + 3 >= len) { ------------------ | Branch (99:11): [True: 48, False: 10.8k] ------------------ 100| 48| return 0; 101| 48| } // minimal bound checking 102| 10.8k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (102:11): [True: 42, False: 10.7k] ------------------ 103| 42| return 0; 104| 42| } 105| 10.7k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (105:11): [True: 36, False: 10.7k] ------------------ 106| 36| return 0; 107| 36| } 108| 10.7k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (108:11): [True: 32, False: 10.7k] ------------------ 109| 32| return 0; 110| 32| } 111| | 112| | // range check 113| 10.7k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 114| 10.7k| (data[pos + 1] & 0b00111111) << 12 | 115| 10.7k| (data[pos + 2] & 0b00111111) << 6 | 116| 10.7k| (data[pos + 3] & 0b00111111); 117| 10.7k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (117:11): [True: 46, False: 10.6k] | Branch (117:35): [True: 24, False: 10.6k] ------------------ 118| 70| return 0; 119| 70| } 120| 10.6k| code_point -= 0x10000; 121| 10.6k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 122| 10.6k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 123| 10.6k| if constexpr (!match_system(big_endian)) { 124| 10.6k| high_surrogate = u16_swap_bytes(high_surrogate); 125| 10.6k| low_surrogate = u16_swap_bytes(low_surrogate); 126| 10.6k| } 127| 10.6k| *utf16_output++ = char16_t(high_surrogate); 128| 10.6k| *utf16_output++ = char16_t(low_surrogate); 129| 10.6k| pos += 4; 130| 10.6k| } else { 131| 394| return 0; 132| 394| } 133| 922k| } 134| 704| return utf16_output - start; 135| 1.67k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1619convert_with_errorsILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEENS_6resultES7_mPDs: 142| 1.73k| char16_t *utf16_output) { 143| 1.73k| size_t pos = 0; 144| 1.73k| char16_t *start{utf16_output}; 145| 2.88M| while (pos < len) { ------------------ | Branch (145:10): [True: 2.88M, False: 718] ------------------ 146| |#if SIMDUTF_CPLUSPLUS23 147| | if !consteval 148| |#endif 149| 2.88M| { 150| | // try to convert the next block of 16 ASCII bytes 151| 2.88M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (151:11): [True: 2.87M, False: 5.79k] ------------------ 152| | // they are ascii 153| 2.87M| uint64_t v1; 154| 2.87M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 155| 2.87M| uint64_t v2; 156| 2.87M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 157| 2.87M| uint64_t v{v1 | v2}; 158| 2.87M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (158:13): [True: 1.36M, False: 1.50M] ------------------ 159| 1.36M| size_t final_pos = pos + 16; 160| 23.2M| while (pos < final_pos) { ------------------ | Branch (160:18): [True: 21.9M, False: 1.36M] ------------------ 161| 21.9M| const char16_t byte = uint8_t(data[pos]); 162| 21.9M| *utf16_output++ = 163| 21.9M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (163:17): [Folded, False: 21.9M] ------------------ 164| 21.9M| pos++; 165| 21.9M| } 166| 1.36M| continue; 167| 1.36M| } 168| 2.87M| } 169| 2.88M| } 170| | 171| 1.51M| auto leading_byte = uint8_t(data[pos]); // leading byte 172| 1.51M| if (leading_byte < 0b10000000) { ------------------ | Branch (172:9): [True: 1.24M, False: 273k] ------------------ 173| | // converting one ASCII byte !!! 174| 1.24M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (174:25): [Folded, False: 1.24M] ------------------ 175| 1.24M| ? char16_t(u16_swap_bytes(leading_byte)) 176| 1.24M| : char16_t(leading_byte); 177| 1.24M| pos++; 178| 1.24M| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (178:16): [True: 179k, False: 94.4k] ------------------ 179| | // We have a two-byte UTF-8, it should become 180| | // a single UTF-16 word. 181| 179k| if (pos + 1 >= len) { ------------------ | Branch (181:11): [True: 54, False: 179k] ------------------ 182| 54| return result(error_code::TOO_SHORT, pos); 183| 54| } // minimal bound checking 184| 179k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (184:11): [True: 144, False: 179k] ------------------ 185| 144| return result(error_code::TOO_SHORT, pos); 186| 144| } 187| | // range check 188| 179k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 189| 179k| (uint8_t(data[pos + 1]) & 0b00111111); 190| 179k| if (code_point < 0x80) { ------------------ | Branch (190:11): [True: 24, False: 179k] ------------------ 191| 24| return result(error_code::OVERLONG, pos); 192| 24| } 193| | if constexpr (!match_system(big_endian)) { 194| | code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 195| | } 196| 179k| *utf16_output++ = char16_t(code_point); 197| 179k| pos += 2; 198| 179k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (198:16): [True: 68.4k, False: 25.9k] ------------------ 199| | // We have a three-byte UTF-8, it should become 200| | // a single UTF-16 word. 201| 68.4k| if (pos + 2 >= len) { ------------------ | Branch (201:11): [True: 30, False: 68.4k] ------------------ 202| 30| return result(error_code::TOO_SHORT, pos); 203| 30| } // minimal bound checking 204| | 205| 68.4k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (205:11): [True: 108, False: 68.2k] ------------------ 206| 108| return result(error_code::TOO_SHORT, pos); 207| 108| } 208| 68.2k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (208:11): [True: 12, False: 68.2k] ------------------ 209| 12| return result(error_code::TOO_SHORT, pos); 210| 12| } 211| | // range check 212| 68.2k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 213| 68.2k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 214| 68.2k| (uint8_t(data[pos + 2]) & 0b00111111); 215| 68.2k| if (code_point < 0x800) { ------------------ | Branch (215:11): [True: 42, False: 68.2k] ------------------ 216| 42| return result(error_code::OVERLONG, pos); 217| 42| } 218| 68.2k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (218:11): [True: 1.00k, False: 67.2k] | Branch (218:34): [True: 18, False: 986] ------------------ 219| 18| return result(error_code::SURROGATE, pos); 220| 18| } 221| | if constexpr (!match_system(big_endian)) { 222| | code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 223| | } 224| 68.2k| *utf16_output++ = char16_t(code_point); 225| 68.2k| pos += 3; 226| 68.2k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (226:16): [True: 25.6k, False: 384] ------------------ 227| | // we have a 4-byte UTF-8 word. 228| 25.6k| if (pos + 3 >= len) { ------------------ | Branch (228:11): [True: 48, False: 25.5k] ------------------ 229| 48| return result(error_code::TOO_SHORT, pos); 230| 48| } // minimal bound checking 231| 25.5k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (231:11): [True: 42, False: 25.5k] ------------------ 232| 42| return result(error_code::TOO_SHORT, pos); 233| 42| } 234| 25.5k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (234:11): [True: 42, False: 25.4k] ------------------ 235| 42| return result(error_code::TOO_SHORT, pos); 236| 42| } 237| 25.4k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (237:11): [True: 12, False: 25.4k] ------------------ 238| 12| return result(error_code::TOO_SHORT, pos); 239| 12| } 240| | 241| | // range check 242| 25.4k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 243| 25.4k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 244| 25.4k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 245| 25.4k| (uint8_t(data[pos + 3]) & 0b00111111); 246| 25.4k| if (code_point <= 0xffff) { ------------------ | Branch (246:11): [True: 36, False: 25.4k] ------------------ 247| 36| return result(error_code::OVERLONG, pos); 248| 36| } 249| 25.4k| if (0x10ffff < code_point) { ------------------ | Branch (249:11): [True: 24, False: 25.4k] ------------------ 250| 24| return result(error_code::TOO_LARGE, pos); 251| 24| } 252| 25.4k| code_point -= 0x10000; 253| 25.4k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 254| 25.4k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 255| | if constexpr (!match_system(big_endian)) { 256| | high_surrogate = u16_swap_bytes(high_surrogate); 257| | low_surrogate = u16_swap_bytes(low_surrogate); 258| | } 259| 25.4k| *utf16_output++ = char16_t(high_surrogate); 260| 25.4k| *utf16_output++ = char16_t(low_surrogate); 261| 25.4k| pos += 4; 262| 25.4k| } else { 263| | // we either have too many continuation bytes or an invalid leading byte 264| 384| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (264:11): [True: 192, False: 192] ------------------ 265| 192| return result(error_code::TOO_LONG, pos); 266| 192| } else { 267| 192| return result(error_code::HEADER_BITS, pos); 268| 192| } 269| 384| } 270| 1.51M| } 271| 718| return result(error_code::SUCCESS, utf16_output - start); 272| 1.73k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1619convert_with_errorsILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEENS_6resultES7_mPDs: 142| 2.04k| char16_t *utf16_output) { 143| 2.04k| size_t pos = 0; 144| 2.04k| char16_t *start{utf16_output}; 145| 2.49M| while (pos < len) { ------------------ | Branch (145:10): [True: 2.49M, False: 928] ------------------ 146| |#if SIMDUTF_CPLUSPLUS23 147| | if !consteval 148| |#endif 149| 2.49M| { 150| | // try to convert the next block of 16 ASCII bytes 151| 2.49M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (151:11): [True: 2.48M, False: 8.98k] ------------------ 152| | // they are ascii 153| 2.48M| uint64_t v1; 154| 2.48M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 155| 2.48M| uint64_t v2; 156| 2.48M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 157| 2.48M| uint64_t v{v1 | v2}; 158| 2.48M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (158:13): [True: 1.32M, False: 1.15M] ------------------ 159| 1.32M| size_t final_pos = pos + 16; 160| 22.5M| while (pos < final_pos) { ------------------ | Branch (160:18): [True: 21.2M, False: 1.32M] ------------------ 161| 21.2M| const char16_t byte = uint8_t(data[pos]); 162| 21.2M| *utf16_output++ = 163| 21.2M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (163:17): [True: 21.2M, Folded] ------------------ 164| 21.2M| pos++; 165| 21.2M| } 166| 1.32M| continue; 167| 1.32M| } 168| 2.48M| } 169| 2.49M| } 170| | 171| 1.16M| auto leading_byte = uint8_t(data[pos]); // leading byte 172| 1.16M| if (leading_byte < 0b10000000) { ------------------ | Branch (172:9): [True: 950k, False: 213k] ------------------ 173| | // converting one ASCII byte !!! 174| 950k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (174:25): [True: 950k, Folded] ------------------ 175| 950k| ? char16_t(u16_swap_bytes(leading_byte)) 176| 950k| : char16_t(leading_byte); 177| 950k| pos++; 178| 950k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (178:16): [True: 139k, False: 74.2k] ------------------ 179| | // We have a two-byte UTF-8, it should become 180| | // a single UTF-16 word. 181| 139k| if (pos + 1 >= len) { ------------------ | Branch (181:11): [True: 30, False: 139k] ------------------ 182| 30| return result(error_code::TOO_SHORT, pos); 183| 30| } // minimal bound checking 184| 139k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (184:11): [True: 156, False: 138k] ------------------ 185| 156| return result(error_code::TOO_SHORT, pos); 186| 156| } 187| | // range check 188| 138k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 189| 138k| (uint8_t(data[pos + 1]) & 0b00111111); 190| 138k| if (code_point < 0x80) { ------------------ | Branch (190:11): [True: 12, False: 138k] ------------------ 191| 12| return result(error_code::OVERLONG, pos); 192| 12| } 193| 138k| if constexpr (!match_system(big_endian)) { 194| 138k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 195| 138k| } 196| 138k| *utf16_output++ = char16_t(code_point); 197| 138k| pos += 2; 198| 138k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (198:16): [True: 64.5k, False: 9.67k] ------------------ 199| | // We have a three-byte UTF-8, it should become 200| | // a single UTF-16 word. 201| 64.5k| if (pos + 2 >= len) { ------------------ | Branch (201:11): [True: 42, False: 64.5k] ------------------ 202| 42| return result(error_code::TOO_SHORT, pos); 203| 42| } // minimal bound checking 204| | 205| 64.5k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (205:11): [True: 102, False: 64.4k] ------------------ 206| 102| return result(error_code::TOO_SHORT, pos); 207| 102| } 208| 64.4k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (208:11): [True: 30, False: 64.4k] ------------------ 209| 30| return result(error_code::TOO_SHORT, pos); 210| 30| } 211| | // range check 212| 64.4k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 213| 64.4k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 214| 64.4k| (uint8_t(data[pos + 2]) & 0b00111111); 215| 64.4k| if (code_point < 0x800) { ------------------ | Branch (215:11): [True: 36, False: 64.3k] ------------------ 216| 36| return result(error_code::OVERLONG, pos); 217| 36| } 218| 64.3k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (218:11): [True: 2.55k, False: 61.8k] | Branch (218:34): [True: 18, False: 2.53k] ------------------ 219| 18| return result(error_code::SURROGATE, pos); 220| 18| } 221| 64.3k| if constexpr (!match_system(big_endian)) { 222| 64.3k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 223| 64.3k| } 224| 64.3k| *utf16_output++ = char16_t(code_point); 225| 64.3k| pos += 3; 226| 64.3k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (226:16): [True: 9.19k, False: 478] ------------------ 227| | // we have a 4-byte UTF-8 word. 228| 9.19k| if (pos + 3 >= len) { ------------------ | Branch (228:11): [True: 60, False: 9.13k] ------------------ 229| 60| return result(error_code::TOO_SHORT, pos); 230| 60| } // minimal bound checking 231| 9.13k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (231:11): [True: 36, False: 9.09k] ------------------ 232| 36| return result(error_code::TOO_SHORT, pos); 233| 36| } 234| 9.09k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (234:11): [True: 36, False: 9.06k] ------------------ 235| 36| return result(error_code::TOO_SHORT, pos); 236| 36| } 237| 9.06k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (237:11): [True: 36, False: 9.02k] ------------------ 238| 36| return result(error_code::TOO_SHORT, pos); 239| 36| } 240| | 241| | // range check 242| 9.02k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 243| 9.02k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 244| 9.02k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 245| 9.02k| (uint8_t(data[pos + 3]) & 0b00111111); 246| 9.02k| if (code_point <= 0xffff) { ------------------ | Branch (246:11): [True: 12, False: 9.01k] ------------------ 247| 12| return result(error_code::OVERLONG, pos); 248| 12| } 249| 9.01k| if (0x10ffff < code_point) { ------------------ | Branch (249:11): [True: 30, False: 8.98k] ------------------ 250| 30| return result(error_code::TOO_LARGE, pos); 251| 30| } 252| 8.98k| code_point -= 0x10000; 253| 8.98k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 254| 8.98k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 255| 8.98k| if constexpr (!match_system(big_endian)) { 256| 8.98k| high_surrogate = u16_swap_bytes(high_surrogate); 257| 8.98k| low_surrogate = u16_swap_bytes(low_surrogate); 258| 8.98k| } 259| 8.98k| *utf16_output++ = char16_t(high_surrogate); 260| 8.98k| *utf16_output++ = char16_t(low_surrogate); 261| 8.98k| pos += 4; 262| 8.98k| } else { 263| | // we either have too many continuation bytes or an invalid leading byte 264| 478| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (264:11): [True: 232, False: 246] ------------------ 265| 232| return result(error_code::TOO_LONG, pos); 266| 246| } else { 267| 246| return result(error_code::HEADER_BITS, pos); 268| 246| } 269| 478| } 270| 1.16M| } 271| 928| return result(error_code::SUCCESS, utf16_output - start); 272| 2.04k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1630rewind_and_convert_with_errorsILNS_10endiannessE0EEENS_6resultEmPKcmPDs: 292| 1.21k| char16_t *utf16_output) { 293| 1.21k| size_t extra_len{0}; 294| | // We potentially need to go back in time and find a leading byte. 295| | // In theory '3' would be sufficient, but sometimes the error can go back 296| | // quite far. 297| 1.21k| size_t how_far_back = prior_bytes; 298| | // size_t how_far_back = 3; // 3 bytes in the past + current position 299| | // if(how_far_back >= prior_bytes) { how_far_back = prior_bytes; } 300| 1.21k| bool found_leading_bytes{false}; 301| | // important: it is i <= how_far_back and not 'i < how_far_back'. 302| 1.29k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (302:22): [True: 1.21k, False: 80] ------------------ 303| 1.21k| unsigned char byte = buf[-static_cast(i)]; 304| 1.21k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 305| 1.21k| if (found_leading_bytes) { ------------------ | Branch (305:9): [True: 1.13k, False: 84] ------------------ 306| 1.13k| if (i > 0 && byte < 128) { ------------------ | Branch (306:11): [True: 4, False: 1.12k] | Branch (306:20): [True: 4, False: 0] ------------------ 307| | // If we had to go back and the leading byte is ascii 308| | // then we can stop right away. 309| 4| return result(error_code::TOO_LONG, 0 - i + 1); 310| 4| } 311| 1.12k| buf -= i; 312| 1.12k| extra_len = i; 313| 1.12k| break; 314| 1.13k| } 315| 1.21k| } 316| | // 317| | // It is possible for this function to return a negative count in its result. 318| | // C++ Standard Section 18.1 defines size_t is in which is described 319| | // in C Standard as . C Standard Section 4.1.5 defines size_t as an 320| | // unsigned integral type of the result of the sizeof operator 321| | // 322| | // An unsigned type will simply wrap round arithmetically (well defined). 323| | // 324| 1.20k| if (!found_leading_bytes) { ------------------ | Branch (324:7): [True: 80, False: 1.12k] ------------------ 325| | // If how_far_back == 3, we may have four consecutive continuation bytes!!! 326| | // [....] [continuation] [continuation] [continuation] | [buf is 327| | // continuation] Or we possibly have a stream that does not start with a 328| | // leading byte. 329| 80| return result(error_code::TOO_LONG, 0 - how_far_back); 330| 80| } 331| 1.12k| result res = convert_with_errors(buf, len + extra_len, utf16_output); 332| 1.12k| if (res.error) { ------------------ | Branch (332:7): [True: 652, False: 476] ------------------ 333| 652| res.count -= extra_len; 334| 652| } 335| 1.12k| return res; 336| 1.20k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1630rewind_and_convert_with_errorsILNS_10endiannessE1EEENS_6resultEmPKcmPDs: 292| 1.40k| char16_t *utf16_output) { 293| 1.40k| size_t extra_len{0}; 294| | // We potentially need to go back in time and find a leading byte. 295| | // In theory '3' would be sufficient, but sometimes the error can go back 296| | // quite far. 297| 1.40k| size_t how_far_back = prior_bytes; 298| | // size_t how_far_back = 3; // 3 bytes in the past + current position 299| | // if(how_far_back >= prior_bytes) { how_far_back = prior_bytes; } 300| 1.40k| bool found_leading_bytes{false}; 301| | // important: it is i <= how_far_back and not 'i < how_far_back'. 302| 1.47k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (302:22): [True: 1.40k, False: 64] ------------------ 303| 1.40k| unsigned char byte = buf[-static_cast(i)]; 304| 1.40k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 305| 1.40k| if (found_leading_bytes) { ------------------ | Branch (305:9): [True: 1.34k, False: 68] ------------------ 306| 1.34k| if (i > 0 && byte < 128) { ------------------ | Branch (306:11): [True: 4, False: 1.33k] | Branch (306:20): [True: 4, False: 0] ------------------ 307| | // If we had to go back and the leading byte is ascii 308| | // then we can stop right away. 309| 4| return result(error_code::TOO_LONG, 0 - i + 1); 310| 4| } 311| 1.33k| buf -= i; 312| 1.33k| extra_len = i; 313| 1.33k| break; 314| 1.34k| } 315| 1.40k| } 316| | // 317| | // It is possible for this function to return a negative count in its result. 318| | // C++ Standard Section 18.1 defines size_t is in which is described 319| | // in C Standard as . C Standard Section 4.1.5 defines size_t as an 320| | // unsigned integral type of the result of the sizeof operator 321| | // 322| | // An unsigned type will simply wrap round arithmetically (well defined). 323| | // 324| 1.40k| if (!found_leading_bytes) { ------------------ | Branch (324:7): [True: 64, False: 1.33k] ------------------ 325| | // If how_far_back == 3, we may have four consecutive continuation bytes!!! 326| | // [....] [continuation] [continuation] [continuation] | [buf is 327| | // continuation] Or we possibly have a stream that does not start with a 328| | // leading byte. 329| 64| return result(error_code::TOO_LONG, 0 - how_far_back); 330| 64| } 331| 1.33k| result res = convert_with_errors(buf, len + extra_len, utf16_output); 332| 1.33k| if (res.error) { ------------------ | Branch (332:7): [True: 720, False: 616] ------------------ 333| 720| res.count -= extra_len; 334| 720| } 335| 1.33k| return res; 336| 1.40k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.35k| char16_t *utf16_output) { 17| 1.35k| size_t pos = 0; 18| 1.35k| char16_t *start{utf16_output}; 19| 5.05M| while (pos < len) { ------------------ | Branch (19:10): [True: 5.05M, False: 1.35k] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 5.05M| { // try to convert the next block of 8 ASCII bytes 24| 5.05M| if (pos + 8 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (24:11): [True: 5.04M, False: 4.40k] ------------------ 25| | // they are ascii 26| 5.04M| uint64_t v; 27| 5.04M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 28| 5.04M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (28:13): [True: 3.20M, False: 1.84M] ------------------ 29| 3.20M| size_t final_pos = pos + 8; 30| 28.8M| while (pos < final_pos) { ------------------ | Branch (30:18): [True: 25.6M, False: 3.20M] ------------------ 31| 25.6M| const char16_t byte = uint8_t(data[pos]); 32| 25.6M| *utf16_output++ = 33| 25.6M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (33:17): [Folded, False: 25.6M] ------------------ 34| 25.6M| pos++; 35| 25.6M| } 36| 3.20M| continue; 37| 3.20M| } 38| 5.04M| } 39| 5.05M| } 40| | 41| 1.84M| auto leading_byte = uint8_t(data[pos]); // leading byte 42| 1.84M| if (leading_byte < 0b10000000) { ------------------ | Branch (42:9): [True: 1.23M, False: 612k] ------------------ 43| | // converting one ASCII byte !!! 44| 1.23M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (44:25): [Folded, False: 1.23M] ------------------ 45| 1.23M| ? char16_t(u16_swap_bytes(leading_byte)) 46| 1.23M| : char16_t(leading_byte); 47| 1.23M| pos++; 48| 1.23M| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (48:16): [True: 362k, False: 249k] ------------------ 49| | // We have a two-byte UTF-8, it should become 50| | // a single UTF-16 word. 51| 362k| if (pos + 1 >= len) { ------------------ | Branch (51:11): [True: 0, False: 362k] ------------------ 52| 0| break; 53| 0| } // minimal bound checking 54| 362k| uint16_t code_point = uint16_t(((leading_byte & 0b00011111) << 6) | 55| 362k| (uint8_t(data[pos + 1]) & 0b00111111)); 56| | if constexpr (!match_system(big_endian)) { 57| | code_point = u16_swap_bytes(uint16_t(code_point)); 58| | } 59| 362k| *utf16_output++ = char16_t(code_point); 60| 362k| pos += 2; 61| 362k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (61:16): [True: 229k, False: 19.7k] ------------------ 62| | // We have a three-byte UTF-8, it should become 63| | // a single UTF-16 word. 64| 229k| if (pos + 2 >= len) { ------------------ | Branch (64:11): [True: 0, False: 229k] ------------------ 65| 0| break; 66| 0| } // minimal bound checking 67| 229k| uint16_t code_point = 68| 229k| uint16_t(((leading_byte & 0b00001111) << 12) | 69| 229k| ((uint8_t(data[pos + 1]) & 0b00111111) << 6) | 70| 229k| (uint8_t(data[pos + 2]) & 0b00111111)); 71| | if constexpr (!match_system(big_endian)) { 72| | code_point = u16_swap_bytes(uint16_t(code_point)); 73| | } 74| 229k| *utf16_output++ = char16_t(code_point); 75| 229k| pos += 3; 76| 229k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (76:16): [True: 19.7k, False: 0] ------------------ 77| | // we have a 4-byte UTF-8 word. 78| 19.7k| if (pos + 3 >= len) { ------------------ | Branch (78:11): [True: 0, False: 19.7k] ------------------ 79| 0| break; 80| 0| } // minimal bound checking 81| 19.7k| uint32_t code_point = ((leading_byte & 0b00000111) << 18) | 82| 19.7k| ((uint8_t(data[pos + 1]) & 0b00111111) << 12) | 83| 19.7k| ((uint8_t(data[pos + 2]) & 0b00111111) << 6) | 84| 19.7k| (uint8_t(data[pos + 3]) & 0b00111111); 85| 19.7k| code_point -= 0x10000; 86| 19.7k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 87| 19.7k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 88| | if constexpr (!match_system(big_endian)) { 89| | high_surrogate = u16_swap_bytes(high_surrogate); 90| | low_surrogate = u16_swap_bytes(low_surrogate); 91| | } 92| 19.7k| *utf16_output++ = char16_t(high_surrogate); 93| 19.7k| *utf16_output++ = char16_t(low_surrogate); 94| 19.7k| pos += 4; 95| 19.7k| } else { 96| | // we may have a continuation but we do not do error checking 97| 0| return 0; 98| 0| } 99| 1.84M| } 100| 1.35k| return utf16_output - start; 101| 1.35k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.53k| char16_t *utf16_output) { 17| 1.53k| size_t pos = 0; 18| 1.53k| char16_t *start{utf16_output}; 19| 4.11M| while (pos < len) { ------------------ | Branch (19:10): [True: 4.11M, False: 1.53k] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 4.11M| { // try to convert the next block of 8 ASCII bytes 24| 4.11M| if (pos + 8 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (24:11): [True: 4.10M, False: 4.72k] ------------------ 25| | // they are ascii 26| 4.10M| uint64_t v; 27| 4.10M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 28| 4.10M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (28:13): [True: 3.21M, False: 895k] ------------------ 29| 3.21M| size_t final_pos = pos + 8; 30| 28.9M| while (pos < final_pos) { ------------------ | Branch (30:18): [True: 25.7M, False: 3.21M] ------------------ 31| 25.7M| const char16_t byte = uint8_t(data[pos]); 32| 25.7M| *utf16_output++ = 33| 25.7M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (33:17): [True: 25.7M, Folded] ------------------ 34| 25.7M| pos++; 35| 25.7M| } 36| 3.21M| continue; 37| 3.21M| } 38| 4.10M| } 39| 4.11M| } 40| | 41| 899k| auto leading_byte = uint8_t(data[pos]); // leading byte 42| 899k| if (leading_byte < 0b10000000) { ------------------ | Branch (42:9): [True: 521k, False: 378k] ------------------ 43| | // converting one ASCII byte !!! 44| 521k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (44:25): [True: 521k, Folded] ------------------ 45| 521k| ? char16_t(u16_swap_bytes(leading_byte)) 46| 521k| : char16_t(leading_byte); 47| 521k| pos++; 48| 521k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (48:16): [True: 292k, False: 85.9k] ------------------ 49| | // We have a two-byte UTF-8, it should become 50| | // a single UTF-16 word. 51| 292k| if (pos + 1 >= len) { ------------------ | Branch (51:11): [True: 0, False: 292k] ------------------ 52| 0| break; 53| 0| } // minimal bound checking 54| 292k| uint16_t code_point = uint16_t(((leading_byte & 0b00011111) << 6) | 55| 292k| (uint8_t(data[pos + 1]) & 0b00111111)); 56| 292k| if constexpr (!match_system(big_endian)) { 57| 292k| code_point = u16_swap_bytes(uint16_t(code_point)); 58| 292k| } 59| 292k| *utf16_output++ = char16_t(code_point); 60| 292k| pos += 2; 61| 292k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (61:16): [True: 56.1k, False: 29.8k] ------------------ 62| | // We have a three-byte UTF-8, it should become 63| | // a single UTF-16 word. 64| 56.1k| if (pos + 2 >= len) { ------------------ | Branch (64:11): [True: 0, False: 56.1k] ------------------ 65| 0| break; 66| 0| } // minimal bound checking 67| 56.1k| uint16_t code_point = 68| 56.1k| uint16_t(((leading_byte & 0b00001111) << 12) | 69| 56.1k| ((uint8_t(data[pos + 1]) & 0b00111111) << 6) | 70| 56.1k| (uint8_t(data[pos + 2]) & 0b00111111)); 71| 56.1k| if constexpr (!match_system(big_endian)) { 72| 56.1k| code_point = u16_swap_bytes(uint16_t(code_point)); 73| 56.1k| } 74| 56.1k| *utf16_output++ = char16_t(code_point); 75| 56.1k| pos += 3; 76| 56.1k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (76:16): [True: 29.8k, False: 0] ------------------ 77| | // we have a 4-byte UTF-8 word. 78| 29.8k| if (pos + 3 >= len) { ------------------ | Branch (78:11): [True: 0, False: 29.8k] ------------------ 79| 0| break; 80| 0| } // minimal bound checking 81| 29.8k| uint32_t code_point = ((leading_byte & 0b00000111) << 18) | 82| 29.8k| ((uint8_t(data[pos + 1]) & 0b00111111) << 12) | 83| 29.8k| ((uint8_t(data[pos + 2]) & 0b00111111) << 6) | 84| 29.8k| (uint8_t(data[pos + 3]) & 0b00111111); 85| 29.8k| code_point -= 0x10000; 86| 29.8k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 87| 29.8k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 88| 29.8k| if constexpr (!match_system(big_endian)) { 89| 29.8k| high_surrogate = u16_swap_bytes(high_surrogate); 90| 29.8k| low_surrogate = u16_swap_bytes(low_surrogate); 91| 29.8k| } 92| 29.8k| *utf16_output++ = char16_t(high_surrogate); 93| 29.8k| *utf16_output++ = char16_t(low_surrogate); 94| 29.8k| pos += 4; 95| 29.8k| } else { 96| | // we may have a continuation but we do not do error checking 97| 0| return 0; 98| 0| } 99| 899k| } 100| 1.53k| return utf16_output - start; 101| 1.53k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf3230rewind_and_convert_with_errorsEmPKcmPDi: 249| 1.48k| char32_t *utf32_output) { 250| 1.48k| size_t extra_len{0}; 251| | // We potentially need to go back in time and find a leading byte. 252| 1.48k| size_t how_far_back = 3; // 3 bytes in the past + current position 253| 1.48k| if (how_far_back > prior_bytes) { ------------------ | Branch (253:7): [True: 800, False: 684] ------------------ 254| 800| how_far_back = prior_bytes; 255| 800| } 256| 1.48k| bool found_leading_bytes{false}; 257| | // important: it is i <= how_far_back and not 'i < how_far_back'. 258| 1.57k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (258:22): [True: 1.48k, False: 88] ------------------ 259| 1.48k| unsigned char byte = buf[-static_cast(i)]; 260| 1.48k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 261| 1.48k| if (found_leading_bytes) { ------------------ | Branch (261:9): [True: 1.39k, False: 92] ------------------ 262| 1.39k| if (i > 0 && byte < 128) { ------------------ | Branch (262:11): [True: 4, False: 1.39k] | Branch (262:20): [True: 4, False: 0] ------------------ 263| | // If we had to go back and the leading byte is ascii 264| | // then we can stop right away. 265| 4| return result(error_code::TOO_LONG, 0 - i + 1); 266| 4| } 267| 1.39k| buf -= i; 268| 1.39k| extra_len = i; 269| 1.39k| break; 270| 1.39k| } 271| 1.48k| } 272| | // 273| | // It is possible for this function to return a negative count in its result. 274| | // C++ Standard Section 18.1 defines size_t is in which is described 275| | // in C Standard as . C Standard Section 4.1.5 defines size_t as an 276| | // unsigned integral type of the result of the sizeof operator 277| | // 278| | // An unsigned type will simply wrap round arithmetically (well defined). 279| | // 280| 1.48k| if (!found_leading_bytes) { ------------------ | Branch (280:7): [True: 88, False: 1.39k] ------------------ 281| | // If how_far_back == 3, we may have four consecutive continuation bytes!!! 282| | // [....] [continuation] [continuation] [continuation] | [buf is 283| | // continuation] Or we possibly have a stream that does not start with a 284| | // leading byte. 285| 88| return result(error_code::TOO_LONG, 0 - how_far_back); 286| 88| } 287| | 288| 1.39k| result res = convert_with_errors(buf, len + extra_len, utf32_output); 289| 1.39k| if (res.error) { ------------------ | Branch (289:7): [True: 776, False: 616] ------------------ 290| 776| res.count -= extra_len; 291| 776| } 292| 1.39k| return res; 293| 1.48k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf3219convert_with_errorsIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEENS_6resultES6_mPDi: 121| 2.13k| char32_t *utf32_output) { 122| 2.13k| size_t pos = 0; 123| 2.13k| char32_t *start{utf32_output}; 124| 2.79M| while (pos < len) { ------------------ | Branch (124:10): [True: 2.79M, False: 928] ------------------ 125| |#if SIMDUTF_CPLUSPLUS23 126| | if !consteval 127| |#endif 128| 2.79M| { 129| | // try to convert the next block of 16 ASCII bytes 130| 2.79M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (130:11): [True: 2.78M, False: 8.18k] ------------------ 131| | // they are ascii 132| 2.78M| uint64_t v1; 133| 2.78M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 134| 2.78M| uint64_t v2; 135| 2.78M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 136| 2.78M| uint64_t v{v1 | v2}; 137| 2.78M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (137:13): [True: 1.61M, False: 1.16M] ------------------ 138| 1.61M| size_t final_pos = pos + 16; 139| 27.5M| while (pos < final_pos) { ------------------ | Branch (139:18): [True: 25.8M, False: 1.61M] ------------------ 140| 25.8M| *utf32_output++ = uint8_t(data[pos]); 141| 25.8M| pos++; 142| 25.8M| } 143| 1.61M| continue; 144| 1.61M| } 145| 2.78M| } 146| 2.79M| } 147| 1.17M| auto leading_byte = uint8_t(data[pos]); // leading byte 148| 1.17M| if (leading_byte < 0b10000000) { ------------------ | Branch (148:9): [True: 946k, False: 229k] ------------------ 149| | // converting one ASCII byte !!! 150| 946k| *utf32_output++ = char32_t(leading_byte); 151| 946k| pos++; 152| 946k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (152:16): [True: 143k, False: 85.3k] ------------------ 153| | // We have a two-byte UTF-8 154| 143k| if (pos + 1 >= len) { ------------------ | Branch (154:11): [True: 60, False: 143k] ------------------ 155| 60| return result(error_code::TOO_SHORT, pos); 156| 60| } // minimal bound checking 157| 143k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (157:11): [True: 150, False: 143k] ------------------ 158| 150| return result(error_code::TOO_SHORT, pos); 159| 150| } 160| | // range check 161| 143k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 162| 143k| (uint8_t(data[pos + 1]) & 0b00111111); 163| 143k| if (code_point < 0x80) { ------------------ | Branch (163:11): [True: 36, False: 143k] ------------------ 164| 36| return result(error_code::OVERLONG, pos); 165| 36| } 166| 143k| *utf32_output++ = char32_t(code_point); 167| 143k| pos += 2; 168| 143k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (168:16): [True: 77.0k, False: 8.31k] ------------------ 169| | // We have a three-byte UTF-8 170| 77.0k| if (pos + 2 >= len) { ------------------ | Branch (170:11): [True: 78, False: 76.9k] ------------------ 171| 78| return result(error_code::TOO_SHORT, pos); 172| 78| } // minimal bound checking 173| | 174| 76.9k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (174:11): [True: 132, False: 76.8k] ------------------ 175| 132| return result(error_code::TOO_SHORT, pos); 176| 132| } 177| 76.8k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (177:11): [True: 18, False: 76.8k] ------------------ 178| 18| return result(error_code::TOO_SHORT, pos); 179| 18| } 180| | // range check 181| 76.8k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 182| 76.8k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 183| 76.8k| (uint8_t(data[pos + 2]) & 0b00111111); 184| 76.8k| if (code_point < 0x800) { ------------------ | Branch (184:11): [True: 36, False: 76.7k] ------------------ 185| 36| return result(error_code::OVERLONG, pos); 186| 36| } 187| 76.7k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (187:11): [True: 3.78k, False: 73.0k] | Branch (187:34): [True: 18, False: 3.76k] ------------------ 188| 18| return result(error_code::SURROGATE, pos); 189| 18| } 190| 76.7k| *utf32_output++ = char32_t(code_point); 191| 76.7k| pos += 3; 192| 76.7k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (192:16): [True: 7.89k, False: 424] ------------------ 193| | // we have a 4-byte UTF-8 word. 194| 7.89k| if (pos + 3 >= len) { ------------------ | Branch (194:11): [True: 36, False: 7.85k] ------------------ 195| 36| return result(error_code::TOO_SHORT, pos); 196| 36| } // minimal bound checking 197| 7.85k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (197:11): [True: 90, False: 7.76k] ------------------ 198| 90| return result(error_code::TOO_SHORT, pos); 199| 90| } 200| 7.76k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (200:11): [True: 36, False: 7.73k] ------------------ 201| 36| return result(error_code::TOO_SHORT, pos); 202| 36| } 203| 7.73k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (203:11): [True: 36, False: 7.69k] ------------------ 204| 36| return result(error_code::TOO_SHORT, pos); 205| 36| } 206| | 207| | // range check 208| 7.69k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 209| 7.69k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 210| 7.69k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 211| 7.69k| (uint8_t(data[pos + 3]) & 0b00111111); 212| 7.69k| if (code_point <= 0xffff) { ------------------ | Branch (212:11): [True: 30, False: 7.66k] ------------------ 213| 30| return result(error_code::OVERLONG, pos); 214| 30| } 215| 7.66k| if (0x10ffff < code_point) { ------------------ | Branch (215:11): [True: 30, False: 7.63k] ------------------ 216| 30| return result(error_code::TOO_LARGE, pos); 217| 30| } 218| 7.63k| *utf32_output++ = char32_t(code_point); 219| 7.63k| pos += 4; 220| 7.63k| } else { 221| | // we either have too many continuation bytes or an invalid leading byte 222| 424| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (222:11): [True: 202, False: 222] ------------------ 223| 202| return result(error_code::TOO_LONG, pos); 224| 222| } else { 225| 222| return result(error_code::HEADER_BITS, pos); 226| 222| } 227| 424| } 228| 1.17M| } 229| 928| return result(error_code::SUCCESS, utf32_output - start); 230| 2.13k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf327convertIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mPDi: 16| 1.72k| char32_t *utf32_output) { 17| 1.72k| size_t pos = 0; 18| 1.72k| char32_t *start{utf32_output}; 19| 2.05M| while (pos < len) { ------------------ | Branch (19:10): [True: 2.05M, False: 788] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 2.05M| { 24| | // try to convert the next block of 16 ASCII bytes 25| 2.05M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (25:11): [True: 2.04M, False: 6.13k] ------------------ 26| | // they are ascii 27| 2.04M| uint64_t v1; 28| 2.04M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 29| 2.04M| uint64_t v2; 30| 2.04M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 31| 2.04M| uint64_t v{v1 | v2}; 32| 2.04M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (32:13): [True: 1.31M, False: 732k] ------------------ 33| 1.31M| size_t final_pos = pos + 16; 34| 22.3M| while (pos < final_pos) { ------------------ | Branch (34:18): [True: 21.0M, False: 1.31M] ------------------ 35| 21.0M| *utf32_output++ = uint8_t(data[pos]); 36| 21.0M| pos++; 37| 21.0M| } 38| 1.31M| continue; 39| 1.31M| } 40| 2.04M| } 41| 2.05M| } 42| 738k| auto leading_byte = uint8_t(data[pos]); // leading byte 43| 738k| if (leading_byte < 0b10000000) { ------------------ | Branch (43:9): [True: 610k, False: 128k] ------------------ 44| | // converting one ASCII byte !!! 45| 610k| *utf32_output++ = char32_t(leading_byte); 46| 610k| pos++; 47| 610k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (47:16): [True: 74.2k, False: 53.8k] ------------------ 48| | // We have a two-byte UTF-8 49| 74.2k| if (pos + 1 >= len) { ------------------ | Branch (49:11): [True: 54, False: 74.2k] ------------------ 50| 54| return 0; 51| 54| } // minimal bound checking 52| 74.2k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (52:11): [True: 116, False: 74.1k] ------------------ 53| 116| return 0; 54| 116| } 55| | // range check 56| 74.1k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 57| 74.1k| (uint8_t(data[pos + 1]) & 0b00111111); 58| 74.1k| if (code_point < 0x80) { ------------------ | Branch (58:11): [True: 22, False: 74.0k] ------------------ 59| 22| return 0; 60| 22| } 61| 74.0k| *utf32_output++ = char32_t(code_point); 62| 74.0k| pos += 2; 63| 74.0k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (63:16): [True: 48.9k, False: 4.86k] ------------------ 64| | // We have a three-byte UTF-8 65| 48.9k| if (pos + 2 >= len) { ------------------ | Branch (65:11): [True: 42, False: 48.9k] ------------------ 66| 42| return 0; 67| 42| } // minimal bound checking 68| | 69| 48.9k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (69:11): [True: 80, False: 48.8k] ------------------ 70| 80| return 0; 71| 80| } 72| 48.8k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (72:11): [True: 40, False: 48.8k] ------------------ 73| 40| return 0; 74| 40| } 75| | // range check 76| 48.8k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 77| 48.8k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 78| 48.8k| (uint8_t(data[pos + 2]) & 0b00111111); 79| 48.8k| if (code_point < 0x800 || (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (79:11): [True: 28, False: 48.7k] | Branch (79:34): [True: 536, False: 48.2k] | Branch (79:57): [True: 16, False: 520] ------------------ 80| 44| return 0; 81| 44| } 82| 48.7k| *utf32_output++ = char32_t(code_point); 83| 48.7k| pos += 3; 84| 48.7k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (84:16): [True: 4.50k, False: 356] ------------------ 85| | // we have a 4-byte UTF-8 word. 86| 4.50k| if (pos + 3 >= len) { ------------------ | Branch (86:11): [True: 24, False: 4.48k] ------------------ 87| 24| return 0; 88| 24| } // minimal bound checking 89| 4.48k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (89:11): [True: 36, False: 4.44k] ------------------ 90| 36| return 0; 91| 36| } 92| 4.44k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (92:11): [True: 20, False: 4.42k] ------------------ 93| 20| return 0; 94| 20| } 95| 4.42k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (95:11): [True: 12, False: 4.41k] ------------------ 96| 12| return 0; 97| 12| } 98| | 99| | // range check 100| 4.41k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 101| 4.41k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 102| 4.41k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 103| 4.41k| (uint8_t(data[pos + 3]) & 0b00111111); 104| 4.41k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (104:11): [True: 50, False: 4.36k] | Branch (104:35): [True: 42, False: 4.32k] ------------------ 105| 92| return 0; 106| 92| } 107| 4.32k| *utf32_output++ = char32_t(code_point); 108| 4.32k| pos += 4; 109| 4.32k| } else { 110| 356| return 0; 111| 356| } 112| 738k| } 113| 788| return utf32_output - start; 114| 1.72k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf3213convert_validIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mPDi: 16| 1.39k| char32_t *utf32_output) { 17| 1.39k| size_t pos = 0; 18| 1.39k| char32_t *start{utf32_output}; 19| 3.85M| while (pos < len) { ------------------ | Branch (19:10): [True: 3.85M, False: 1.39k] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 3.85M| { 24| | // try to convert the next block of 8 ASCII bytes 25| 3.85M| if (pos + 8 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (25:11): [True: 3.85M, False: 4.28k] ------------------ 26| | // they are ascii 27| 3.85M| uint64_t v; 28| 3.85M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 29| 3.85M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (29:13): [True: 3.21M, False: 634k] ------------------ 30| 3.21M| size_t final_pos = pos + 8; 31| 28.9M| while (pos < final_pos) { ------------------ | Branch (31:18): [True: 25.7M, False: 3.21M] ------------------ 32| 25.7M| *utf32_output++ = uint8_t(data[pos]); 33| 25.7M| pos++; 34| 25.7M| } 35| 3.21M| continue; 36| 3.21M| } 37| 3.85M| } 38| 3.85M| } 39| 638k| auto leading_byte = uint8_t(data[pos]); // leading byte 40| 638k| if (leading_byte < 0b10000000) { ------------------ | Branch (40:9): [True: 372k, False: 265k] ------------------ 41| | // converting one ASCII byte !!! 42| 372k| *utf32_output++ = char32_t(leading_byte); 43| 372k| pos++; 44| 372k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (44:16): [True: 194k, False: 71.7k] ------------------ 45| | // We have a two-byte UTF-8 46| 194k| if (pos + 1 >= len) { ------------------ | Branch (46:11): [True: 0, False: 194k] ------------------ 47| 0| break; 48| 0| } // minimal bound checking 49| 194k| *utf32_output++ = char32_t(((leading_byte & 0b00011111) << 6) | 50| 194k| (uint8_t(data[pos + 1]) & 0b00111111)); 51| 194k| pos += 2; 52| 194k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (52:16): [True: 45.2k, False: 26.5k] ------------------ 53| | // We have a three-byte UTF-8 54| 45.2k| if (pos + 2 >= len) { ------------------ | Branch (54:11): [True: 0, False: 45.2k] ------------------ 55| 0| break; 56| 0| } // minimal bound checking 57| 45.2k| *utf32_output++ = char32_t(((leading_byte & 0b00001111) << 12) | 58| 45.2k| ((uint8_t(data[pos + 1]) & 0b00111111) << 6) | 59| 45.2k| (uint8_t(data[pos + 2]) & 0b00111111)); 60| 45.2k| pos += 3; 61| 45.2k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (61:16): [True: 26.5k, False: 0] ------------------ 62| | // we have a 4-byte UTF-8 word. 63| 26.5k| if (pos + 3 >= len) { ------------------ | Branch (63:11): [True: 0, False: 26.5k] ------------------ 64| 0| break; 65| 0| } // minimal bound checking 66| 26.5k| uint32_t code_word = ((leading_byte & 0b00000111) << 18) | 67| 26.5k| ((uint8_t(data[pos + 1]) & 0b00111111) << 12) | 68| 26.5k| ((uint8_t(data[pos + 2]) & 0b00111111) << 6) | 69| 26.5k| (uint8_t(data[pos + 3]) & 0b00111111); 70| 26.5k| *utf32_output++ = char32_t(code_word); 71| 26.5k| pos += 4; 72| 26.5k| } else { 73| | // we may have a continuation but we do not do error checking 74| 0| return 0; 75| 0| } 76| 638k| } 77| 1.39k| return utf32_output - start; 78| 1.39k|} _ZNK7simdutf8fallback14implementation13validate_utf8EPKcm: 37| 3.48k|implementation::validate_utf8(const char *buf, size_t len) const noexcept { 38| 3.48k| return scalar::utf8::validate(buf, len); 39| 3.48k|} _ZNK7simdutf8fallback14implementation25validate_utf8_with_errorsEPKcm: 44| 3.48k| const char *buf, size_t len) const noexcept { 45| 3.48k| return scalar::utf8::validate_with_errors(buf, len); 46| 3.48k|} _ZNK7simdutf8fallback14implementation16validate_utf16leEPKDsm: 77| 1.35k| size_t len) const noexcept { 78| 1.35k| return scalar::utf16::validate(buf, len); 79| 1.35k|} _ZNK7simdutf8fallback14implementation16validate_utf16beEPKDsm: 85| 1.33k| size_t len) const noexcept { 86| 1.33k| return scalar::utf16::validate(buf, len); 87| 1.33k|} _ZNK7simdutf8fallback14implementation28validate_utf16le_with_errorsEPKDsm: 90| 1.35k| const char16_t *buf, size_t len) const noexcept { 91| 1.35k| return scalar::utf16::validate_with_errors(buf, len); 92| 1.35k|} _ZNK7simdutf8fallback14implementation28validate_utf16be_with_errorsEPKDsm: 95| 1.33k| const char16_t *buf, size_t len) const noexcept { 96| 1.33k| return scalar::utf16::validate_with_errors(buf, len); 97| 1.33k|} _ZNK7simdutf8fallback14implementation14validate_utf32EPKDim: 114| 2.63k|implementation::validate_utf32(const char32_t *buf, size_t len) const noexcept { 115| 2.63k| return scalar::utf32::validate(buf, len); 116| 2.63k|} _ZNK7simdutf8fallback14implementation26validate_utf32_with_errorsEPKDim: 121| 2.63k| const char32_t *buf, size_t len) const noexcept { 122| 2.63k| return scalar::utf32::validate_with_errors(buf, len); 123| 2.63k|} _ZNK7simdutf8fallback14implementation22convert_latin1_to_utf8EPKcmPc: 128| 604| const char *buf, size_t len, char *utf8_output) const noexcept { 129| 604| return scalar::latin1_to_utf8::convert(buf, len, utf8_output); 130| 604|} _ZNK7simdutf8fallback14implementation25convert_latin1_to_utf16leEPKcmPDs: 135| 76| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 136| 76| return scalar::latin1_to_utf16::convert(buf, len, 137| 76| utf16_output); 138| 76|} _ZNK7simdutf8fallback14implementation25convert_latin1_to_utf16beEPKcmPDs: 141| 96| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 142| 96| return scalar::latin1_to_utf16::convert(buf, len, 143| 96| utf16_output); 144| 96|} _ZNK7simdutf8fallback14implementation23convert_latin1_to_utf32EPKcmPDi: 149| 124| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 150| 124| return scalar::latin1_to_utf32::convert(buf, len, utf32_output); 151| 124|} _ZNK7simdutf8fallback14implementation22convert_utf8_to_latin1EPKcmPc: 156| 586| const char *buf, size_t len, char *latin1_output) const noexcept { 157| 586| return scalar::utf8_to_latin1::convert(buf, len, latin1_output); 158| 586|} _ZNK7simdutf8fallback14implementation34convert_utf8_to_latin1_with_errorsEPKcmPc: 161| 512| const char *buf, size_t len, char *latin1_output) const noexcept { 162| 512| return scalar::utf8_to_latin1::convert_with_errors(buf, len, latin1_output); 163| 512|} _ZNK7simdutf8fallback14implementation23convert_utf8_to_utf16leEPKcmPDs: 173| 776| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 174| 776| return scalar::utf8_to_utf16::convert(buf, len, 175| 776| utf16_output); 176| 776|} _ZNK7simdutf8fallback14implementation23convert_utf8_to_utf16beEPKcmPDs: 179| 794| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 180| 794| return scalar::utf8_to_utf16::convert(buf, len, 181| 794| utf16_output); 182| 794|} _ZNK7simdutf8fallback14implementation35convert_utf8_to_utf16le_with_errorsEPKcmPDs: 185| 610| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 186| 610| return scalar::utf8_to_utf16::convert_with_errors( 187| 610| buf, len, utf16_output); 188| 610|} _ZNK7simdutf8fallback14implementation35convert_utf8_to_utf16be_with_errorsEPKcmPDs: 191| 706| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 192| 706| return scalar::utf8_to_utf16::convert_with_errors( 193| 706| buf, len, utf16_output); 194| 706|} _ZNK7simdutf8fallback14implementation29convert_valid_utf8_to_utf16leEPKcmPDs: 197| 450| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 198| 450| return scalar::utf8_to_utf16::convert_valid(buf, len, 199| 450| utf16_output); 200| 450|} _ZNK7simdutf8fallback14implementation29convert_valid_utf8_to_utf16beEPKcmPDs: 203| 510| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 204| 510| return scalar::utf8_to_utf16::convert_valid(buf, len, 205| 510| utf16_output); 206| 510|} _ZNK7simdutf8fallback14implementation21convert_utf8_to_utf32EPKcmPDi: 211| 758| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 212| 758| return scalar::utf8_to_utf32::convert(buf, len, utf32_output); 213| 758|} _ZNK7simdutf8fallback14implementation33convert_utf8_to_utf32_with_errorsEPKcmPDi: 216| 746| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 217| 746| return scalar::utf8_to_utf32::convert_with_errors(buf, len, utf32_output); 218| 746|} _ZNK7simdutf8fallback14implementation27convert_valid_utf8_to_utf32EPKcmPDi: 221| 464| const char *input, size_t size, char32_t *utf32_output) const noexcept { 222| 464| return scalar::utf8_to_utf32::convert_valid(input, size, utf32_output); 223| 464|} _ZNK7simdutf8fallback14implementation25convert_utf16le_to_latin1EPKDsmPc: 228| 124| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 229| 124| return scalar::utf16_to_latin1::convert(buf, len, 230| 124| latin1_output); 231| 124|} _ZNK7simdutf8fallback14implementation25convert_utf16be_to_latin1EPKDsmPc: 234| 114| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 235| 114| return scalar::utf16_to_latin1::convert(buf, len, 236| 114| latin1_output); 237| 114|} _ZNK7simdutf8fallback14implementation37convert_utf16le_to_latin1_with_errorsEPKDsmPc: 241| 278| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 242| 278| return scalar::utf16_to_latin1::convert_with_errors( 243| 278| buf, len, latin1_output); 244| 278|} _ZNK7simdutf8fallback14implementation37convert_utf16be_to_latin1_with_errorsEPKDsmPc: 248| 266| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 249| 266| return scalar::utf16_to_latin1::convert_with_errors( 250| 266| buf, len, latin1_output); 251| 266|} _ZNK7simdutf8fallback14implementation23convert_utf16le_to_utf8EPKDsmPc: 268| 436| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 269| 436| return scalar::utf16_to_utf8::convert(buf, len, 270| 436| utf8_output); 271| 436|} _ZNK7simdutf8fallback14implementation23convert_utf16be_to_utf8EPKDsmPc: 274| 346| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 275| 346| return scalar::utf16_to_utf8::convert(buf, len, utf8_output); 276| 346|} _ZNK7simdutf8fallback14implementation35convert_utf16le_to_utf8_with_errorsEPKDsmPc: 279| 642| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 280| 642| return scalar::utf16_to_utf8::convert_with_errors( 281| 642| buf, len, utf8_output); 282| 642|} _ZNK7simdutf8fallback14implementation35convert_utf16be_to_utf8_with_errorsEPKDsmPc: 285| 654| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 286| 654| return scalar::utf16_to_utf8::convert_with_errors( 287| 654| buf, len, utf8_output); 288| 654|} _ZNK7simdutf8fallback14implementation29convert_valid_utf16le_to_utf8EPKDsmPc: 291| 406| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 292| 406| return scalar::utf16_to_utf8::convert_valid(buf, len, 293| 406| utf8_output); 294| 406|} _ZNK7simdutf8fallback14implementation29convert_valid_utf16be_to_utf8EPKDsmPc: 297| 452| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 298| 452| return scalar::utf16_to_utf8::convert_valid(buf, len, 299| 452| utf8_output); 300| 452|} _ZNK7simdutf8fallback14implementation23convert_utf32_to_latin1EPKDimPc: 305| 228| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 306| 228| return scalar::utf32_to_latin1::convert(buf, len, latin1_output); 307| 228|} _ZNK7simdutf8fallback14implementation35convert_utf32_to_latin1_with_errorsEPKDimPc: 310| 436| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 311| 436| return scalar::utf32_to_latin1::convert_with_errors(buf, len, latin1_output); 312| 436|} _ZNK7simdutf8fallback14implementation21convert_utf32_to_utf8EPKDimPc: 322| 648| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 323| 648| return scalar::utf32_to_utf8::convert(buf, len, utf8_output); 324| 648|} _ZNK7simdutf8fallback14implementation33convert_utf32_to_utf8_with_errorsEPKDimPc: 327| 832| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 328| 832| return scalar::utf32_to_utf8::convert_with_errors(buf, len, utf8_output); 329| 832|} _ZNK7simdutf8fallback14implementation27convert_valid_utf32_to_utf8EPKDimPc: 332| 216| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 333| 216| return scalar::utf32_to_utf8::convert_valid(buf, len, utf8_output); 334| 216|} _ZNK7simdutf8fallback14implementation24convert_utf32_to_utf16leEPKDimPDs: 339| 542| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 340| 542| return scalar::utf32_to_utf16::convert(buf, len, 341| 542| utf16_output); 342| 542|} _ZNK7simdutf8fallback14implementation24convert_utf32_to_utf16beEPKDimPDs: 345| 540| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 346| 540| return scalar::utf32_to_utf16::convert(buf, len, 347| 540| utf16_output); 348| 540|} _ZNK7simdutf8fallback14implementation36convert_utf32_to_utf16le_with_errorsEPKDimPDs: 351| 674| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 352| 674| return scalar::utf32_to_utf16::convert_with_errors( 353| 674| buf, len, utf16_output); 354| 674|} _ZNK7simdutf8fallback14implementation36convert_utf32_to_utf16be_with_errorsEPKDimPDs: 357| 694| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 358| 694| return scalar::utf32_to_utf16::convert_with_errors( 359| 694| buf, len, utf16_output); 360| 694|} _ZNK7simdutf8fallback14implementation30convert_valid_utf32_to_utf16leEPKDimPDs: 363| 280| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 364| 280| return scalar::utf32_to_utf16::convert_valid( 365| 280| buf, len, utf16_output); 366| 280|} _ZNK7simdutf8fallback14implementation30convert_valid_utf32_to_utf16beEPKDimPDs: 369| 134| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 370| 134| return scalar::utf32_to_utf16::convert_valid(buf, len, 371| 134| utf16_output); 372| 134|} _ZNK7simdutf8fallback14implementation24convert_utf16le_to_utf32EPKDsmPDi: 375| 326| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 376| 326| return scalar::utf16_to_utf32::convert(buf, len, 377| 326| utf32_output); 378| 326|} _ZNK7simdutf8fallback14implementation24convert_utf16be_to_utf32EPKDsmPDi: 381| 280| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 382| 280| return scalar::utf16_to_utf32::convert(buf, len, 383| 280| utf32_output); 384| 280|} _ZNK7simdutf8fallback14implementation36convert_utf16le_to_utf32_with_errorsEPKDsmPDi: 387| 334| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 388| 334| return scalar::utf16_to_utf32::convert_with_errors( 389| 334| buf, len, utf32_output); 390| 334|} _ZNK7simdutf8fallback14implementation36convert_utf16be_to_utf32_with_errorsEPKDsmPDi: 393| 346| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 394| 346| return scalar::utf16_to_utf32::convert_with_errors( 395| 346| buf, len, utf32_output); 396| 346|} _ZNK7simdutf8fallback14implementation30convert_valid_utf16le_to_utf32EPKDsmPDi: 399| 128| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 400| 128| return scalar::utf16_to_utf32::convert_valid( 401| 128| buf, len, utf32_output); 402| 128|} _ZNK7simdutf8fallback14implementation30convert_valid_utf16be_to_utf32EPKDsmPDi: 405| 142| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 406| 142| return scalar::utf16_to_utf32::convert_valid(buf, len, 407| 142| utf32_output); 408| 142|} _ZNK7simdutf8fallback14implementation13count_utf16leEPKDsm: 419| 1.35k| const char16_t *input, size_t length) const noexcept { 420| 1.35k| return scalar::utf16::count_code_points(input, length); 421| 1.35k|} _ZNK7simdutf8fallback14implementation13count_utf16beEPKDsm: 424| 1.33k| const char16_t *input, size_t length) const noexcept { 425| 1.33k| return scalar::utf16::count_code_points(input, length); 426| 1.33k|} _ZNK7simdutf8fallback14implementation10count_utf8EPKcm: 431| 3.48k|implementation::count_utf8(const char *input, size_t length) const noexcept { 432| 3.48k| return scalar::utf8::count_code_points(input, length); 433| 3.48k|} _ZNK7simdutf8fallback14implementation23latin1_length_from_utf8EPKcm: 438| 549| const char *buf, size_t len) const noexcept { 439| 549| return scalar::utf8::count_code_points(buf, len); 440| 549|} _ZNK7simdutf8fallback14implementation23utf8_length_from_latin1EPKcm: 445| 302| const char *input, size_t length) const noexcept { 446| 302| return scalar::latin1_to_utf8::utf8_length_from_latin1(input, length); 447| 302|} _ZNK7simdutf8fallback14implementation24utf8_length_from_utf16leEPKDsm: 452| 754| const char16_t *input, size_t length) const noexcept { 453| 754| return scalar::utf16::utf8_length_from_utf16(input, 454| 754| length); 455| 754|} _ZNK7simdutf8fallback14implementation24utf8_length_from_utf16beEPKDsm: 458| 748| const char16_t *input, size_t length) const noexcept { 459| 748| return scalar::utf16::utf8_length_from_utf16(input, length); 460| 748|} _ZNK7simdutf8fallback14implementation25utf32_length_from_utf16leEPKDsm: 465| 400| const char16_t *input, size_t length) const noexcept { 466| 400| return scalar::utf16::utf32_length_from_utf16(input, 467| 400| length); 468| 400|} _ZNK7simdutf8fallback14implementation25utf32_length_from_utf16beEPKDsm: 471| 400| const char16_t *input, size_t length) const noexcept { 472| 400| return scalar::utf16::utf32_length_from_utf16(input, length); 473| 400|} _ZNK7simdutf8fallback14implementation22utf16_length_from_utf8EPKcm: 478| 1.94k| const char *input, size_t length) const noexcept { 479| 1.94k| return scalar::utf8::utf16_length_from_utf8(input, length); 480| 1.94k|} _ZNK7simdutf8fallback14implementation22utf8_length_from_utf32EPKDim: 513| 860| const char32_t *input, size_t length) const noexcept { 514| 860| return scalar::utf32::utf8_length_from_utf32(input, length); 515| 860|} _ZNK7simdutf8fallback14implementation23utf16_length_from_utf32EPKDim: 520| 1.44k| const char32_t *input, size_t length) const noexcept { 521| 1.44k| return scalar::utf32::utf16_length_from_utf32(input, length); 522| 1.44k|} _ZNK7simdutf8fallback14implementation22utf32_length_from_utf8EPKcm: 527| 994| const char *input, size_t length) const noexcept { 528| 994| return scalar::utf8::count_code_points(input, length); 529| 994|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE13get_remainderEPh: 57| 6.11k|buf_block_reader::get_remainder(uint8_t *dst) const { 58| 6.11k| if (len == idx) { ------------------ | Branch (58:7): [True: 32, False: 6.08k] ------------------ 59| 32| return 0; 60| 32| } // memcpy(dst, null, 0) will trigger an error with some sanitizers 61| 6.08k| std::memset(dst, 0x20, 62| 6.08k| STEP_SIZE); // std::memset STEP_SIZE because it is more efficient 63| | // to write out 8 or 16 bytes at once. 64| 6.08k| std::memcpy(dst, buf + idx, len - idx); 65| 6.08k| return len - idx; 66| 6.11k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE13get_remainderEPh: 57| 6.11k|buf_block_reader::get_remainder(uint8_t *dst) const { 58| 6.11k| if (len == idx) { ------------------ | Branch (58:7): [True: 32, False: 6.08k] ------------------ 59| 32| return 0; 60| 32| } // memcpy(dst, null, 0) will trigger an error with some sanitizers 61| 6.08k| std::memset(dst, 0x20, 62| 6.08k| STEP_SIZE); // std::memset STEP_SIZE because it is more efficient 63| | // to write out 8 or 16 bytes at once. 64| 6.08k| std::memcpy(dst, buf + idx, len - idx); 65| 6.08k| return len - idx; 66| 6.11k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EEC2EPKhm: 36| 6.97k| : buf{_buf}, len{_len}, lenminusstep{len < STEP_SIZE ? 0 : len - STEP_SIZE}, ------------------ | Branch (36:42): [True: 2.64k, False: 4.32k] ------------------ 37| 6.97k| idx{0} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE14has_full_blockEv: 45| 4.87M|simdutf_really_inline bool buf_block_reader::has_full_block() const { 46| 4.87M| return idx < lenminusstep; 47| 4.87M|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE10full_blockEv: 51| 4.86M|buf_block_reader::full_block() const { 52| 4.86M| return &buf[idx]; 53| 4.86M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE7advanceEv: 69| 4.87M|simdutf_really_inline void buf_block_reader::advance() { 70| 4.87M| idx += STEP_SIZE; 71| 4.87M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EEC2EPKhm: 36| 6.97k| : buf{_buf}, len{_len}, lenminusstep{len < STEP_SIZE ? 0 : len - STEP_SIZE}, ------------------ | Branch (36:42): [True: 2.64k, False: 4.32k] ------------------ 37| 6.97k| idx{0} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE14has_full_blockEv: 45| 4.87M|simdutf_really_inline bool buf_block_reader::has_full_block() const { 46| 4.87M| return idx < lenminusstep; 47| 4.87M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE10full_blockEv: 51| 4.86M|buf_block_reader::full_block() const { 52| 4.86M| return &buf[idx]; 53| 4.86M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE7advanceEv: 69| 4.87M|simdutf_really_inline void buf_block_reader::advance() { 70| 4.87M| idx += STEP_SIZE; 71| 4.87M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE0EEEmPKDsm: 8| 1.75k| size_t size) { 9| 1.75k| size_t pos = 0; 10| 1.75k| size_t count = 0; 11| 1.15M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 1.15M, False: 1.75k] ------------------ 12| 1.15M| simd16x32 input(reinterpret_cast(in + pos)); 13| | if constexpr (!match_system(big_endian)) { 14| | input.swap_bytes(); 15| | } 16| 1.15M| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 1.15M| count += count_ones(not_pair) / 2; 18| 1.15M| } 19| 1.75k| return count + 20| 1.75k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.75k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE1EEEmPKDsm: 8| 1.73k| size_t size) { 9| 1.73k| size_t pos = 0; 10| 1.73k| size_t count = 0; 11| 1.04M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 1.03M, False: 1.73k] ------------------ 12| 1.03M| simd16x32 input(reinterpret_cast(in + pos)); 13| 1.03M| if constexpr (!match_system(big_endian)) { 14| 1.03M| input.swap_bytes(); 15| 1.03M| } 16| 1.03M| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 1.03M| count += count_ones(not_pair) / 2; 18| 1.03M| } 19| 1.73k| return count + 20| 1.73k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.73k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 51| 400| size_t size) { 52| 400| return count_code_points(in, size); 53| 400|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 51| 400| size_t size) { 52| 400| return count_code_points(in, size); 53| 400|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE0EEEmPKDsm: 8| 1.75k| size_t size) { 9| 1.75k| size_t pos = 0; 10| 1.75k| size_t count = 0; 11| 1.15M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 1.15M, False: 1.75k] ------------------ 12| 1.15M| simd16x32 input(reinterpret_cast(in + pos)); 13| | if constexpr (!match_system(big_endian)) { 14| | input.swap_bytes(); 15| | } 16| 1.15M| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 1.15M| count += count_ones(not_pair) / 2; 18| 1.15M| } 19| 1.75k| return count + 20| 1.75k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.75k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE1EEEmPKDsm: 8| 1.73k| size_t size) { 9| 1.73k| size_t pos = 0; 10| 1.73k| size_t count = 0; 11| 1.04M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 1.03M, False: 1.73k] ------------------ 12| 1.03M| simd16x32 input(reinterpret_cast(in + pos)); 13| 1.03M| if constexpr (!match_system(big_endian)) { 14| 1.03M| input.swap_bytes(); 15| 1.03M| } 16| 1.03M| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 1.03M| count += count_ones(not_pair) / 2; 18| 1.03M| } 19| 1.73k| return count + 20| 1.73k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.73k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 51| 400| size_t size) { 52| 400| return count_code_points(in, size); 53| 400|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 51| 400| size_t size) { 52| 400| return count_code_points(in, size); 53| 400|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE0EEEmPKDsm: 10| 754| size_t size) { 11| 754| size_t pos = 0; 12| | 13| 754| using vector_u16 = simd16; 14| 754| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 754| const auto one = vector_u16::splat(1); 17| | 18| 754| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 754| size_t count = size / N * N; 22| | 23| | // in a single iteration the increment is 0, 1 or 2, despite we have 24| | // three additions 25| 754| constexpr size_t max_iterations = 65535 / 2; 26| 754| size_t iteration = max_iterations; 27| | 28| 1.05M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 1.05M, False: 754] ------------------ 29| 1.05M| auto input = vector_u16::load(reinterpret_cast(in + pos)); 30| | if constexpr (!match_system(big_endian)) { 31| | input = input.swap_bytes(); 32| | } 33| | // 0xd800 .. 0xdbff - low surrogate 34| | // 0xdc00 .. 0xdfff - high surrogate 35| 1.05M| const auto is_surrogate = ((input & uint16_t(0xf800)) == uint16_t(0xd800)); 36| | 37| | // c0 - chars that yield 2- or 3-byte UTF-8 codes 38| 1.05M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 1.05M| const auto c1 = min(input & uint16_t(0xf800), one); 42| | 43| | /* 44| | Explanation how the counting works. 45| | 46| | In the case of a non-surrogate character we count: 47| | * always 1 -- see how `count` is initialized above; 48| | * c0 = 1 if the current char yields 2 or 3 bytes; 49| | * c1 = 1 if the current char yields 3 bytes. 50| | 51| | Thus, we always have correct count for the current char: 52| | from 1, 2 or 3 bytes. 53| | 54| | A trickier part is how we count surrogate pairs. Whether 55| | we encounter a surrogate (low or high), we count it as 56| | 3 chars and then minus 1 (`is_surrogate` is -1 or 0). 57| | Each surrogate char yields 2. A surrogate pair, that 58| | is a low surrogate followed by a high one, yields 59| | the expected 4 bytes. 60| | 61| | It also correctly handles cases when low surrogate is 62| | processed by the this loop, but high surrogate is counted 63| | by the scalar procedure. The scalar procedure uses exactly 64| | the described approach, thanks to that for valid UTF-16 65| | strings it always count correctly. 66| | */ 67| 1.05M| v_count += c0; 68| 1.05M| v_count += c1; 69| 1.05M| v_count += vector_u16(is_surrogate); 70| | 71| 1.05M| iteration -= 1; 72| 1.05M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 19, False: 1.05M] ------------------ 73| 19| count += v_count.sum(); 74| 19| v_count = vector_u16::zero(); 75| 19| iteration = max_iterations; 76| 19| } 77| 1.05M| } 78| | 79| 754| if (iteration > 0) { ------------------ | Branch (79:7): [True: 754, False: 0] ------------------ 80| 754| count += v_count.sum(); 81| 754| } 82| | 83| 754| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 754| size - pos); 85| 754|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE1EEEmPKDsm: 10| 748| size_t size) { 11| 748| size_t pos = 0; 12| | 13| 748| using vector_u16 = simd16; 14| 748| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 748| const auto one = vector_u16::splat(1); 17| | 18| 748| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 748| size_t count = size / N * N; 22| | 23| | // in a single iteration the increment is 0, 1 or 2, despite we have 24| | // three additions 25| 748| constexpr size_t max_iterations = 65535 / 2; 26| 748| size_t iteration = max_iterations; 27| | 28| 949k| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 949k, False: 748] ------------------ 29| 949k| auto input = vector_u16::load(reinterpret_cast(in + pos)); 30| 949k| if constexpr (!match_system(big_endian)) { 31| 949k| input = input.swap_bytes(); 32| 949k| } 33| | // 0xd800 .. 0xdbff - low surrogate 34| | // 0xdc00 .. 0xdfff - high surrogate 35| 949k| const auto is_surrogate = ((input & uint16_t(0xf800)) == uint16_t(0xd800)); 36| | 37| | // c0 - chars that yield 2- or 3-byte UTF-8 codes 38| 949k| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 949k| const auto c1 = min(input & uint16_t(0xf800), one); 42| | 43| | /* 44| | Explanation how the counting works. 45| | 46| | In the case of a non-surrogate character we count: 47| | * always 1 -- see how `count` is initialized above; 48| | * c0 = 1 if the current char yields 2 or 3 bytes; 49| | * c1 = 1 if the current char yields 3 bytes. 50| | 51| | Thus, we always have correct count for the current char: 52| | from 1, 2 or 3 bytes. 53| | 54| | A trickier part is how we count surrogate pairs. Whether 55| | we encounter a surrogate (low or high), we count it as 56| | 3 chars and then minus 1 (`is_surrogate` is -1 or 0). 57| | Each surrogate char yields 2. A surrogate pair, that 58| | is a low surrogate followed by a high one, yields 59| | the expected 4 bytes. 60| | 61| | It also correctly handles cases when low surrogate is 62| | processed by the this loop, but high surrogate is counted 63| | by the scalar procedure. The scalar procedure uses exactly 64| | the described approach, thanks to that for valid UTF-16 65| | strings it always count correctly. 66| | */ 67| 949k| v_count += c0; 68| 949k| v_count += c1; 69| 949k| v_count += vector_u16(is_surrogate); 70| | 71| 949k| iteration -= 1; 72| 949k| if (iteration == 0) { ------------------ | Branch (72:9): [True: 18, False: 949k] ------------------ 73| 18| count += v_count.sum(); 74| 18| v_count = vector_u16::zero(); 75| 18| iteration = max_iterations; 76| 18| } 77| 949k| } 78| | 79| 748| if (iteration > 0) { ------------------ | Branch (79:7): [True: 748, False: 0] ------------------ 80| 748| count += v_count.sum(); 81| 748| } 82| | 83| 748| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 748| size - pos); 85| 748|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE0EEEmPKDsm: 10| 754| size_t size) { 11| 754| size_t pos = 0; 12| | 13| 754| using vector_u16 = simd16; 14| 754| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 754| const auto one = vector_u16::splat(1); 17| | 18| 754| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 754| size_t count = size / N * N; 22| | 23| | // in a single iteration the increment is 0, 1 or 2, despite we have 24| | // three additions 25| 754| constexpr size_t max_iterations = 65535 / 2; 26| 754| size_t iteration = max_iterations; 27| | 28| 2.11M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 2.11M, False: 754] ------------------ 29| 2.11M| auto input = vector_u16::load(reinterpret_cast(in + pos)); 30| | if constexpr (!match_system(big_endian)) { 31| | input = input.swap_bytes(); 32| | } 33| | // 0xd800 .. 0xdbff - low surrogate 34| | // 0xdc00 .. 0xdfff - high surrogate 35| 2.11M| const auto is_surrogate = ((input & uint16_t(0xf800)) == uint16_t(0xd800)); 36| | 37| | // c0 - chars that yield 2- or 3-byte UTF-8 codes 38| 2.11M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 2.11M| const auto c1 = min(input & uint16_t(0xf800), one); 42| | 43| | /* 44| | Explanation how the counting works. 45| | 46| | In the case of a non-surrogate character we count: 47| | * always 1 -- see how `count` is initialized above; 48| | * c0 = 1 if the current char yields 2 or 3 bytes; 49| | * c1 = 1 if the current char yields 3 bytes. 50| | 51| | Thus, we always have correct count for the current char: 52| | from 1, 2 or 3 bytes. 53| | 54| | A trickier part is how we count surrogate pairs. Whether 55| | we encounter a surrogate (low or high), we count it as 56| | 3 chars and then minus 1 (`is_surrogate` is -1 or 0). 57| | Each surrogate char yields 2. A surrogate pair, that 58| | is a low surrogate followed by a high one, yields 59| | the expected 4 bytes. 60| | 61| | It also correctly handles cases when low surrogate is 62| | processed by the this loop, but high surrogate is counted 63| | by the scalar procedure. The scalar procedure uses exactly 64| | the described approach, thanks to that for valid UTF-16 65| | strings it always count correctly. 66| | */ 67| 2.11M| v_count += c0; 68| 2.11M| v_count += c1; 69| 2.11M| v_count += vector_u16(is_surrogate); 70| | 71| 2.11M| iteration -= 1; 72| 2.11M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 50, False: 2.11M] ------------------ 73| 50| count += v_count.sum(); 74| 50| v_count = vector_u16::zero(); 75| 50| iteration = max_iterations; 76| 50| } 77| 2.11M| } 78| | 79| 754| if (iteration > 0) { ------------------ | Branch (79:7): [True: 754, False: 0] ------------------ 80| 754| count += v_count.sum(); 81| 754| } 82| | 83| 754| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 754| size - pos); 85| 754|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE1EEEmPKDsm: 10| 748| size_t size) { 11| 748| size_t pos = 0; 12| | 13| 748| using vector_u16 = simd16; 14| 748| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 748| const auto one = vector_u16::splat(1); 17| | 18| 748| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 748| size_t count = size / N * N; 22| | 23| | // in a single iteration the increment is 0, 1 or 2, despite we have 24| | // three additions 25| 748| constexpr size_t max_iterations = 65535 / 2; 26| 748| size_t iteration = max_iterations; 27| | 28| 1.89M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 1.89M, False: 748] ------------------ 29| 1.89M| auto input = vector_u16::load(reinterpret_cast(in + pos)); 30| 1.89M| if constexpr (!match_system(big_endian)) { 31| 1.89M| input = input.swap_bytes(); 32| 1.89M| } 33| | // 0xd800 .. 0xdbff - low surrogate 34| | // 0xdc00 .. 0xdfff - high surrogate 35| 1.89M| const auto is_surrogate = ((input & uint16_t(0xf800)) == uint16_t(0xd800)); 36| | 37| | // c0 - chars that yield 2- or 3-byte UTF-8 codes 38| 1.89M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 1.89M| const auto c1 = min(input & uint16_t(0xf800), one); 42| | 43| | /* 44| | Explanation how the counting works. 45| | 46| | In the case of a non-surrogate character we count: 47| | * always 1 -- see how `count` is initialized above; 48| | * c0 = 1 if the current char yields 2 or 3 bytes; 49| | * c1 = 1 if the current char yields 3 bytes. 50| | 51| | Thus, we always have correct count for the current char: 52| | from 1, 2 or 3 bytes. 53| | 54| | A trickier part is how we count surrogate pairs. Whether 55| | we encounter a surrogate (low or high), we count it as 56| | 3 chars and then minus 1 (`is_surrogate` is -1 or 0). 57| | Each surrogate char yields 2. A surrogate pair, that 58| | is a low surrogate followed by a high one, yields 59| | the expected 4 bytes. 60| | 61| | It also correctly handles cases when low surrogate is 62| | processed by the this loop, but high surrogate is counted 63| | by the scalar procedure. The scalar procedure uses exactly 64| | the described approach, thanks to that for valid UTF-16 65| | strings it always count correctly. 66| | */ 67| 1.89M| v_count += c0; 68| 1.89M| v_count += c1; 69| 1.89M| v_count += vector_u16(is_surrogate); 70| | 71| 1.89M| iteration -= 1; 72| 1.89M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 46, False: 1.89M] ------------------ 73| 46| count += v_count.sum(); 74| 46| v_count = vector_u16::zero(); 75| 46| iteration = max_iterations; 76| 46| } 77| 1.89M| } 78| | 79| 748| if (iteration > 0) { ------------------ | Branch (79:7): [True: 748, False: 0] ------------------ 80| 748| count += v_count.sum(); 81| 748| } 82| | 83| 748| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 748| size - pos); 85| 748|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf3222utf8_length_from_utf32EPKDim: 11| 860| size_t length) { 12| 860| using vector_u32 = simd32; 13| | 14| 860| const char32_t *start = input; 15| | 16| | // we add up to three ones in a single iteration (see the vectorized loop in 17| | // section #2 below) 18| 860| const size_t max_increment = 3; 19| | 20| 860| const size_t N = vector_u32::ELEMENTS; 21| | 22| |#if SIMDUTF_SIMD_HAS_UNSIGNED_CMP 23| | const auto v_0000007f = vector_u32::splat(0x0000007f); 24| | const auto v_000007ff = vector_u32::splat(0x000007ff); 25| | const auto v_0000ffff = vector_u32::splat(0x0000ffff); 26| |#else 27| 860| const auto v_ffffff80 = vector_u32::splat(0xffffff80); 28| 860| const auto v_fffff800 = vector_u32::splat(0xfffff800); 29| 860| const auto v_ffff0000 = vector_u32::splat(0xffff0000); 30| 860| const auto one = vector_u32::splat(1); 31| 860|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 32| | 33| 860| size_t counter = 0; 34| | 35| | // 1. vectorized loop unrolled 4 times 36| 860| { 37| | // we use vector of uint32 counters, this is why this limit is used 38| 860| const size_t max_iterations = 39| 860| std::numeric_limits::max() / (max_increment * 4); 40| 860| size_t blocks = length / (N * 4); 41| 860| length -= blocks * (N * 4); 42| 1.05k| while (blocks != 0) { ------------------ | Branch (42:12): [True: 198, False: 860] ------------------ 43| 198| const size_t iterations = min(blocks, max_iterations); 44| 198| blocks -= iterations; 45| | 46| 198| simd32 acc = vector_u32::zero(); 47| 87.1k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (47:26): [True: 86.9k, False: 198] ------------------ 48| 86.9k| const auto in0 = vector_u32(input + 0 * N); 49| 86.9k| const auto in1 = vector_u32(input + 1 * N); 50| 86.9k| const auto in2 = vector_u32(input + 2 * N); 51| 86.9k| const auto in3 = vector_u32(input + 3 * N); 52| | 53| |#if SIMDUTF_SIMD_HAS_UNSIGNED_CMP 54| | acc -= as_vector_u32(in0 > v_0000007f); 55| | acc -= as_vector_u32(in1 > v_0000007f); 56| | acc -= as_vector_u32(in2 > v_0000007f); 57| | acc -= as_vector_u32(in3 > v_0000007f); 58| | 59| | acc -= as_vector_u32(in0 > v_000007ff); 60| | acc -= as_vector_u32(in1 > v_000007ff); 61| | acc -= as_vector_u32(in2 > v_000007ff); 62| | acc -= as_vector_u32(in3 > v_000007ff); 63| | 64| | acc -= as_vector_u32(in0 > v_0000ffff); 65| | acc -= as_vector_u32(in1 > v_0000ffff); 66| | acc -= as_vector_u32(in2 > v_0000ffff); 67| | acc -= as_vector_u32(in3 > v_0000ffff); 68| |#else 69| 86.9k| acc += min(one, in0 & v_ffffff80); 70| 86.9k| acc += min(one, in1 & v_ffffff80); 71| 86.9k| acc += min(one, in2 & v_ffffff80); 72| 86.9k| acc += min(one, in3 & v_ffffff80); 73| | 74| 86.9k| acc += min(one, in0 & v_fffff800); 75| 86.9k| acc += min(one, in1 & v_fffff800); 76| 86.9k| acc += min(one, in2 & v_fffff800); 77| 86.9k| acc += min(one, in3 & v_fffff800); 78| | 79| 86.9k| acc += min(one, in0 & v_ffff0000); 80| 86.9k| acc += min(one, in1 & v_ffff0000); 81| 86.9k| acc += min(one, in2 & v_ffff0000); 82| 86.9k| acc += min(one, in3 & v_ffff0000); 83| 86.9k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 84| | 85| 86.9k| input += 4 * N; 86| 86.9k| } 87| | 88| 198| counter += acc.sum(); 89| 198| } 90| 860| } 91| | 92| | // 2. vectorized loop for tail 93| 860| { 94| 860| const size_t max_iterations = 95| 860| std::numeric_limits::max() / max_increment; 96| 860| size_t blocks = length / N; 97| 860| length -= blocks * N; 98| 1.28k| while (blocks != 0) { ------------------ | Branch (98:12): [True: 425, False: 860] ------------------ 99| 425| const size_t iterations = min(blocks, max_iterations); 100| 425| blocks -= iterations; 101| | 102| 425| auto acc = vector_u32::zero(); 103| 1.43k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (103:26): [True: 1.01k, False: 425] ------------------ 104| 1.01k| const auto in = vector_u32(input); 105| | 106| |#if SIMDUTF_SIMD_HAS_UNSIGNED_CMP 107| | acc -= as_vector_u32(in > v_0000007f); 108| | acc -= as_vector_u32(in > v_000007ff); 109| | acc -= as_vector_u32(in > v_0000ffff); 110| |#else 111| 1.01k| acc += min(one, in & v_ffffff80); 112| 1.01k| acc += min(one, in & v_fffff800); 113| 1.01k| acc += min(one, in & v_ffff0000); 114| 1.01k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 115| | 116| 1.01k| input += N; 117| 1.01k| } 118| | 119| 425| counter += acc.sum(); 120| 425| } 121| 860| } 122| | 123| 860| const size_t consumed = input - start; 124| 860| if (consumed != 0) { ------------------ | Branch (124:7): [True: 468, False: 392] ------------------ 125| | // We don't count 0th bytes in the vectorized loops above, this 126| | // is why we need to count them in the end. 127| 468| counter += consumed; 128| 468| } 129| | 130| 860| return counter + scalar::utf32::utf8_length_from_utf32(input, length); 131| 860|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf323minImEET_S4_S4_: 8| 623|template T min(T a, T b) { return a <= b ? a : b; } ------------------ | Branch (8:48): [True: 623, False: 0] ------------------ simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf3222utf8_length_from_utf32EPKDim: 11| 860| size_t length) { 12| 860| using vector_u32 = simd32; 13| | 14| 860| const char32_t *start = input; 15| | 16| | // we add up to three ones in a single iteration (see the vectorized loop in 17| | // section #2 below) 18| 860| const size_t max_increment = 3; 19| | 20| 860| const size_t N = vector_u32::ELEMENTS; 21| | 22| |#if SIMDUTF_SIMD_HAS_UNSIGNED_CMP 23| | const auto v_0000007f = vector_u32::splat(0x0000007f); 24| | const auto v_000007ff = vector_u32::splat(0x000007ff); 25| | const auto v_0000ffff = vector_u32::splat(0x0000ffff); 26| |#else 27| 860| const auto v_ffffff80 = vector_u32::splat(0xffffff80); 28| 860| const auto v_fffff800 = vector_u32::splat(0xfffff800); 29| 860| const auto v_ffff0000 = vector_u32::splat(0xffff0000); 30| 860| const auto one = vector_u32::splat(1); 31| 860|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 32| | 33| 860| size_t counter = 0; 34| | 35| | // 1. vectorized loop unrolled 4 times 36| 860| { 37| | // we use vector of uint32 counters, this is why this limit is used 38| 860| const size_t max_iterations = 39| 860| std::numeric_limits::max() / (max_increment * 4); 40| 860| size_t blocks = length / (N * 4); 41| 860| length -= blocks * (N * 4); 42| 1.29k| while (blocks != 0) { ------------------ | Branch (42:12): [True: 431, False: 860] ------------------ 43| 431| const size_t iterations = min(blocks, max_iterations); 44| 431| blocks -= iterations; 45| | 46| 431| simd32 acc = vector_u32::zero(); 47| 174k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (47:26): [True: 174k, False: 431] ------------------ 48| 174k| const auto in0 = vector_u32(input + 0 * N); 49| 174k| const auto in1 = vector_u32(input + 1 * N); 50| 174k| const auto in2 = vector_u32(input + 2 * N); 51| 174k| const auto in3 = vector_u32(input + 3 * N); 52| | 53| |#if SIMDUTF_SIMD_HAS_UNSIGNED_CMP 54| | acc -= as_vector_u32(in0 > v_0000007f); 55| | acc -= as_vector_u32(in1 > v_0000007f); 56| | acc -= as_vector_u32(in2 > v_0000007f); 57| | acc -= as_vector_u32(in3 > v_0000007f); 58| | 59| | acc -= as_vector_u32(in0 > v_000007ff); 60| | acc -= as_vector_u32(in1 > v_000007ff); 61| | acc -= as_vector_u32(in2 > v_000007ff); 62| | acc -= as_vector_u32(in3 > v_000007ff); 63| | 64| | acc -= as_vector_u32(in0 > v_0000ffff); 65| | acc -= as_vector_u32(in1 > v_0000ffff); 66| | acc -= as_vector_u32(in2 > v_0000ffff); 67| | acc -= as_vector_u32(in3 > v_0000ffff); 68| |#else 69| 174k| acc += min(one, in0 & v_ffffff80); 70| 174k| acc += min(one, in1 & v_ffffff80); 71| 174k| acc += min(one, in2 & v_ffffff80); 72| 174k| acc += min(one, in3 & v_ffffff80); 73| | 74| 174k| acc += min(one, in0 & v_fffff800); 75| 174k| acc += min(one, in1 & v_fffff800); 76| 174k| acc += min(one, in2 & v_fffff800); 77| 174k| acc += min(one, in3 & v_fffff800); 78| | 79| 174k| acc += min(one, in0 & v_ffff0000); 80| 174k| acc += min(one, in1 & v_ffff0000); 81| 174k| acc += min(one, in2 & v_ffff0000); 82| 174k| acc += min(one, in3 & v_ffff0000); 83| 174k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 84| | 85| 174k| input += 4 * N; 86| 174k| } 87| | 88| 431| counter += acc.sum(); 89| 431| } 90| 860| } 91| | 92| | // 2. vectorized loop for tail 93| 860| { 94| 860| const size_t max_iterations = 95| 860| std::numeric_limits::max() / max_increment; 96| 860| size_t blocks = length / N; 97| 860| length -= blocks * N; 98| 1.32k| while (blocks != 0) { ------------------ | Branch (98:12): [True: 464, False: 860] ------------------ 99| 464| const size_t iterations = min(blocks, max_iterations); 100| 464| blocks -= iterations; 101| | 102| 464| auto acc = vector_u32::zero(); 103| 1.58k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (103:26): [True: 1.12k, False: 464] ------------------ 104| 1.12k| const auto in = vector_u32(input); 105| | 106| |#if SIMDUTF_SIMD_HAS_UNSIGNED_CMP 107| | acc -= as_vector_u32(in > v_0000007f); 108| | acc -= as_vector_u32(in > v_000007ff); 109| | acc -= as_vector_u32(in > v_0000ffff); 110| |#else 111| 1.12k| acc += min(one, in & v_ffffff80); 112| 1.12k| acc += min(one, in & v_fffff800); 113| 1.12k| acc += min(one, in & v_ffff0000); 114| 1.12k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 115| | 116| 1.12k| input += N; 117| 1.12k| } 118| | 119| 464| counter += acc.sum(); 120| 464| } 121| 860| } 122| | 123| 860| const size_t consumed = input - start; 124| 860| if (consumed != 0) { ------------------ | Branch (124:7): [True: 505, False: 355] ------------------ 125| | // We don't count 0th bytes in the vectorized loops above, this 126| | // is why we need to count them in the end. 127| 505| counter += consumed; 128| 505| } 129| | 130| 860| return counter + scalar::utf32::utf8_length_from_utf32(input, length); 131| 860|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf323minImEET_S4_S4_: 8| 895|template T min(T a, T b) { return a <= b ? a : b; } ------------------ | Branch (8:48): [True: 895, False: 0] ------------------ simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14utf826count_code_points_bytemaskEPKcm: 21| 4.03k|count_code_points_bytemask(const char *in, size_t size) { 22| 4.03k| using vector_i8 = simd8; 23| 4.03k| using vector_u8 = simd8; 24| 4.03k| using vector_u64 = simd64; 25| | 26| 4.03k| constexpr size_t N = vector_i8::SIZE; 27| 4.03k| constexpr size_t max_iterations = 255 / 4; 28| | 29| 4.03k| size_t pos = 0; 30| 4.03k| size_t count = 0; 31| | 32| 4.03k| auto counters = vector_u64::zero(); 33| 4.03k| auto local = vector_u8::zero(); 34| 4.03k| size_t iterations = 0; 35| 1.69M| for (; pos + 4 * N <= size; pos += 4 * N) { ------------------ | Branch (35:10): [True: 1.68M, False: 4.03k] ------------------ 36| 1.68M| const auto input0 = 37| 1.68M| simd8::load(reinterpret_cast(in + pos + 0 * N)); 38| 1.68M| const auto input1 = 39| 1.68M| simd8::load(reinterpret_cast(in + pos + 1 * N)); 40| 1.68M| const auto input2 = 41| 1.68M| simd8::load(reinterpret_cast(in + pos + 2 * N)); 42| 1.68M| const auto input3 = 43| 1.68M| simd8::load(reinterpret_cast(in + pos + 3 * N)); 44| 1.68M| const auto mask0 = input0 > int8_t(-65); 45| 1.68M| const auto mask1 = input1 > int8_t(-65); 46| 1.68M| const auto mask2 = input2 > int8_t(-65); 47| 1.68M| const auto mask3 = input3 > int8_t(-65); 48| | 49| 1.68M| local -= vector_u8(mask0); 50| 1.68M| local -= vector_u8(mask1); 51| 1.68M| local -= vector_u8(mask2); 52| 1.68M| local -= vector_u8(mask3); 53| | 54| 1.68M| iterations += 1; 55| 1.68M| if (iterations == max_iterations) { ------------------ | Branch (55:9): [True: 26.5k, False: 1.66M] ------------------ 56| 26.5k| counters += sum_8bytes(local); 57| 26.5k| local = vector_u8::zero(); 58| 26.5k| iterations = 0; 59| 26.5k| } 60| 1.68M| } 61| | 62| 4.03k| if (iterations > 0) { ------------------ | Branch (62:7): [True: 1.52k, False: 2.51k] ------------------ 63| 1.52k| count += local.sum_bytes(); 64| 1.52k| } 65| | 66| 4.03k| count += counters.sum(); 67| | 68| 4.03k| return count + scalar::utf8::count_code_points(in + pos, size - pos); 69| 4.03k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14utf817count_code_pointsEPKcm: 8| 994|simdutf_really_inline size_t count_code_points(const char *in, size_t size) { 9| 994| size_t pos = 0; 10| 994| size_t count = 0; 11| 753k| for (; pos + 64 <= size; pos += 64) { ------------------ | Branch (11:10): [True: 752k, False: 994] ------------------ 12| 752k| simd8x64 input(reinterpret_cast(in + pos)); 13| 752k| uint64_t utf8_continuation_mask = input.gt(-65); 14| 752k| count += count_ones(utf8_continuation_mask); 15| 752k| } 16| 994| return count + scalar::utf8::count_code_points(in + pos, size - pos); 17| 994|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14utf826count_code_points_bytemaskEPKcm: 21| 4.03k|count_code_points_bytemask(const char *in, size_t size) { 22| 4.03k| using vector_i8 = simd8; 23| 4.03k| using vector_u8 = simd8; 24| 4.03k| using vector_u64 = simd64; 25| | 26| 4.03k| constexpr size_t N = vector_i8::SIZE; 27| 4.03k| constexpr size_t max_iterations = 255 / 4; 28| | 29| 4.03k| size_t pos = 0; 30| 4.03k| size_t count = 0; 31| | 32| 4.03k| auto counters = vector_u64::zero(); 33| 4.03k| auto local = vector_u8::zero(); 34| 4.03k| size_t iterations = 0; 35| 3.38M| for (; pos + 4 * N <= size; pos += 4 * N) { ------------------ | Branch (35:10): [True: 3.37M, False: 4.03k] ------------------ 36| 3.37M| const auto input0 = 37| 3.37M| simd8::load(reinterpret_cast(in + pos + 0 * N)); 38| 3.37M| const auto input1 = 39| 3.37M| simd8::load(reinterpret_cast(in + pos + 1 * N)); 40| 3.37M| const auto input2 = 41| 3.37M| simd8::load(reinterpret_cast(in + pos + 2 * N)); 42| 3.37M| const auto input3 = 43| 3.37M| simd8::load(reinterpret_cast(in + pos + 3 * N)); 44| 3.37M| const auto mask0 = input0 > int8_t(-65); 45| 3.37M| const auto mask1 = input1 > int8_t(-65); 46| 3.37M| const auto mask2 = input2 > int8_t(-65); 47| 3.37M| const auto mask3 = input3 > int8_t(-65); 48| | 49| 3.37M| local -= vector_u8(mask0); 50| 3.37M| local -= vector_u8(mask1); 51| 3.37M| local -= vector_u8(mask2); 52| 3.37M| local -= vector_u8(mask3); 53| | 54| 3.37M| iterations += 1; 55| 3.37M| if (iterations == max_iterations) { ------------------ | Branch (55:9): [True: 53.3k, False: 3.32M] ------------------ 56| 53.3k| counters += sum_8bytes(local); 57| 53.3k| local = vector_u8::zero(); 58| 53.3k| iterations = 0; 59| 53.3k| } 60| 3.37M| } 61| | 62| 4.03k| if (iterations > 0) { ------------------ | Branch (62:7): [True: 2.47k, False: 1.56k] ------------------ 63| 2.47k| count += local.sum_bytes(); 64| 2.47k| } 65| | 66| 4.03k| count += counters.sum(); 67| | 68| 4.03k| return count + scalar::utf8::count_code_points(in + pos, size - pos); 69| 4.03k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14utf817count_code_pointsEPKcm: 8| 994|simdutf_really_inline size_t count_code_points(const char *in, size_t size) { 9| 994| size_t pos = 0; 10| 994| size_t count = 0; 11| 753k| for (; pos + 64 <= size; pos += 64) { ------------------ | Branch (11:10): [True: 752k, False: 994] ------------------ 12| 752k| simd8x64 input(reinterpret_cast(in + pos)); 13| 752k| uint64_t utf8_continuation_mask = input.gt(-65); 14| 752k| count += count_ones(utf8_continuation_mask); 15| 752k| } 16| 994| return count + scalar::utf8::count_code_points(in + pos, size - pos); 17| 994|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14utf831utf16_length_from_utf8_bytemaskEPKcm: 9| 1.94k| size_t size) { 10| 1.94k| using vector_i8 = simd8; 11| 1.94k| using vector_u8 = simd8; 12| 1.94k| using vector_u64 = simd64; 13| | 14| 1.94k| constexpr size_t N = vector_i8::SIZE; 15| 1.94k| constexpr size_t max_iterations = 255 / 2; 16| | 17| 1.94k| auto counters = vector_u64::zero(); 18| 1.94k| auto local = vector_u8::zero(); 19| | 20| 1.94k| size_t iterations = 0; 21| 1.94k| size_t pos = 0; 22| 1.94k| size_t count = 0; 23| 3.04M| for (; pos + N <= size; pos += N) { ------------------ | Branch (23:10): [True: 3.04M, False: 1.94k] ------------------ 24| 3.04M| const auto input = 25| 3.04M| vector_i8::load(reinterpret_cast(in + pos)); 26| | 27| 3.04M| const auto continuation = input > int8_t(-65); 28| 3.04M| const auto utf_4bytes = vector_u8(input.value) >= uint8_t(240); 29| | 30| 3.04M| local -= vector_u8(continuation); 31| 3.04M| local -= vector_u8(utf_4bytes); 32| | 33| 3.04M| iterations += 1; 34| 3.04M| if (iterations == max_iterations) { ------------------ | Branch (34:9): [True: 23.8k, False: 3.01M] ------------------ 35| 23.8k| counters += sum_8bytes(local); 36| 23.8k| local = vector_u8::zero(); 37| 23.8k| iterations = 0; 38| 23.8k| } 39| 3.04M| } 40| | 41| 1.94k| if (iterations > 0) { ------------------ | Branch (41:7): [True: 1.25k, False: 692] ------------------ 42| 1.25k| count += local.sum_bytes(); 43| 1.25k| } 44| | 45| 1.94k| count += counters.sum(); 46| | 47| 1.94k| return count + scalar::utf8::utf16_length_from_utf8(in + pos, size - pos); 48| 1.94k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14utf831utf16_length_from_utf8_bytemaskEPKcm: 9| 1.94k| size_t size) { 10| 1.94k| using vector_i8 = simd8; 11| 1.94k| using vector_u8 = simd8; 12| 1.94k| using vector_u64 = simd64; 13| | 14| 1.94k| constexpr size_t N = vector_i8::SIZE; 15| 1.94k| constexpr size_t max_iterations = 255 / 2; 16| | 17| 1.94k| auto counters = vector_u64::zero(); 18| 1.94k| auto local = vector_u8::zero(); 19| | 20| 1.94k| size_t iterations = 0; 21| 1.94k| size_t pos = 0; 22| 1.94k| size_t count = 0; 23| 6.08M| for (; pos + N <= size; pos += N) { ------------------ | Branch (23:10): [True: 6.08M, False: 1.94k] ------------------ 24| 6.08M| const auto input = 25| 6.08M| vector_i8::load(reinterpret_cast(in + pos)); 26| | 27| 6.08M| const auto continuation = input > int8_t(-65); 28| 6.08M| const auto utf_4bytes = vector_u8(input.value) >= uint8_t(240); 29| | 30| 6.08M| local -= vector_u8(continuation); 31| 6.08M| local -= vector_u8(utf_4bytes); 32| | 33| 6.08M| iterations += 1; 34| 6.08M| if (iterations == max_iterations) { ------------------ | Branch (34:9): [True: 47.7k, False: 6.03M] ------------------ 35| 47.7k| counters += sum_8bytes(local); 36| 47.7k| local = vector_u8::zero(); 37| 47.7k| iterations = 0; 38| 47.7k| } 39| 6.08M| } 40| | 41| 1.94k| if (iterations > 0) { ------------------ | Branch (41:7): [True: 1.37k, False: 573] ------------------ 42| 1.37k| count += local.sum_bytes(); 43| 1.37k| } 44| | 45| 1.94k| count += counters.sum(); 46| | 47| 1.94k| return count + scalar::utf8::utf16_length_from_utf8(in + pos, size - pos); 48| 1.94k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoderC2Ev: 101| 1.09k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder7convertEPKcmPc: 115| 586| char *latin1_output) { 116| 586| size_t pos = 0; 117| 586| char *start{latin1_output}; 118| | // In the worst case, we have the haswell kernel which can cause an overflow 119| | // of 8 bytes when calling convert_masked_utf8_to_latin1. If you skip the 120| | // last 16 bytes, and if the data is valid, then it is entirely safe because 121| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 122| | // generally assume that you have valid UTF-8 input, so we are going to go 123| | // back from the end counting 16 leading bytes, to give us a good margin. 124| 586| size_t leading_byte = 0; 125| 586| size_t margin = size; 126| 9.85k| for (; margin > 0 && leading_byte < 16; margin--) { ------------------ | Branch (126:12): [True: 9.71k, False: 140] | Branch (126:26): [True: 9.26k, False: 446] ------------------ 127| 9.26k| leading_byte += (int8_t(in[margin - 1]) > 128| 9.26k| -65); // twos complement of -65 is 1011 1111 ... 129| 9.26k| } 130| | // If the input is long enough, then we have that margin-1 is the eight last 131| | // leading byte. 132| 586| const size_t safety_margin = size - margin + 1; // to avoid overruns! 133| 657k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (133:12): [True: 656k, False: 556] ------------------ 134| 656k| simd8x64 input(reinterpret_cast(in + pos)); 135| 656k| if (input.is_ascii()) { ------------------ | Branch (135:11): [True: 606k, False: 50.5k] ------------------ 136| 606k| input.store((int8_t *)latin1_output); 137| 606k| latin1_output += 64; 138| 606k| pos += 64; 139| 606k| } else { 140| | // you might think that a for-loop would work, but under Visual Studio, 141| | // it is not good enough. 142| 50.5k| static_assert( 143| 50.5k| (simd8x64::NUM_CHUNKS == 2) || 144| 50.5k| (simd8x64::NUM_CHUNKS == 4), 145| 50.5k| "We support either two or four chunks per 64-byte block."); 146| 50.5k| auto zero = simd8{uint8_t(0)}; 147| 50.5k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 148| 50.5k| this->check_utf8_bytes(input.chunks[0], zero); 149| 50.5k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 150| | } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 151| | this->check_utf8_bytes(input.chunks[0], zero); 152| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 153| | this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 154| | this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 155| | } 156| 50.5k| uint64_t utf8_continuation_mask = 157| 50.5k| input.lt(-65 + 1); // -64 is 1100 0000 in twos complement. Note: in 158| | // this case, we also have ASCII to account for. 159| 50.5k| if (utf8_continuation_mask & 1) { ------------------ | Branch (159:13): [True: 30, False: 50.4k] ------------------ 160| 30| return 0; // error 161| 30| } 162| 50.4k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 163| 50.4k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 164| | // We process in blocks of up to 12 bytes except possibly 165| | // for fast paths which may process up to 16 bytes. For the 166| | // slow path to work, we should have at least 12 input bytes left. 167| 50.4k| size_t max_starting_point = (pos + 64) - 12; 168| | // Next loop is going to run at least five times. 169| 501k| while (pos < max_starting_point) { ------------------ | Branch (169:16): [True: 450k, False: 50.4k] ------------------ 170| | // Performance note: our ability to compute 'consumed' and 171| | // then shift and recompute is critical. If there is a 172| | // latency of, say, 4 cycles on getting 'consumed', then 173| | // the inner loop might have a total latency of about 6 cycles. 174| | // Yet we process between 6 to 12 inputs bytes, thus we get 175| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 176| | // for this section of the code. Hence, there is a limit 177| | // to how much we can further increase this latency before 178| | // it seriously harms performance. 179| 450k| size_t consumed = convert_masked_utf8_to_latin1( 180| 450k| in + pos, utf8_end_of_code_point_mask, latin1_output); 181| 450k| pos += consumed; 182| 450k| utf8_end_of_code_point_mask >>= consumed; 183| 450k| } 184| | // At this point there may remain between 0 and 12 bytes in the 185| | // 64-byte block. These bytes will be processed again. So we have an 186| | // 80% efficiency (in the worst case). In practice we expect an 187| | // 85% to 90% efficiency. 188| 50.4k| } 189| 656k| } 190| 556| if (errors()) { ------------------ | Branch (190:9): [True: 232, False: 324] ------------------ 191| 232| return 0; 192| 232| } 193| 324| if (pos < size) { ------------------ | Branch (193:9): [True: 320, False: 4] ------------------ 194| 320| size_t howmany = 195| 320| scalar::utf8_to_latin1::convert(in + pos, size - pos, latin1_output); 196| 320| if (howmany == 0) { ------------------ | Branch (196:11): [True: 152, False: 168] ------------------ 197| 152| return 0; 198| 152| } 199| 168| latin1_output += howmany; 200| 168| } 201| 172| return latin1_output - start; 202| 324| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 106| 132k| const simd8 prev_input) { 107| | // Flip prev1...prev3 so we can easily determine if they are 2+, 3+ or 4+ 108| | // lead bytes (2, 3, 4-byte leads become large positive numbers instead of 109| | // small negative numbers) 110| 132k| simd8 prev1 = input.prev<1>(prev_input); 111| 132k| this->error |= check_special_cases(input, prev1); 112| 132k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin119check_special_casesENS1_4simd5simd8IhEES5_: 8| 132k|check_special_cases(const simd8 input, const simd8 prev1) { 9| | // For UTF-8 to Latin 1, we can allow any ASCII character, and any 10| | // continuation byte, but the non-ASCII leading bytes must be 0b11000011 or 11| | // 0b11000010 and nothing else. 12| | // 13| | // Bit 0 = Too Short (lead byte/ASCII followed by lead byte/ASCII) 14| | // Bit 1 = Too Long (ASCII followed by continuation) 15| | // Bit 2 = Overlong 3-byte 16| | // Bit 4 = Surrogate 17| | // Bit 5 = Overlong 2-byte 18| | // Bit 7 = Two Continuations 19| 132k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 20| | // 11______ 11______ 21| 132k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 22| 132k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 23| 132k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 24| 132k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 25| 132k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 26| 132k| constexpr const uint8_t TOO_LARGE = 1 << 3; // 11110100 1001____ 27| | // 11110100 101_____ 28| | // 11110101 1001____ 29| | // 11110101 101_____ 30| | // 1111011_ 1001____ 31| | // 1111011_ 101_____ 32| | // 11111___ 1001____ 33| | // 11111___ 101_____ 34| 132k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 35| | // 11110101 1000____ 36| | // 1111011_ 1000____ 37| | // 11111___ 1000____ 38| 132k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 39| 132k| constexpr const uint8_t FORBIDDEN = 0xff; 40| | 41| 132k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 42| | // 0_______ ________ 43| 132k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 44| 132k| TOO_LONG, 45| | // 10______ ________ 46| 132k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 47| | // 1100____ ________ 48| 132k| TOO_SHORT | OVERLONG_2, 49| | // 1101____ ________ 50| 132k| FORBIDDEN, 51| | // 1110____ ________ 52| 132k| FORBIDDEN, 53| | // 1111____ ________ 54| 132k| FORBIDDEN); 55| 132k| constexpr const uint8_t CARRY = 56| 132k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 57| 132k| const simd8 byte_1_low = 58| 132k| (prev1 & 0x0F) 59| 132k| .lookup_16( 60| | // ____0000 ________ 61| 132k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 62| | // ____0001 ________ 63| 132k| CARRY | OVERLONG_2, 64| | // ____001_ ________ 65| 132k| CARRY, CARRY, 66| | 67| | // ____0100 ________ 68| 132k| FORBIDDEN, 69| | // ____0101 ________ 70| 132k| FORBIDDEN, 71| | // ____011_ ________ 72| 132k| FORBIDDEN, FORBIDDEN, 73| | 74| | // ____1___ ________ 75| 132k| FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, 76| | // ____1101 ________ 77| 132k| FORBIDDEN, FORBIDDEN, FORBIDDEN); 78| 132k| const simd8 byte_2_high = input.shr<4>().lookup_16( 79| | // ________ 0_______ 80| 132k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 81| 132k| TOO_SHORT, TOO_SHORT, 82| | 83| | // ________ 1000____ 84| 132k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 85| 132k| OVERLONG_4, 86| | // ________ 1001____ 87| 132k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 88| | // ________ 101_____ 89| 132k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 90| 132k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 91| | 92| | // ________ 11______ 93| 132k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 94| 132k| return (byte_1_high & byte_1_low & byte_2_high); 95| 132k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder6errorsEv: 308| 16.4k| simdutf_really_inline bool errors() const { 309| 16.4k| return this->error.any_bits_set_anywhere(); 310| 16.4k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder19convert_with_errorsEPKcmPc: 205| 512| char *latin1_output) { 206| 512| size_t pos = 0; 207| 512| char *start{latin1_output}; 208| | // In the worst case, we have the haswell kernel which can cause an overflow 209| | // of 8 bytes when calling convert_masked_utf8_to_latin1. If you skip the 210| | // last 16 bytes, and if the data is valid, then it is entirely safe because 211| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 212| | // generally assume that you have valid UTF-8 input, so we are going to go 213| | // back from the end counting 8 leading bytes, to give us a good margin. 214| 512| size_t leading_byte = 0; 215| 512| size_t margin = size; 216| 4.89k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (216:12): [True: 4.77k, False: 120] | Branch (216:26): [True: 4.37k, False: 392] ------------------ 217| 4.37k| leading_byte += (int8_t(in[margin - 1]) > -65); 218| 4.37k| } 219| | // If the input is long enough, then we have that margin-1 is the eight last 220| | // leading byte. 221| 512| const size_t safety_margin = size - margin + 1; // to avoid overruns! 222| 313k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (222:12): [True: 312k, False: 418] ------------------ 223| 312k| simd8x64 input(reinterpret_cast(in + pos)); 224| 312k| if (input.is_ascii()) { ------------------ | Branch (224:11): [True: 297k, False: 15.5k] ------------------ 225| 297k| input.store((int8_t *)latin1_output); 226| 297k| latin1_output += 64; 227| 297k| pos += 64; 228| 297k| } else { 229| | // you might think that a for-loop would work, but under Visual Studio, 230| | // it is not good enough. 231| 15.5k| static_assert( 232| 15.5k| (simd8x64::NUM_CHUNKS == 2) || 233| 15.5k| (simd8x64::NUM_CHUNKS == 4), 234| 15.5k| "We support either two or four chunks per 64-byte block."); 235| 15.5k| auto zero = simd8{uint8_t(0)}; 236| 15.5k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 237| 15.5k| this->check_utf8_bytes(input.chunks[0], zero); 238| 15.5k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 239| | } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 240| | this->check_utf8_bytes(input.chunks[0], zero); 241| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 242| | this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 243| | this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 244| | } 245| 15.5k| if (errors()) { ------------------ | Branch (245:13): [True: 94, False: 15.4k] ------------------ 246| | // rewind_and_convert_with_errors will seek a potential error from 247| | // in+pos onward, with the ability to go back up to pos bytes, and 248| | // read size-pos bytes forward. 249| 94| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 250| 94| pos, in + pos, size - pos, latin1_output); 251| 94| res.count += pos; 252| 94| return res; 253| 94| } 254| 15.4k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 255| 15.4k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 256| 15.4k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 257| | // We process in blocks of up to 12 bytes except possibly 258| | // for fast paths which may process up to 16 bytes. For the 259| | // slow path to work, we should have at least 12 input bytes left. 260| 15.4k| size_t max_starting_point = (pos + 64) - 12; 261| | // Next loop is going to run at least five times. 262| 153k| while (pos < max_starting_point) { ------------------ | Branch (262:16): [True: 138k, False: 15.4k] ------------------ 263| | // Performance note: our ability to compute 'consumed' and 264| | // then shift and recompute is critical. If there is a 265| | // latency of, say, 4 cycles on getting 'consumed', then 266| | // the inner loop might have a total latency of about 6 cycles. 267| | // Yet we process between 6 to 12 inputs bytes, thus we get 268| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 269| | // for this section of the code. Hence, there is a limit 270| | // to how much we can further increase this latency before 271| | // it seriously harms performance. 272| 138k| size_t consumed = convert_masked_utf8_to_latin1( 273| 138k| in + pos, utf8_end_of_code_point_mask, latin1_output); 274| 138k| pos += consumed; 275| 138k| utf8_end_of_code_point_mask >>= consumed; 276| 138k| } 277| | // At this point there may remain between 0 and 12 bytes in the 278| | // 64-byte block. These bytes will be processed again. So we have an 279| | // 80% efficiency (in the worst case). In practice we expect an 280| | // 85% to 90% efficiency. 281| 15.4k| } 282| 312k| } 283| 418| if (errors()) { ------------------ | Branch (283:9): [True: 0, False: 418] ------------------ 284| | // rewind_and_convert_with_errors will seek a potential error from in+pos 285| | // onward, with the ability to go back up to pos bytes, and read size-pos 286| | // bytes forward. 287| 0| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 288| 0| pos, in + pos, size - pos, latin1_output); 289| 0| res.count += pos; 290| 0| return res; 291| 0| } 292| 418| if (pos < size) { ------------------ | Branch (292:9): [True: 414, False: 4] ------------------ 293| | // rewind_and_convert_with_errors will seek a potential error from in+pos 294| | // onward, with the ability to go back up to pos bytes, and read size-pos 295| | // bytes forward. 296| 414| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 297| 414| pos, in + pos, size - pos, latin1_output); 298| 414| if (res.error) { // In case of error, we want the error position ------------------ | Branch (298:11): [True: 192, False: 222] ------------------ 299| 192| res.count += pos; 300| 192| return res; 301| 222| } else { // In case of success, we want the number of word written 302| 222| latin1_output += res.count; 303| 222| } 304| 414| } 305| 226| return result(error_code::SUCCESS, latin1_output - start); 306| 418| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoderC2Ev: 101| 1.09k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder7convertEPKcmPc: 115| 586| char *latin1_output) { 116| 586| size_t pos = 0; 117| 586| char *start{latin1_output}; 118| | // In the worst case, we have the haswell kernel which can cause an overflow 119| | // of 8 bytes when calling convert_masked_utf8_to_latin1. If you skip the 120| | // last 16 bytes, and if the data is valid, then it is entirely safe because 121| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 122| | // generally assume that you have valid UTF-8 input, so we are going to go 123| | // back from the end counting 16 leading bytes, to give us a good margin. 124| 586| size_t leading_byte = 0; 125| 586| size_t margin = size; 126| 9.85k| for (; margin > 0 && leading_byte < 16; margin--) { ------------------ | Branch (126:12): [True: 9.71k, False: 140] | Branch (126:26): [True: 9.26k, False: 446] ------------------ 127| 9.26k| leading_byte += (int8_t(in[margin - 1]) > 128| 9.26k| -65); // twos complement of -65 is 1011 1111 ... 129| 9.26k| } 130| | // If the input is long enough, then we have that margin-1 is the eight last 131| | // leading byte. 132| 586| const size_t safety_margin = size - margin + 1; // to avoid overruns! 133| 657k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (133:12): [True: 656k, False: 556] ------------------ 134| 656k| simd8x64 input(reinterpret_cast(in + pos)); 135| 656k| if (input.is_ascii()) { ------------------ | Branch (135:11): [True: 606k, False: 50.5k] ------------------ 136| 606k| input.store((int8_t *)latin1_output); 137| 606k| latin1_output += 64; 138| 606k| pos += 64; 139| 606k| } else { 140| | // you might think that a for-loop would work, but under Visual Studio, 141| | // it is not good enough. 142| 50.5k| static_assert( 143| 50.5k| (simd8x64::NUM_CHUNKS == 2) || 144| 50.5k| (simd8x64::NUM_CHUNKS == 4), 145| 50.5k| "We support either two or four chunks per 64-byte block."); 146| 50.5k| auto zero = simd8{uint8_t(0)}; 147| | if constexpr (simd8x64::NUM_CHUNKS == 2) { 148| | this->check_utf8_bytes(input.chunks[0], zero); 149| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 150| 50.5k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 151| 50.5k| this->check_utf8_bytes(input.chunks[0], zero); 152| 50.5k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 153| 50.5k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 154| 50.5k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 155| 50.5k| } 156| 50.5k| uint64_t utf8_continuation_mask = 157| 50.5k| input.lt(-65 + 1); // -64 is 1100 0000 in twos complement. Note: in 158| | // this case, we also have ASCII to account for. 159| 50.5k| if (utf8_continuation_mask & 1) { ------------------ | Branch (159:13): [True: 30, False: 50.4k] ------------------ 160| 30| return 0; // error 161| 30| } 162| 50.4k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 163| 50.4k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 164| | // We process in blocks of up to 12 bytes except possibly 165| | // for fast paths which may process up to 16 bytes. For the 166| | // slow path to work, we should have at least 12 input bytes left. 167| 50.4k| size_t max_starting_point = (pos + 64) - 12; 168| | // Next loop is going to run at least five times. 169| 501k| while (pos < max_starting_point) { ------------------ | Branch (169:16): [True: 450k, False: 50.4k] ------------------ 170| | // Performance note: our ability to compute 'consumed' and 171| | // then shift and recompute is critical. If there is a 172| | // latency of, say, 4 cycles on getting 'consumed', then 173| | // the inner loop might have a total latency of about 6 cycles. 174| | // Yet we process between 6 to 12 inputs bytes, thus we get 175| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 176| | // for this section of the code. Hence, there is a limit 177| | // to how much we can further increase this latency before 178| | // it seriously harms performance. 179| 450k| size_t consumed = convert_masked_utf8_to_latin1( 180| 450k| in + pos, utf8_end_of_code_point_mask, latin1_output); 181| 450k| pos += consumed; 182| 450k| utf8_end_of_code_point_mask >>= consumed; 183| 450k| } 184| | // At this point there may remain between 0 and 12 bytes in the 185| | // 64-byte block. These bytes will be processed again. So we have an 186| | // 80% efficiency (in the worst case). In practice we expect an 187| | // 85% to 90% efficiency. 188| 50.4k| } 189| 656k| } 190| 556| if (errors()) { ------------------ | Branch (190:9): [True: 232, False: 324] ------------------ 191| 232| return 0; 192| 232| } 193| 324| if (pos < size) { ------------------ | Branch (193:9): [True: 320, False: 4] ------------------ 194| 320| size_t howmany = 195| 320| scalar::utf8_to_latin1::convert(in + pos, size - pos, latin1_output); 196| 320| if (howmany == 0) { ------------------ | Branch (196:11): [True: 152, False: 168] ------------------ 197| 152| return 0; 198| 152| } 199| 168| latin1_output += howmany; 200| 168| } 201| 172| return latin1_output - start; 202| 324| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 106| 264k| const simd8 prev_input) { 107| | // Flip prev1...prev3 so we can easily determine if they are 2+, 3+ or 4+ 108| | // lead bytes (2, 3, 4-byte leads become large positive numbers instead of 109| | // small negative numbers) 110| 264k| simd8 prev1 = input.prev<1>(prev_input); 111| 264k| this->error |= check_special_cases(input, prev1); 112| 264k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin119check_special_casesENS1_4simd5simd8IhEES5_: 8| 264k|check_special_cases(const simd8 input, const simd8 prev1) { 9| | // For UTF-8 to Latin 1, we can allow any ASCII character, and any 10| | // continuation byte, but the non-ASCII leading bytes must be 0b11000011 or 11| | // 0b11000010 and nothing else. 12| | // 13| | // Bit 0 = Too Short (lead byte/ASCII followed by lead byte/ASCII) 14| | // Bit 1 = Too Long (ASCII followed by continuation) 15| | // Bit 2 = Overlong 3-byte 16| | // Bit 4 = Surrogate 17| | // Bit 5 = Overlong 2-byte 18| | // Bit 7 = Two Continuations 19| 264k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 20| | // 11______ 11______ 21| 264k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 22| 264k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 23| 264k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 24| 264k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 25| 264k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 26| 264k| constexpr const uint8_t TOO_LARGE = 1 << 3; // 11110100 1001____ 27| | // 11110100 101_____ 28| | // 11110101 1001____ 29| | // 11110101 101_____ 30| | // 1111011_ 1001____ 31| | // 1111011_ 101_____ 32| | // 11111___ 1001____ 33| | // 11111___ 101_____ 34| 264k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 35| | // 11110101 1000____ 36| | // 1111011_ 1000____ 37| | // 11111___ 1000____ 38| 264k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 39| 264k| constexpr const uint8_t FORBIDDEN = 0xff; 40| | 41| 264k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 42| | // 0_______ ________ 43| 264k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 44| 264k| TOO_LONG, 45| | // 10______ ________ 46| 264k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 47| | // 1100____ ________ 48| 264k| TOO_SHORT | OVERLONG_2, 49| | // 1101____ ________ 50| 264k| FORBIDDEN, 51| | // 1110____ ________ 52| 264k| FORBIDDEN, 53| | // 1111____ ________ 54| 264k| FORBIDDEN); 55| 264k| constexpr const uint8_t CARRY = 56| 264k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 57| 264k| const simd8 byte_1_low = 58| 264k| (prev1 & 0x0F) 59| 264k| .lookup_16( 60| | // ____0000 ________ 61| 264k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 62| | // ____0001 ________ 63| 264k| CARRY | OVERLONG_2, 64| | // ____001_ ________ 65| 264k| CARRY, CARRY, 66| | 67| | // ____0100 ________ 68| 264k| FORBIDDEN, 69| | // ____0101 ________ 70| 264k| FORBIDDEN, 71| | // ____011_ ________ 72| 264k| FORBIDDEN, FORBIDDEN, 73| | 74| | // ____1___ ________ 75| 264k| FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, 76| | // ____1101 ________ 77| 264k| FORBIDDEN, FORBIDDEN, FORBIDDEN); 78| 264k| const simd8 byte_2_high = input.shr<4>().lookup_16( 79| | // ________ 0_______ 80| 264k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 81| 264k| TOO_SHORT, TOO_SHORT, 82| | 83| | // ________ 1000____ 84| 264k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 85| 264k| OVERLONG_4, 86| | // ________ 1001____ 87| 264k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 88| | // ________ 101_____ 89| 264k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 90| 264k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 91| | 92| | // ________ 11______ 93| 264k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 94| 264k| return (byte_1_high & byte_1_low & byte_2_high); 95| 264k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder6errorsEv: 308| 16.4k| simdutf_really_inline bool errors() const { 309| 16.4k| return this->error.any_bits_set_anywhere(); 310| 16.4k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder19convert_with_errorsEPKcmPc: 205| 512| char *latin1_output) { 206| 512| size_t pos = 0; 207| 512| char *start{latin1_output}; 208| | // In the worst case, we have the haswell kernel which can cause an overflow 209| | // of 8 bytes when calling convert_masked_utf8_to_latin1. If you skip the 210| | // last 16 bytes, and if the data is valid, then it is entirely safe because 211| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 212| | // generally assume that you have valid UTF-8 input, so we are going to go 213| | // back from the end counting 8 leading bytes, to give us a good margin. 214| 512| size_t leading_byte = 0; 215| 512| size_t margin = size; 216| 4.89k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (216:12): [True: 4.77k, False: 120] | Branch (216:26): [True: 4.37k, False: 392] ------------------ 217| 4.37k| leading_byte += (int8_t(in[margin - 1]) > -65); 218| 4.37k| } 219| | // If the input is long enough, then we have that margin-1 is the eight last 220| | // leading byte. 221| 512| const size_t safety_margin = size - margin + 1; // to avoid overruns! 222| 313k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (222:12): [True: 312k, False: 418] ------------------ 223| 312k| simd8x64 input(reinterpret_cast(in + pos)); 224| 312k| if (input.is_ascii()) { ------------------ | Branch (224:11): [True: 297k, False: 15.5k] ------------------ 225| 297k| input.store((int8_t *)latin1_output); 226| 297k| latin1_output += 64; 227| 297k| pos += 64; 228| 297k| } else { 229| | // you might think that a for-loop would work, but under Visual Studio, 230| | // it is not good enough. 231| 15.5k| static_assert( 232| 15.5k| (simd8x64::NUM_CHUNKS == 2) || 233| 15.5k| (simd8x64::NUM_CHUNKS == 4), 234| 15.5k| "We support either two or four chunks per 64-byte block."); 235| 15.5k| auto zero = simd8{uint8_t(0)}; 236| | if constexpr (simd8x64::NUM_CHUNKS == 2) { 237| | this->check_utf8_bytes(input.chunks[0], zero); 238| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 239| 15.5k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 240| 15.5k| this->check_utf8_bytes(input.chunks[0], zero); 241| 15.5k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 242| 15.5k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 243| 15.5k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 244| 15.5k| } 245| 15.5k| if (errors()) { ------------------ | Branch (245:13): [True: 94, False: 15.4k] ------------------ 246| | // rewind_and_convert_with_errors will seek a potential error from 247| | // in+pos onward, with the ability to go back up to pos bytes, and 248| | // read size-pos bytes forward. 249| 94| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 250| 94| pos, in + pos, size - pos, latin1_output); 251| 94| res.count += pos; 252| 94| return res; 253| 94| } 254| 15.4k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 255| 15.4k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 256| 15.4k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 257| | // We process in blocks of up to 12 bytes except possibly 258| | // for fast paths which may process up to 16 bytes. For the 259| | // slow path to work, we should have at least 12 input bytes left. 260| 15.4k| size_t max_starting_point = (pos + 64) - 12; 261| | // Next loop is going to run at least five times. 262| 153k| while (pos < max_starting_point) { ------------------ | Branch (262:16): [True: 138k, False: 15.4k] ------------------ 263| | // Performance note: our ability to compute 'consumed' and 264| | // then shift and recompute is critical. If there is a 265| | // latency of, say, 4 cycles on getting 'consumed', then 266| | // the inner loop might have a total latency of about 6 cycles. 267| | // Yet we process between 6 to 12 inputs bytes, thus we get 268| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 269| | // for this section of the code. Hence, there is a limit 270| | // to how much we can further increase this latency before 271| | // it seriously harms performance. 272| 138k| size_t consumed = convert_masked_utf8_to_latin1( 273| 138k| in + pos, utf8_end_of_code_point_mask, latin1_output); 274| 138k| pos += consumed; 275| 138k| utf8_end_of_code_point_mask >>= consumed; 276| 138k| } 277| | // At this point there may remain between 0 and 12 bytes in the 278| | // 64-byte block. These bytes will be processed again. So we have an 279| | // 80% efficiency (in the worst case). In practice we expect an 280| | // 85% to 90% efficiency. 281| 15.4k| } 282| 312k| } 283| 418| if (errors()) { ------------------ | Branch (283:9): [True: 0, False: 418] ------------------ 284| | // rewind_and_convert_with_errors will seek a potential error from in+pos 285| | // onward, with the ability to go back up to pos bytes, and read size-pos 286| | // bytes forward. 287| 0| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 288| 0| pos, in + pos, size - pos, latin1_output); 289| 0| res.count += pos; 290| 0| return res; 291| 0| } 292| 418| if (pos < size) { ------------------ | Branch (292:9): [True: 414, False: 4] ------------------ 293| | // rewind_and_convert_with_errors will seek a potential error from in+pos 294| | // onward, with the ability to go back up to pos bytes, and read size-pos 295| | // bytes forward. 296| 414| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 297| 414| pos, in + pos, size - pos, latin1_output); 298| 414| if (res.error) { // In case of error, we want the error position ------------------ | Branch (298:11): [True: 192, False: 222] ------------------ 299| 192| res.count += pos; 300| 192| return res; 301| 222| } else { // In case of success, we want the number of word written 302| 222| latin1_output += res.count; 303| 222| } 304| 414| } 305| 226| return result(error_code::SUCCESS, latin1_output - start); 306| 418| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoderC2Ev: 114| 2.88k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoderC2Ev: 114| 2.88k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE0EEEmPKcmPDs: 130| 776| char16_t *utf16_output) { 131| 776| size_t pos = 0; 132| 776| char16_t *start{utf16_output}; 133| | // In the worst case, we have the haswell kernel which can cause an overflow 134| | // of 8 bytes when calling convert_masked_utf8_to_utf16. If you skip the 135| | // last 16 bytes, and if the data is valid, then it is entirely safe because 136| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 137| | // generally assume that you have valid UTF-8 input, so we are going to go 138| | // back from the end counting 8 leading bytes, to give us a good margin. 139| 776| size_t leading_byte = 0; 140| 776| size_t margin = size; 141| 7.21k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 6.96k, False: 252] | Branch (141:26): [True: 6.44k, False: 524] ------------------ 142| 6.44k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 6.44k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 776| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 502k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 501k, False: 756] ------------------ 148| 501k| simd8x64 input(reinterpret_cast(in + pos)); 149| 501k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 444k, False: 56.9k] ------------------ 150| 444k| input.store_ascii_as_utf16(utf16_output); 151| 444k| utf16_output += 64; 152| 444k| pos += 64; 153| 444k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 56.9k| static_assert( 157| 56.9k| (simd8x64::NUM_CHUNKS == 2) || 158| 56.9k| (simd8x64::NUM_CHUNKS == 4), 159| 56.9k| "We support either two or four chunks per 64-byte block."); 160| 56.9k| auto zero = simd8{uint8_t(0)}; 161| 56.9k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 162| 56.9k| this->check_utf8_bytes(input.chunks[0], zero); 163| 56.9k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 164| | } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 165| | this->check_utf8_bytes(input.chunks[0], zero); 166| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 167| | this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 168| | this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 169| | } 170| 56.9k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 56.9k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 20, False: 56.9k] ------------------ 172| 20| return 0; // error 173| 20| } 174| 56.9k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 56.9k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 176| | // We process in blocks of up to 12 bytes except possibly 177| | // for fast paths which may process up to 16 bytes. For the 178| | // slow path to work, we should have at least 12 input bytes left. 179| 56.9k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 558k| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 501k, False: 56.9k] ------------------ 182| | // Performance note: our ability to compute 'consumed' and 183| | // then shift and recompute is critical. If there is a 184| | // latency of, say, 4 cycles on getting 'consumed', then 185| | // the inner loop might have a total latency of about 6 cycles. 186| | // Yet we process between 6 to 12 inputs bytes, thus we get 187| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 188| | // for this section of the code. Hence, there is a limit 189| | // to how much we can further increase this latency before 190| | // it seriously harms performance. 191| 501k| size_t consumed = convert_masked_utf8_to_utf16( 192| 501k| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 501k| pos += consumed; 194| 501k| utf8_end_of_code_point_mask >>= consumed; 195| 501k| } 196| | // At this point there may remain between 0 and 12 bytes in the 197| | // 64-byte block. These bytes will be processed again. So we have an 198| | // 80% efficiency (in the worst case). In practice we expect an 199| | // 85% to 90% efficiency. 200| 56.9k| } 201| 501k| } 202| 756| if (errors()) { ------------------ | Branch (202:9): [True: 308, False: 448] ------------------ 203| 308| return 0; 204| 308| } 205| 448| if (pos < size) { ------------------ | Branch (205:9): [True: 446, False: 2] ------------------ 206| 446| size_t howmany = scalar::utf8_to_utf16::convert( 207| 446| in + pos, size - pos, utf16_output); 208| 446| if (howmany == 0) { ------------------ | Branch (208:11): [True: 204, False: 242] ------------------ 209| 204| return 0; 210| 204| } 211| 242| utf16_output += howmany; 212| 242| } 213| 244| return utf16_output - start; 214| 448| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 803k| const simd8 prev_input) { 120| | // Flip prev1...prev3 so we can easily determine if they are 2+, 3+ or 4+ 121| | // lead bytes (2, 3, 4-byte leads become large positive numbers instead of 122| | // small negative numbers) 123| 803k| simd8 prev1 = input.prev<1>(prev_input); 124| 803k| simd8 sc = check_special_cases(input, prev1); 125| 803k| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 803k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1619check_special_casesENS1_4simd5simd8IhEES5_: 8| 803k|check_special_cases(const simd8 input, const simd8 prev1) { 9| | // Bit 0 = Too Short (lead byte/ASCII followed by lead byte/ASCII) 10| | // Bit 1 = Too Long (ASCII followed by continuation) 11| | // Bit 2 = Overlong 3-byte 12| | // Bit 4 = Surrogate 13| | // Bit 5 = Overlong 2-byte 14| | // Bit 7 = Two Continuations 15| 803k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 803k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 803k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 803k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 803k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 803k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 803k| constexpr const uint8_t TOO_LARGE = 1 << 3; // 11110100 1001____ 23| | // 11110100 101_____ 24| | // 11110101 1001____ 25| | // 11110101 101_____ 26| | // 1111011_ 1001____ 27| | // 1111011_ 101_____ 28| | // 11111___ 1001____ 29| | // 11111___ 101_____ 30| 803k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 803k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 803k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 803k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 803k| TOO_LONG, 40| | // 10______ ________ 41| 803k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 803k| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 803k| TOO_SHORT, 46| | // 1110____ ________ 47| 803k| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 803k| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 803k| constexpr const uint8_t CARRY = 51| 803k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 803k| const simd8 byte_1_low = 53| 803k| (prev1 & 0x0F) 54| 803k| .lookup_16( 55| | // ____0000 ________ 56| 803k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 803k| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 803k| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 803k| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 803k| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 803k| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 803k| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 803k| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 803k| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 803k| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 803k| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 803k| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 803k| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 803k| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 803k| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 803k| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 803k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 803k| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 803k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 803k| OVERLONG_4, 88| | // ________ 1001____ 89| 803k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 803k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 803k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 803k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 803k| return (byte_1_high & byte_1_low & byte_2_high); 97| 803k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1623check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 803k| const simd8 sc) { 102| 803k| simd8 prev2 = input.prev<2>(prev_input); 103| 803k| simd8 prev3 = input.prev<3>(prev_input); 104| 803k| simd8 must23 = 105| 803k| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 803k| simd8 must23_80 = must23 & uint8_t(0x80); 107| 803k| return must23_80 ^ sc; 108| 803k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder6errorsEv: 324| 166k| simdutf_really_inline bool errors() const { 325| 166k| return this->error.any_bits_set_anywhere(); 326| 166k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE1EEEmPKcmPDs: 130| 794| char16_t *utf16_output) { 131| 794| size_t pos = 0; 132| 794| char16_t *start{utf16_output}; 133| | // In the worst case, we have the haswell kernel which can cause an overflow 134| | // of 8 bytes when calling convert_masked_utf8_to_utf16. If you skip the 135| | // last 16 bytes, and if the data is valid, then it is entirely safe because 136| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 137| | // generally assume that you have valid UTF-8 input, so we are going to go 138| | // back from the end counting 8 leading bytes, to give us a good margin. 139| 794| size_t leading_byte = 0; 140| 794| size_t margin = size; 141| 7.34k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 7.10k, False: 246] | Branch (141:26): [True: 6.55k, False: 548] ------------------ 142| 6.55k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 6.55k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 794| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 674k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 674k, False: 770] ------------------ 148| 674k| simd8x64 input(reinterpret_cast(in + pos)); 149| 674k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 492k, False: 181k] ------------------ 150| 492k| input.store_ascii_as_utf16(utf16_output); 151| 492k| utf16_output += 64; 152| 492k| pos += 64; 153| 492k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 181k| static_assert( 157| 181k| (simd8x64::NUM_CHUNKS == 2) || 158| 181k| (simd8x64::NUM_CHUNKS == 4), 159| 181k| "We support either two or four chunks per 64-byte block."); 160| 181k| auto zero = simd8{uint8_t(0)}; 161| 181k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 162| 181k| this->check_utf8_bytes(input.chunks[0], zero); 163| 181k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 164| | } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 165| | this->check_utf8_bytes(input.chunks[0], zero); 166| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 167| | this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 168| | this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 169| | } 170| 181k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 181k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 24, False: 181k] ------------------ 172| 24| return 0; // error 173| 24| } 174| 181k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 181k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 176| | // We process in blocks of up to 12 bytes except possibly 177| | // for fast paths which may process up to 16 bytes. For the 178| | // slow path to work, we should have at least 12 input bytes left. 179| 181k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 1.79M| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 1.61M, False: 181k] ------------------ 182| | // Performance note: our ability to compute 'consumed' and 183| | // then shift and recompute is critical. If there is a 184| | // latency of, say, 4 cycles on getting 'consumed', then 185| | // the inner loop might have a total latency of about 6 cycles. 186| | // Yet we process between 6 to 12 inputs bytes, thus we get 187| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 188| | // for this section of the code. Hence, there is a limit 189| | // to how much we can further increase this latency before 190| | // it seriously harms performance. 191| 1.61M| size_t consumed = convert_masked_utf8_to_utf16( 192| 1.61M| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 1.61M| pos += consumed; 194| 1.61M| utf8_end_of_code_point_mask >>= consumed; 195| 1.61M| } 196| | // At this point there may remain between 0 and 12 bytes in the 197| | // 64-byte block. These bytes will be processed again. So we have an 198| | // 80% efficiency (in the worst case). In practice we expect an 199| | // 85% to 90% efficiency. 200| 181k| } 201| 674k| } 202| 770| if (errors()) { ------------------ | Branch (202:9): [True: 328, False: 442] ------------------ 203| 328| return 0; 204| 328| } 205| 442| if (pos < size) { ------------------ | Branch (205:9): [True: 440, False: 2] ------------------ 206| 440| size_t howmany = scalar::utf8_to_utf16::convert( 207| 440| in + pos, size - pos, utf16_output); 208| 440| if (howmany == 0) { ------------------ | Branch (208:11): [True: 206, False: 234] ------------------ 209| 206| return 0; 210| 206| } 211| 234| utf16_output += howmany; 212| 234| } 213| 236| return utf16_output - start; 214| 442| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKcmPDs: 218| 610| char16_t *utf16_output) { 219| 610| size_t pos = 0; 220| 610| char16_t *start{utf16_output}; 221| | // In the worst case, we have the haswell kernel which can cause an overflow 222| | // of 8 bytes when calling convert_masked_utf8_to_utf16. If you skip the 223| | // last 16 bytes, and if the data is valid, then it is entirely safe because 224| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 225| | // generally assume that you have valid UTF-8 input, so we are going to go 226| | // back from the end counting 8 leading bytes, to give us a good margin. 227| 610| size_t leading_byte = 0; 228| 610| size_t margin = size; 229| 5.37k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 5.13k, False: 242] | Branch (229:26): [True: 4.76k, False: 368] ------------------ 230| 4.76k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 4.76k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 610| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 384k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 384k, False: 522] ------------------ 236| 384k| simd8x64 input(reinterpret_cast(in + pos)); 237| 384k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 294k, False: 89.8k] ------------------ 238| 294k| input.store_ascii_as_utf16(utf16_output); 239| 294k| utf16_output += 64; 240| 294k| pos += 64; 241| 294k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 89.8k| static_assert( 245| 89.8k| (simd8x64::NUM_CHUNKS == 2) || 246| 89.8k| (simd8x64::NUM_CHUNKS == 4), 247| 89.8k| "We support either two or four chunks per 64-byte block."); 248| 89.8k| auto zero = simd8{uint8_t(0)}; 249| 89.8k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 250| 89.8k| this->check_utf8_bytes(input.chunks[0], zero); 251| 89.8k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 252| | } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 253| | this->check_utf8_bytes(input.chunks[0], zero); 254| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 255| | this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 256| | this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 257| | } 258| 89.8k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 89.8k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 88, False: 89.7k] | Branch (259:25): [True: 0, False: 89.7k] ------------------ 260| | // rewind_and_convert_with_errors will seek a potential error from 261| | // in+pos onward, with the ability to go back up to pos bytes, and 262| | // read size-pos bytes forward. 263| 88| result res = 264| 88| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 88| pos, in + pos, size - pos, utf16_output); 266| 88| res.count += pos; 267| 88| return res; 268| 88| } 269| 89.7k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 89.7k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 271| | // We process in blocks of up to 12 bytes except possibly 272| | // for fast paths which may process up to 16 bytes. For the 273| | // slow path to work, we should have at least 12 input bytes left. 274| 89.7k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 885k| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 795k, False: 89.7k] ------------------ 277| | // Performance note: our ability to compute 'consumed' and 278| | // then shift and recompute is critical. If there is a 279| | // latency of, say, 4 cycles on getting 'consumed', then 280| | // the inner loop might have a total latency of about 6 cycles. 281| | // Yet we process between 6 to 12 inputs bytes, thus we get 282| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 283| | // for this section of the code. Hence, there is a limit 284| | // to how much we can further increase this latency before 285| | // it seriously harms performance. 286| 795k| size_t consumed = convert_masked_utf8_to_utf16( 287| 795k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 795k| pos += consumed; 289| 795k| utf8_end_of_code_point_mask >>= consumed; 290| 795k| } 291| | // At this point there may remain between 0 and 12 bytes in the 292| | // 64-byte block. These bytes will be processed again. So we have an 293| | // 80% efficiency (in the worst case). In practice we expect an 294| | // 85% to 90% efficiency. 295| 89.7k| } 296| 384k| } 297| 522| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 522] ------------------ 298| | // rewind_and_convert_with_errors will seek a potential error from in+pos 299| | // onward, with the ability to go back up to pos bytes, and read size-pos 300| | // bytes forward. 301| 0| result res = 302| 0| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 303| 0| pos, in + pos, size - pos, utf16_output); 304| 0| res.count += pos; 305| 0| return res; 306| 0| } 307| 522| if (pos < size) { ------------------ | Branch (307:9): [True: 518, False: 4] ------------------ 308| | // rewind_and_convert_with_errors will seek a potential error from in+pos 309| | // onward, with the ability to go back up to pos bytes, and read size-pos 310| | // bytes forward. 311| 518| result res = 312| 518| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 518| pos, in + pos, size - pos, utf16_output); 314| 518| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 280, False: 238] ------------------ 315| 280| res.count += pos; 316| 280| return res; 317| 280| } else { // In case of success, we want the number of word written 318| 238| utf16_output += res.count; 319| 238| } 320| 518| } 321| 242| return result(error_code::SUCCESS, utf16_output - start); 322| 522| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKcmPDs: 218| 706| char16_t *utf16_output) { 219| 706| size_t pos = 0; 220| 706| char16_t *start{utf16_output}; 221| | // In the worst case, we have the haswell kernel which can cause an overflow 222| | // of 8 bytes when calling convert_masked_utf8_to_utf16. If you skip the 223| | // last 16 bytes, and if the data is valid, then it is entirely safe because 224| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 225| | // generally assume that you have valid UTF-8 input, so we are going to go 226| | // back from the end counting 8 leading bytes, to give us a good margin. 227| 706| size_t leading_byte = 0; 228| 706| size_t margin = size; 229| 6.20k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 5.96k, False: 236] | Branch (229:26): [True: 5.49k, False: 470] ------------------ 230| 5.49k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 5.49k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 706| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 365k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 365k, False: 634] ------------------ 236| 365k| simd8x64 input(reinterpret_cast(in + pos)); 237| 365k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 291k, False: 73.6k] ------------------ 238| 291k| input.store_ascii_as_utf16(utf16_output); 239| 291k| utf16_output += 64; 240| 291k| pos += 64; 241| 291k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 73.6k| static_assert( 245| 73.6k| (simd8x64::NUM_CHUNKS == 2) || 246| 73.6k| (simd8x64::NUM_CHUNKS == 4), 247| 73.6k| "We support either two or four chunks per 64-byte block."); 248| 73.6k| auto zero = simd8{uint8_t(0)}; 249| 73.6k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 250| 73.6k| this->check_utf8_bytes(input.chunks[0], zero); 251| 73.6k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 252| | } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 253| | this->check_utf8_bytes(input.chunks[0], zero); 254| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 255| | this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 256| | this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 257| | } 258| 73.6k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 73.6k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 72, False: 73.5k] | Branch (259:25): [True: 0, False: 73.5k] ------------------ 260| | // rewind_and_convert_with_errors will seek a potential error from 261| | // in+pos onward, with the ability to go back up to pos bytes, and 262| | // read size-pos bytes forward. 263| 72| result res = 264| 72| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 72| pos, in + pos, size - pos, utf16_output); 266| 72| res.count += pos; 267| 72| return res; 268| 72| } 269| 73.5k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 73.5k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 271| | // We process in blocks of up to 12 bytes except possibly 272| | // for fast paths which may process up to 16 bytes. For the 273| | // slow path to work, we should have at least 12 input bytes left. 274| 73.5k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 725k| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 651k, False: 73.5k] ------------------ 277| | // Performance note: our ability to compute 'consumed' and 278| | // then shift and recompute is critical. If there is a 279| | // latency of, say, 4 cycles on getting 'consumed', then 280| | // the inner loop might have a total latency of about 6 cycles. 281| | // Yet we process between 6 to 12 inputs bytes, thus we get 282| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 283| | // for this section of the code. Hence, there is a limit 284| | // to how much we can further increase this latency before 285| | // it seriously harms performance. 286| 651k| size_t consumed = convert_masked_utf8_to_utf16( 287| 651k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 651k| pos += consumed; 289| 651k| utf8_end_of_code_point_mask >>= consumed; 290| 651k| } 291| | // At this point there may remain between 0 and 12 bytes in the 292| | // 64-byte block. These bytes will be processed again. So we have an 293| | // 80% efficiency (in the worst case). In practice we expect an 294| | // 85% to 90% efficiency. 295| 73.5k| } 296| 365k| } 297| 634| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 634] ------------------ 298| | // rewind_and_convert_with_errors will seek a potential error from in+pos 299| | // onward, with the ability to go back up to pos bytes, and read size-pos 300| | // bytes forward. 301| 0| result res = 302| 0| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 303| 0| pos, in + pos, size - pos, utf16_output); 304| 0| res.count += pos; 305| 0| return res; 306| 0| } 307| 634| if (pos < size) { ------------------ | Branch (307:9): [True: 630, False: 4] ------------------ 308| | // rewind_and_convert_with_errors will seek a potential error from in+pos 309| | // onward, with the ability to go back up to pos bytes, and read size-pos 310| | // bytes forward. 311| 630| result res = 312| 630| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 630| pos, in + pos, size - pos, utf16_output); 314| 630| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 322, False: 308] ------------------ 315| 322| res.count += pos; 316| 322| return res; 317| 322| } else { // In case of success, we want the number of word written 318| 308| utf16_output += res.count; 319| 308| } 320| 630| } 321| 312| return result(error_code::SUCCESS, utf16_output - start); 322| 634| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE0EEEmPKcmPDs: 130| 776| char16_t *utf16_output) { 131| 776| size_t pos = 0; 132| 776| char16_t *start{utf16_output}; 133| | // In the worst case, we have the haswell kernel which can cause an overflow 134| | // of 8 bytes when calling convert_masked_utf8_to_utf16. If you skip the 135| | // last 16 bytes, and if the data is valid, then it is entirely safe because 136| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 137| | // generally assume that you have valid UTF-8 input, so we are going to go 138| | // back from the end counting 8 leading bytes, to give us a good margin. 139| 776| size_t leading_byte = 0; 140| 776| size_t margin = size; 141| 7.21k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 6.96k, False: 252] | Branch (141:26): [True: 6.44k, False: 524] ------------------ 142| 6.44k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 6.44k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 776| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 502k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 501k, False: 756] ------------------ 148| 501k| simd8x64 input(reinterpret_cast(in + pos)); 149| 501k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 444k, False: 56.9k] ------------------ 150| 444k| input.store_ascii_as_utf16(utf16_output); 151| 444k| utf16_output += 64; 152| 444k| pos += 64; 153| 444k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 56.9k| static_assert( 157| 56.9k| (simd8x64::NUM_CHUNKS == 2) || 158| 56.9k| (simd8x64::NUM_CHUNKS == 4), 159| 56.9k| "We support either two or four chunks per 64-byte block."); 160| 56.9k| auto zero = simd8{uint8_t(0)}; 161| | if constexpr (simd8x64::NUM_CHUNKS == 2) { 162| | this->check_utf8_bytes(input.chunks[0], zero); 163| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 164| 56.9k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 165| 56.9k| this->check_utf8_bytes(input.chunks[0], zero); 166| 56.9k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 167| 56.9k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 168| 56.9k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 169| 56.9k| } 170| 56.9k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 56.9k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 20, False: 56.9k] ------------------ 172| 20| return 0; // error 173| 20| } 174| 56.9k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 56.9k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 176| | // We process in blocks of up to 12 bytes except possibly 177| | // for fast paths which may process up to 16 bytes. For the 178| | // slow path to work, we should have at least 12 input bytes left. 179| 56.9k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 558k| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 501k, False: 56.9k] ------------------ 182| | // Performance note: our ability to compute 'consumed' and 183| | // then shift and recompute is critical. If there is a 184| | // latency of, say, 4 cycles on getting 'consumed', then 185| | // the inner loop might have a total latency of about 6 cycles. 186| | // Yet we process between 6 to 12 inputs bytes, thus we get 187| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 188| | // for this section of the code. Hence, there is a limit 189| | // to how much we can further increase this latency before 190| | // it seriously harms performance. 191| 501k| size_t consumed = convert_masked_utf8_to_utf16( 192| 501k| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 501k| pos += consumed; 194| 501k| utf8_end_of_code_point_mask >>= consumed; 195| 501k| } 196| | // At this point there may remain between 0 and 12 bytes in the 197| | // 64-byte block. These bytes will be processed again. So we have an 198| | // 80% efficiency (in the worst case). In practice we expect an 199| | // 85% to 90% efficiency. 200| 56.9k| } 201| 501k| } 202| 756| if (errors()) { ------------------ | Branch (202:9): [True: 308, False: 448] ------------------ 203| 308| return 0; 204| 308| } 205| 448| if (pos < size) { ------------------ | Branch (205:9): [True: 446, False: 2] ------------------ 206| 446| size_t howmany = scalar::utf8_to_utf16::convert( 207| 446| in + pos, size - pos, utf16_output); 208| 446| if (howmany == 0) { ------------------ | Branch (208:11): [True: 204, False: 242] ------------------ 209| 204| return 0; 210| 204| } 211| 242| utf16_output += howmany; 212| 242| } 213| 244| return utf16_output - start; 214| 448| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 1.60M| const simd8 prev_input) { 120| | // Flip prev1...prev3 so we can easily determine if they are 2+, 3+ or 4+ 121| | // lead bytes (2, 3, 4-byte leads become large positive numbers instead of 122| | // small negative numbers) 123| 1.60M| simd8 prev1 = input.prev<1>(prev_input); 124| 1.60M| simd8 sc = check_special_cases(input, prev1); 125| 1.60M| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 1.60M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1619check_special_casesENS1_4simd5simd8IhEES5_: 8| 1.60M|check_special_cases(const simd8 input, const simd8 prev1) { 9| | // Bit 0 = Too Short (lead byte/ASCII followed by lead byte/ASCII) 10| | // Bit 1 = Too Long (ASCII followed by continuation) 11| | // Bit 2 = Overlong 3-byte 12| | // Bit 4 = Surrogate 13| | // Bit 5 = Overlong 2-byte 14| | // Bit 7 = Two Continuations 15| 1.60M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 1.60M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 1.60M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 1.60M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 1.60M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 1.60M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 1.60M| constexpr const uint8_t TOO_LARGE = 1 << 3; // 11110100 1001____ 23| | // 11110100 101_____ 24| | // 11110101 1001____ 25| | // 11110101 101_____ 26| | // 1111011_ 1001____ 27| | // 1111011_ 101_____ 28| | // 11111___ 1001____ 29| | // 11111___ 101_____ 30| 1.60M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 1.60M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 1.60M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 1.60M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 1.60M| TOO_LONG, 40| | // 10______ ________ 41| 1.60M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 1.60M| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 1.60M| TOO_SHORT, 46| | // 1110____ ________ 47| 1.60M| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 1.60M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 1.60M| constexpr const uint8_t CARRY = 51| 1.60M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 1.60M| const simd8 byte_1_low = 53| 1.60M| (prev1 & 0x0F) 54| 1.60M| .lookup_16( 55| | // ____0000 ________ 56| 1.60M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 1.60M| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 1.60M| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 1.60M| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 1.60M| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 1.60M| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 1.60M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 1.60M| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 1.60M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 1.60M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 1.60M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 1.60M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 1.60M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 1.60M| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 1.60M| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 1.60M| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 1.60M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 1.60M| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 1.60M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 1.60M| OVERLONG_4, 88| | // ________ 1001____ 89| 1.60M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 1.60M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 1.60M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 1.60M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 1.60M| return (byte_1_high & byte_1_low & byte_2_high); 97| 1.60M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1623check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 1.60M| const simd8 sc) { 102| 1.60M| simd8 prev2 = input.prev<2>(prev_input); 103| 1.60M| simd8 prev3 = input.prev<3>(prev_input); 104| 1.60M| simd8 must23 = 105| 1.60M| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 1.60M| simd8 must23_80 = must23 & uint8_t(0x80); 107| 1.60M| return must23_80 ^ sc; 108| 1.60M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder6errorsEv: 324| 166k| simdutf_really_inline bool errors() const { 325| 166k| return this->error.any_bits_set_anywhere(); 326| 166k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE1EEEmPKcmPDs: 130| 794| char16_t *utf16_output) { 131| 794| size_t pos = 0; 132| 794| char16_t *start{utf16_output}; 133| | // In the worst case, we have the haswell kernel which can cause an overflow 134| | // of 8 bytes when calling convert_masked_utf8_to_utf16. If you skip the 135| | // last 16 bytes, and if the data is valid, then it is entirely safe because 136| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 137| | // generally assume that you have valid UTF-8 input, so we are going to go 138| | // back from the end counting 8 leading bytes, to give us a good margin. 139| 794| size_t leading_byte = 0; 140| 794| size_t margin = size; 141| 7.34k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 7.10k, False: 246] | Branch (141:26): [True: 6.55k, False: 548] ------------------ 142| 6.55k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 6.55k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 794| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 674k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 674k, False: 770] ------------------ 148| 674k| simd8x64 input(reinterpret_cast(in + pos)); 149| 674k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 492k, False: 181k] ------------------ 150| 492k| input.store_ascii_as_utf16(utf16_output); 151| 492k| utf16_output += 64; 152| 492k| pos += 64; 153| 492k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 181k| static_assert( 157| 181k| (simd8x64::NUM_CHUNKS == 2) || 158| 181k| (simd8x64::NUM_CHUNKS == 4), 159| 181k| "We support either two or four chunks per 64-byte block."); 160| 181k| auto zero = simd8{uint8_t(0)}; 161| | if constexpr (simd8x64::NUM_CHUNKS == 2) { 162| | this->check_utf8_bytes(input.chunks[0], zero); 163| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 164| 181k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 165| 181k| this->check_utf8_bytes(input.chunks[0], zero); 166| 181k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 167| 181k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 168| 181k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 169| 181k| } 170| 181k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 181k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 24, False: 181k] ------------------ 172| 24| return 0; // error 173| 24| } 174| 181k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 181k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 176| | // We process in blocks of up to 12 bytes except possibly 177| | // for fast paths which may process up to 16 bytes. For the 178| | // slow path to work, we should have at least 12 input bytes left. 179| 181k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 1.79M| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 1.61M, False: 181k] ------------------ 182| | // Performance note: our ability to compute 'consumed' and 183| | // then shift and recompute is critical. If there is a 184| | // latency of, say, 4 cycles on getting 'consumed', then 185| | // the inner loop might have a total latency of about 6 cycles. 186| | // Yet we process between 6 to 12 inputs bytes, thus we get 187| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 188| | // for this section of the code. Hence, there is a limit 189| | // to how much we can further increase this latency before 190| | // it seriously harms performance. 191| 1.61M| size_t consumed = convert_masked_utf8_to_utf16( 192| 1.61M| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 1.61M| pos += consumed; 194| 1.61M| utf8_end_of_code_point_mask >>= consumed; 195| 1.61M| } 196| | // At this point there may remain between 0 and 12 bytes in the 197| | // 64-byte block. These bytes will be processed again. So we have an 198| | // 80% efficiency (in the worst case). In practice we expect an 199| | // 85% to 90% efficiency. 200| 181k| } 201| 674k| } 202| 770| if (errors()) { ------------------ | Branch (202:9): [True: 328, False: 442] ------------------ 203| 328| return 0; 204| 328| } 205| 442| if (pos < size) { ------------------ | Branch (205:9): [True: 440, False: 2] ------------------ 206| 440| size_t howmany = scalar::utf8_to_utf16::convert( 207| 440| in + pos, size - pos, utf16_output); 208| 440| if (howmany == 0) { ------------------ | Branch (208:11): [True: 206, False: 234] ------------------ 209| 206| return 0; 210| 206| } 211| 234| utf16_output += howmany; 212| 234| } 213| 236| return utf16_output - start; 214| 442| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKcmPDs: 218| 610| char16_t *utf16_output) { 219| 610| size_t pos = 0; 220| 610| char16_t *start{utf16_output}; 221| | // In the worst case, we have the haswell kernel which can cause an overflow 222| | // of 8 bytes when calling convert_masked_utf8_to_utf16. If you skip the 223| | // last 16 bytes, and if the data is valid, then it is entirely safe because 224| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 225| | // generally assume that you have valid UTF-8 input, so we are going to go 226| | // back from the end counting 8 leading bytes, to give us a good margin. 227| 610| size_t leading_byte = 0; 228| 610| size_t margin = size; 229| 5.37k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 5.13k, False: 242] | Branch (229:26): [True: 4.76k, False: 368] ------------------ 230| 4.76k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 4.76k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 610| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 384k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 384k, False: 522] ------------------ 236| 384k| simd8x64 input(reinterpret_cast(in + pos)); 237| 384k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 294k, False: 89.8k] ------------------ 238| 294k| input.store_ascii_as_utf16(utf16_output); 239| 294k| utf16_output += 64; 240| 294k| pos += 64; 241| 294k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 89.8k| static_assert( 245| 89.8k| (simd8x64::NUM_CHUNKS == 2) || 246| 89.8k| (simd8x64::NUM_CHUNKS == 4), 247| 89.8k| "We support either two or four chunks per 64-byte block."); 248| 89.8k| auto zero = simd8{uint8_t(0)}; 249| | if constexpr (simd8x64::NUM_CHUNKS == 2) { 250| | this->check_utf8_bytes(input.chunks[0], zero); 251| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 252| 89.8k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 253| 89.8k| this->check_utf8_bytes(input.chunks[0], zero); 254| 89.8k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 255| 89.8k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 256| 89.8k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 257| 89.8k| } 258| 89.8k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 89.8k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 88, False: 89.7k] | Branch (259:25): [True: 0, False: 89.7k] ------------------ 260| | // rewind_and_convert_with_errors will seek a potential error from 261| | // in+pos onward, with the ability to go back up to pos bytes, and 262| | // read size-pos bytes forward. 263| 88| result res = 264| 88| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 88| pos, in + pos, size - pos, utf16_output); 266| 88| res.count += pos; 267| 88| return res; 268| 88| } 269| 89.7k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 89.7k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 271| | // We process in blocks of up to 12 bytes except possibly 272| | // for fast paths which may process up to 16 bytes. For the 273| | // slow path to work, we should have at least 12 input bytes left. 274| 89.7k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 885k| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 795k, False: 89.7k] ------------------ 277| | // Performance note: our ability to compute 'consumed' and 278| | // then shift and recompute is critical. If there is a 279| | // latency of, say, 4 cycles on getting 'consumed', then 280| | // the inner loop might have a total latency of about 6 cycles. 281| | // Yet we process between 6 to 12 inputs bytes, thus we get 282| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 283| | // for this section of the code. Hence, there is a limit 284| | // to how much we can further increase this latency before 285| | // it seriously harms performance. 286| 795k| size_t consumed = convert_masked_utf8_to_utf16( 287| 795k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 795k| pos += consumed; 289| 795k| utf8_end_of_code_point_mask >>= consumed; 290| 795k| } 291| | // At this point there may remain between 0 and 12 bytes in the 292| | // 64-byte block. These bytes will be processed again. So we have an 293| | // 80% efficiency (in the worst case). In practice we expect an 294| | // 85% to 90% efficiency. 295| 89.7k| } 296| 384k| } 297| 522| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 522] ------------------ 298| | // rewind_and_convert_with_errors will seek a potential error from in+pos 299| | // onward, with the ability to go back up to pos bytes, and read size-pos 300| | // bytes forward. 301| 0| result res = 302| 0| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 303| 0| pos, in + pos, size - pos, utf16_output); 304| 0| res.count += pos; 305| 0| return res; 306| 0| } 307| 522| if (pos < size) { ------------------ | Branch (307:9): [True: 518, False: 4] ------------------ 308| | // rewind_and_convert_with_errors will seek a potential error from in+pos 309| | // onward, with the ability to go back up to pos bytes, and read size-pos 310| | // bytes forward. 311| 518| result res = 312| 518| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 518| pos, in + pos, size - pos, utf16_output); 314| 518| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 280, False: 238] ------------------ 315| 280| res.count += pos; 316| 280| return res; 317| 280| } else { // In case of success, we want the number of word written 318| 238| utf16_output += res.count; 319| 238| } 320| 518| } 321| 242| return result(error_code::SUCCESS, utf16_output - start); 322| 522| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKcmPDs: 218| 706| char16_t *utf16_output) { 219| 706| size_t pos = 0; 220| 706| char16_t *start{utf16_output}; 221| | // In the worst case, we have the haswell kernel which can cause an overflow 222| | // of 8 bytes when calling convert_masked_utf8_to_utf16. If you skip the 223| | // last 16 bytes, and if the data is valid, then it is entirely safe because 224| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 225| | // generally assume that you have valid UTF-8 input, so we are going to go 226| | // back from the end counting 8 leading bytes, to give us a good margin. 227| 706| size_t leading_byte = 0; 228| 706| size_t margin = size; 229| 6.20k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 5.96k, False: 236] | Branch (229:26): [True: 5.49k, False: 470] ------------------ 230| 5.49k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 5.49k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 706| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 365k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 365k, False: 634] ------------------ 236| 365k| simd8x64 input(reinterpret_cast(in + pos)); 237| 365k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 291k, False: 73.6k] ------------------ 238| 291k| input.store_ascii_as_utf16(utf16_output); 239| 291k| utf16_output += 64; 240| 291k| pos += 64; 241| 291k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 73.6k| static_assert( 245| 73.6k| (simd8x64::NUM_CHUNKS == 2) || 246| 73.6k| (simd8x64::NUM_CHUNKS == 4), 247| 73.6k| "We support either two or four chunks per 64-byte block."); 248| 73.6k| auto zero = simd8{uint8_t(0)}; 249| | if constexpr (simd8x64::NUM_CHUNKS == 2) { 250| | this->check_utf8_bytes(input.chunks[0], zero); 251| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 252| 73.6k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 253| 73.6k| this->check_utf8_bytes(input.chunks[0], zero); 254| 73.6k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 255| 73.6k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 256| 73.6k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 257| 73.6k| } 258| 73.6k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 73.6k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 72, False: 73.5k] | Branch (259:25): [True: 0, False: 73.5k] ------------------ 260| | // rewind_and_convert_with_errors will seek a potential error from 261| | // in+pos onward, with the ability to go back up to pos bytes, and 262| | // read size-pos bytes forward. 263| 72| result res = 264| 72| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 72| pos, in + pos, size - pos, utf16_output); 266| 72| res.count += pos; 267| 72| return res; 268| 72| } 269| 73.5k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 73.5k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 271| | // We process in blocks of up to 12 bytes except possibly 272| | // for fast paths which may process up to 16 bytes. For the 273| | // slow path to work, we should have at least 12 input bytes left. 274| 73.5k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 725k| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 651k, False: 73.5k] ------------------ 277| | // Performance note: our ability to compute 'consumed' and 278| | // then shift and recompute is critical. If there is a 279| | // latency of, say, 4 cycles on getting 'consumed', then 280| | // the inner loop might have a total latency of about 6 cycles. 281| | // Yet we process between 6 to 12 inputs bytes, thus we get 282| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 283| | // for this section of the code. Hence, there is a limit 284| | // to how much we can further increase this latency before 285| | // it seriously harms performance. 286| 651k| size_t consumed = convert_masked_utf8_to_utf16( 287| 651k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 651k| pos += consumed; 289| 651k| utf8_end_of_code_point_mask >>= consumed; 290| 651k| } 291| | // At this point there may remain between 0 and 12 bytes in the 292| | // 64-byte block. These bytes will be processed again. So we have an 293| | // 80% efficiency (in the worst case). In practice we expect an 294| | // 85% to 90% efficiency. 295| 73.5k| } 296| 365k| } 297| 634| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 634] ------------------ 298| | // rewind_and_convert_with_errors will seek a potential error from in+pos 299| | // onward, with the ability to go back up to pos bytes, and read size-pos 300| | // bytes forward. 301| 0| result res = 302| 0| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 303| 0| pos, in + pos, size - pos, utf16_output); 304| 0| res.count += pos; 305| 0| return res; 306| 0| } 307| 634| if (pos < size) { ------------------ | Branch (307:9): [True: 630, False: 4] ------------------ 308| | // rewind_and_convert_with_errors will seek a potential error from in+pos 309| | // onward, with the ability to go back up to pos bytes, and read size-pos 310| | // bytes forward. 311| 630| result res = 312| 630| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 630| pos, in + pos, size - pos, utf16_output); 314| 630| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 322, False: 308] ------------------ 315| 322| res.count += pos; 316| 322| return res; 317| 322| } else { // In case of success, we want the number of word written 318| 308| utf16_output += res.count; 319| 308| } 320| 630| } 321| 312| return result(error_code::SUCCESS, utf16_output - start); 322| 634| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE0EEEmPKcmPDs: 10| 450| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 450| size_t pos = 0; 14| 450| char16_t *start{utf16_output}; 15| 450| const size_t safety_margin = 16; // to avoid overruns! 16| 473k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 472k, False: 450] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 472k| simd8x64 in(reinterpret_cast(input + pos)); 20| 472k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 261k, False: 211k] ------------------ 21| 261k| in.store_ascii_as_utf16(utf16_output); 22| 261k| utf16_output += 64; 23| 261k| pos += 64; 24| 261k| } else { 25| | // Slow path. We hope that the compiler will recognize that this is a slow 26| | // path. Anything that is not a continuation mask is a 'leading byte', 27| | // that is, the start of a new code point. 28| 211k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 211k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 32| | // The *start* of code points is not so useful, rather, we want the *end* 33| | // of code points. 34| 211k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 35| | // We process in blocks of up to 12 bytes except possibly 36| | // for fast paths which may process up to 16 bytes. For the 37| | // slow path to work, we should have at least 12 input bytes left. 38| 211k| size_t max_starting_point = (pos + 64) - 12; 39| | // Next loop is going to run at least five times when using solely 40| | // the slow/regular path, and at least four times if there are fast paths. 41| 2.09M| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 1.88M, False: 211k] ------------------ 42| | // Performance note: our ability to compute 'consumed' and 43| | // then shift and recompute is critical. If there is a 44| | // latency of, say, 4 cycles on getting 'consumed', then 45| | // the inner loop might have a total latency of about 6 cycles. 46| | // Yet we process between 6 to 12 inputs bytes, thus we get 47| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 48| | // for this section of the code. Hence, there is a limit 49| | // to how much we can further increase this latency before 50| | // it seriously harms performance. 51| | // 52| | // Thus we may allow convert_masked_utf8_to_utf16 to process 53| | // more bytes at a time under a fast-path mode where 16 bytes 54| | // are consumed at once (e.g., when encountering ASCII). 55| 1.88M| size_t consumed = convert_masked_utf8_to_utf16( 56| 1.88M| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 1.88M| pos += consumed; 58| 1.88M| utf8_end_of_code_point_mask >>= consumed; 59| 1.88M| } 60| | // At this point there may remain between 0 and 12 bytes in the 61| | // 64-byte block. These bytes will be processed again. So we have an 62| | // 80% efficiency (in the worst case). In practice we expect an 63| | // 85% to 90% efficiency. 64| 211k| } 65| 472k| } 66| 450| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 450| input + pos, size - pos, utf16_output); 68| 450| return utf16_output - start; 69| 450|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE1EEEmPKcmPDs: 10| 510| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 510| size_t pos = 0; 14| 510| char16_t *start{utf16_output}; 15| 510| const size_t safety_margin = 16; // to avoid overruns! 16| 435k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 434k, False: 510] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 434k| simd8x64 in(reinterpret_cast(input + pos)); 20| 434k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 349k, False: 85.6k] ------------------ 21| 349k| in.store_ascii_as_utf16(utf16_output); 22| 349k| utf16_output += 64; 23| 349k| pos += 64; 24| 349k| } else { 25| | // Slow path. We hope that the compiler will recognize that this is a slow 26| | // path. Anything that is not a continuation mask is a 'leading byte', 27| | // that is, the start of a new code point. 28| 85.6k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 85.6k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 32| | // The *start* of code points is not so useful, rather, we want the *end* 33| | // of code points. 34| 85.6k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 35| | // We process in blocks of up to 12 bytes except possibly 36| | // for fast paths which may process up to 16 bytes. For the 37| | // slow path to work, we should have at least 12 input bytes left. 38| 85.6k| size_t max_starting_point = (pos + 64) - 12; 39| | // Next loop is going to run at least five times when using solely 40| | // the slow/regular path, and at least four times if there are fast paths. 41| 819k| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 734k, False: 85.6k] ------------------ 42| | // Performance note: our ability to compute 'consumed' and 43| | // then shift and recompute is critical. If there is a 44| | // latency of, say, 4 cycles on getting 'consumed', then 45| | // the inner loop might have a total latency of about 6 cycles. 46| | // Yet we process between 6 to 12 inputs bytes, thus we get 47| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 48| | // for this section of the code. Hence, there is a limit 49| | // to how much we can further increase this latency before 50| | // it seriously harms performance. 51| | // 52| | // Thus we may allow convert_masked_utf8_to_utf16 to process 53| | // more bytes at a time under a fast-path mode where 16 bytes 54| | // are consumed at once (e.g., when encountering ASCII). 55| 734k| size_t consumed = convert_masked_utf8_to_utf16( 56| 734k| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 734k| pos += consumed; 58| 734k| utf8_end_of_code_point_mask >>= consumed; 59| 734k| } 60| | // At this point there may remain between 0 and 12 bytes in the 61| | // 64-byte block. These bytes will be processed again. So we have an 62| | // 80% efficiency (in the worst case). In practice we expect an 63| | // 85% to 90% efficiency. 64| 85.6k| } 65| 434k| } 66| 510| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 510| input + pos, size - pos, utf16_output); 68| 510| return utf16_output - start; 69| 510|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE0EEEmPKcmPDs: 10| 450| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 450| size_t pos = 0; 14| 450| char16_t *start{utf16_output}; 15| 450| const size_t safety_margin = 16; // to avoid overruns! 16| 473k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 472k, False: 450] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 472k| simd8x64 in(reinterpret_cast(input + pos)); 20| 472k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 261k, False: 211k] ------------------ 21| 261k| in.store_ascii_as_utf16(utf16_output); 22| 261k| utf16_output += 64; 23| 261k| pos += 64; 24| 261k| } else { 25| | // Slow path. We hope that the compiler will recognize that this is a slow 26| | // path. Anything that is not a continuation mask is a 'leading byte', 27| | // that is, the start of a new code point. 28| 211k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 211k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 32| | // The *start* of code points is not so useful, rather, we want the *end* 33| | // of code points. 34| 211k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 35| | // We process in blocks of up to 12 bytes except possibly 36| | // for fast paths which may process up to 16 bytes. For the 37| | // slow path to work, we should have at least 12 input bytes left. 38| 211k| size_t max_starting_point = (pos + 64) - 12; 39| | // Next loop is going to run at least five times when using solely 40| | // the slow/regular path, and at least four times if there are fast paths. 41| 2.09M| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 1.88M, False: 211k] ------------------ 42| | // Performance note: our ability to compute 'consumed' and 43| | // then shift and recompute is critical. If there is a 44| | // latency of, say, 4 cycles on getting 'consumed', then 45| | // the inner loop might have a total latency of about 6 cycles. 46| | // Yet we process between 6 to 12 inputs bytes, thus we get 47| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 48| | // for this section of the code. Hence, there is a limit 49| | // to how much we can further increase this latency before 50| | // it seriously harms performance. 51| | // 52| | // Thus we may allow convert_masked_utf8_to_utf16 to process 53| | // more bytes at a time under a fast-path mode where 16 bytes 54| | // are consumed at once (e.g., when encountering ASCII). 55| 1.88M| size_t consumed = convert_masked_utf8_to_utf16( 56| 1.88M| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 1.88M| pos += consumed; 58| 1.88M| utf8_end_of_code_point_mask >>= consumed; 59| 1.88M| } 60| | // At this point there may remain between 0 and 12 bytes in the 61| | // 64-byte block. These bytes will be processed again. So we have an 62| | // 80% efficiency (in the worst case). In practice we expect an 63| | // 85% to 90% efficiency. 64| 211k| } 65| 472k| } 66| 450| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 450| input + pos, size - pos, utf16_output); 68| 450| return utf16_output - start; 69| 450|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE1EEEmPKcmPDs: 10| 510| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 510| size_t pos = 0; 14| 510| char16_t *start{utf16_output}; 15| 510| const size_t safety_margin = 16; // to avoid overruns! 16| 435k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 434k, False: 510] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 434k| simd8x64 in(reinterpret_cast(input + pos)); 20| 434k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 349k, False: 85.6k] ------------------ 21| 349k| in.store_ascii_as_utf16(utf16_output); 22| 349k| utf16_output += 64; 23| 349k| pos += 64; 24| 349k| } else { 25| | // Slow path. We hope that the compiler will recognize that this is a slow 26| | // path. Anything that is not a continuation mask is a 'leading byte', 27| | // that is, the start of a new code point. 28| 85.6k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 85.6k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 32| | // The *start* of code points is not so useful, rather, we want the *end* 33| | // of code points. 34| 85.6k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 35| | // We process in blocks of up to 12 bytes except possibly 36| | // for fast paths which may process up to 16 bytes. For the 37| | // slow path to work, we should have at least 12 input bytes left. 38| 85.6k| size_t max_starting_point = (pos + 64) - 12; 39| | // Next loop is going to run at least five times when using solely 40| | // the slow/regular path, and at least four times if there are fast paths. 41| 819k| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 734k, False: 85.6k] ------------------ 42| | // Performance note: our ability to compute 'consumed' and 43| | // then shift and recompute is critical. If there is a 44| | // latency of, say, 4 cycles on getting 'consumed', then 45| | // the inner loop might have a total latency of about 6 cycles. 46| | // Yet we process between 6 to 12 inputs bytes, thus we get 47| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 48| | // for this section of the code. Hence, there is a limit 49| | // to how much we can further increase this latency before 50| | // it seriously harms performance. 51| | // 52| | // Thus we may allow convert_masked_utf8_to_utf16 to process 53| | // more bytes at a time under a fast-path mode where 16 bytes 54| | // are consumed at once (e.g., when encountering ASCII). 55| 734k| size_t consumed = convert_masked_utf8_to_utf16( 56| 734k| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 734k| pos += consumed; 58| 734k| utf8_end_of_code_point_mask >>= consumed; 59| 734k| } 60| | // At this point there may remain between 0 and 12 bytes in the 61| | // 64-byte block. These bytes will be processed again. So we have an 62| | // 80% efficiency (in the worst case). In practice we expect an 63| | // 85% to 90% efficiency. 64| 85.6k| } 65| 434k| } 66| 510| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 510| input + pos, size - pos, utf16_output); 68| 510| return utf16_output - start; 69| 510|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoderC2Ev: 114| 1.50k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder7convertEPKcmPDi: 129| 758| char32_t *utf32_output) { 130| 758| size_t pos = 0; 131| 758| char32_t *start{utf32_output}; 132| | // In the worst case, we have the haswell kernel which can cause an overflow 133| | // of 8 words when calling convert_masked_utf8_to_utf32. If you skip the 134| | // last 16 bytes, and if the data is valid, then it is entirely safe because 135| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 136| | // generally assume that you have valid UTF-8 input, so we are going to go 137| | // back from the end counting 16 leading bytes, to give us a good margin. 138| 758| size_t leading_byte = 0; 139| 758| size_t margin = size; 140| 7.04k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (140:12): [True: 6.79k, False: 250] | Branch (140:26): [True: 6.28k, False: 508] ------------------ 141| 6.28k| leading_byte += (int8_t(in[margin - 1]) > -65); 142| 6.28k| } 143| | // If the input is long enough, then we have that margin-1 is the fourth 144| | // last leading byte. 145| 758| const size_t safety_margin = size - margin + 1; // to avoid overruns! 146| 624k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (146:12): [True: 624k, False: 738] ------------------ 147| 624k| simd8x64 input(reinterpret_cast(in + pos)); 148| 624k| if (input.is_ascii()) { ------------------ | Branch (148:11): [True: 475k, False: 148k] ------------------ 149| 475k| input.store_ascii_as_utf32(utf32_output); 150| 475k| utf32_output += 64; 151| 475k| pos += 64; 152| 475k| } else { 153| | // you might think that a for-loop would work, but under Visual Studio, 154| | // it is not good enough. 155| 148k| static_assert( 156| 148k| (simd8x64::NUM_CHUNKS == 2) || 157| 148k| (simd8x64::NUM_CHUNKS == 4), 158| 148k| "We support either two or four chunks per 64-byte block."); 159| 148k| auto zero = simd8{uint8_t(0)}; 160| 148k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 161| 148k| this->check_utf8_bytes(input.chunks[0], zero); 162| 148k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 163| | } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 164| | this->check_utf8_bytes(input.chunks[0], zero); 165| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 166| | this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 167| | this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 168| | } 169| 148k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 170| 148k| if (utf8_continuation_mask & 1) { ------------------ | Branch (170:13): [True: 20, False: 148k] ------------------ 171| 20| return 0; // we have an error 172| 20| } 173| 148k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 174| 148k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 175| | // We process in blocks of up to 12 bytes except possibly 176| | // for fast paths which may process up to 16 bytes. For the 177| | // slow path to work, we should have at least 12 input bytes left. 178| 148k| size_t max_starting_point = (pos + 64) - 12; 179| | // Next loop is going to run at least five times. 180| 1.47M| while (pos < max_starting_point) { ------------------ | Branch (180:16): [True: 1.32M, False: 148k] ------------------ 181| | // Performance note: our ability to compute 'consumed' and 182| | // then shift and recompute is critical. If there is a 183| | // latency of, say, 4 cycles on getting 'consumed', then 184| | // the inner loop might have a total latency of about 6 cycles. 185| | // Yet we process between 6 to 12 inputs bytes, thus we get 186| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 187| | // for this section of the code. Hence, there is a limit 188| | // to how much we can further increase this latency before 189| | // it seriously harms performance. 190| 1.32M| size_t consumed = convert_masked_utf8_to_utf32( 191| 1.32M| in + pos, utf8_end_of_code_point_mask, utf32_output); 192| 1.32M| pos += consumed; 193| 1.32M| utf8_end_of_code_point_mask >>= consumed; 194| 1.32M| } 195| | // At this point there may remain between 0 and 12 bytes in the 196| | // 64-byte block. These bytes will be processed again. So we have an 197| | // 80% efficiency (in the worst case). In practice we expect an 198| | // 85% to 90% efficiency. 199| 148k| } 200| 624k| } 201| 738| if (errors()) { ------------------ | Branch (201:9): [True: 252, False: 486] ------------------ 202| 252| return 0; 203| 252| } 204| 486| if (pos < size) { ------------------ | Branch (204:9): [True: 484, False: 2] ------------------ 205| 484| size_t howmany = 206| 484| scalar::utf8_to_utf32::convert(in + pos, size - pos, utf32_output); 207| 484| if (howmany == 0) { ------------------ | Branch (207:11): [True: 222, False: 262] ------------------ 208| 222| return 0; 209| 222| } 210| 262| utf32_output += howmany; 211| 262| } 212| 264| return utf32_output - start; 213| 486| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 438k| const simd8 prev_input) { 120| | // Flip prev1...prev3 so we can easily determine if they are 2+, 3+ or 4+ 121| | // lead bytes (2, 3, 4-byte leads become large positive numbers instead of 122| | // small negative numbers) 123| 438k| simd8 prev1 = input.prev<1>(prev_input); 124| 438k| simd8 sc = check_special_cases(input, prev1); 125| 438k| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 438k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3219check_special_casesENS1_4simd5simd8IhEES5_: 8| 438k|check_special_cases(const simd8 input, const simd8 prev1) { 9| | // Bit 0 = Too Short (lead byte/ASCII followed by lead byte/ASCII) 10| | // Bit 1 = Too Long (ASCII followed by continuation) 11| | // Bit 2 = Overlong 3-byte 12| | // Bit 4 = Surrogate 13| | // Bit 5 = Overlong 2-byte 14| | // Bit 7 = Two Continuations 15| 438k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 438k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 438k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 438k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 438k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 438k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 438k| constexpr const uint8_t TOO_LARGE = 1 << 3; // 11110100 1001____ 23| | // 11110100 101_____ 24| | // 11110101 1001____ 25| | // 11110101 101_____ 26| | // 1111011_ 1001____ 27| | // 1111011_ 101_____ 28| | // 11111___ 1001____ 29| | // 11111___ 101_____ 30| 438k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 438k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 438k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 438k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 438k| TOO_LONG, 40| | // 10______ ________ 41| 438k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 438k| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 438k| TOO_SHORT, 46| | // 1110____ ________ 47| 438k| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 438k| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 438k| constexpr const uint8_t CARRY = 51| 438k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 438k| const simd8 byte_1_low = 53| 438k| (prev1 & 0x0F) 54| 438k| .lookup_16( 55| | // ____0000 ________ 56| 438k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 438k| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 438k| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 438k| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 438k| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 438k| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 438k| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 438k| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 438k| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 438k| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 438k| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 438k| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 438k| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 438k| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 438k| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 438k| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 438k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 438k| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 438k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 438k| OVERLONG_4, 88| | // ________ 1001____ 89| 438k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 438k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 438k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 438k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 438k| return (byte_1_high & byte_1_low & byte_2_high); 97| 438k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3223check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 438k| const simd8 sc) { 102| 438k| simd8 prev2 = input.prev<2>(prev_input); 103| 438k| simd8 prev3 = input.prev<3>(prev_input); 104| 438k| simd8 must23 = 105| 438k| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 438k| simd8 must23_80 = must23 & uint8_t(0x80); 107| 438k| return must23_80 ^ sc; 108| 438k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder6errorsEv: 310| 72.3k| simdutf_really_inline bool errors() const { 311| 72.3k| return this->error.any_bits_set_anywhere(); 312| 72.3k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder19convert_with_errorsEPKcmPDi: 216| 746| char32_t *utf32_output) { 217| 746| size_t pos = 0; 218| 746| char32_t *start{utf32_output}; 219| | // In the worst case, we have the haswell kernel which can cause an overflow 220| | // of 8 bytes when calling convert_masked_utf8_to_utf32. If you skip the 221| | // last 16 bytes, and if the data is valid, then it is entirely safe because 222| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 223| | // generally assume that you have valid UTF-8 input, so we are going to go 224| | // back from the end counting 8 leading bytes, to give us a good margin. 225| 746| size_t leading_byte = 0; 226| 746| size_t margin = size; 227| 6.90k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (227:12): [True: 6.65k, False: 254] | Branch (227:26): [True: 6.15k, False: 492] ------------------ 228| 6.15k| leading_byte += (int8_t(in[margin - 1]) > -65); 229| 6.15k| } 230| | // If the input is long enough, then we have that margin-1 is the fourth 231| | // last leading byte. 232| 746| const size_t safety_margin = size - margin + 1; // to avoid overruns! 233| 438k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (233:12): [True: 437k, False: 662] ------------------ 234| 437k| simd8x64 input(reinterpret_cast(in + pos)); 235| 437k| if (input.is_ascii()) { ------------------ | Branch (235:11): [True: 366k, False: 70.9k] ------------------ 236| 366k| input.store_ascii_as_utf32(utf32_output); 237| 366k| utf32_output += 64; 238| 366k| pos += 64; 239| 366k| } else { 240| | // you might think that a for-loop would work, but under Visual Studio, 241| | // it is not good enough. 242| 70.9k| static_assert( 243| 70.9k| (simd8x64::NUM_CHUNKS == 2) || 244| 70.9k| (simd8x64::NUM_CHUNKS == 4), 245| 70.9k| "We support either two or four chunks per 64-byte block."); 246| 70.9k| auto zero = simd8{uint8_t(0)}; 247| 70.9k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 248| 70.9k| this->check_utf8_bytes(input.chunks[0], zero); 249| 70.9k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 250| | } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 251| | this->check_utf8_bytes(input.chunks[0], zero); 252| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 253| | this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 254| | this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 255| | } 256| 70.9k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 257| 70.9k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (257:13): [True: 84, False: 70.8k] | Branch (257:25): [True: 0, False: 70.8k] ------------------ 258| 84| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 259| 84| pos, in + pos, size - pos, utf32_output); 260| 84| res.count += pos; 261| 84| return res; 262| 84| } 263| 70.8k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 264| 70.8k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 265| | // We process in blocks of up to 12 bytes except possibly 266| | // for fast paths which may process up to 16 bytes. For the 267| | // slow path to work, we should have at least 12 input bytes left. 268| 70.8k| size_t max_starting_point = (pos + 64) - 12; 269| | // Next loop is going to run at least five times. 270| 696k| while (pos < max_starting_point) { ------------------ | Branch (270:16): [True: 625k, False: 70.8k] ------------------ 271| | // Performance note: our ability to compute 'consumed' and 272| | // then shift and recompute is critical. If there is a 273| | // latency of, say, 4 cycles on getting 'consumed', then 274| | // the inner loop might have a total latency of about 6 cycles. 275| | // Yet we process between 6 to 12 inputs bytes, thus we get 276| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 277| | // for this section of the code. Hence, there is a limit 278| | // to how much we can further increase this latency before 279| | // it seriously harms performance. 280| 625k| size_t consumed = convert_masked_utf8_to_utf32( 281| 625k| in + pos, utf8_end_of_code_point_mask, utf32_output); 282| 625k| pos += consumed; 283| 625k| utf8_end_of_code_point_mask >>= consumed; 284| 625k| } 285| | // At this point there may remain between 0 and 12 bytes in the 286| | // 64-byte block. These bytes will be processed again. So we have an 287| | // 80% efficiency (in the worst case). In practice we expect an 288| | // 85% to 90% efficiency. 289| 70.8k| } 290| 437k| } 291| 662| if (errors()) { ------------------ | Branch (291:9): [True: 0, False: 662] ------------------ 292| 0| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 293| 0| pos, in + pos, size - pos, utf32_output); 294| 0| res.count += pos; 295| 0| return res; 296| 0| } 297| 662| if (pos < size) { ------------------ | Branch (297:9): [True: 658, False: 4] ------------------ 298| 658| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 299| 658| pos, in + pos, size - pos, utf32_output); 300| 658| if (res.error) { // In case of error, we want the error position ------------------ | Branch (300:11): [True: 350, False: 308] ------------------ 301| 350| res.count += pos; 302| 350| return res; 303| 350| } else { // In case of success, we want the number of word written 304| 308| utf32_output += res.count; 305| 308| } 306| 658| } 307| 312| return result(error_code::SUCCESS, utf32_output - start); 308| 662| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoderC2Ev: 114| 1.50k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder7convertEPKcmPDi: 129| 758| char32_t *utf32_output) { 130| 758| size_t pos = 0; 131| 758| char32_t *start{utf32_output}; 132| | // In the worst case, we have the haswell kernel which can cause an overflow 133| | // of 8 words when calling convert_masked_utf8_to_utf32. If you skip the 134| | // last 16 bytes, and if the data is valid, then it is entirely safe because 135| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 136| | // generally assume that you have valid UTF-8 input, so we are going to go 137| | // back from the end counting 16 leading bytes, to give us a good margin. 138| 758| size_t leading_byte = 0; 139| 758| size_t margin = size; 140| 7.04k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (140:12): [True: 6.79k, False: 250] | Branch (140:26): [True: 6.28k, False: 508] ------------------ 141| 6.28k| leading_byte += (int8_t(in[margin - 1]) > -65); 142| 6.28k| } 143| | // If the input is long enough, then we have that margin-1 is the fourth 144| | // last leading byte. 145| 758| const size_t safety_margin = size - margin + 1; // to avoid overruns! 146| 624k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (146:12): [True: 624k, False: 738] ------------------ 147| 624k| simd8x64 input(reinterpret_cast(in + pos)); 148| 624k| if (input.is_ascii()) { ------------------ | Branch (148:11): [True: 475k, False: 148k] ------------------ 149| 475k| input.store_ascii_as_utf32(utf32_output); 150| 475k| utf32_output += 64; 151| 475k| pos += 64; 152| 475k| } else { 153| | // you might think that a for-loop would work, but under Visual Studio, 154| | // it is not good enough. 155| 148k| static_assert( 156| 148k| (simd8x64::NUM_CHUNKS == 2) || 157| 148k| (simd8x64::NUM_CHUNKS == 4), 158| 148k| "We support either two or four chunks per 64-byte block."); 159| 148k| auto zero = simd8{uint8_t(0)}; 160| | if constexpr (simd8x64::NUM_CHUNKS == 2) { 161| | this->check_utf8_bytes(input.chunks[0], zero); 162| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 163| 148k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 164| 148k| this->check_utf8_bytes(input.chunks[0], zero); 165| 148k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 166| 148k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 167| 148k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 168| 148k| } 169| 148k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 170| 148k| if (utf8_continuation_mask & 1) { ------------------ | Branch (170:13): [True: 20, False: 148k] ------------------ 171| 20| return 0; // we have an error 172| 20| } 173| 148k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 174| 148k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 175| | // We process in blocks of up to 12 bytes except possibly 176| | // for fast paths which may process up to 16 bytes. For the 177| | // slow path to work, we should have at least 12 input bytes left. 178| 148k| size_t max_starting_point = (pos + 64) - 12; 179| | // Next loop is going to run at least five times. 180| 1.47M| while (pos < max_starting_point) { ------------------ | Branch (180:16): [True: 1.32M, False: 148k] ------------------ 181| | // Performance note: our ability to compute 'consumed' and 182| | // then shift and recompute is critical. If there is a 183| | // latency of, say, 4 cycles on getting 'consumed', then 184| | // the inner loop might have a total latency of about 6 cycles. 185| | // Yet we process between 6 to 12 inputs bytes, thus we get 186| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 187| | // for this section of the code. Hence, there is a limit 188| | // to how much we can further increase this latency before 189| | // it seriously harms performance. 190| 1.32M| size_t consumed = convert_masked_utf8_to_utf32( 191| 1.32M| in + pos, utf8_end_of_code_point_mask, utf32_output); 192| 1.32M| pos += consumed; 193| 1.32M| utf8_end_of_code_point_mask >>= consumed; 194| 1.32M| } 195| | // At this point there may remain between 0 and 12 bytes in the 196| | // 64-byte block. These bytes will be processed again. So we have an 197| | // 80% efficiency (in the worst case). In practice we expect an 198| | // 85% to 90% efficiency. 199| 148k| } 200| 624k| } 201| 738| if (errors()) { ------------------ | Branch (201:9): [True: 252, False: 486] ------------------ 202| 252| return 0; 203| 252| } 204| 486| if (pos < size) { ------------------ | Branch (204:9): [True: 484, False: 2] ------------------ 205| 484| size_t howmany = 206| 484| scalar::utf8_to_utf32::convert(in + pos, size - pos, utf32_output); 207| 484| if (howmany == 0) { ------------------ | Branch (207:11): [True: 222, False: 262] ------------------ 208| 222| return 0; 209| 222| } 210| 262| utf32_output += howmany; 211| 262| } 212| 264| return utf32_output - start; 213| 486| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 877k| const simd8 prev_input) { 120| | // Flip prev1...prev3 so we can easily determine if they are 2+, 3+ or 4+ 121| | // lead bytes (2, 3, 4-byte leads become large positive numbers instead of 122| | // small negative numbers) 123| 877k| simd8 prev1 = input.prev<1>(prev_input); 124| 877k| simd8 sc = check_special_cases(input, prev1); 125| 877k| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 877k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3219check_special_casesENS1_4simd5simd8IhEES5_: 8| 877k|check_special_cases(const simd8 input, const simd8 prev1) { 9| | // Bit 0 = Too Short (lead byte/ASCII followed by lead byte/ASCII) 10| | // Bit 1 = Too Long (ASCII followed by continuation) 11| | // Bit 2 = Overlong 3-byte 12| | // Bit 4 = Surrogate 13| | // Bit 5 = Overlong 2-byte 14| | // Bit 7 = Two Continuations 15| 877k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 877k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 877k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 877k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 877k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 877k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 877k| constexpr const uint8_t TOO_LARGE = 1 << 3; // 11110100 1001____ 23| | // 11110100 101_____ 24| | // 11110101 1001____ 25| | // 11110101 101_____ 26| | // 1111011_ 1001____ 27| | // 1111011_ 101_____ 28| | // 11111___ 1001____ 29| | // 11111___ 101_____ 30| 877k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 877k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 877k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 877k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 877k| TOO_LONG, 40| | // 10______ ________ 41| 877k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 877k| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 877k| TOO_SHORT, 46| | // 1110____ ________ 47| 877k| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 877k| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 877k| constexpr const uint8_t CARRY = 51| 877k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 877k| const simd8 byte_1_low = 53| 877k| (prev1 & 0x0F) 54| 877k| .lookup_16( 55| | // ____0000 ________ 56| 877k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 877k| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 877k| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 877k| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 877k| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 877k| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 877k| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 877k| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 877k| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 877k| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 877k| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 877k| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 877k| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 877k| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 877k| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 877k| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 877k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 877k| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 877k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 877k| OVERLONG_4, 88| | // ________ 1001____ 89| 877k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 877k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 877k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 877k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 877k| return (byte_1_high & byte_1_low & byte_2_high); 97| 877k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3223check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 877k| const simd8 sc) { 102| 877k| simd8 prev2 = input.prev<2>(prev_input); 103| 877k| simd8 prev3 = input.prev<3>(prev_input); 104| 877k| simd8 must23 = 105| 877k| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 877k| simd8 must23_80 = must23 & uint8_t(0x80); 107| 877k| return must23_80 ^ sc; 108| 877k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder6errorsEv: 310| 72.3k| simdutf_really_inline bool errors() const { 311| 72.3k| return this->error.any_bits_set_anywhere(); 312| 72.3k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder19convert_with_errorsEPKcmPDi: 216| 746| char32_t *utf32_output) { 217| 746| size_t pos = 0; 218| 746| char32_t *start{utf32_output}; 219| | // In the worst case, we have the haswell kernel which can cause an overflow 220| | // of 8 bytes when calling convert_masked_utf8_to_utf32. If you skip the 221| | // last 16 bytes, and if the data is valid, then it is entirely safe because 222| | // 16 UTF-8 bytes generate much more than 8 bytes. However, you cannot 223| | // generally assume that you have valid UTF-8 input, so we are going to go 224| | // back from the end counting 8 leading bytes, to give us a good margin. 225| 746| size_t leading_byte = 0; 226| 746| size_t margin = size; 227| 6.90k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (227:12): [True: 6.65k, False: 254] | Branch (227:26): [True: 6.15k, False: 492] ------------------ 228| 6.15k| leading_byte += (int8_t(in[margin - 1]) > -65); 229| 6.15k| } 230| | // If the input is long enough, then we have that margin-1 is the fourth 231| | // last leading byte. 232| 746| const size_t safety_margin = size - margin + 1; // to avoid overruns! 233| 438k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (233:12): [True: 437k, False: 662] ------------------ 234| 437k| simd8x64 input(reinterpret_cast(in + pos)); 235| 437k| if (input.is_ascii()) { ------------------ | Branch (235:11): [True: 366k, False: 70.9k] ------------------ 236| 366k| input.store_ascii_as_utf32(utf32_output); 237| 366k| utf32_output += 64; 238| 366k| pos += 64; 239| 366k| } else { 240| | // you might think that a for-loop would work, but under Visual Studio, 241| | // it is not good enough. 242| 70.9k| static_assert( 243| 70.9k| (simd8x64::NUM_CHUNKS == 2) || 244| 70.9k| (simd8x64::NUM_CHUNKS == 4), 245| 70.9k| "We support either two or four chunks per 64-byte block."); 246| 70.9k| auto zero = simd8{uint8_t(0)}; 247| | if constexpr (simd8x64::NUM_CHUNKS == 2) { 248| | this->check_utf8_bytes(input.chunks[0], zero); 249| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 250| 70.9k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 251| 70.9k| this->check_utf8_bytes(input.chunks[0], zero); 252| 70.9k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 253| 70.9k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 254| 70.9k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 255| 70.9k| } 256| 70.9k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 257| 70.9k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (257:13): [True: 84, False: 70.8k] | Branch (257:25): [True: 0, False: 70.8k] ------------------ 258| 84| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 259| 84| pos, in + pos, size - pos, utf32_output); 260| 84| res.count += pos; 261| 84| return res; 262| 84| } 263| 70.8k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 264| 70.8k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 265| | // We process in blocks of up to 12 bytes except possibly 266| | // for fast paths which may process up to 16 bytes. For the 267| | // slow path to work, we should have at least 12 input bytes left. 268| 70.8k| size_t max_starting_point = (pos + 64) - 12; 269| | // Next loop is going to run at least five times. 270| 696k| while (pos < max_starting_point) { ------------------ | Branch (270:16): [True: 625k, False: 70.8k] ------------------ 271| | // Performance note: our ability to compute 'consumed' and 272| | // then shift and recompute is critical. If there is a 273| | // latency of, say, 4 cycles on getting 'consumed', then 274| | // the inner loop might have a total latency of about 6 cycles. 275| | // Yet we process between 6 to 12 inputs bytes, thus we get 276| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 277| | // for this section of the code. Hence, there is a limit 278| | // to how much we can further increase this latency before 279| | // it seriously harms performance. 280| 625k| size_t consumed = convert_masked_utf8_to_utf32( 281| 625k| in + pos, utf8_end_of_code_point_mask, utf32_output); 282| 625k| pos += consumed; 283| 625k| utf8_end_of_code_point_mask >>= consumed; 284| 625k| } 285| | // At this point there may remain between 0 and 12 bytes in the 286| | // 64-byte block. These bytes will be processed again. So we have an 287| | // 80% efficiency (in the worst case). In practice we expect an 288| | // 85% to 90% efficiency. 289| 70.8k| } 290| 437k| } 291| 662| if (errors()) { ------------------ | Branch (291:9): [True: 0, False: 662] ------------------ 292| 0| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 293| 0| pos, in + pos, size - pos, utf32_output); 294| 0| res.count += pos; 295| 0| return res; 296| 0| } 297| 662| if (pos < size) { ------------------ | Branch (297:9): [True: 658, False: 4] ------------------ 298| 658| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 299| 658| pos, in + pos, size - pos, utf32_output); 300| 658| if (res.error) { // In case of error, we want the error position ------------------ | Branch (300:11): [True: 350, False: 308] ------------------ 301| 350| res.count += pos; 302| 350| return res; 303| 350| } else { // In case of success, we want the number of word written 304| 308| utf32_output += res.count; 305| 308| } 306| 658| } 307| 312| return result(error_code::SUCCESS, utf32_output - start); 308| 662| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3213convert_validEPKcmPDi: 9| 464| char32_t *utf32_output) noexcept { 10| 464| size_t pos = 0; 11| 464| char32_t *start{utf32_output}; 12| 464| const size_t safety_margin = 16; // to avoid overruns! 13| 428k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (13:10): [True: 427k, False: 464] ------------------ 14| 427k| simd8x64 in(reinterpret_cast(input + pos)); 15| 427k| if (in.is_ascii()) { ------------------ | Branch (15:9): [True: 355k, False: 72.5k] ------------------ 16| 355k| in.store_ascii_as_utf32(utf32_output); 17| 355k| utf32_output += 64; 18| 355k| pos += 64; 19| 355k| } else { 20| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 21| | // byte 22| 72.5k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 23| 72.5k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 24| 72.5k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 25| 72.5k| size_t max_starting_point = (pos + 64) - 12; 26| 704k| while (pos < max_starting_point) { ------------------ | Branch (26:14): [True: 632k, False: 72.5k] ------------------ 27| 632k| size_t consumed = convert_masked_utf8_to_utf32( 28| 632k| input + pos, utf8_end_of_code_point_mask, utf32_output); 29| 632k| pos += consumed; 30| 632k| utf8_end_of_code_point_mask >>= consumed; 31| 632k| } 32| 72.5k| } 33| 427k| } 34| 464| utf32_output += scalar::utf8_to_utf32::convert_valid(input + pos, size - pos, 35| 464| utf32_output); 36| 464| return utf32_output - start; 37| 464|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3213convert_validEPKcmPDi: 9| 464| char32_t *utf32_output) noexcept { 10| 464| size_t pos = 0; 11| 464| char32_t *start{utf32_output}; 12| 464| const size_t safety_margin = 16; // to avoid overruns! 13| 428k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (13:10): [True: 427k, False: 464] ------------------ 14| 427k| simd8x64 in(reinterpret_cast(input + pos)); 15| 427k| if (in.is_ascii()) { ------------------ | Branch (15:9): [True: 355k, False: 72.5k] ------------------ 16| 355k| in.store_ascii_as_utf32(utf32_output); 17| 355k| utf32_output += 64; 18| 355k| pos += 64; 19| 355k| } else { 20| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 21| | // byte 22| 72.5k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 23| 72.5k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 24| 72.5k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 25| 72.5k| size_t max_starting_point = (pos + 64) - 12; 26| 704k| while (pos < max_starting_point) { ------------------ | Branch (26:14): [True: 632k, False: 72.5k] ------------------ 27| 632k| size_t consumed = convert_masked_utf8_to_utf32( 28| 632k| input + pos, utf8_end_of_code_point_mask, utf32_output); 29| 632k| pos += consumed; 30| 632k| utf8_end_of_code_point_mask >>= consumed; 31| 632k| } 32| 72.5k| } 33| 427k| } 34| 464| utf32_output += scalar::utf8_to_utf32::convert_valid(input + pos, size - pos, 35| 464| utf32_output); 36| 464| return utf32_output - start; 37| 464|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker16check_next_inputERKNS1_4simd8simd8x64IhEE: 187| 4.87M| simdutf_really_inline void check_next_input(const simd8x64 &input) { 188| 4.87M| if (simdutf_likely(is_ascii(input))) { ------------------ | | 90| 4.87M| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.99M, False: 881k] | | ------------------ ------------------ 189| 3.99M| this->error |= this->prev_incomplete; 190| 3.99M| } else { 191| | // you might think that a for-loop would work, but under Visual Studio, it 192| | // is not good enough. 193| 881k| static_assert((simd8x64::NUM_CHUNKS == 2) || 194| 881k| (simd8x64::NUM_CHUNKS == 4), 195| 881k| "We support either two or four chunks per 64-byte block."); 196| 881k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 197| 881k| this->check_utf8_bytes(input.chunks[0], this->prev_input_block); 198| 881k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 199| | } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 200| | this->check_utf8_bytes(input.chunks[0], this->prev_input_block); 201| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 202| | this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 203| | this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 204| | } 205| 881k| this->prev_incomplete = 206| 881k| is_incomplete(input.chunks[simd8x64::NUM_CHUNKS - 1]); 207| 881k| this->prev_input_block = input.chunks[simd8x64::NUM_CHUNKS - 1]; 208| 881k| } 209| 4.87M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker16check_utf8_bytesENS1_4simd5simd8IhEES6_: 169| 1.76M| const simd8 prev_input) { 170| | // Flip prev1...prev3 so we can easily determine if they are 2+, 3+ or 4+ 171| | // lead bytes (2, 3, 4-byte leads become large positive numbers instead of 172| | // small negative numbers) 173| 1.76M| simd8 prev1 = input.prev<1>(prev_input); 174| 1.76M| simd8 sc = check_special_cases(input, prev1); 175| 1.76M| this->error |= check_multibyte_lengths(input, prev_input, sc); 176| 1.76M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation19check_special_casesENS1_4simd5simd8IhEES5_: 9| 1.76M|check_special_cases(const simd8 input, const simd8 prev1) { 10| | // Bit 0 = Too Short (lead byte/ASCII followed by lead byte/ASCII) 11| | // Bit 1 = Too Long (ASCII followed by continuation) 12| | // Bit 2 = Overlong 3-byte 13| | // Bit 4 = Surrogate 14| | // Bit 5 = Overlong 2-byte 15| | // Bit 7 = Two Continuations 16| 1.76M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 17| | // 11______ 11______ 18| 1.76M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 19| 1.76M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 20| 1.76M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 21| 1.76M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 22| 1.76M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 23| 1.76M| constexpr const uint8_t TOO_LARGE = 1 << 3; // 11110100 1001____ 24| | // 11110100 101_____ 25| | // 11110101 1001____ 26| | // 11110101 101_____ 27| | // 1111011_ 1001____ 28| | // 1111011_ 101_____ 29| | // 11111___ 1001____ 30| | // 11111___ 101_____ 31| 1.76M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 32| | // 11110101 1000____ 33| | // 1111011_ 1000____ 34| | // 11111___ 1000____ 35| 1.76M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 36| | 37| 1.76M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 38| | // 0_______ ________ 39| 1.76M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 40| 1.76M| TOO_LONG, 41| | // 10______ ________ 42| 1.76M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 43| | // 1100____ ________ 44| 1.76M| TOO_SHORT | OVERLONG_2, 45| | // 1101____ ________ 46| 1.76M| TOO_SHORT, 47| | // 1110____ ________ 48| 1.76M| TOO_SHORT | OVERLONG_3 | SURROGATE, 49| | // 1111____ ________ 50| 1.76M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 51| 1.76M| constexpr const uint8_t CARRY = 52| 1.76M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 53| 1.76M| const simd8 byte_1_low = 54| 1.76M| (prev1 & 0x0F) 55| 1.76M| .lookup_16( 56| | // ____0000 ________ 57| 1.76M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 58| | // ____0001 ________ 59| 1.76M| CARRY | OVERLONG_2, 60| | // ____001_ ________ 61| 1.76M| CARRY, CARRY, 62| | 63| | // ____0100 ________ 64| 1.76M| CARRY | TOO_LARGE, 65| | // ____0101 ________ 66| 1.76M| CARRY | TOO_LARGE | TOO_LARGE_1000, 67| | // ____011_ ________ 68| 1.76M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| 1.76M| CARRY | TOO_LARGE | TOO_LARGE_1000, 70| | 71| | // ____1___ ________ 72| 1.76M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 1.76M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 1.76M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 1.76M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| 1.76M| CARRY | TOO_LARGE | TOO_LARGE_1000, 77| | // ____1101 ________ 78| 1.76M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 79| 1.76M| CARRY | TOO_LARGE | TOO_LARGE_1000, 80| 1.76M| CARRY | TOO_LARGE | TOO_LARGE_1000); 81| 1.76M| const simd8 byte_2_high = input.shr<4>().lookup_16( 82| | // ________ 0_______ 83| 1.76M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 84| 1.76M| TOO_SHORT, TOO_SHORT, 85| | 86| | // ________ 1000____ 87| 1.76M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 88| 1.76M| OVERLONG_4, 89| | // ________ 1001____ 90| 1.76M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 91| | // ________ 101_____ 92| 1.76M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| 1.76M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 94| | 95| | // ________ 11______ 96| 1.76M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 97| 1.76M| return (byte_1_high & byte_1_low & byte_2_high); 98| 1.76M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation23check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 102| 1.76M| const simd8 sc) { 103| 1.76M| simd8 prev2 = input.prev<2>(prev_input); 104| 1.76M| simd8 prev3 = input.prev<3>(prev_input); 105| 1.76M| simd8 must23 = 106| 1.76M| simd8(must_be_2_3_continuation(prev2, prev3)); 107| 1.76M| simd8 must23_80 = must23 & uint8_t(0x80); 108| 1.76M| return must23_80 ^ sc; 109| 1.76M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation13is_incompleteENS1_4simd5simd8IhEE: 115| 881k|simdutf_really_inline simd8 is_incomplete(const simd8 input) { 116| | // If the previous input's last 3 bytes match this, they're too short (they 117| | // ended at EOF): 118| | // ... 1111____ 111_____ 11______ 119| 881k| static const uint8_t max_array[32] = {255, 120| 881k| 255, 121| 881k| 255, 122| 881k| 255, 123| 881k| 255, 124| 881k| 255, 125| 881k| 255, 126| 881k| 255, 127| 881k| 255, 128| 881k| 255, 129| 881k| 255, 130| 881k| 255, 131| 881k| 255, 132| 881k| 255, 133| 881k| 255, 134| 881k| 255, 135| 881k| 255, 136| 881k| 255, 137| 881k| 255, 138| 881k| 255, 139| 881k| 255, 140| 881k| 255, 141| 881k| 255, 142| 881k| 255, 143| 881k| 255, 144| 881k| 255, 145| 881k| 255, 146| 881k| 255, 147| 881k| 255, 148| 881k| 0b11110000u - 1, 149| 881k| 0b11100000u - 1, 150| 881k| 0b11000000u - 1}; 151| 881k| const simd8 max_value( 152| 881k| &max_array[sizeof(max_array) - sizeof(simd8)]); 153| 881k| return input.gt_bits(max_value); 154| 881k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker9check_eofEv: 181| 6.11k| simdutf_really_inline void check_eof() { 182| | // If the previous block had incomplete UTF-8 characters at the end, an 183| | // ASCII block can't possibly finish them. 184| 6.11k| this->error |= this->prev_incomplete; 185| 6.11k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker6errorsEv: 212| 2.04M| simdutf_really_inline bool errors() const { 213| 2.04M| return this->error.any_bits_set_anywhere(); 214| 2.04M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker16check_next_inputERKNS1_4simd8simd8x64IhEE: 187| 4.87M| simdutf_really_inline void check_next_input(const simd8x64 &input) { 188| 4.87M| if (simdutf_likely(is_ascii(input))) { ------------------ | | 90| 4.87M| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.99M, False: 881k] | | ------------------ ------------------ 189| 3.99M| this->error |= this->prev_incomplete; 190| 3.99M| } else { 191| | // you might think that a for-loop would work, but under Visual Studio, it 192| | // is not good enough. 193| 881k| static_assert((simd8x64::NUM_CHUNKS == 2) || 194| 881k| (simd8x64::NUM_CHUNKS == 4), 195| 881k| "We support either two or four chunks per 64-byte block."); 196| | if constexpr (simd8x64::NUM_CHUNKS == 2) { 197| | this->check_utf8_bytes(input.chunks[0], this->prev_input_block); 198| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 199| 881k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 200| 881k| this->check_utf8_bytes(input.chunks[0], this->prev_input_block); 201| 881k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 202| 881k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 203| 881k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 204| 881k| } 205| 881k| this->prev_incomplete = 206| 881k| is_incomplete(input.chunks[simd8x64::NUM_CHUNKS - 1]); 207| 881k| this->prev_input_block = input.chunks[simd8x64::NUM_CHUNKS - 1]; 208| 881k| } 209| 4.87M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker16check_utf8_bytesENS1_4simd5simd8IhEES6_: 169| 3.52M| const simd8 prev_input) { 170| | // Flip prev1...prev3 so we can easily determine if they are 2+, 3+ or 4+ 171| | // lead bytes (2, 3, 4-byte leads become large positive numbers instead of 172| | // small negative numbers) 173| 3.52M| simd8 prev1 = input.prev<1>(prev_input); 174| 3.52M| simd8 sc = check_special_cases(input, prev1); 175| 3.52M| this->error |= check_multibyte_lengths(input, prev_input, sc); 176| 3.52M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation19check_special_casesENS1_4simd5simd8IhEES5_: 9| 3.52M|check_special_cases(const simd8 input, const simd8 prev1) { 10| | // Bit 0 = Too Short (lead byte/ASCII followed by lead byte/ASCII) 11| | // Bit 1 = Too Long (ASCII followed by continuation) 12| | // Bit 2 = Overlong 3-byte 13| | // Bit 4 = Surrogate 14| | // Bit 5 = Overlong 2-byte 15| | // Bit 7 = Two Continuations 16| 3.52M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 17| | // 11______ 11______ 18| 3.52M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 19| 3.52M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 20| 3.52M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 21| 3.52M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 22| 3.52M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 23| 3.52M| constexpr const uint8_t TOO_LARGE = 1 << 3; // 11110100 1001____ 24| | // 11110100 101_____ 25| | // 11110101 1001____ 26| | // 11110101 101_____ 27| | // 1111011_ 1001____ 28| | // 1111011_ 101_____ 29| | // 11111___ 1001____ 30| | // 11111___ 101_____ 31| 3.52M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 32| | // 11110101 1000____ 33| | // 1111011_ 1000____ 34| | // 11111___ 1000____ 35| 3.52M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 36| | 37| 3.52M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 38| | // 0_______ ________ 39| 3.52M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 40| 3.52M| TOO_LONG, 41| | // 10______ ________ 42| 3.52M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 43| | // 1100____ ________ 44| 3.52M| TOO_SHORT | OVERLONG_2, 45| | // 1101____ ________ 46| 3.52M| TOO_SHORT, 47| | // 1110____ ________ 48| 3.52M| TOO_SHORT | OVERLONG_3 | SURROGATE, 49| | // 1111____ ________ 50| 3.52M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 51| 3.52M| constexpr const uint8_t CARRY = 52| 3.52M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 53| 3.52M| const simd8 byte_1_low = 54| 3.52M| (prev1 & 0x0F) 55| 3.52M| .lookup_16( 56| | // ____0000 ________ 57| 3.52M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 58| | // ____0001 ________ 59| 3.52M| CARRY | OVERLONG_2, 60| | // ____001_ ________ 61| 3.52M| CARRY, CARRY, 62| | 63| | // ____0100 ________ 64| 3.52M| CARRY | TOO_LARGE, 65| | // ____0101 ________ 66| 3.52M| CARRY | TOO_LARGE | TOO_LARGE_1000, 67| | // ____011_ ________ 68| 3.52M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| 3.52M| CARRY | TOO_LARGE | TOO_LARGE_1000, 70| | 71| | // ____1___ ________ 72| 3.52M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 3.52M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 3.52M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 3.52M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| 3.52M| CARRY | TOO_LARGE | TOO_LARGE_1000, 77| | // ____1101 ________ 78| 3.52M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 79| 3.52M| CARRY | TOO_LARGE | TOO_LARGE_1000, 80| 3.52M| CARRY | TOO_LARGE | TOO_LARGE_1000); 81| 3.52M| const simd8 byte_2_high = input.shr<4>().lookup_16( 82| | // ________ 0_______ 83| 3.52M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 84| 3.52M| TOO_SHORT, TOO_SHORT, 85| | 86| | // ________ 1000____ 87| 3.52M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 88| 3.52M| OVERLONG_4, 89| | // ________ 1001____ 90| 3.52M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 91| | // ________ 101_____ 92| 3.52M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| 3.52M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 94| | 95| | // ________ 11______ 96| 3.52M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 97| 3.52M| return (byte_1_high & byte_1_low & byte_2_high); 98| 3.52M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation23check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 102| 3.52M| const simd8 sc) { 103| 3.52M| simd8 prev2 = input.prev<2>(prev_input); 104| 3.52M| simd8 prev3 = input.prev<3>(prev_input); 105| 3.52M| simd8 must23 = 106| 3.52M| simd8(must_be_2_3_continuation(prev2, prev3)); 107| 3.52M| simd8 must23_80 = must23 & uint8_t(0x80); 108| 3.52M| return must23_80 ^ sc; 109| 3.52M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation13is_incompleteENS1_4simd5simd8IhEE: 115| 881k|simdutf_really_inline simd8 is_incomplete(const simd8 input) { 116| | // If the previous input's last 3 bytes match this, they're too short (they 117| | // ended at EOF): 118| | // ... 1111____ 111_____ 11______ 119| 881k| static const uint8_t max_array[32] = {255, 120| 881k| 255, 121| 881k| 255, 122| 881k| 255, 123| 881k| 255, 124| 881k| 255, 125| 881k| 255, 126| 881k| 255, 127| 881k| 255, 128| 881k| 255, 129| 881k| 255, 130| 881k| 255, 131| 881k| 255, 132| 881k| 255, 133| 881k| 255, 134| 881k| 255, 135| 881k| 255, 136| 881k| 255, 137| 881k| 255, 138| 881k| 255, 139| 881k| 255, 140| 881k| 255, 141| 881k| 255, 142| 881k| 255, 143| 881k| 255, 144| 881k| 255, 145| 881k| 255, 146| 881k| 255, 147| 881k| 255, 148| 881k| 0b11110000u - 1, 149| 881k| 0b11100000u - 1, 150| 881k| 0b11000000u - 1}; 151| 881k| const simd8 max_value( 152| 881k| &max_array[sizeof(max_array) - sizeof(simd8)]); 153| 881k| return input.gt_bits(max_value); 154| 881k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker9check_eofEv: 181| 6.11k| simdutf_really_inline void check_eof() { 182| | // If the previous block had incomplete UTF-8 characters at the end, an 183| | // ASCII block can't possibly finish them. 184| 6.11k| this->error |= this->prev_incomplete; 185| 6.11k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker6errorsEv: 212| 2.04M| simdutf_really_inline bool errors() const { 213| 2.04M| return this->error.any_bits_set_anywhere(); 214| 2.04M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation21generic_validate_utf8EPKcm: 27| 3.48k|bool generic_validate_utf8(const char *input, size_t length) { 28| 3.48k| return generic_validate_utf8( 29| 3.48k| reinterpret_cast(input), length); 30| 3.48k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation21generic_validate_utf8INS2_12utf8_checkerEEEbPKhm: 10| 3.48k|bool generic_validate_utf8(const uint8_t *input, size_t length) { 11| 3.48k| checker c{}; 12| 3.48k| buf_block_reader<64> reader(input, length); 13| 2.82M| while (reader.has_full_block()) { ------------------ | Branch (13:10): [True: 2.82M, False: 3.48k] ------------------ 14| 2.82M| simd::simd8x64 in(reader.full_block()); 15| 2.82M| c.check_next_input(in); 16| 2.82M| reader.advance(); 17| 2.82M| } 18| 3.48k| uint8_t block[64]{}; 19| 3.48k| reader.get_remainder(block); 20| 3.48k| simd::simd8x64 in(block); 21| 3.48k| c.check_next_input(in); 22| 3.48k| reader.advance(); 23| 3.48k| c.check_eof(); 24| 3.48k| return !c.errors(); 25| 3.48k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsEPKcm: 76| 3.48k|result generic_validate_utf8_with_errors(const char *input, size_t length) { 77| 3.48k| return generic_validate_utf8_with_errors( 78| 3.48k| reinterpret_cast(input), length); 79| 3.48k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsINS2_12utf8_checkerEEENS_6resultEPKhm: 36| 3.48k|result generic_validate_utf8_with_errors(const uint8_t *input, size_t length) { 37| 3.48k| checker c{}; 38| 3.48k| buf_block_reader<64> reader(input, length); 39| 3.48k| size_t count{0}; 40| 2.04M| while (reader.has_full_block()) { ------------------ | Branch (40:10): [True: 2.04M, False: 2.62k] ------------------ 41| 2.04M| simd::simd8x64 in(reader.full_block()); 42| 2.04M| c.check_next_input(in); 43| 2.04M| if (c.errors()) { ------------------ | Branch (43:9): [True: 858, False: 2.03M] ------------------ 44| 858| if (count != 0) { ------------------ | Branch (44:11): [True: 233, False: 625] ------------------ 45| 233| count--; 46| 233| } // Sometimes the error is only detected in the next chunk 47| 858| result res = scalar::utf8::rewind_and_validate_with_errors( 48| 858| reinterpret_cast(input), 49| 858| reinterpret_cast(input + count), length - count); 50| 858| res.count += count; 51| 858| return res; 52| 858| } 53| 2.03M| reader.advance(); 54| 2.03M| count += 64; 55| 2.03M| } 56| 2.62k| uint8_t block[64]{}; 57| 2.62k| reader.get_remainder(block); 58| 2.62k| simd::simd8x64 in(block); 59| 2.62k| c.check_next_input(in); 60| 2.62k| reader.advance(); 61| 2.62k| c.check_eof(); 62| 2.62k| if (c.errors()) { ------------------ | Branch (62:7): [True: 901, False: 1.72k] ------------------ 63| 901| if (count != 0) { ------------------ | Branch (63:9): [True: 234, False: 667] ------------------ 64| 234| count--; 65| 234| } // Sometimes the error is only detected in the next chunk 66| 901| result res = scalar::utf8::rewind_and_validate_with_errors( 67| 901| reinterpret_cast(input), 68| 901| reinterpret_cast(input) + count, length - count); 69| 901| res.count += count; 70| 901| return res; 71| 1.72k| } else { 72| 1.72k| return result(error_code::SUCCESS, length); 73| 1.72k| } 74| 2.62k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation21generic_validate_utf8EPKcm: 27| 3.48k|bool generic_validate_utf8(const char *input, size_t length) { 28| 3.48k| return generic_validate_utf8( 29| 3.48k| reinterpret_cast(input), length); 30| 3.48k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation21generic_validate_utf8INS2_12utf8_checkerEEEbPKhm: 10| 3.48k|bool generic_validate_utf8(const uint8_t *input, size_t length) { 11| 3.48k| checker c{}; 12| 3.48k| buf_block_reader<64> reader(input, length); 13| 2.82M| while (reader.has_full_block()) { ------------------ | Branch (13:10): [True: 2.82M, False: 3.48k] ------------------ 14| 2.82M| simd::simd8x64 in(reader.full_block()); 15| 2.82M| c.check_next_input(in); 16| 2.82M| reader.advance(); 17| 2.82M| } 18| 3.48k| uint8_t block[64]{}; 19| 3.48k| reader.get_remainder(block); 20| 3.48k| simd::simd8x64 in(block); 21| 3.48k| c.check_next_input(in); 22| 3.48k| reader.advance(); 23| 3.48k| c.check_eof(); 24| 3.48k| return !c.errors(); 25| 3.48k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsEPKcm: 76| 3.48k|result generic_validate_utf8_with_errors(const char *input, size_t length) { 77| 3.48k| return generic_validate_utf8_with_errors( 78| 3.48k| reinterpret_cast(input), length); 79| 3.48k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsINS2_12utf8_checkerEEENS_6resultEPKhm: 36| 3.48k|result generic_validate_utf8_with_errors(const uint8_t *input, size_t length) { 37| 3.48k| checker c{}; 38| 3.48k| buf_block_reader<64> reader(input, length); 39| 3.48k| size_t count{0}; 40| 2.04M| while (reader.has_full_block()) { ------------------ | Branch (40:10): [True: 2.04M, False: 2.62k] ------------------ 41| 2.04M| simd::simd8x64 in(reader.full_block()); 42| 2.04M| c.check_next_input(in); 43| 2.04M| if (c.errors()) { ------------------ | Branch (43:9): [True: 858, False: 2.03M] ------------------ 44| 858| if (count != 0) { ------------------ | Branch (44:11): [True: 233, False: 625] ------------------ 45| 233| count--; 46| 233| } // Sometimes the error is only detected in the next chunk 47| 858| result res = scalar::utf8::rewind_and_validate_with_errors( 48| 858| reinterpret_cast(input), 49| 858| reinterpret_cast(input + count), length - count); 50| 858| res.count += count; 51| 858| return res; 52| 858| } 53| 2.03M| reader.advance(); 54| 2.03M| count += 64; 55| 2.03M| } 56| 2.62k| uint8_t block[64]{}; 57| 2.62k| reader.get_remainder(block); 58| 2.62k| simd::simd8x64 in(block); 59| 2.62k| c.check_next_input(in); 60| 2.62k| reader.advance(); 61| 2.62k| c.check_eof(); 62| 2.62k| if (c.errors()) { ------------------ | Branch (62:7): [True: 901, False: 1.72k] ------------------ 63| 901| if (count != 0) { ------------------ | Branch (63:9): [True: 234, False: 667] ------------------ 64| 234| count--; 65| 234| } // Sometimes the error is only detected in the next chunk 66| 901| result res = scalar::utf8::rewind_and_validate_with_errors( 67| 901| reinterpret_cast(input), 68| 901| reinterpret_cast(input) + count, length - count); 69| 901| res.count += count; 70| 901| return res; 71| 1.72k| } else { 72| 1.72k| return result(error_code::SUCCESS, length); 73| 1.72k| } 74| 2.62k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE0EEEKNS_6resultEPKDsm: 49| 2.69k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.69k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.69k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 17, False: 2.67k] | | ------------------ ------------------ 51| 17| return result(error_code::SUCCESS, 0); 52| 17| } 53| | 54| 2.67k| const char16_t *start = input; 55| 2.67k| const char16_t *end = input + size; 56| | 57| 2.67k| const auto v_d8 = simd8::splat(0xd8); 58| 2.67k| const auto v_f8 = simd8::splat(0xf8); 59| 2.67k| const auto v_fc = simd8::splat(0xfc); 60| 2.67k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.54M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.54M, False: 2.50k] ------------------ 63| | // 0. Load data: since the validation takes into account only higher 64| | // byte of each word, we compress the two vectors into one which 65| | // consists only the higher bytes. 66| 2.54M| auto in0 = simd16(input); 67| 2.54M| auto in1 = 68| 2.54M| simd16(input + simd16::SIZE / sizeof(char16_t)); 69| | 70| | // Function `utf16_gather_high_bytes` consumes two vectors of UTF-16 71| | // and yields a single vector having only higher bytes of characters. 72| 2.54M| const auto in = utf16_gather_high_bytes(in0, in1); 73| | 74| | // 1. Check whether we have any 0xD800..DFFF word (0b1101'1xxx'yyyy'yyyy). 75| 2.54M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.54M| const uint16_t surrogates_bitmask = 77| 2.54M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.54M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.52M, False: 19.6k] ------------------ 79| 2.52M| input += 16; 80| 2.52M| } else { 81| | // 2. We have some surrogates that have to be distinguished: 82| | // - low surrogates: 0b1101'10xx'yyyy'yyyy (0xD800..0xDBFF) 83| | // - high surrogates: 0b1101'11xx'yyyy'yyyy (0xDC00..0xDFFF) 84| | // 85| | // Fact: high surrogate has 11th bit set (3rd bit in the higher byte) 86| | 87| | // V - non-surrogate code units 88| | // V = not surrogates_wordmask 89| 19.6k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 19.6k| const auto vH = (in & v_fc) == v_dc; 93| 19.6k| const uint16_t H = static_cast(vH.to_bitmask()); 94| | 95| | // L - word mask for low surrogates 96| | // L = not H and surrogates_wordmask 97| 19.6k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 19.6k| const uint16_t a = static_cast( 100| 19.6k| L & (H >> 1)); // A low surrogate must be followed by high one. 101| | // (A low surrogate placed in the 7th register's word 102| | // is an exception we handle.) 103| 19.6k| const uint16_t b = static_cast( 104| 19.6k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 19.6k| const uint16_t c = static_cast( 107| 19.6k| V | a | b); // Combine all the masks into the final one. 108| | 109| 19.6k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 14.7k, False: 4.94k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 14.7k| input += 16; 113| 14.7k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 4.77k, False: 176] ------------------ 114| | // The 15 lower code units of the input register contains valid UTF-16. 115| | // The 15th word may be either a low or high surrogate. It the next 116| | // iteration we 1) check if the low surrogate is followed by a high 117| | // one, 2) reject sole high surrogate. 118| 4.77k| input += 15; 119| 4.77k| } else { 120| 176| return result(error_code::SURROGATE, input - start); 121| 176| } 122| 19.6k| } 123| 2.54M| } 124| | 125| 2.50k| return result(error_code::SUCCESS, input - start); 126| 2.67k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE1EEEKNS_6resultEPKDsm: 49| 2.66k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.66k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.66k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 2.64k] | | ------------------ ------------------ 51| 16| return result(error_code::SUCCESS, 0); 52| 16| } 53| | 54| 2.64k| const char16_t *start = input; 55| 2.64k| const char16_t *end = input + size; 56| | 57| 2.64k| const auto v_d8 = simd8::splat(0xd8); 58| 2.64k| const auto v_f8 = simd8::splat(0xf8); 59| 2.64k| const auto v_fc = simd8::splat(0xfc); 60| 2.64k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.61M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.61M, False: 2.44k] ------------------ 63| | // 0. Load data: since the validation takes into account only higher 64| | // byte of each word, we compress the two vectors into one which 65| | // consists only the higher bytes. 66| 2.61M| auto in0 = simd16(input); 67| 2.61M| auto in1 = 68| 2.61M| simd16(input + simd16::SIZE / sizeof(char16_t)); 69| | 70| | // Function `utf16_gather_high_bytes` consumes two vectors of UTF-16 71| | // and yields a single vector having only higher bytes of characters. 72| 2.61M| const auto in = utf16_gather_high_bytes(in0, in1); 73| | 74| | // 1. Check whether we have any 0xD800..DFFF word (0b1101'1xxx'yyyy'yyyy). 75| 2.61M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.61M| const uint16_t surrogates_bitmask = 77| 2.61M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.61M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.59M, False: 18.0k] ------------------ 79| 2.59M| input += 16; 80| 2.59M| } else { 81| | // 2. We have some surrogates that have to be distinguished: 82| | // - low surrogates: 0b1101'10xx'yyyy'yyyy (0xD800..0xDBFF) 83| | // - high surrogates: 0b1101'11xx'yyyy'yyyy (0xDC00..0xDFFF) 84| | // 85| | // Fact: high surrogate has 11th bit set (3rd bit in the higher byte) 86| | 87| | // V - non-surrogate code units 88| | // V = not surrogates_wordmask 89| 18.0k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 18.0k| const auto vH = (in & v_fc) == v_dc; 93| 18.0k| const uint16_t H = static_cast(vH.to_bitmask()); 94| | 95| | // L - word mask for low surrogates 96| | // L = not H and surrogates_wordmask 97| 18.0k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 18.0k| const uint16_t a = static_cast( 100| 18.0k| L & (H >> 1)); // A low surrogate must be followed by high one. 101| | // (A low surrogate placed in the 7th register's word 102| | // is an exception we handle.) 103| 18.0k| const uint16_t b = static_cast( 104| 18.0k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 18.0k| const uint16_t c = static_cast( 107| 18.0k| V | a | b); // Combine all the masks into the final one. 108| | 109| 18.0k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 13.4k, False: 4.56k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 13.4k| input += 16; 113| 13.4k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 4.36k, False: 204] ------------------ 114| | // The 15 lower code units of the input register contains valid UTF-16. 115| | // The 15th word may be either a low or high surrogate. It the next 116| | // iteration we 1) check if the low surrogate is followed by a high 117| | // one, 2) reject sole high surrogate. 118| 4.36k| input += 15; 119| 4.36k| } else { 120| 204| return result(error_code::SURROGATE, input - start); 121| 204| } 122| 18.0k| } 123| 2.61M| } 124| | 125| 2.44k| return result(error_code::SUCCESS, input - start); 126| 2.64k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE0EEEKNS_6resultEPKDsm: 49| 2.69k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.69k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.69k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 17, False: 2.67k] | | ------------------ ------------------ 51| 17| return result(error_code::SUCCESS, 0); 52| 17| } 53| | 54| 2.67k| const char16_t *start = input; 55| 2.67k| const char16_t *end = input + size; 56| | 57| 2.67k| const auto v_d8 = simd8::splat(0xd8); 58| 2.67k| const auto v_f8 = simd8::splat(0xf8); 59| 2.67k| const auto v_fc = simd8::splat(0xfc); 60| 2.67k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.54M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.54M, False: 2.39k] ------------------ 63| | // 0. Load data: since the validation takes into account only higher 64| | // byte of each word, we compress the two vectors into one which 65| | // consists only the higher bytes. 66| 2.54M| auto in0 = simd16(input); 67| 2.54M| auto in1 = 68| 2.54M| simd16(input + simd16::SIZE / sizeof(char16_t)); 69| | 70| | // Function `utf16_gather_high_bytes` consumes two vectors of UTF-16 71| | // and yields a single vector having only higher bytes of characters. 72| 2.54M| const auto in = utf16_gather_high_bytes(in0, in1); 73| | 74| | // 1. Check whether we have any 0xD800..DFFF word (0b1101'1xxx'yyyy'yyyy). 75| 2.54M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.54M| const uint16_t surrogates_bitmask = 77| 2.54M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.54M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.52M, False: 20.1k] ------------------ 79| 2.52M| input += 16; 80| 2.52M| } else { 81| | // 2. We have some surrogates that have to be distinguished: 82| | // - low surrogates: 0b1101'10xx'yyyy'yyyy (0xD800..0xDBFF) 83| | // - high surrogates: 0b1101'11xx'yyyy'yyyy (0xDC00..0xDFFF) 84| | // 85| | // Fact: high surrogate has 11th bit set (3rd bit in the higher byte) 86| | 87| | // V - non-surrogate code units 88| | // V = not surrogates_wordmask 89| 20.1k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 20.1k| const auto vH = (in & v_fc) == v_dc; 93| 20.1k| const uint16_t H = static_cast(vH.to_bitmask()); 94| | 95| | // L - word mask for low surrogates 96| | // L = not H and surrogates_wordmask 97| 20.1k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 20.1k| const uint16_t a = static_cast( 100| 20.1k| L & (H >> 1)); // A low surrogate must be followed by high one. 101| | // (A low surrogate placed in the 7th register's word 102| | // is an exception we handle.) 103| 20.1k| const uint16_t b = static_cast( 104| 20.1k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 20.1k| const uint16_t c = static_cast( 107| 20.1k| V | a | b); // Combine all the masks into the final one. 108| | 109| 20.1k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 15.0k, False: 5.11k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 15.0k| input += 16; 113| 15.0k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 4.83k, False: 278] ------------------ 114| | // The 15 lower code units of the input register contains valid UTF-16. 115| | // The 15th word may be either a low or high surrogate. It the next 116| | // iteration we 1) check if the low surrogate is followed by a high 117| | // one, 2) reject sole high surrogate. 118| 4.83k| input += 15; 119| 4.83k| } else { 120| 278| return result(error_code::SURROGATE, input - start); 121| 278| } 122| 20.1k| } 123| 2.54M| } 124| | 125| 2.39k| return result(error_code::SUCCESS, input - start); 126| 2.67k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE1EEEKNS_6resultEPKDsm: 49| 2.66k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.66k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.66k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 2.64k] | | ------------------ ------------------ 51| 16| return result(error_code::SUCCESS, 0); 52| 16| } 53| | 54| 2.64k| const char16_t *start = input; 55| 2.64k| const char16_t *end = input + size; 56| | 57| 2.64k| const auto v_d8 = simd8::splat(0xd8); 58| 2.64k| const auto v_f8 = simd8::splat(0xf8); 59| 2.64k| const auto v_fc = simd8::splat(0xfc); 60| 2.64k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.61M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.61M, False: 2.33k] ------------------ 63| | // 0. Load data: since the validation takes into account only higher 64| | // byte of each word, we compress the two vectors into one which 65| | // consists only the higher bytes. 66| 2.61M| auto in0 = simd16(input); 67| 2.61M| auto in1 = 68| 2.61M| simd16(input + simd16::SIZE / sizeof(char16_t)); 69| | 70| | // Function `utf16_gather_high_bytes` consumes two vectors of UTF-16 71| | // and yields a single vector having only higher bytes of characters. 72| 2.61M| const auto in = utf16_gather_high_bytes(in0, in1); 73| | 74| | // 1. Check whether we have any 0xD800..DFFF word (0b1101'1xxx'yyyy'yyyy). 75| 2.61M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.61M| const uint16_t surrogates_bitmask = 77| 2.61M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.61M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.59M, False: 18.5k] ------------------ 79| 2.59M| input += 16; 80| 2.59M| } else { 81| | // 2. We have some surrogates that have to be distinguished: 82| | // - low surrogates: 0b1101'10xx'yyyy'yyyy (0xD800..0xDBFF) 83| | // - high surrogates: 0b1101'11xx'yyyy'yyyy (0xDC00..0xDFFF) 84| | // 85| | // Fact: high surrogate has 11th bit set (3rd bit in the higher byte) 86| | 87| | // V - non-surrogate code units 88| | // V = not surrogates_wordmask 89| 18.5k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 18.5k| const auto vH = (in & v_fc) == v_dc; 93| 18.5k| const uint16_t H = static_cast(vH.to_bitmask()); 94| | 95| | // L - word mask for low surrogates 96| | // L = not H and surrogates_wordmask 97| 18.5k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 18.5k| const uint16_t a = static_cast( 100| 18.5k| L & (H >> 1)); // A low surrogate must be followed by high one. 101| | // (A low surrogate placed in the 7th register's word 102| | // is an exception we handle.) 103| 18.5k| const uint16_t b = static_cast( 104| 18.5k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 18.5k| const uint16_t c = static_cast( 107| 18.5k| V | a | b); // Combine all the masks into the final one. 108| | 109| 18.5k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 13.7k, False: 4.78k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 13.7k| input += 16; 113| 13.7k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 4.47k, False: 310] ------------------ 114| | // The 15 lower code units of the input register contains valid UTF-16. 115| | // The 15th word may be either a low or high surrogate. It the next 116| | // iteration we 1) check if the low surrogate is followed by a high 117| | // one, 2) reject sole high surrogate. 118| 4.47k| input += 15; 119| 4.47k| } else { 120| 310| return result(error_code::SURROGATE, input - start); 121| 310| } 122| 18.5k| } 123| 2.61M| } 124| | 125| 2.33k| return result(error_code::SUCCESS, input - start); 126| 2.64k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf328validateEPKDim: 6| 2.63k|simdutf_really_inline bool validate(const char32_t *input, size_t size) { 7| 2.63k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.63k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 22, False: 2.61k] | | ------------------ ------------------ 8| | // empty input is valid UTF-32. protect the implementation from 9| | // handling nullptr 10| 22| return true; 11| 22| } 12| | 13| 2.61k| const char32_t *end = input + size; 14| | 15| 2.61k| using vector_u32 = simd32; 16| | 17| 2.61k| const auto standardmax = vector_u32::splat(0x10ffff); 18| 2.61k| const auto offset = vector_u32::splat(0xffff2000); 19| 2.61k| const auto standardoffsetmax = vector_u32::splat(0xfffff7ff); 20| 2.61k| auto currentmax = vector_u32::zero(); 21| 2.61k| auto currentoffsetmax = vector_u32::zero(); 22| | 23| 2.61k| constexpr size_t N = vector_u32::ELEMENTS; 24| | 25| 1.12M| while (input + N < end) { ------------------ | Branch (25:10): [True: 1.11M, False: 2.61k] ------------------ 26| 1.11M| auto in = vector_u32(input); 27| 1.11M| if constexpr (!match_system(endianness::BIG)) { 28| 1.11M| in.swap_bytes(); 29| 1.11M| } 30| | 31| 1.11M| currentmax = max(currentmax, in); 32| 1.11M| currentoffsetmax = max(currentoffsetmax, in + offset); 33| 1.11M| input += N; 34| 1.11M| } 35| | 36| 2.61k| const auto too_large = currentmax > standardmax; 37| 2.61k| if (too_large.any()) { ------------------ | Branch (37:7): [True: 968, False: 1.64k] ------------------ 38| 968| return false; 39| 968| } 40| | 41| 1.64k| const auto surrogate = currentoffsetmax > standardoffsetmax; 42| 1.64k| if (surrogate.any()) { ------------------ | Branch (42:7): [True: 27, False: 1.61k] ------------------ 43| 27| return false; 44| 27| } 45| | 46| 1.61k| return scalar::utf32::validate(input, end - input); 47| 1.64k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf3220validate_with_errorsEPKDim: 50| 2.63k| size_t size) { 51| 2.63k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.63k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 22, False: 2.61k] | | ------------------ ------------------ 52| | // empty input is valid UTF-32. protect the implementation from 53| | // handling nullptr 54| 22| return result(error_code::SUCCESS, 0); 55| 22| } 56| | 57| 2.61k| const char32_t *start = input; 58| 2.61k| const char32_t *end = input + size; 59| | 60| 2.61k| using vector_u32 = simd32; 61| | 62| 2.61k| const auto standardmax = vector_u32::splat(0x10ffff + 1); 63| 2.61k| const auto surrogate_mask = vector_u32::splat(0xfffff800); 64| 2.61k| const auto surrogate_byte = vector_u32::splat(0x0000d800); 65| | 66| 2.61k| constexpr size_t N = vector_u32::ELEMENTS; 67| | 68| 250k| while (input + N < end) { ------------------ | Branch (68:10): [True: 248k, False: 1.61k] ------------------ 69| 248k| auto in = vector_u32(input); 70| 248k| if constexpr (!match_system(endianness::BIG)) { 71| 248k| in.swap_bytes(); 72| 248k| } 73| | 74| 248k| const auto too_large = in >= standardmax; 75| 248k| const auto surrogate = (in & surrogate_mask) == surrogate_byte; 76| | 77| 248k| const auto combined = too_large | surrogate; 78| 248k| if (simdutf_unlikely(combined.any())) { ------------------ | | 93| 248k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 995, False: 247k] | | ------------------ ------------------ 79| 995| const size_t consumed = input - start; 80| 995| auto sr = scalar::utf32::validate_with_errors(input, end - input); 81| 995| sr.count += consumed; 82| | 83| 995| return sr; 84| 995| } 85| | 86| 247k| input += N; 87| 247k| } 88| | 89| 1.61k| const size_t consumed = input - start; 90| 1.61k| auto sr = scalar::utf32::validate_with_errors(input, end - input); 91| 1.61k| sr.count += consumed; 92| | 93| 1.61k| return sr; 94| 2.61k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf328validateEPKDim: 6| 2.63k|simdutf_really_inline bool validate(const char32_t *input, size_t size) { 7| 2.63k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.63k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 22, False: 2.61k] | | ------------------ ------------------ 8| | // empty input is valid UTF-32. protect the implementation from 9| | // handling nullptr 10| 22| return true; 11| 22| } 12| | 13| 2.61k| const char32_t *end = input + size; 14| | 15| 2.61k| using vector_u32 = simd32; 16| | 17| 2.61k| const auto standardmax = vector_u32::splat(0x10ffff); 18| 2.61k| const auto offset = vector_u32::splat(0xffff2000); 19| 2.61k| const auto standardoffsetmax = vector_u32::splat(0xfffff7ff); 20| 2.61k| auto currentmax = vector_u32::zero(); 21| 2.61k| auto currentoffsetmax = vector_u32::zero(); 22| | 23| 2.61k| constexpr size_t N = vector_u32::ELEMENTS; 24| | 25| 2.24M| while (input + N < end) { ------------------ | Branch (25:10): [True: 2.23M, False: 2.61k] ------------------ 26| 2.23M| auto in = vector_u32(input); 27| 2.23M| if constexpr (!match_system(endianness::BIG)) { 28| 2.23M| in.swap_bytes(); 29| 2.23M| } 30| | 31| 2.23M| currentmax = max(currentmax, in); 32| 2.23M| currentoffsetmax = max(currentoffsetmax, in + offset); 33| 2.23M| input += N; 34| 2.23M| } 35| | 36| 2.61k| const auto too_large = currentmax > standardmax; 37| 2.61k| if (too_large.any()) { ------------------ | Branch (37:7): [True: 1.15k, False: 1.45k] ------------------ 38| 1.15k| return false; 39| 1.15k| } 40| | 41| 1.45k| const auto surrogate = currentoffsetmax > standardoffsetmax; 42| 1.45k| if (surrogate.any()) { ------------------ | Branch (42:7): [True: 35, False: 1.42k] ------------------ 43| 35| return false; 44| 35| } 45| | 46| 1.42k| return scalar::utf32::validate(input, end - input); 47| 1.45k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf3220validate_with_errorsEPKDim: 50| 2.63k| size_t size) { 51| 2.63k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.63k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 22, False: 2.61k] | | ------------------ ------------------ 52| | // empty input is valid UTF-32. protect the implementation from 53| | // handling nullptr 54| 22| return result(error_code::SUCCESS, 0); 55| 22| } 56| | 57| 2.61k| const char32_t *start = input; 58| 2.61k| const char32_t *end = input + size; 59| | 60| 2.61k| using vector_u32 = simd32; 61| | 62| 2.61k| const auto standardmax = vector_u32::splat(0x10ffff + 1); 63| 2.61k| const auto surrogate_mask = vector_u32::splat(0xfffff800); 64| 2.61k| const auto surrogate_byte = vector_u32::splat(0x0000d800); 65| | 66| 2.61k| constexpr size_t N = vector_u32::ELEMENTS; 67| | 68| 499k| while (input + N < end) { ------------------ | Branch (68:10): [True: 497k, False: 1.42k] ------------------ 69| 497k| auto in = vector_u32(input); 70| 497k| if constexpr (!match_system(endianness::BIG)) { 71| 497k| in.swap_bytes(); 72| 497k| } 73| | 74| 497k| const auto too_large = in >= standardmax; 75| 497k| const auto surrogate = (in & surrogate_mask) == surrogate_byte; 76| | 77| 497k| const auto combined = too_large | surrogate; 78| 497k| if (simdutf_unlikely(combined.any())) { ------------------ | | 93| 497k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 1.19k, False: 496k] | | ------------------ ------------------ 79| 1.19k| const size_t consumed = input - start; 80| 1.19k| auto sr = scalar::utf32::validate_with_errors(input, end - input); 81| 1.19k| sr.count += consumed; 82| | 83| 1.19k| return sr; 84| 1.19k| } 85| | 86| 496k| input += N; 87| 496k| } 88| | 89| 1.42k| const size_t consumed = input - start; 90| 1.42k| auto sr = scalar::utf32::validate_with_errors(input, end - input); 91| 1.42k| sr.count += consumed; 92| | 93| 1.42k| return sr; 94| 2.61k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_latin1_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 76| char16_t *utf16_output) { 5| 76| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| | 7| 76| size_t i = 0; 8| 93.4k| for (; i < rounded_len; i += 16) { ------------------ | Branch (8:10): [True: 93.3k, False: 76] ------------------ 9| | // Load 16 bytes from the address (input + i) into a xmm register 10| 93.3k| const __m128i latin1 = 11| 93.3k| _mm_loadu_si128(reinterpret_cast(latin1_input + i)); 12| | 13| | // Zero extend each byte in `in` to word 14| 93.3k| __m256i utf16 = _mm256_cvtepu8_epi16(latin1); 15| | 16| 93.3k| if (big_endian) { ------------------ | Branch (16:9): [Folded, False: 93.3k] ------------------ 17| 0| const __m128i swap128 = 18| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 19| 0| const __m256i swap = _mm256_set_m128i(swap128, swap128); ------------------ | | 4| 0| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 0| _mm256_castsi128_si256(xmm2), 2) ------------------ 20| 0| utf16 = _mm256_shuffle_epi8(utf16, swap); 21| 0| } 22| | 23| | // Store the contents of xmm1 into the address pointed by (output + i) 24| 93.3k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output + i), utf16); 25| 93.3k| } 26| | 27| 76| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 28| 76|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_latin1_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 96| char16_t *utf16_output) { 5| 96| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| | 7| 96| size_t i = 0; 8| 190k| for (; i < rounded_len; i += 16) { ------------------ | Branch (8:10): [True: 190k, False: 96] ------------------ 9| | // Load 16 bytes from the address (input + i) into a xmm register 10| 190k| const __m128i latin1 = 11| 190k| _mm_loadu_si128(reinterpret_cast(latin1_input + i)); 12| | 13| | // Zero extend each byte in `in` to word 14| 190k| __m256i utf16 = _mm256_cvtepu8_epi16(latin1); 15| | 16| 190k| if (big_endian) { ------------------ | Branch (16:9): [True: 190k, Folded] ------------------ 17| 190k| const __m128i swap128 = 18| 190k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 19| 190k| const __m256i swap = _mm256_set_m128i(swap128, swap128); ------------------ | | 4| 190k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 190k| _mm256_castsi128_si256(xmm2), 2) ------------------ 20| 190k| utf16 = _mm256_shuffle_epi8(utf16, swap); 21| 190k| } 22| | 23| | // Store the contents of xmm1 into the address pointed by (output + i) 24| 190k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output + i), utf16); 25| 190k| } 26| | 27| 96| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 28| 96|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_latin1_to_utf32EPKcmPDi: 3| 124| char32_t *utf32_output) { 4| 124| size_t rounded_len = ((len | 7) ^ 7); // Round down to nearest multiple of 8 5| | 6| 867k| for (size_t i = 0; i < rounded_len; i += 8) { ------------------ | Branch (6:22): [True: 867k, False: 124] ------------------ 7| | // Load 8 Latin1 characters into a 64-bit register 8| 867k| __m128i in = _mm_loadl_epi64((__m128i *)&buf[i]); 9| | 10| | // Zero extend each set of 8 Latin1 characters to 8 32-bit integers using 11| | // vpmovzxbd 12| 867k| __m256i out = _mm256_cvtepu8_epi32(in); 13| | 14| | // Store the results back to memory 15| 867k| _mm256_storeu_si256((__m256i *)&utf32_output[i], out); 16| 867k| } 17| | 18| | // return pointers pointing to where we left off 19| 124| return std::make_pair(buf + rounded_len, utf32_output + rounded_len); 20| 124|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_latin1_to_utf8EPKcmPc: 3| 604| char *utf8_output) { 4| 604| const char *end = latin1_input + len; 5| 604| const __m256i v_0000 = _mm256_setzero_si256(); 6| 604| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 7| 604| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 8| 604| const size_t safety_margin = 12; 9| | 10| 3.35M| while (end - latin1_input >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (10:10): [True: 3.35M, False: 604] ------------------ 11| 3.35M| __m128i in8 = _mm_loadu_si128((__m128i *)latin1_input); 12| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 13| 3.35M| const __m128i v_80 = _mm_set1_epi8((char)0x80); 14| 3.35M| if (_mm_testz_si128(in8, v_80)) { // ASCII fast path!!!! ------------------ | Branch (14:9): [True: 2.90M, False: 445k] ------------------ 15| | // 1. store (16 bytes) 16| 2.90M| _mm_storeu_si128((__m128i *)utf8_output, in8); 17| | // 2. adjust pointers 18| 2.90M| latin1_input += 16; 19| 2.90M| utf8_output += 16; 20| 2.90M| continue; // we are done for this round! 21| 2.90M| } 22| | // We proceed only with the first 16 bytes. 23| 445k| const __m256i in = _mm256_cvtepu8_epi16((in8)); 24| | 25| | // 1. prepare 2-byte values 26| | // input 16-bit word : [0000|0000|aabb|bbbb] x 8 27| | // expected output : [1100|00aa|10bb|bbbb] x 8 28| 445k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 29| 445k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 30| | 31| | // t0 = [0000|00aa|bbbb|bb00] 32| 445k| const __m256i t0 = _mm256_slli_epi16(in, 2); 33| | // t1 = [0000|00aa|0000|0000] 34| 445k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 35| | // t2 = [0000|0000|00bb|bbbb] 36| 445k| const __m256i t2 = _mm256_and_si256(in, v_003f); 37| | // t3 = [000a|aaaa|00bb|bbbb] 38| 445k| const __m256i t3 = _mm256_or_si256(t1, t2); 39| | // t4 = [1100|00aa|10bb|bbbb] 40| 445k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 41| | 42| | // 2. merge ASCII and 2-byte codewords 43| | 44| | // no bits set above 7th bit 45| 445k| const __m256i one_byte_bytemask = 46| 445k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 47| 445k| const uint32_t one_byte_bitmask = 48| 445k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 49| | 50| 445k| const __m256i utf8_unpacked = _mm256_blendv_epi8(t4, in, one_byte_bytemask); 51| | 52| | // 3. prepare bitmask for 8-bit lookup 53| 445k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 54| 445k| const uint32_t M1 = M0 >> 7; 55| 445k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 56| | // 4. pack the bytes 57| | 58| 445k| const uint8_t *row = 59| 445k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 60| 445k| const uint8_t *row_2 = 61| 445k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 16)] 62| 445k| [0]; 63| | 64| 445k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 65| 445k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 66| | 67| 445k| const __m256i utf8_packed = _mm256_shuffle_epi8( 68| 445k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 445k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 445k| _mm256_castsi128_si256(xmm2), 2) ------------------ 69| | // 5. store bytes 70| 445k| _mm_storeu_si128((__m128i *)utf8_output, 71| 445k| _mm256_castsi256_si128(utf8_packed)); 72| 445k| utf8_output += row[0]; 73| 445k| _mm_storeu_si128((__m128i *)utf8_output, 74| 445k| _mm256_extractf128_si256(utf8_packed, 1)); 75| 445k| utf8_output += row_2[0]; 76| | 77| | // 6. adjust pointers 78| 445k| latin1_input += 16; 79| 445k| continue; 80| | 81| 3.35M| } // while 82| 604| return std::make_pair(latin1_input, utf8_output); 83| 604|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_utf16_to_latin1ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 124| char *latin1_output) { 5| 124| const char16_t *end = buf + len; 6| 366| while (end - buf >= 32) { ------------------ | Branch (6:10): [True: 264, False: 102] ------------------ 7| | // Load 16 UTF-16 characters into 256-bit AVX2 register 8| 264| __m256i in0 = _mm256_loadu_si256(reinterpret_cast(buf)); 9| 264| __m256i in1 = 10| 264| _mm256_loadu_si256(reinterpret_cast(buf + 16)); 11| | 12| | if constexpr (!match_system(big_endian)) { 13| | const __m256i swap = _mm256_setr_epi8( 14| | 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 15| | 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 16| | in0 = _mm256_shuffle_epi8(in0, swap); 17| | in1 = _mm256_shuffle_epi8(in1, swap); 18| | } 19| | 20| 264| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 21| 264| if (_mm256_testz_si256(_mm256_or_si256(in0, in1), high_byte_mask)) { ------------------ | Branch (21:9): [True: 242, False: 22] ------------------ 22| | // Pack 16-bit characters into 8-bit and store in latin1_output 23| 242| const __m256i packed = _mm256_packus_epi16(in0, in1); 24| | 25| 242| const __m256i result = _mm256_permute4x64_epi64(packed, 0b11011000); 26| | 27| 242| _mm256_storeu_si256(reinterpret_cast<__m256i *>(latin1_output), result); 28| | // Adjust pointers for the next iteration 29| 242| buf += 32; 30| 242| latin1_output += 32; 31| 242| } else { 32| 22| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 33| 22| } 34| 264| } // while 35| 102| return std::make_pair(buf, latin1_output); 36| 124|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_utf16_to_latin1ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 114| char *latin1_output) { 5| 114| const char16_t *end = buf + len; 6| 858| while (end - buf >= 32) { ------------------ | Branch (6:10): [True: 768, False: 90] ------------------ 7| | // Load 16 UTF-16 characters into 256-bit AVX2 register 8| 768| __m256i in0 = _mm256_loadu_si256(reinterpret_cast(buf)); 9| 768| __m256i in1 = 10| 768| _mm256_loadu_si256(reinterpret_cast(buf + 16)); 11| | 12| 768| if constexpr (!match_system(big_endian)) { 13| 768| const __m256i swap = _mm256_setr_epi8( 14| 768| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 15| 768| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 16| 768| in0 = _mm256_shuffle_epi8(in0, swap); 17| 768| in1 = _mm256_shuffle_epi8(in1, swap); 18| 768| } 19| | 20| 768| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 21| 768| if (_mm256_testz_si256(_mm256_or_si256(in0, in1), high_byte_mask)) { ------------------ | Branch (21:9): [True: 744, False: 24] ------------------ 22| | // Pack 16-bit characters into 8-bit and store in latin1_output 23| 744| const __m256i packed = _mm256_packus_epi16(in0, in1); 24| | 25| 744| const __m256i result = _mm256_permute4x64_epi64(packed, 0b11011000); 26| | 27| 744| _mm256_storeu_si256(reinterpret_cast<__m256i *>(latin1_output), result); 28| | // Adjust pointers for the next iteration 29| 744| buf += 32; 30| 744| latin1_output += 32; 31| 744| } else { 32| 24| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 33| 24| } 34| 768| } // while 35| 90| return std::make_pair(buf, latin1_output); 36| 114|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_140avx2_convert_utf16_to_latin1_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 41| 278| char *latin1_output) { 42| 278| const char16_t *start = buf; 43| 278| const char16_t *end = buf + len; 44| 410k| while (end - buf >= 16) { ------------------ | Branch (44:10): [True: 410k, False: 162] ------------------ 45| 410k| __m256i in = _mm256_loadu_si256(reinterpret_cast(buf)); 46| | 47| | if constexpr (!match_system(big_endian)) { 48| | const __m256i swap = _mm256_setr_epi8( 49| | 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 50| | 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 51| | in = _mm256_shuffle_epi8(in, swap); 52| | } 53| | 54| 410k| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 55| 410k| if (_mm256_testz_si256(in, high_byte_mask)) { ------------------ | Branch (55:9): [True: 410k, False: 116] ------------------ 56| 410k| __m128i lo = _mm256_extractf128_si256(in, 0); 57| 410k| __m128i hi = _mm256_extractf128_si256(in, 1); 58| 410k| __m128i latin1_packed_lo = _mm_packus_epi16(lo, lo); 59| 410k| __m128i latin1_packed_hi = _mm_packus_epi16(hi, hi); 60| 410k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 61| 410k| latin1_packed_lo); 62| 410k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output + 8), 63| 410k| latin1_packed_hi); 64| 410k| buf += 16; 65| 410k| latin1_output += 16; 66| 410k| } else { 67| | // Fallback to scalar code for handling errors 68| 528| for (int k = 0; k < 16; k++) { ------------------ | Branch (68:23): [True: 528, False: 0] ------------------ 69| 528| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 70| 528| if (word <= 0xff) { ------------------ | Branch (70:13): [True: 412, False: 116] ------------------ 71| 412| *latin1_output++ = char(word); 72| 412| } else { 73| 116| return std::make_pair( 74| 116| result{error_code::TOO_LARGE, (size_t)(buf - start + k)}, 75| 116| latin1_output); 76| 116| } 77| 528| } 78| 0| buf += 16; 79| 0| } 80| 410k| } // while 81| 162| return std::make_pair(result{error_code::SUCCESS, (size_t)(buf - start)}, 82| 162| latin1_output); 83| 278|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_140avx2_convert_utf16_to_latin1_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 41| 266| char *latin1_output) { 42| 266| const char16_t *start = buf; 43| 266| const char16_t *end = buf + len; 44| 522k| while (end - buf >= 16) { ------------------ | Branch (44:10): [True: 522k, False: 152] ------------------ 45| 522k| __m256i in = _mm256_loadu_si256(reinterpret_cast(buf)); 46| | 47| 522k| if constexpr (!match_system(big_endian)) { 48| 522k| const __m256i swap = _mm256_setr_epi8( 49| 522k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 50| 522k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 51| 522k| in = _mm256_shuffle_epi8(in, swap); 52| 522k| } 53| | 54| 522k| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 55| 522k| if (_mm256_testz_si256(in, high_byte_mask)) { ------------------ | Branch (55:9): [True: 522k, False: 114] ------------------ 56| 522k| __m128i lo = _mm256_extractf128_si256(in, 0); 57| 522k| __m128i hi = _mm256_extractf128_si256(in, 1); 58| 522k| __m128i latin1_packed_lo = _mm_packus_epi16(lo, lo); 59| 522k| __m128i latin1_packed_hi = _mm_packus_epi16(hi, hi); 60| 522k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 61| 522k| latin1_packed_lo); 62| 522k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output + 8), 63| 522k| latin1_packed_hi); 64| 522k| buf += 16; 65| 522k| latin1_output += 16; 66| 522k| } else { 67| | // Fallback to scalar code for handling errors 68| 494| for (int k = 0; k < 16; k++) { ------------------ | Branch (68:23): [True: 494, False: 0] ------------------ 69| 494| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 70| 494| if (word <= 0xff) { ------------------ | Branch (70:13): [True: 380, False: 114] ------------------ 71| 380| *latin1_output++ = char(word); 72| 380| } else { 73| 114| return std::make_pair( 74| 114| result{error_code::TOO_LARGE, (size_t)(buf - start + k)}, 75| 114| latin1_output); 76| 114| } 77| 494| } 78| 0| buf += 16; 79| 0| } 80| 522k| } // while 81| 152| return std::make_pair(result{error_code::SUCCESS, (size_t)(buf - start)}, 82| 152| latin1_output); 83| 266|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf16_to_utf32ILNS_10endiannessE0EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 454| char32_t *utf32_output) { 58| 454| const char16_t *end = buf + len; 59| 454| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 454| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| | 62| 274k| while (end - buf >= 16) { ------------------ | Branch (62:10): [True: 273k, False: 336] ------------------ 63| 273k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 64| 273k| if (big_endian) { ------------------ | Branch (64:9): [Folded, False: 273k] ------------------ 65| 0| const __m256i swap = _mm256_setr_epi8( 66| 0| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 67| 0| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 68| 0| in = _mm256_shuffle_epi8(in, swap); 69| 0| } 70| | 71| | // 1. Check if there are any surrogate word in the input chunk. 72| | // We have also deal with situation when there is a surrogate word 73| | // at the end of a chunk. 74| 273k| const __m256i surrogates_bytemask = 75| 273k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_d800); 76| | 77| | // bitmask = 0x0000 if there are no surrogates 78| | // = 0xc000 if the last word is a surrogate 79| 273k| const uint32_t surrogates_bitmask = 80| 273k| static_cast(_mm256_movemask_epi8(surrogates_bytemask)); 81| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 82| | // However, it is likely an uncommon occurrence. 83| 273k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (83:9): [True: 268k, False: 5.60k] ------------------ 84| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 85| | // units 86| 268k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 87| 268k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 88| 268k| _mm256_storeu_si256( 89| 268k| reinterpret_cast<__m256i *>(utf32_output + 8), 90| 268k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 91| 268k| utf32_output += 16; 92| 268k| buf += 16; 93| | // surrogate pair(s) in a register 94| 268k| } else { 95| | // Let us do a scalar fallback. 96| | // It may seem wasteful to use scalar code, but being efficient with SIMD 97| | // in the presence of surrogate pairs may require non-trivial tables. 98| 5.60k| size_t forward = 15; 99| 5.60k| size_t k = 0; 100| 5.60k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (100:11): [True: 0, False: 5.60k] ------------------ 101| 0| forward = size_t(end - buf - 1); 102| 0| } 103| 82.2k| for (; k < forward; k++) { ------------------ | Branch (103:14): [True: 76.7k, False: 5.48k] ------------------ 104| 76.7k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 105| 76.7k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (105:13): [True: 70.1k, False: 6.66k] ------------------ 106| | // No surrogate pair 107| 70.1k| *utf32_output++ = char32_t(word); 108| 70.1k| } else { 109| | // must be a surrogate pair 110| 6.66k| uint16_t diff = uint16_t(word - 0xD800); 111| 6.66k| uint16_t next_word = 112| 6.66k| scalar::utf16::swap_if_needed(buf[k + 1]); 113| 6.66k| k++; 114| 6.66k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 115| 6.66k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (115:15): [True: 118, False: 6.54k] ------------------ 116| 118| return std::make_pair(nullptr, utf32_output); 117| 118| } 118| 6.54k| uint32_t value = (diff << 10) + diff2 + 0x10000; 119| 6.54k| *utf32_output++ = char32_t(value); 120| 6.54k| } 121| 76.7k| } 122| 5.48k| buf += k; 123| 5.48k| } 124| 273k| } // while 125| 336| return std::make_pair(buf, utf32_output); 126| 454|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf16_to_utf32ILNS_10endiannessE1EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 422| char32_t *utf32_output) { 58| 422| const char16_t *end = buf + len; 59| 422| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 422| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| | 62| 369k| while (end - buf >= 16) { ------------------ | Branch (62:10): [True: 369k, False: 340] ------------------ 63| 369k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 64| 369k| if (big_endian) { ------------------ | Branch (64:9): [True: 369k, Folded] ------------------ 65| 369k| const __m256i swap = _mm256_setr_epi8( 66| 369k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 67| 369k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 68| 369k| in = _mm256_shuffle_epi8(in, swap); 69| 369k| } 70| | 71| | // 1. Check if there are any surrogate word in the input chunk. 72| | // We have also deal with situation when there is a surrogate word 73| | // at the end of a chunk. 74| 369k| const __m256i surrogates_bytemask = 75| 369k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_d800); 76| | 77| | // bitmask = 0x0000 if there are no surrogates 78| | // = 0xc000 if the last word is a surrogate 79| 369k| const uint32_t surrogates_bitmask = 80| 369k| static_cast(_mm256_movemask_epi8(surrogates_bytemask)); 81| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 82| | // However, it is likely an uncommon occurrence. 83| 369k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (83:9): [True: 367k, False: 1.46k] ------------------ 84| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 85| | // units 86| 367k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 87| 367k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 88| 367k| _mm256_storeu_si256( 89| 367k| reinterpret_cast<__m256i *>(utf32_output + 8), 90| 367k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 91| 367k| utf32_output += 16; 92| 367k| buf += 16; 93| | // surrogate pair(s) in a register 94| 367k| } else { 95| | // Let us do a scalar fallback. 96| | // It may seem wasteful to use scalar code, but being efficient with SIMD 97| | // in the presence of surrogate pairs may require non-trivial tables. 98| 1.46k| size_t forward = 15; 99| 1.46k| size_t k = 0; 100| 1.46k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (100:11): [True: 0, False: 1.46k] ------------------ 101| 0| forward = size_t(end - buf - 1); 102| 0| } 103| 20.6k| for (; k < forward; k++) { ------------------ | Branch (103:14): [True: 19.2k, False: 1.38k] ------------------ 104| 19.2k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 105| 19.2k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (105:13): [True: 16.9k, False: 2.24k] ------------------ 106| | // No surrogate pair 107| 16.9k| *utf32_output++ = char32_t(word); 108| 16.9k| } else { 109| | // must be a surrogate pair 110| 2.24k| uint16_t diff = uint16_t(word - 0xD800); 111| 2.24k| uint16_t next_word = 112| 2.24k| scalar::utf16::swap_if_needed(buf[k + 1]); 113| 2.24k| k++; 114| 2.24k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 115| 2.24k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (115:15): [True: 82, False: 2.16k] ------------------ 116| 82| return std::make_pair(nullptr, utf32_output); 117| 82| } 118| 2.16k| uint32_t value = (diff << 10) + diff2 + 0x10000; 119| 2.16k| *utf32_output++ = char32_t(value); 120| 2.16k| } 121| 19.2k| } 122| 1.38k| buf += k; 123| 1.38k| } 124| 369k| } // while 125| 340| return std::make_pair(buf, utf32_output); 126| 422|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf16_to_utf32_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 138| 334| char32_t *utf32_output) { 139| 334| const char16_t *start = buf; 140| 334| const char16_t *end = buf + len; 141| 334| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 142| 334| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 143| | 144| 160k| while (end - buf >= 16) { ------------------ | Branch (144:10): [True: 160k, False: 226] ------------------ 145| 160k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 146| 160k| if (big_endian) { ------------------ | Branch (146:9): [Folded, False: 160k] ------------------ 147| 0| const __m256i swap = _mm256_setr_epi8( 148| 0| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 149| 0| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 150| 0| in = _mm256_shuffle_epi8(in, swap); 151| 0| } 152| | 153| | // 1. Check if there are any surrogate word in the input chunk. 154| | // We have also deal with situation when there is a surrogate word 155| | // at the end of a chunk. 156| 160k| const __m256i surrogates_bytemask = 157| 160k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_d800); 158| | 159| | // bitmask = 0x0000 if there are no surrogates 160| | // = 0xc000 if the last word is a surrogate 161| 160k| const uint32_t surrogates_bitmask = 162| 160k| static_cast(_mm256_movemask_epi8(surrogates_bytemask)); 163| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 164| | // However, it is likely an uncommon occurrence. 165| 160k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (165:9): [True: 154k, False: 5.22k] ------------------ 166| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 167| | // units 168| 154k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 169| 154k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 170| 154k| _mm256_storeu_si256( 171| 154k| reinterpret_cast<__m256i *>(utf32_output + 8), 172| 154k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 173| 154k| utf32_output += 16; 174| 154k| buf += 16; 175| | // surrogate pair(s) in a register 176| 154k| } else { 177| | // Let us do a scalar fallback. 178| | // It may seem wasteful to use scalar code, but being efficient with SIMD 179| | // in the presence of surrogate pairs may require non-trivial tables. 180| 5.22k| size_t forward = 15; 181| 5.22k| size_t k = 0; 182| 5.22k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (182:11): [True: 0, False: 5.22k] ------------------ 183| 0| forward = size_t(end - buf - 1); 184| 0| } 185| 73.9k| for (; k < forward; k++) { ------------------ | Branch (185:14): [True: 68.8k, False: 5.11k] ------------------ 186| 68.8k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 187| 68.8k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (187:13): [True: 59.3k, False: 9.44k] ------------------ 188| | // No surrogate pair 189| 59.3k| *utf32_output++ = char32_t(word); 190| 59.3k| } else { 191| | // must be a surrogate pair 192| 9.44k| uint16_t diff = uint16_t(word - 0xD800); 193| 9.44k| uint16_t next_word = 194| 9.44k| scalar::utf16::swap_if_needed(buf[k + 1]); 195| 9.44k| k++; 196| 9.44k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 197| 9.44k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (197:15): [True: 108, False: 9.33k] ------------------ 198| 108| return std::make_pair( 199| 108| result(error_code::SURROGATE, buf - start + k - 1), 200| 108| utf32_output); 201| 108| } 202| 9.33k| uint32_t value = (diff << 10) + diff2 + 0x10000; 203| 9.33k| *utf32_output++ = char32_t(value); 204| 9.33k| } 205| 68.8k| } 206| 5.11k| buf += k; 207| 5.11k| } 208| 160k| } // while 209| 226| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 210| 334|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf16_to_utf32_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 138| 346| char32_t *utf32_output) { 139| 346| const char16_t *start = buf; 140| 346| const char16_t *end = buf + len; 141| 346| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 142| 346| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 143| | 144| 151k| while (end - buf >= 16) { ------------------ | Branch (144:10): [True: 151k, False: 254] ------------------ 145| 151k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 146| 151k| if (big_endian) { ------------------ | Branch (146:9): [True: 151k, Folded] ------------------ 147| 151k| const __m256i swap = _mm256_setr_epi8( 148| 151k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 149| 151k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 150| 151k| in = _mm256_shuffle_epi8(in, swap); 151| 151k| } 152| | 153| | // 1. Check if there are any surrogate word in the input chunk. 154| | // We have also deal with situation when there is a surrogate word 155| | // at the end of a chunk. 156| 151k| const __m256i surrogates_bytemask = 157| 151k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_d800); 158| | 159| | // bitmask = 0x0000 if there are no surrogates 160| | // = 0xc000 if the last word is a surrogate 161| 151k| const uint32_t surrogates_bitmask = 162| 151k| static_cast(_mm256_movemask_epi8(surrogates_bytemask)); 163| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 164| | // However, it is likely an uncommon occurrence. 165| 151k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (165:9): [True: 146k, False: 5.28k] ------------------ 166| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 167| | // units 168| 146k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 169| 146k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 170| 146k| _mm256_storeu_si256( 171| 146k| reinterpret_cast<__m256i *>(utf32_output + 8), 172| 146k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 173| 146k| utf32_output += 16; 174| 146k| buf += 16; 175| | // surrogate pair(s) in a register 176| 146k| } else { 177| | // Let us do a scalar fallback. 178| | // It may seem wasteful to use scalar code, but being efficient with SIMD 179| | // in the presence of surrogate pairs may require non-trivial tables. 180| 5.28k| size_t forward = 15; 181| 5.28k| size_t k = 0; 182| 5.28k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (182:11): [True: 0, False: 5.28k] ------------------ 183| 0| forward = size_t(end - buf - 1); 184| 0| } 185| 74.7k| for (; k < forward; k++) { ------------------ | Branch (185:14): [True: 69.5k, False: 5.18k] ------------------ 186| 69.5k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 187| 69.5k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (187:13): [True: 59.7k, False: 9.78k] ------------------ 188| | // No surrogate pair 189| 59.7k| *utf32_output++ = char32_t(word); 190| 59.7k| } else { 191| | // must be a surrogate pair 192| 9.78k| uint16_t diff = uint16_t(word - 0xD800); 193| 9.78k| uint16_t next_word = 194| 9.78k| scalar::utf16::swap_if_needed(buf[k + 1]); 195| 9.78k| k++; 196| 9.78k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 197| 9.78k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (197:15): [True: 92, False: 9.69k] ------------------ 198| 92| return std::make_pair( 199| 92| result(error_code::SURROGATE, buf - start + k - 1), 200| 92| utf32_output); 201| 92| } 202| 9.69k| uint32_t value = (diff << 10) + diff2 + 0x10000; 203| 9.69k| *utf32_output++ = char32_t(value); 204| 9.69k| } 205| 69.5k| } 206| 5.18k| buf += k; 207| 5.18k| } 208| 151k| } // while 209| 254| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 210| 346|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_126avx2_convert_utf16_to_utf8ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 842|avx2_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| 842| const char16_t *end = buf + len; 58| 842| const __m256i v_0000 = _mm256_setzero_si256(); 59| 842| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 842| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| 842| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 62| 842| const size_t safety_margin = 63| 842| 12; // to avoid overruns, see issue 64| | // https://github.com/simdutf/simdutf/issues/92 65| | 66| 923k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (66:10): [True: 922k, False: 724] ------------------ 67| 922k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 68| 922k| if (big_endian) { ------------------ | Branch (68:9): [Folded, False: 922k] ------------------ 69| 0| const __m256i swap = _mm256_setr_epi8( 70| 0| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 71| 0| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 72| 0| in = _mm256_shuffle_epi8(in, swap); 73| 0| } 74| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 75| 922k| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 76| 922k| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 713k, False: 209k] ------------------ 77| | // 1. pack the bytes 78| 713k| const __m128i utf8_packed = _mm_packus_epi16( 79| 713k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 80| | // 2. store (16 bytes) 81| 713k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 82| | // 3. adjust pointers 83| 713k| buf += 16; 84| 713k| utf8_output += 16; 85| 713k| continue; // we are done for this round! 86| 713k| } 87| | // no bits set above 7th bit 88| 209k| const __m256i one_byte_bytemask = 89| 209k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 90| 209k| const uint32_t one_byte_bitmask = 91| 209k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 92| | 93| | // no bits set above 11th bit 94| 209k| const __m256i one_or_two_bytes_bytemask = 95| 209k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 96| 209k| const uint32_t one_or_two_bytes_bitmask = 97| 209k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 98| 209k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (98:9): [True: 126k, False: 83.2k] ------------------ 99| | 100| | // 1. prepare 2-byte values 101| | // input 16-bit word : [0000|0aaa|aabb|bbbb] x 8 102| | // expected output : [110a|aaaa|10bb|bbbb] x 8 103| 126k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 104| 126k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 105| | 106| | // t0 = [000a|aaaa|bbbb|bb00] 107| 126k| const __m256i t0 = _mm256_slli_epi16(in, 2); 108| | // t1 = [000a|aaaa|0000|0000] 109| 126k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 110| | // t2 = [0000|0000|00bb|bbbb] 111| 126k| const __m256i t2 = _mm256_and_si256(in, v_003f); 112| | // t3 = [000a|aaaa|00bb|bbbb] 113| 126k| const __m256i t3 = _mm256_or_si256(t1, t2); 114| | // t4 = [110a|aaaa|10bb|bbbb] 115| 126k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 116| | 117| | // 2. merge ASCII and 2-byte codewords 118| 126k| const __m256i utf8_unpacked = 119| 126k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 120| | 121| | // 3. prepare bitmask for 8-bit lookup 122| 126k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 123| 126k| const uint32_t M1 = M0 >> 7; 124| 126k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 125| | // 4. pack the bytes 126| | 127| 126k| const uint8_t *row = 128| 126k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 129| 126k| const uint8_t *row_2 = 130| 126k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 131| 126k| 16)][0]; 132| | 133| 126k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 134| 126k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 135| | 136| 126k| const __m256i utf8_packed = _mm256_shuffle_epi8( 137| 126k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 126k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 126k| _mm256_castsi128_si256(xmm2), 2) ------------------ 138| | // 5. store bytes 139| 126k| _mm_storeu_si128((__m128i *)utf8_output, 140| 126k| _mm256_castsi256_si128(utf8_packed)); 141| 126k| utf8_output += row[0]; 142| 126k| _mm_storeu_si128((__m128i *)utf8_output, 143| 126k| _mm256_extractf128_si256(utf8_packed, 1)); 144| 126k| utf8_output += row_2[0]; 145| | 146| | // 6. adjust pointers 147| 126k| buf += 16; 148| 126k| continue; 149| 126k| } 150| | // 1. Check if there are any surrogate word in the input chunk. 151| | // We have also deal with situation when there is a surrogate word 152| | // at the end of a chunk. 153| 83.2k| const __m256i surrogates_bytemask = 154| 83.2k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_d800); 155| | 156| | // bitmask = 0x0000 if there are no surrogates 157| | // = 0xc000 if the last word is a surrogate 158| 83.2k| const uint32_t surrogates_bitmask = 159| 83.2k| static_cast(_mm256_movemask_epi8(surrogates_bytemask)); 160| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 161| | // However, it is likely an uncommon occurrence. 162| 83.2k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (162:9): [True: 76.0k, False: 7.18k] ------------------ 163| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 164| 76.0k| const __m256i dup_even = _mm256_setr_epi16( 165| 76.0k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 166| 76.0k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 167| | 168| | /* In this branch we handle three cases: 169| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 170| | single UFT-8 byte 171| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - two 172| | UTF-8 bytes 173| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 174| | three UTF-8 bytes 175| | 176| | We expand the input word (16-bit) into two code units (32-bit), thus 177| | we have room for four bytes. However, we need five distinct bit 178| | layouts. Note that the last byte in cases #2 and #3 is the same. 179| | 180| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 181| | in register t2. 182| | 183| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 184| | either byte 1 for case #2 or byte 2 for case #3. Note that they 185| | differ by exactly one bit. 186| | 187| | Finally from these two code units we build proper UTF-8 sequence, taking 188| | into account the case (i.e, the number of bytes to write). 189| | */ 190| | /** 191| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 192| | * t2 => [0ccc|cccc] [10cc|cccc] 193| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 194| | */ 195| 76.0k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 196| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 197| 76.0k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 198| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 199| 76.0k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 195| 76.0k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 200| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 201| 76.0k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 195| 76.0k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 202| | 203| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 204| 76.0k| const __m256i s0 = _mm256_srli_epi16(in, 4); 205| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 206| 76.0k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 195| 76.0k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 207| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 208| 76.0k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 195| 76.0k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 209| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 210| 76.0k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 195| 76.0k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 211| 76.0k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 212| 76.0k| simdutf_vec(0b0100000000000000)); ------------------ | | 195| 76.0k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 213| 76.0k| const __m256i s4 = _mm256_xor_si256(s3, m0); 214| 76.0k|#undef simdutf_vec 215| | 216| | // 4. expand code units 16-bit => 32-bit 217| 76.0k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 218| 76.0k| const __m256i out1 = _mm256_unpackhi_epi16(t2, s4); 219| | 220| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 221| 76.0k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 222| 76.0k| (one_or_two_bytes_bitmask & 0xaaaaaaaa); 223| | // Due to the wider registers, the following path is less likely to be 224| | // useful. 225| | /*if(mask == 0) { 226| | // We only have three-byte code units. Use fast path. 227| | const __m256i shuffle = 228| | _mm256_setr_epi8(2,3,1,6,7,5,10,11,9,14,15,13,-1,-1,-1,-1, 229| | 2,3,1,6,7,5,10,11,9,14,15,13,-1,-1,-1,-1); const __m256i utf8_0 = 230| | _mm256_shuffle_epi8(out0, shuffle); const __m256i utf8_1 = 231| | _mm256_shuffle_epi8(out1, shuffle); 232| | _mm_storeu_si128((__m128i*)utf8_output, _mm256_castsi256_si128(utf8_0)); 233| | utf8_output += 12; 234| | _mm_storeu_si128((__m128i*)utf8_output, _mm256_castsi256_si128(utf8_1)); 235| | utf8_output += 12; 236| | _mm_storeu_si128((__m128i*)utf8_output, 237| | _mm256_extractf128_si256(utf8_0,1)); utf8_output += 12; 238| | _mm_storeu_si128((__m128i*)utf8_output, 239| | _mm256_extractf128_si256(utf8_1,1)); utf8_output += 12; buf += 16; 240| | continue; 241| | }*/ 242| 76.0k| const uint8_t mask0 = uint8_t(mask); 243| 76.0k| const uint8_t *row0 = 244| 76.0k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 245| 76.0k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 246| 76.0k| const __m128i utf8_0 = 247| 76.0k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 248| | 249| 76.0k| const uint8_t mask1 = static_cast(mask >> 8); 250| 76.0k| const uint8_t *row1 = 251| 76.0k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 252| 76.0k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 253| 76.0k| const __m128i utf8_1 = 254| 76.0k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 255| | 256| 76.0k| const uint8_t mask2 = static_cast(mask >> 16); 257| 76.0k| const uint8_t *row2 = 258| 76.0k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 259| 76.0k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 260| 76.0k| const __m128i utf8_2 = 261| 76.0k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 262| | 263| 76.0k| const uint8_t mask3 = static_cast(mask >> 24); 264| 76.0k| const uint8_t *row3 = 265| 76.0k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 266| 76.0k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 267| 76.0k| const __m128i utf8_3 = 268| 76.0k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 269| | 270| 76.0k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 271| 76.0k| utf8_output += row0[0]; 272| 76.0k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 273| 76.0k| utf8_output += row1[0]; 274| 76.0k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 275| 76.0k| utf8_output += row2[0]; 276| 76.0k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 277| 76.0k| utf8_output += row3[0]; 278| 76.0k| buf += 16; 279| | // surrogate pair(s) in a register 280| 76.0k| } else { 281| | // Let us do a scalar fallback. 282| | // It may seem wasteful to use scalar code, but being efficient with SIMD 283| | // in the presence of surrogate pairs may require non-trivial tables. 284| 7.18k| size_t forward = 15; 285| 7.18k| size_t k = 0; 286| 7.18k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (286:11): [True: 0, False: 7.18k] ------------------ 287| 0| forward = size_t(end - buf - 1); 288| 0| } 289| 106k| for (; k < forward; k++) { ------------------ | Branch (289:14): [True: 99.1k, False: 7.06k] ------------------ 290| 99.1k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 291| 99.1k| if ((word & 0xFF80) == 0) { ------------------ | Branch (291:13): [True: 14.5k, False: 84.5k] ------------------ 292| 14.5k| *utf8_output++ = char(word); 293| 84.5k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (293:20): [True: 4.35k, False: 80.2k] ------------------ 294| 4.35k| *utf8_output++ = char((word >> 6) | 0b11000000); 295| 4.35k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 296| 80.2k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (296:20): [True: 72.2k, False: 7.94k] ------------------ 297| 72.2k| *utf8_output++ = char((word >> 12) | 0b11100000); 298| 72.2k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 299| 72.2k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 300| 72.2k| } else { 301| | // must be a surrogate pair 302| 7.94k| uint16_t diff = uint16_t(word - 0xD800); 303| 7.94k| uint16_t next_word = 304| 7.94k| scalar::utf16::swap_if_needed(buf[k + 1]); 305| 7.94k| k++; 306| 7.94k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 307| 7.94k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (307:15): [True: 118, False: 7.82k] ------------------ 308| 118| return std::make_pair(nullptr, utf8_output); 309| 118| } 310| 7.82k| uint32_t value = (diff << 10) + diff2 + 0x10000; 311| 7.82k| *utf8_output++ = char((value >> 18) | 0b11110000); 312| 7.82k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 313| 7.82k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 314| 7.82k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 315| 7.82k| } 316| 99.1k| } 317| 7.06k| buf += k; 318| 7.06k| } 319| 83.2k| } // while 320| 724| return std::make_pair(buf, utf8_output); 321| 842|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_126avx2_convert_utf16_to_utf8ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 798|avx2_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| 798| const char16_t *end = buf + len; 58| 798| const __m256i v_0000 = _mm256_setzero_si256(); 59| 798| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 798| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| 798| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 62| 798| const size_t safety_margin = 63| 798| 12; // to avoid overruns, see issue 64| | // https://github.com/simdutf/simdutf/issues/92 65| | 66| 957k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (66:10): [True: 956k, False: 692] ------------------ 67| 956k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 68| 956k| if (big_endian) { ------------------ | Branch (68:9): [True: 956k, Folded] ------------------ 69| 956k| const __m256i swap = _mm256_setr_epi8( 70| 956k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 71| 956k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 72| 956k| in = _mm256_shuffle_epi8(in, swap); 73| 956k| } 74| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 75| 956k| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 76| 956k| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 702k, False: 253k] ------------------ 77| | // 1. pack the bytes 78| 702k| const __m128i utf8_packed = _mm_packus_epi16( 79| 702k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 80| | // 2. store (16 bytes) 81| 702k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 82| | // 3. adjust pointers 83| 702k| buf += 16; 84| 702k| utf8_output += 16; 85| 702k| continue; // we are done for this round! 86| 702k| } 87| | // no bits set above 7th bit 88| 253k| const __m256i one_byte_bytemask = 89| 253k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 90| 253k| const uint32_t one_byte_bitmask = 91| 253k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 92| | 93| | // no bits set above 11th bit 94| 253k| const __m256i one_or_two_bytes_bytemask = 95| 253k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 96| 253k| const uint32_t one_or_two_bytes_bitmask = 97| 253k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 98| 253k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (98:9): [True: 13.2k, False: 240k] ------------------ 99| | 100| | // 1. prepare 2-byte values 101| | // input 16-bit word : [0000|0aaa|aabb|bbbb] x 8 102| | // expected output : [110a|aaaa|10bb|bbbb] x 8 103| 13.2k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 104| 13.2k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 105| | 106| | // t0 = [000a|aaaa|bbbb|bb00] 107| 13.2k| const __m256i t0 = _mm256_slli_epi16(in, 2); 108| | // t1 = [000a|aaaa|0000|0000] 109| 13.2k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 110| | // t2 = [0000|0000|00bb|bbbb] 111| 13.2k| const __m256i t2 = _mm256_and_si256(in, v_003f); 112| | // t3 = [000a|aaaa|00bb|bbbb] 113| 13.2k| const __m256i t3 = _mm256_or_si256(t1, t2); 114| | // t4 = [110a|aaaa|10bb|bbbb] 115| 13.2k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 116| | 117| | // 2. merge ASCII and 2-byte codewords 118| 13.2k| const __m256i utf8_unpacked = 119| 13.2k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 120| | 121| | // 3. prepare bitmask for 8-bit lookup 122| 13.2k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 123| 13.2k| const uint32_t M1 = M0 >> 7; 124| 13.2k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 125| | // 4. pack the bytes 126| | 127| 13.2k| const uint8_t *row = 128| 13.2k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 129| 13.2k| const uint8_t *row_2 = 130| 13.2k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 131| 13.2k| 16)][0]; 132| | 133| 13.2k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 134| 13.2k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 135| | 136| 13.2k| const __m256i utf8_packed = _mm256_shuffle_epi8( 137| 13.2k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 13.2k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 13.2k| _mm256_castsi128_si256(xmm2), 2) ------------------ 138| | // 5. store bytes 139| 13.2k| _mm_storeu_si128((__m128i *)utf8_output, 140| 13.2k| _mm256_castsi256_si128(utf8_packed)); 141| 13.2k| utf8_output += row[0]; 142| 13.2k| _mm_storeu_si128((__m128i *)utf8_output, 143| 13.2k| _mm256_extractf128_si256(utf8_packed, 1)); 144| 13.2k| utf8_output += row_2[0]; 145| | 146| | // 6. adjust pointers 147| 13.2k| buf += 16; 148| 13.2k| continue; 149| 13.2k| } 150| | // 1. Check if there are any surrogate word in the input chunk. 151| | // We have also deal with situation when there is a surrogate word 152| | // at the end of a chunk. 153| 240k| const __m256i surrogates_bytemask = 154| 240k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_d800); 155| | 156| | // bitmask = 0x0000 if there are no surrogates 157| | // = 0xc000 if the last word is a surrogate 158| 240k| const uint32_t surrogates_bitmask = 159| 240k| static_cast(_mm256_movemask_epi8(surrogates_bytemask)); 160| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 161| | // However, it is likely an uncommon occurrence. 162| 240k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (162:9): [True: 236k, False: 4.17k] ------------------ 163| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 164| 236k| const __m256i dup_even = _mm256_setr_epi16( 165| 236k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 166| 236k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 167| | 168| | /* In this branch we handle three cases: 169| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 170| | single UFT-8 byte 171| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - two 172| | UTF-8 bytes 173| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 174| | three UTF-8 bytes 175| | 176| | We expand the input word (16-bit) into two code units (32-bit), thus 177| | we have room for four bytes. However, we need five distinct bit 178| | layouts. Note that the last byte in cases #2 and #3 is the same. 179| | 180| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 181| | in register t2. 182| | 183| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 184| | either byte 1 for case #2 or byte 2 for case #3. Note that they 185| | differ by exactly one bit. 186| | 187| | Finally from these two code units we build proper UTF-8 sequence, taking 188| | into account the case (i.e, the number of bytes to write). 189| | */ 190| | /** 191| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 192| | * t2 => [0ccc|cccc] [10cc|cccc] 193| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 194| | */ 195| 236k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 196| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 197| 236k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 198| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 199| 236k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 195| 236k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 200| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 201| 236k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 195| 236k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 202| | 203| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 204| 236k| const __m256i s0 = _mm256_srli_epi16(in, 4); 205| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 206| 236k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 195| 236k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 207| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 208| 236k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 195| 236k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 209| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 210| 236k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 195| 236k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 211| 236k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 212| 236k| simdutf_vec(0b0100000000000000)); ------------------ | | 195| 236k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 213| 236k| const __m256i s4 = _mm256_xor_si256(s3, m0); 214| 236k|#undef simdutf_vec 215| | 216| | // 4. expand code units 16-bit => 32-bit 217| 236k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 218| 236k| const __m256i out1 = _mm256_unpackhi_epi16(t2, s4); 219| | 220| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 221| 236k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 222| 236k| (one_or_two_bytes_bitmask & 0xaaaaaaaa); 223| | // Due to the wider registers, the following path is less likely to be 224| | // useful. 225| | /*if(mask == 0) { 226| | // We only have three-byte code units. Use fast path. 227| | const __m256i shuffle = 228| | _mm256_setr_epi8(2,3,1,6,7,5,10,11,9,14,15,13,-1,-1,-1,-1, 229| | 2,3,1,6,7,5,10,11,9,14,15,13,-1,-1,-1,-1); const __m256i utf8_0 = 230| | _mm256_shuffle_epi8(out0, shuffle); const __m256i utf8_1 = 231| | _mm256_shuffle_epi8(out1, shuffle); 232| | _mm_storeu_si128((__m128i*)utf8_output, _mm256_castsi256_si128(utf8_0)); 233| | utf8_output += 12; 234| | _mm_storeu_si128((__m128i*)utf8_output, _mm256_castsi256_si128(utf8_1)); 235| | utf8_output += 12; 236| | _mm_storeu_si128((__m128i*)utf8_output, 237| | _mm256_extractf128_si256(utf8_0,1)); utf8_output += 12; 238| | _mm_storeu_si128((__m128i*)utf8_output, 239| | _mm256_extractf128_si256(utf8_1,1)); utf8_output += 12; buf += 16; 240| | continue; 241| | }*/ 242| 236k| const uint8_t mask0 = uint8_t(mask); 243| 236k| const uint8_t *row0 = 244| 236k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 245| 236k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 246| 236k| const __m128i utf8_0 = 247| 236k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 248| | 249| 236k| const uint8_t mask1 = static_cast(mask >> 8); 250| 236k| const uint8_t *row1 = 251| 236k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 252| 236k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 253| 236k| const __m128i utf8_1 = 254| 236k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 255| | 256| 236k| const uint8_t mask2 = static_cast(mask >> 16); 257| 236k| const uint8_t *row2 = 258| 236k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 259| 236k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 260| 236k| const __m128i utf8_2 = 261| 236k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 262| | 263| 236k| const uint8_t mask3 = static_cast(mask >> 24); 264| 236k| const uint8_t *row3 = 265| 236k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 266| 236k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 267| 236k| const __m128i utf8_3 = 268| 236k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 269| | 270| 236k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 271| 236k| utf8_output += row0[0]; 272| 236k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 273| 236k| utf8_output += row1[0]; 274| 236k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 275| 236k| utf8_output += row2[0]; 276| 236k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 277| 236k| utf8_output += row3[0]; 278| 236k| buf += 16; 279| | // surrogate pair(s) in a register 280| 236k| } else { 281| | // Let us do a scalar fallback. 282| | // It may seem wasteful to use scalar code, but being efficient with SIMD 283| | // in the presence of surrogate pairs may require non-trivial tables. 284| 4.17k| size_t forward = 15; 285| 4.17k| size_t k = 0; 286| 4.17k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (286:11): [True: 0, False: 4.17k] ------------------ 287| 0| forward = size_t(end - buf - 1); 288| 0| } 289| 61.0k| for (; k < forward; k++) { ------------------ | Branch (289:14): [True: 56.9k, False: 4.07k] ------------------ 290| 56.9k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 291| 56.9k| if ((word & 0xFF80) == 0) { ------------------ | Branch (291:13): [True: 8.41k, False: 48.5k] ------------------ 292| 8.41k| *utf8_output++ = char(word); 293| 48.5k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (293:20): [True: 2.49k, False: 46.0k] ------------------ 294| 2.49k| *utf8_output++ = char((word >> 6) | 0b11000000); 295| 2.49k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 296| 46.0k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (296:20): [True: 41.0k, False: 4.95k] ------------------ 297| 41.0k| *utf8_output++ = char((word >> 12) | 0b11100000); 298| 41.0k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 299| 41.0k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 300| 41.0k| } else { 301| | // must be a surrogate pair 302| 4.95k| uint16_t diff = uint16_t(word - 0xD800); 303| 4.95k| uint16_t next_word = 304| 4.95k| scalar::utf16::swap_if_needed(buf[k + 1]); 305| 4.95k| k++; 306| 4.95k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 307| 4.95k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (307:15): [True: 106, False: 4.85k] ------------------ 308| 106| return std::make_pair(nullptr, utf8_output); 309| 106| } 310| 4.85k| uint32_t value = (diff << 10) + diff2 + 0x10000; 311| 4.85k| *utf8_output++ = char((value >> 18) | 0b11110000); 312| 4.85k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 313| 4.85k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 314| 4.85k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 315| 4.85k| } 316| 56.9k| } 317| 4.07k| buf += k; 318| 4.07k| } 319| 240k| } // while 320| 692| return std::make_pair(buf, utf8_output); 321| 798|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_138avx2_convert_utf16_to_utf8_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 333| 642| char *utf8_output) { 334| 642| const char16_t *start = buf; 335| 642| const char16_t *end = buf + len; 336| | 337| 642| const __m256i v_0000 = _mm256_setzero_si256(); 338| 642| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 339| 642| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 340| 642| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 341| 642| const size_t safety_margin = 342| 642| 12; // to avoid overruns, see issue 343| | // https://github.com/simdutf/simdutf/issues/92 344| | 345| 646k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (345:10): [True: 646k, False: 484] ------------------ 346| 646k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 347| 646k| if (big_endian) { ------------------ | Branch (347:9): [Folded, False: 646k] ------------------ 348| 0| const __m256i swap = _mm256_setr_epi8( 349| 0| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 350| 0| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 351| 0| in = _mm256_shuffle_epi8(in, swap); 352| 0| } 353| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 354| 646k| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 355| 646k| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (355:9): [True: 308k, False: 337k] ------------------ 356| | // 1. pack the bytes 357| 308k| const __m128i utf8_packed = _mm_packus_epi16( 358| 308k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 359| | // 2. store (16 bytes) 360| 308k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 361| | // 3. adjust pointers 362| 308k| buf += 16; 363| 308k| utf8_output += 16; 364| 308k| continue; // we are done for this round! 365| 308k| } 366| | // no bits set above 7th bit 367| 337k| const __m256i one_byte_bytemask = 368| 337k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 369| 337k| const uint32_t one_byte_bitmask = 370| 337k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 371| | 372| | // no bits set above 11th bit 373| 337k| const __m256i one_or_two_bytes_bytemask = 374| 337k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 375| 337k| const uint32_t one_or_two_bytes_bitmask = 376| 337k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 377| 337k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (377:9): [True: 62.4k, False: 275k] ------------------ 378| | 379| | // 1. prepare 2-byte values 380| | // input 16-bit word : [0000|0aaa|aabb|bbbb] x 8 381| | // expected output : [110a|aaaa|10bb|bbbb] x 8 382| 62.4k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 383| 62.4k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 384| | 385| | // t0 = [000a|aaaa|bbbb|bb00] 386| 62.4k| const __m256i t0 = _mm256_slli_epi16(in, 2); 387| | // t1 = [000a|aaaa|0000|0000] 388| 62.4k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 389| | // t2 = [0000|0000|00bb|bbbb] 390| 62.4k| const __m256i t2 = _mm256_and_si256(in, v_003f); 391| | // t3 = [000a|aaaa|00bb|bbbb] 392| 62.4k| const __m256i t3 = _mm256_or_si256(t1, t2); 393| | // t4 = [110a|aaaa|10bb|bbbb] 394| 62.4k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 395| | 396| | // 2. merge ASCII and 2-byte codewords 397| 62.4k| const __m256i utf8_unpacked = 398| 62.4k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 399| | 400| | // 3. prepare bitmask for 8-bit lookup 401| 62.4k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 402| 62.4k| const uint32_t M1 = M0 >> 7; 403| 62.4k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 404| | // 4. pack the bytes 405| | 406| 62.4k| const uint8_t *row = 407| 62.4k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 408| 62.4k| const uint8_t *row_2 = 409| 62.4k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 410| 62.4k| 16)][0]; 411| | 412| 62.4k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 413| 62.4k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 414| | 415| 62.4k| const __m256i utf8_packed = _mm256_shuffle_epi8( 416| 62.4k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 62.4k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 62.4k| _mm256_castsi128_si256(xmm2), 2) ------------------ 417| | // 5. store bytes 418| 62.4k| _mm_storeu_si128((__m128i *)utf8_output, 419| 62.4k| _mm256_castsi256_si128(utf8_packed)); 420| 62.4k| utf8_output += row[0]; 421| 62.4k| _mm_storeu_si128((__m128i *)utf8_output, 422| 62.4k| _mm256_extractf128_si256(utf8_packed, 1)); 423| 62.4k| utf8_output += row_2[0]; 424| | 425| | // 6. adjust pointers 426| 62.4k| buf += 16; 427| 62.4k| continue; 428| 62.4k| } 429| | // 1. Check if there are any surrogate word in the input chunk. 430| | // We have also deal with situation when there is a surrogate word 431| | // at the end of a chunk. 432| 275k| const __m256i surrogates_bytemask = 433| 275k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_d800); 434| | 435| | // bitmask = 0x0000 if there are no surrogates 436| | // = 0xc000 if the last word is a surrogate 437| 275k| const uint32_t surrogates_bitmask = 438| 275k| static_cast(_mm256_movemask_epi8(surrogates_bytemask)); 439| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 440| | // However, it is likely an uncommon occurrence. 441| 275k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (441:9): [True: 272k, False: 2.72k] ------------------ 442| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 443| 272k| const __m256i dup_even = _mm256_setr_epi16( 444| 272k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 445| 272k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 446| | 447| | /* In this branch we handle three cases: 448| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 449| | single UFT-8 byte 450| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - two 451| | UTF-8 bytes 452| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 453| | three UTF-8 bytes 454| | 455| | We expand the input word (16-bit) into two code units (32-bit), thus 456| | we have room for four bytes. However, we need five distinct bit 457| | layouts. Note that the last byte in cases #2 and #3 is the same. 458| | 459| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 460| | in register t2. 461| | 462| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 463| | either byte 1 for case #2 or byte 2 for case #3. Note that they 464| | differ by exactly one bit. 465| | 466| | Finally from these two code units we build proper UTF-8 sequence, taking 467| | into account the case (i.e, the number of bytes to write). 468| | */ 469| | /** 470| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 471| | * t2 => [0ccc|cccc] [10cc|cccc] 472| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 473| | */ 474| 272k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 475| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 476| 272k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 477| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 478| 272k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 474| 272k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 479| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 480| 272k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 474| 272k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 481| | 482| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 483| 272k| const __m256i s0 = _mm256_srli_epi16(in, 4); 484| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 485| 272k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 474| 272k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 486| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 487| 272k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 474| 272k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 488| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 489| 272k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 474| 272k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 490| 272k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 491| 272k| simdutf_vec(0b0100000000000000)); ------------------ | | 474| 272k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 492| 272k| const __m256i s4 = _mm256_xor_si256(s3, m0); 493| 272k|#undef simdutf_vec 494| | 495| | // 4. expand code units 16-bit => 32-bit 496| 272k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 497| 272k| const __m256i out1 = _mm256_unpackhi_epi16(t2, s4); 498| | 499| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 500| 272k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 501| 272k| (one_or_two_bytes_bitmask & 0xaaaaaaaa); 502| | // Due to the wider registers, the following path is less likely to be 503| | // useful. 504| | /*if(mask == 0) { 505| | // We only have three-byte code units. Use fast path. 506| | const __m256i shuffle = 507| | _mm256_setr_epi8(2,3,1,6,7,5,10,11,9,14,15,13,-1,-1,-1,-1, 508| | 2,3,1,6,7,5,10,11,9,14,15,13,-1,-1,-1,-1); const __m256i utf8_0 = 509| | _mm256_shuffle_epi8(out0, shuffle); const __m256i utf8_1 = 510| | _mm256_shuffle_epi8(out1, shuffle); 511| | _mm_storeu_si128((__m128i*)utf8_output, _mm256_castsi256_si128(utf8_0)); 512| | utf8_output += 12; 513| | _mm_storeu_si128((__m128i*)utf8_output, _mm256_castsi256_si128(utf8_1)); 514| | utf8_output += 12; 515| | _mm_storeu_si128((__m128i*)utf8_output, 516| | _mm256_extractf128_si256(utf8_0,1)); utf8_output += 12; 517| | _mm_storeu_si128((__m128i*)utf8_output, 518| | _mm256_extractf128_si256(utf8_1,1)); utf8_output += 12; buf += 16; 519| | continue; 520| | }*/ 521| 272k| const uint8_t mask0 = uint8_t(mask); 522| 272k| const uint8_t *row0 = 523| 272k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 524| 272k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 525| 272k| const __m128i utf8_0 = 526| 272k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 527| | 528| 272k| const uint8_t mask1 = static_cast(mask >> 8); 529| 272k| const uint8_t *row1 = 530| 272k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 531| 272k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 532| 272k| const __m128i utf8_1 = 533| 272k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 534| | 535| 272k| const uint8_t mask2 = static_cast(mask >> 16); 536| 272k| const uint8_t *row2 = 537| 272k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 538| 272k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 539| 272k| const __m128i utf8_2 = 540| 272k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 541| | 542| 272k| const uint8_t mask3 = static_cast(mask >> 24); 543| 272k| const uint8_t *row3 = 544| 272k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 545| 272k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 546| 272k| const __m128i utf8_3 = 547| 272k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 548| | 549| 272k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 550| 272k| utf8_output += row0[0]; 551| 272k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 552| 272k| utf8_output += row1[0]; 553| 272k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 554| 272k| utf8_output += row2[0]; 555| 272k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 556| 272k| utf8_output += row3[0]; 557| 272k| buf += 16; 558| | // surrogate pair(s) in a register 559| 272k| } else { 560| | // Let us do a scalar fallback. 561| | // It may seem wasteful to use scalar code, but being efficient with SIMD 562| | // in the presence of surrogate pairs may require non-trivial tables. 563| 2.72k| size_t forward = 15; 564| 2.72k| size_t k = 0; 565| 2.72k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (565:11): [True: 0, False: 2.72k] ------------------ 566| 0| forward = size_t(end - buf - 1); 567| 0| } 568| 39.2k| for (; k < forward; k++) { ------------------ | Branch (568:14): [True: 36.6k, False: 2.56k] ------------------ 569| 36.6k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 570| 36.6k| if ((word & 0xFF80) == 0) { ------------------ | Branch (570:13): [True: 6.43k, False: 30.2k] ------------------ 571| 6.43k| *utf8_output++ = char(word); 572| 30.2k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (572:20): [True: 1.80k, False: 28.4k] ------------------ 573| 1.80k| *utf8_output++ = char((word >> 6) | 0b11000000); 574| 1.80k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 575| 28.4k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (575:20): [True: 25.3k, False: 3.06k] ------------------ 576| 25.3k| *utf8_output++ = char((word >> 12) | 0b11100000); 577| 25.3k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 578| 25.3k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 579| 25.3k| } else { 580| | // must be a surrogate pair 581| 3.06k| uint16_t diff = uint16_t(word - 0xD800); 582| 3.06k| uint16_t next_word = 583| 3.06k| scalar::utf16::swap_if_needed(buf[k + 1]); 584| 3.06k| k++; 585| 3.06k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 586| 3.06k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (586:15): [True: 158, False: 2.90k] ------------------ 587| 158| return std::make_pair( 588| 158| result(error_code::SURROGATE, buf - start + k - 1), 589| 158| utf8_output); 590| 158| } 591| 2.90k| uint32_t value = (diff << 10) + diff2 + 0x10000; 592| 2.90k| *utf8_output++ = char((value >> 18) | 0b11110000); 593| 2.90k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 594| 2.90k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 595| 2.90k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 596| 2.90k| } 597| 36.6k| } 598| 2.56k| buf += k; 599| 2.56k| } 600| 275k| } // while 601| 484| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 602| 642|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_138avx2_convert_utf16_to_utf8_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 333| 654| char *utf8_output) { 334| 654| const char16_t *start = buf; 335| 654| const char16_t *end = buf + len; 336| | 337| 654| const __m256i v_0000 = _mm256_setzero_si256(); 338| 654| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 339| 654| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 340| 654| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 341| 654| const size_t safety_margin = 342| 654| 12; // to avoid overruns, see issue 343| | // https://github.com/simdutf/simdutf/issues/92 344| | 345| 487k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (345:10): [True: 486k, False: 474] ------------------ 346| 486k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 347| 486k| if (big_endian) { ------------------ | Branch (347:9): [True: 486k, Folded] ------------------ 348| 486k| const __m256i swap = _mm256_setr_epi8( 349| 486k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 350| 486k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 351| 486k| in = _mm256_shuffle_epi8(in, swap); 352| 486k| } 353| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 354| 486k| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 355| 486k| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (355:9): [True: 368k, False: 117k] ------------------ 356| | // 1. pack the bytes 357| 368k| const __m128i utf8_packed = _mm_packus_epi16( 358| 368k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 359| | // 2. store (16 bytes) 360| 368k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 361| | // 3. adjust pointers 362| 368k| buf += 16; 363| 368k| utf8_output += 16; 364| 368k| continue; // we are done for this round! 365| 368k| } 366| | // no bits set above 7th bit 367| 117k| const __m256i one_byte_bytemask = 368| 117k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 369| 117k| const uint32_t one_byte_bitmask = 370| 117k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 371| | 372| | // no bits set above 11th bit 373| 117k| const __m256i one_or_two_bytes_bytemask = 374| 117k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 375| 117k| const uint32_t one_or_two_bytes_bitmask = 376| 117k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 377| 117k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (377:9): [True: 25.8k, False: 92.0k] ------------------ 378| | 379| | // 1. prepare 2-byte values 380| | // input 16-bit word : [0000|0aaa|aabb|bbbb] x 8 381| | // expected output : [110a|aaaa|10bb|bbbb] x 8 382| 25.8k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 383| 25.8k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 384| | 385| | // t0 = [000a|aaaa|bbbb|bb00] 386| 25.8k| const __m256i t0 = _mm256_slli_epi16(in, 2); 387| | // t1 = [000a|aaaa|0000|0000] 388| 25.8k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 389| | // t2 = [0000|0000|00bb|bbbb] 390| 25.8k| const __m256i t2 = _mm256_and_si256(in, v_003f); 391| | // t3 = [000a|aaaa|00bb|bbbb] 392| 25.8k| const __m256i t3 = _mm256_or_si256(t1, t2); 393| | // t4 = [110a|aaaa|10bb|bbbb] 394| 25.8k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 395| | 396| | // 2. merge ASCII and 2-byte codewords 397| 25.8k| const __m256i utf8_unpacked = 398| 25.8k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 399| | 400| | // 3. prepare bitmask for 8-bit lookup 401| 25.8k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 402| 25.8k| const uint32_t M1 = M0 >> 7; 403| 25.8k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 404| | // 4. pack the bytes 405| | 406| 25.8k| const uint8_t *row = 407| 25.8k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 408| 25.8k| const uint8_t *row_2 = 409| 25.8k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 410| 25.8k| 16)][0]; 411| | 412| 25.8k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 413| 25.8k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 414| | 415| 25.8k| const __m256i utf8_packed = _mm256_shuffle_epi8( 416| 25.8k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 25.8k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 25.8k| _mm256_castsi128_si256(xmm2), 2) ------------------ 417| | // 5. store bytes 418| 25.8k| _mm_storeu_si128((__m128i *)utf8_output, 419| 25.8k| _mm256_castsi256_si128(utf8_packed)); 420| 25.8k| utf8_output += row[0]; 421| 25.8k| _mm_storeu_si128((__m128i *)utf8_output, 422| 25.8k| _mm256_extractf128_si256(utf8_packed, 1)); 423| 25.8k| utf8_output += row_2[0]; 424| | 425| | // 6. adjust pointers 426| 25.8k| buf += 16; 427| 25.8k| continue; 428| 25.8k| } 429| | // 1. Check if there are any surrogate word in the input chunk. 430| | // We have also deal with situation when there is a surrogate word 431| | // at the end of a chunk. 432| 92.0k| const __m256i surrogates_bytemask = 433| 92.0k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_d800); 434| | 435| | // bitmask = 0x0000 if there are no surrogates 436| | // = 0xc000 if the last word is a surrogate 437| 92.0k| const uint32_t surrogates_bitmask = 438| 92.0k| static_cast(_mm256_movemask_epi8(surrogates_bytemask)); 439| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 440| | // However, it is likely an uncommon occurrence. 441| 92.0k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (441:9): [True: 84.2k, False: 7.81k] ------------------ 442| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 443| 84.2k| const __m256i dup_even = _mm256_setr_epi16( 444| 84.2k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 445| 84.2k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 446| | 447| | /* In this branch we handle three cases: 448| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 449| | single UFT-8 byte 450| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - two 451| | UTF-8 bytes 452| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 453| | three UTF-8 bytes 454| | 455| | We expand the input word (16-bit) into two code units (32-bit), thus 456| | we have room for four bytes. However, we need five distinct bit 457| | layouts. Note that the last byte in cases #2 and #3 is the same. 458| | 459| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 460| | in register t2. 461| | 462| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 463| | either byte 1 for case #2 or byte 2 for case #3. Note that they 464| | differ by exactly one bit. 465| | 466| | Finally from these two code units we build proper UTF-8 sequence, taking 467| | into account the case (i.e, the number of bytes to write). 468| | */ 469| | /** 470| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 471| | * t2 => [0ccc|cccc] [10cc|cccc] 472| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 473| | */ 474| 84.2k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 475| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 476| 84.2k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 477| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 478| 84.2k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 474| 84.2k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 479| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 480| 84.2k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 474| 84.2k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 481| | 482| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 483| 84.2k| const __m256i s0 = _mm256_srli_epi16(in, 4); 484| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 485| 84.2k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 474| 84.2k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 486| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 487| 84.2k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 474| 84.2k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 488| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 489| 84.2k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 474| 84.2k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 490| 84.2k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 491| 84.2k| simdutf_vec(0b0100000000000000)); ------------------ | | 474| 84.2k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 492| 84.2k| const __m256i s4 = _mm256_xor_si256(s3, m0); 493| 84.2k|#undef simdutf_vec 494| | 495| | // 4. expand code units 16-bit => 32-bit 496| 84.2k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 497| 84.2k| const __m256i out1 = _mm256_unpackhi_epi16(t2, s4); 498| | 499| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 500| 84.2k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 501| 84.2k| (one_or_two_bytes_bitmask & 0xaaaaaaaa); 502| | // Due to the wider registers, the following path is less likely to be 503| | // useful. 504| | /*if(mask == 0) { 505| | // We only have three-byte code units. Use fast path. 506| | const __m256i shuffle = 507| | _mm256_setr_epi8(2,3,1,6,7,5,10,11,9,14,15,13,-1,-1,-1,-1, 508| | 2,3,1,6,7,5,10,11,9,14,15,13,-1,-1,-1,-1); const __m256i utf8_0 = 509| | _mm256_shuffle_epi8(out0, shuffle); const __m256i utf8_1 = 510| | _mm256_shuffle_epi8(out1, shuffle); 511| | _mm_storeu_si128((__m128i*)utf8_output, _mm256_castsi256_si128(utf8_0)); 512| | utf8_output += 12; 513| | _mm_storeu_si128((__m128i*)utf8_output, _mm256_castsi256_si128(utf8_1)); 514| | utf8_output += 12; 515| | _mm_storeu_si128((__m128i*)utf8_output, 516| | _mm256_extractf128_si256(utf8_0,1)); utf8_output += 12; 517| | _mm_storeu_si128((__m128i*)utf8_output, 518| | _mm256_extractf128_si256(utf8_1,1)); utf8_output += 12; buf += 16; 519| | continue; 520| | }*/ 521| 84.2k| const uint8_t mask0 = uint8_t(mask); 522| 84.2k| const uint8_t *row0 = 523| 84.2k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 524| 84.2k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 525| 84.2k| const __m128i utf8_0 = 526| 84.2k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 527| | 528| 84.2k| const uint8_t mask1 = static_cast(mask >> 8); 529| 84.2k| const uint8_t *row1 = 530| 84.2k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 531| 84.2k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 532| 84.2k| const __m128i utf8_1 = 533| 84.2k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 534| | 535| 84.2k| const uint8_t mask2 = static_cast(mask >> 16); 536| 84.2k| const uint8_t *row2 = 537| 84.2k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 538| 84.2k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 539| 84.2k| const __m128i utf8_2 = 540| 84.2k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 541| | 542| 84.2k| const uint8_t mask3 = static_cast(mask >> 24); 543| 84.2k| const uint8_t *row3 = 544| 84.2k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 545| 84.2k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 546| 84.2k| const __m128i utf8_3 = 547| 84.2k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 548| | 549| 84.2k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 550| 84.2k| utf8_output += row0[0]; 551| 84.2k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 552| 84.2k| utf8_output += row1[0]; 553| 84.2k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 554| 84.2k| utf8_output += row2[0]; 555| 84.2k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 556| 84.2k| utf8_output += row3[0]; 557| 84.2k| buf += 16; 558| | // surrogate pair(s) in a register 559| 84.2k| } else { 560| | // Let us do a scalar fallback. 561| | // It may seem wasteful to use scalar code, but being efficient with SIMD 562| | // in the presence of surrogate pairs may require non-trivial tables. 563| 7.81k| size_t forward = 15; 564| 7.81k| size_t k = 0; 565| 7.81k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (565:11): [True: 0, False: 7.81k] ------------------ 566| 0| forward = size_t(end - buf - 1); 567| 0| } 568| 115k| for (; k < forward; k++) { ------------------ | Branch (568:14): [True: 107k, False: 7.63k] ------------------ 569| 107k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 570| 107k| if ((word & 0xFF80) == 0) { ------------------ | Branch (570:13): [True: 36.5k, False: 70.8k] ------------------ 571| 36.5k| *utf8_output++ = char(word); 572| 70.8k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (572:20): [True: 5.30k, False: 65.5k] ------------------ 573| 5.30k| *utf8_output++ = char((word >> 6) | 0b11000000); 574| 5.30k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 575| 65.5k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (575:20): [True: 56.7k, False: 8.74k] ------------------ 576| 56.7k| *utf8_output++ = char((word >> 12) | 0b11100000); 577| 56.7k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 578| 56.7k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 579| 56.7k| } else { 580| | // must be a surrogate pair 581| 8.74k| uint16_t diff = uint16_t(word - 0xD800); 582| 8.74k| uint16_t next_word = 583| 8.74k| scalar::utf16::swap_if_needed(buf[k + 1]); 584| 8.74k| k++; 585| 8.74k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 586| 8.74k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (586:15): [True: 180, False: 8.56k] ------------------ 587| 180| return std::make_pair( 588| 180| result(error_code::SURROGATE, buf - start + k - 1), 589| 180| utf8_output); 590| 180| } 591| 8.56k| uint32_t value = (diff << 10) + diff2 + 0x10000; 592| 8.56k| *utf8_output++ = char((value >> 18) | 0b11110000); 593| 8.56k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 594| 8.56k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 595| 8.56k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 596| 8.56k| } 597| 107k| } 598| 7.63k| buf += k; 599| 7.63k| } 600| 92.0k| } // while 601| 474| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 602| 654|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_utf32_to_latin1EPKDimPc: 3| 228| char *latin1_output) { 4| 228| const size_t rounded_len = 5| 228| len & ~0x1F; // Round down to nearest multiple of 32 6| | 7| 228| const __m256i high_bytes_mask = _mm256_set1_epi32(0xFFFFFF00); 8| | 9| 16.6k| for (size_t i = 0; i < rounded_len; i += 4 * 8) { ------------------ | Branch (9:22): [True: 16.5k, False: 182] ------------------ 10| 16.5k| __m256i a = _mm256_loadu_si256((__m256i *)(buf + 0 * 8)); 11| 16.5k| __m256i b = _mm256_loadu_si256((__m256i *)(buf + 1 * 8)); 12| 16.5k| __m256i c = _mm256_loadu_si256((__m256i *)(buf + 2 * 8)); 13| 16.5k| __m256i d = _mm256_loadu_si256((__m256i *)(buf + 3 * 8)); 14| | 15| 16.5k| const __m256i check_combined = 16| 16.5k| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 17| | 18| 16.5k| if (!_mm256_testz_si256(check_combined, high_bytes_mask)) { ------------------ | Branch (18:9): [True: 46, False: 16.4k] ------------------ 19| 46| return std::make_pair(nullptr, latin1_output); 20| 46| } 21| | 22| 16.4k| b = _mm256_slli_epi32(b, 1 * 8); 23| 16.4k| c = _mm256_slli_epi32(c, 2 * 8); 24| 16.4k| d = _mm256_slli_epi32(d, 3 * 8); 25| | 26| | // clang-format off 27| | 28| | // a = [.. .. .. a7|.. .. .. a6|.. .. .. a5|.. .. .. a4||.. .. .. a3|.. .. .. a2|.. .. .. a1|.. .. .. a0] 29| | // b = [.. .. b7 ..|.. .. b6 ..|.. .. b5 ..|.. .. b4 ..||.. .. b3 ..|.. .. b2 ..|.. .. b1 ..|.. .. b0 ..] 30| | // c = [.. c7 .. ..|.. c6 .. ..|.. c5 .. ..|.. c4 .. ..||.. c3 .. ..|.. c2 .. ..|.. c1 .. ..|.. c0 .. ..] 31| | // d = [d7 .. .. ..|d6 .. .. ..|d5 .. .. ..|d4 .. .. ..||d3 .. .. ..|d2 .. .. ..|d1 .. .. ..|d0 .. .. ..] 32| | 33| | // t0 = [d7 c7 b7 a7|d6 c6 b6 a6|d5 c5 b5 a5|d4 c4 b4 a4||d3 c3 b3 a3|d2 c2 b2 a2|d1 c1 b1 a1|d0 c0 b0 a0] 34| 16.4k| const __m256i t0 = 35| 16.4k| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 36| | 37| | // shuffle bytes within 128-bit lanes 38| | // t1 = [d7 d6 d5 d4|c7 c6 c5 c4|b7 b6 b5 b4|a7 a6 a5 a4||d3 d2 d1 d0|c3 c2 c1 c0|b3 b2 b1 b0|a3 a2 a1 a0] 39| 16.4k| const __m256i shuffle_bytes = 40| 16.4k| _mm256_setr_epi8(0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15, 41| 16.4k| 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15); 42| | 43| 16.4k| const __m256i t1 = _mm256_shuffle_epi8(t0, shuffle_bytes); 44| | 45| | // reshuffle dwords 46| | // t2 = [d7 d6 d5 d4|d3 d2 d1 d0|c7 c6 c5 c4|c3 c2 c1 c0||b7 b6 b5 b4|b3 b2 b1 b0|a7 a6 a5 a4|a3 a2 a1 a0] 47| 16.4k| const __m256i shuffle_dwords = _mm256_setr_epi32(0, 4, 1, 5, 2, 6, 3, 7); 48| 16.4k| const __m256i t2 = _mm256_permutevar8x32_epi32(t1, shuffle_dwords); 49| |// clang format on 50| | 51| 16.4k| _mm256_storeu_si256((__m256i *)latin1_output, t2); 52| | 53| 16.4k| latin1_output += 32; 54| 16.4k| buf += 32; 55| 16.4k| } 56| | 57| 182| return std::make_pair(buf, latin1_output); 58| 228|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_140avx2_convert_utf32_to_latin1_with_errorsEPKDimPc: 62| 436| char *latin1_output) { 63| 436| const size_t rounded_len = 64| 436| len & ~0x1F; // Round down to nearest multiple of 32 65| | 66| 436| const char32_t *start = buf; 67| | 68| 436| const __m256i high_bytes_mask = _mm256_set1_epi32(0xFFFFFF00); 69| | 70| 17.0k| for (size_t i = 0; i < rounded_len; i += 4 * 8) { ------------------ | Branch (70:22): [True: 16.7k, False: 312] ------------------ 71| 16.7k| __m256i a = _mm256_loadu_si256((__m256i *)(buf + 0 * 8)); 72| 16.7k| __m256i b = _mm256_loadu_si256((__m256i *)(buf + 1 * 8)); 73| 16.7k| __m256i c = _mm256_loadu_si256((__m256i *)(buf + 2 * 8)); 74| 16.7k| __m256i d = _mm256_loadu_si256((__m256i *)(buf + 3 * 8)); 75| | 76| 16.7k| const __m256i check_combined = 77| 16.7k| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 78| | 79| 16.7k| if (!_mm256_testz_si256(check_combined, high_bytes_mask)) { ------------------ | Branch (79:9): [True: 124, False: 16.6k] ------------------ 80| | // Fallback to scalar code for handling errors 81| 590| for (int k = 0; k < 4 * 8; k++) { ------------------ | Branch (81:23): [True: 590, False: 0] ------------------ 82| 590| char32_t codepoint = buf[k]; 83| 590| if (codepoint <= 0xFF) { ------------------ | Branch (83:13): [True: 466, False: 124] ------------------ 84| 466| *latin1_output++ = static_cast(codepoint); 85| 466| } else { 86| 124| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 87| 124| latin1_output); 88| 124| } 89| 590| } 90| 124| } 91| | 92| 16.6k| b = _mm256_slli_epi32(b, 1 * 8); 93| 16.6k| c = _mm256_slli_epi32(c, 2 * 8); 94| 16.6k| d = _mm256_slli_epi32(d, 3 * 8); 95| | 96| 16.6k| const __m256i t0 = 97| 16.6k| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 98| | 99| 16.6k| const __m256i shuffle_bytes = 100| 16.6k| _mm256_setr_epi8(0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15, 101| 16.6k| 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15); 102| | 103| 16.6k| const __m256i t1 = _mm256_shuffle_epi8(t0, shuffle_bytes); 104| | 105| 16.6k| const __m256i shuffle_dwords = _mm256_setr_epi32(0, 4, 1, 5, 2, 6, 3, 7); 106| 16.6k| const __m256i t2 = _mm256_permutevar8x32_epi32(t1, shuffle_dwords); 107| | 108| 16.6k| _mm256_storeu_si256((__m256i *)latin1_output, t2); 109| | 110| 16.6k| latin1_output += 32; 111| 16.6k| buf += 32; 112| 16.6k| } 113| | 114| 312| return std::make_pair(result(error_code::SUCCESS, buf - start), 115| 312| latin1_output); 116| 436|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf32_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKDiPDsEES7_mS8_: 4| 822| char16_t *utf16_output) { 5| 822| const char32_t *end = buf + len; 6| | 7| 822| const size_t safety_margin = 8| 822| 12; // to avoid overruns, see issue 9| | // https://github.com/simdutf/simdutf/issues/92 10| 822| __m256i forbidden_bytemask = _mm256_setzero_si256(); 11| | 12| 822| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 13| 822| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 14| 822| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 15| | 16| 151k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (16:10): [True: 151k, False: 662] ------------------ 17| 151k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 18| | 19| 151k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 151k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 147k, False: 4.26k] | | ------------------ ------------------ 20| | // no bits set above 16th bit <=> can pack to UTF16 21| | // without surrogate pairs 22| 147k| forbidden_bytemask = _mm256_or_si256( 23| 147k| forbidden_bytemask, 24| 147k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800)); 25| | 26| 147k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 27| 147k| _mm256_extractf128_si256(in, 1)); 28| 147k| if (big_endian) { ------------------ | Branch (28:11): [Folded, False: 147k] ------------------ 29| 0| const __m128i swap = 30| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 31| 0| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 32| 0| } 33| 147k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 34| 147k| utf16_output += 8; 35| 147k| buf += 8; 36| 147k| } else { 37| 4.26k| size_t forward = 7; 38| 4.26k| size_t k = 0; 39| 4.26k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (39:11): [True: 0, False: 4.26k] ------------------ 40| 0| forward = size_t(end - buf - 1); 41| 0| } 42| 33.3k| for (; k < forward; k++) { ------------------ | Branch (42:14): [True: 29.2k, False: 4.10k] ------------------ 43| 29.2k| uint32_t word = buf[k]; 44| 29.2k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (44:13): [True: 8.53k, False: 20.6k] ------------------ 45| | // will not generate a surrogate pair 46| 8.53k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (46:15): [True: 820, False: 7.71k] | Branch (46:33): [True: 24, False: 796] ------------------ 47| 24| return std::make_pair(nullptr, utf16_output); 48| 24| } 49| 8.50k| *utf16_output++ = 50| 8.50k| big_endian ------------------ | Branch (50:15): [Folded, False: 8.50k] ------------------ 51| 8.50k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 52| 8.50k| : char16_t(word); 53| 20.6k| } else { 54| | // will generate a surrogate pair 55| 20.6k| if (word > 0x10FFFF) { ------------------ | Branch (55:15): [True: 136, False: 20.5k] ------------------ 56| 136| return std::make_pair(nullptr, utf16_output); 57| 136| } 58| 20.5k| word -= 0x10000; 59| 20.5k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 60| 20.5k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 61| 20.5k| if (big_endian) { ------------------ | Branch (61:15): [Folded, False: 20.5k] ------------------ 62| 0| high_surrogate = 63| 0| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 64| 0| low_surrogate = 65| 0| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 66| 0| } 67| 20.5k| *utf16_output++ = char16_t(high_surrogate); 68| 20.5k| *utf16_output++ = char16_t(low_surrogate); 69| 20.5k| } 70| 29.2k| } 71| 4.10k| buf += k; 72| 4.10k| } 73| 151k| } 74| | 75| | // check for invalid input 76| 662| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (76:7): [True: 42, False: 620] ------------------ 77| 42| return std::make_pair(nullptr, utf16_output); 78| 42| } 79| | 80| 620| return std::make_pair(buf, utf16_output); 81| 662|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf32_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKDiPDsEES7_mS8_: 4| 674| char16_t *utf16_output) { 5| 674| const char32_t *end = buf + len; 6| | 7| 674| const size_t safety_margin = 8| 674| 12; // to avoid overruns, see issue 9| | // https://github.com/simdutf/simdutf/issues/92 10| 674| __m256i forbidden_bytemask = _mm256_setzero_si256(); 11| | 12| 674| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 13| 674| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 14| 674| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 15| | 16| 18.0k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (16:10): [True: 17.5k, False: 488] ------------------ 17| 17.5k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 18| | 19| 17.5k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 17.5k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 14.0k, False: 3.50k] | | ------------------ ------------------ 20| | // no bits set above 16th bit <=> can pack to UTF16 21| | // without surrogate pairs 22| 14.0k| forbidden_bytemask = _mm256_or_si256( 23| 14.0k| forbidden_bytemask, 24| 14.0k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800)); 25| | 26| 14.0k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 27| 14.0k| _mm256_extractf128_si256(in, 1)); 28| 14.0k| if (big_endian) { ------------------ | Branch (28:11): [True: 14.0k, Folded] ------------------ 29| 14.0k| const __m128i swap = 30| 14.0k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 31| 14.0k| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 32| 14.0k| } 33| 14.0k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 34| 14.0k| utf16_output += 8; 35| 14.0k| buf += 8; 36| 14.0k| } else { 37| 3.50k| size_t forward = 7; 38| 3.50k| size_t k = 0; 39| 3.50k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (39:11): [True: 0, False: 3.50k] ------------------ 40| 0| forward = size_t(end - buf - 1); 41| 0| } 42| 27.1k| for (; k < forward; k++) { ------------------ | Branch (42:14): [True: 23.8k, False: 3.32k] ------------------ 43| 23.8k| uint32_t word = buf[k]; 44| 23.8k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (44:13): [True: 2.63k, False: 21.1k] ------------------ 45| | // will not generate a surrogate pair 46| 2.63k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (46:15): [True: 440, False: 2.19k] | Branch (46:33): [True: 4, False: 436] ------------------ 47| 4| return std::make_pair(nullptr, utf16_output); 48| 4| } 49| 2.62k| *utf16_output++ = 50| 2.62k| big_endian ------------------ | Branch (50:15): [True: 2.62k, Folded] ------------------ 51| 2.62k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 52| 2.62k| : char16_t(word); 53| 21.1k| } else { 54| | // will generate a surrogate pair 55| 21.1k| if (word > 0x10FFFF) { ------------------ | Branch (55:15): [True: 182, False: 21.0k] ------------------ 56| 182| return std::make_pair(nullptr, utf16_output); 57| 182| } 58| 21.0k| word -= 0x10000; 59| 21.0k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 60| 21.0k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 61| 21.0k| if (big_endian) { ------------------ | Branch (61:15): [True: 21.0k, Folded] ------------------ 62| 21.0k| high_surrogate = 63| 21.0k| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 64| 21.0k| low_surrogate = 65| 21.0k| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 66| 21.0k| } 67| 21.0k| *utf16_output++ = char16_t(high_surrogate); 68| 21.0k| *utf16_output++ = char16_t(low_surrogate); 69| 21.0k| } 70| 23.8k| } 71| 3.32k| buf += k; 72| 3.32k| } 73| 17.5k| } 74| | 75| | // check for invalid input 76| 488| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (76:7): [True: 34, False: 454] ------------------ 77| 34| return std::make_pair(nullptr, utf16_output); 78| 34| } 79| | 80| 454| return std::make_pair(buf, utf16_output); 81| 488|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf32_to_utf16_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 86| 674| char16_t *utf16_output) { 87| 674| const char32_t *start = buf; 88| 674| const char32_t *end = buf + len; 89| | 90| 674| const size_t safety_margin = 91| 674| 12; // to avoid overruns, see issue 92| | // https://github.com/simdutf/simdutf/issues/92 93| | 94| 674| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 95| 674| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 96| 674| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 97| | 98| 68.4k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (98:10): [True: 67.9k, False: 502] ------------------ 99| 67.9k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 100| | 101| 67.9k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 67.9k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 66.5k, False: 1.43k] | | ------------------ ------------------ 102| | // no bits set above 16th bit <=> can pack to UTF16 without surrogate 103| | // pairs 104| 66.5k| const __m256i forbidden_bytemask = 105| 66.5k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800); 106| 66.5k| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != ------------------ | Branch (106:11): [True: 32, False: 66.5k] ------------------ 107| 66.5k| 0x0) { 108| 32| return std::make_pair(result(error_code::SURROGATE, buf - start), 109| 32| utf16_output); 110| 32| } 111| | 112| 66.5k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 113| 66.5k| _mm256_extractf128_si256(in, 1)); 114| 66.5k| if (big_endian) { ------------------ | Branch (114:11): [Folded, False: 66.5k] ------------------ 115| 0| const __m128i swap = 116| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 117| 0| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 118| 0| } 119| 66.5k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 120| 66.5k| utf16_output += 8; 121| 66.5k| buf += 8; 122| 66.5k| } else { 123| 1.43k| size_t forward = 7; 124| 1.43k| size_t k = 0; 125| 1.43k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (125:11): [True: 0, False: 1.43k] ------------------ 126| 0| forward = size_t(end - buf - 1); 127| 0| } 128| 10.8k| for (; k < forward; k++) { ------------------ | Branch (128:14): [True: 9.50k, False: 1.29k] ------------------ 129| 9.50k| uint32_t word = buf[k]; 130| 9.50k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (130:13): [True: 4.44k, False: 5.06k] ------------------ 131| | // will not generate a surrogate pair 132| 4.44k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (132:15): [True: 580, False: 3.86k] | Branch (132:33): [True: 4, False: 576] ------------------ 133| 4| return std::make_pair( 134| 4| result(error_code::SURROGATE, buf - start + k), utf16_output); 135| 4| } 136| 4.44k| *utf16_output++ = 137| 4.44k| big_endian ------------------ | Branch (137:15): [Folded, False: 4.44k] ------------------ 138| 4.44k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 139| 4.44k| : char16_t(word); 140| 5.06k| } else { 141| | // will generate a surrogate pair 142| 5.06k| if (word > 0x10FFFF) { ------------------ | Branch (142:15): [True: 136, False: 4.92k] ------------------ 143| 136| return std::make_pair( 144| 136| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 145| 136| } 146| 4.92k| word -= 0x10000; 147| 4.92k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 148| 4.92k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 149| 4.92k| if (big_endian) { ------------------ | Branch (149:15): [Folded, False: 4.92k] ------------------ 150| 0| high_surrogate = 151| 0| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 152| 0| low_surrogate = 153| 0| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 154| 0| } 155| 4.92k| *utf16_output++ = char16_t(high_surrogate); 156| 4.92k| *utf16_output++ = char16_t(low_surrogate); 157| 4.92k| } 158| 9.50k| } 159| 1.29k| buf += k; 160| 1.29k| } 161| 67.9k| } 162| | 163| 502| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 164| 674|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf32_to_utf16_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 86| 694| char16_t *utf16_output) { 87| 694| const char32_t *start = buf; 88| 694| const char32_t *end = buf + len; 89| | 90| 694| const size_t safety_margin = 91| 694| 12; // to avoid overruns, see issue 92| | // https://github.com/simdutf/simdutf/issues/92 93| | 94| 694| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 95| 694| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 96| 694| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 97| | 98| 4.70k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (98:10): [True: 4.17k, False: 528] ------------------ 99| 4.17k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 100| | 101| 4.17k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 4.17k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.29k, False: 878] | | ------------------ ------------------ 102| | // no bits set above 16th bit <=> can pack to UTF16 without surrogate 103| | // pairs 104| 3.29k| const __m256i forbidden_bytemask = 105| 3.29k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800); 106| 3.29k| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != ------------------ | Branch (106:11): [True: 6, False: 3.28k] ------------------ 107| 3.29k| 0x0) { 108| 6| return std::make_pair(result(error_code::SURROGATE, buf - start), 109| 6| utf16_output); 110| 6| } 111| | 112| 3.28k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 113| 3.28k| _mm256_extractf128_si256(in, 1)); 114| 3.28k| if (big_endian) { ------------------ | Branch (114:11): [True: 3.28k, Folded] ------------------ 115| 3.28k| const __m128i swap = 116| 3.28k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 117| 3.28k| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 118| 3.28k| } 119| 3.28k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 120| 3.28k| utf16_output += 8; 121| 3.28k| buf += 8; 122| 3.28k| } else { 123| 878| size_t forward = 7; 124| 878| size_t k = 0; 125| 878| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (125:11): [True: 0, False: 878] ------------------ 126| 0| forward = size_t(end - buf - 1); 127| 0| } 128| 6.28k| for (; k < forward; k++) { ------------------ | Branch (128:14): [True: 5.56k, False: 718] ------------------ 129| 5.56k| uint32_t word = buf[k]; 130| 5.56k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (130:13): [True: 2.66k, False: 2.89k] ------------------ 131| | // will not generate a surrogate pair 132| 2.66k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (132:15): [True: 662, False: 2.00k] | Branch (132:33): [True: 8, False: 654] ------------------ 133| 8| return std::make_pair( 134| 8| result(error_code::SURROGATE, buf - start + k), utf16_output); 135| 8| } 136| 2.65k| *utf16_output++ = 137| 2.65k| big_endian ------------------ | Branch (137:15): [True: 2.65k, Folded] ------------------ 138| 2.65k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 139| 2.65k| : char16_t(word); 140| 2.89k| } else { 141| | // will generate a surrogate pair 142| 2.89k| if (word > 0x10FFFF) { ------------------ | Branch (142:15): [True: 152, False: 2.74k] ------------------ 143| 152| return std::make_pair( 144| 152| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 145| 152| } 146| 2.74k| word -= 0x10000; 147| 2.74k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 148| 2.74k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 149| 2.74k| if (big_endian) { ------------------ | Branch (149:15): [True: 2.74k, Folded] ------------------ 150| 2.74k| high_surrogate = 151| 2.74k| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 152| 2.74k| low_surrogate = 153| 2.74k| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 154| 2.74k| } 155| 2.74k| *utf16_output++ = char16_t(high_surrogate); 156| 2.74k| *utf16_output++ = char16_t(low_surrogate); 157| 2.74k| } 158| 5.56k| } 159| 718| buf += k; 160| 718| } 161| 4.17k| } 162| | 163| 528| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 164| 694|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_126avx2_convert_utf32_to_utf8EPKDimPc: 2| 864|avx2_convert_utf32_to_utf8(const char32_t *buf, size_t len, char *utf8_output) { 3| 864| const char32_t *end = buf + len; 4| 864| const __m256i v_0000 = _mm256_setzero_si256(); 5| 864| const __m256i v_ffff0000 = _mm256_set1_epi32((uint32_t)0xffff0000); 6| 864| const __m256i v_ff80 = _mm256_set1_epi16((uint16_t)0xff80); 7| 864| const __m256i v_f800 = _mm256_set1_epi16((uint16_t)0xf800); 8| 864| const __m256i v_c080 = _mm256_set1_epi16((uint16_t)0xc080); 9| 864| const __m256i v_7fffffff = _mm256_set1_epi32((uint32_t)0x7fffffff); 10| 864| __m256i running_max = _mm256_setzero_si256(); 11| 864| __m256i forbidden_bytemask = _mm256_setzero_si256(); 12| | 13| 864| const size_t safety_margin = 14| 864| 12; // to avoid overruns, see issue 15| | // https://github.com/simdutf/simdutf/issues/92 16| | 17| 39.2k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (17:10): [True: 38.5k, False: 650] ------------------ 18| 38.5k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 19| 38.5k| __m256i nextin = _mm256_loadu_si256((__m256i *)buf + 1); 20| 38.5k| running_max = _mm256_max_epu32(_mm256_max_epu32(in, running_max), nextin); 21| | 22| | // Pack 32-bit UTF-32 code units to 16-bit UTF-16 code units with unsigned 23| | // saturation 24| 38.5k| __m256i in_16 = _mm256_packus_epi32(_mm256_and_si256(in, v_7fffffff), 25| 38.5k| _mm256_and_si256(nextin, v_7fffffff)); 26| 38.5k| in_16 = _mm256_permute4x64_epi64(in_16, 0b11011000); 27| | 28| | // Try to apply UTF-16 => UTF-8 routine on 256 bits 29| | // (haswell/avx2_convert_utf16_to_utf8.cpp) 30| | 31| 38.5k| if (_mm256_testz_si256(in_16, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (31:9): [True: 33.9k, False: 4.60k] ------------------ 32| | // 1. pack the bytes 33| 33.9k| const __m128i utf8_packed = _mm_packus_epi16( 34| 33.9k| _mm256_castsi256_si128(in_16), _mm256_extractf128_si256(in_16, 1)); 35| | // 2. store (16 bytes) 36| 33.9k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 37| | // 3. adjust pointers 38| 33.9k| buf += 16; 39| 33.9k| utf8_output += 16; 40| 33.9k| continue; // we are done for this round! 41| 33.9k| } 42| | // no bits set above 7th bit 43| 4.60k| const __m256i one_byte_bytemask = 44| 4.60k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_ff80), v_0000); 45| 4.60k| const uint32_t one_byte_bitmask = 46| 4.60k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 47| | 48| | // no bits set above 11th bit 49| 4.60k| const __m256i one_or_two_bytes_bytemask = 50| 4.60k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_0000); 51| 4.60k| const uint32_t one_or_two_bytes_bitmask = 52| 4.60k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 53| 4.60k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (53:9): [True: 1.76k, False: 2.83k] ------------------ 54| | // 1. prepare 2-byte values 55| | // input 16-bit word : [0000|0aaa|aabb|bbbb] x 8 56| | // expected output : [110a|aaaa|10bb|bbbb] x 8 57| 1.76k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 58| 1.76k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 59| | 60| | // t0 = [000a|aaaa|bbbb|bb00] 61| 1.76k| const __m256i t0 = _mm256_slli_epi16(in_16, 2); 62| | // t1 = [000a|aaaa|0000|0000] 63| 1.76k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 64| | // t2 = [0000|0000|00bb|bbbb] 65| 1.76k| const __m256i t2 = _mm256_and_si256(in_16, v_003f); 66| | // t3 = [000a|aaaa|00bb|bbbb] 67| 1.76k| const __m256i t3 = _mm256_or_si256(t1, t2); 68| | // t4 = [110a|aaaa|10bb|bbbb] 69| 1.76k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 70| | 71| | // 2. merge ASCII and 2-byte codewords 72| 1.76k| const __m256i utf8_unpacked = 73| 1.76k| _mm256_blendv_epi8(t4, in_16, one_byte_bytemask); 74| | 75| | // 3. prepare bitmask for 8-bit lookup 76| 1.76k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 77| 1.76k| const uint32_t M1 = M0 >> 7; 78| 1.76k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 79| | // 4. pack the bytes 80| | 81| 1.76k| const uint8_t *row = 82| 1.76k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 83| 1.76k| const uint8_t *row_2 = 84| 1.76k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 85| 1.76k| 16)][0]; 86| | 87| 1.76k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 88| 1.76k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 89| | 90| 1.76k| const __m256i utf8_packed = _mm256_shuffle_epi8( 91| 1.76k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 1.76k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 1.76k| _mm256_castsi128_si256(xmm2), 2) ------------------ 92| | // 5. store bytes 93| 1.76k| _mm_storeu_si128((__m128i *)utf8_output, 94| 1.76k| _mm256_castsi256_si128(utf8_packed)); 95| 1.76k| utf8_output += row[0]; 96| 1.76k| _mm_storeu_si128((__m128i *)utf8_output, 97| 1.76k| _mm256_extractf128_si256(utf8_packed, 1)); 98| 1.76k| utf8_output += row_2[0]; 99| | 100| | // 6. adjust pointers 101| 1.76k| buf += 16; 102| 1.76k| continue; 103| 1.76k| } 104| | // Must check for overflow in packing 105| 2.83k| const __m256i saturation_bytemask = _mm256_cmpeq_epi32( 106| 2.83k| _mm256_and_si256(_mm256_or_si256(in, nextin), v_ffff0000), v_0000); 107| 2.83k| const uint32_t saturation_bitmask = 108| 2.83k| static_cast(_mm256_movemask_epi8(saturation_bytemask)); 109| 2.83k| if (saturation_bitmask == 0xffffffff) { ------------------ | Branch (109:9): [True: 1.04k, False: 1.78k] ------------------ 110| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 111| 1.04k| const __m256i v_d800 = _mm256_set1_epi16((uint16_t)0xd800); 112| 1.04k| forbidden_bytemask = _mm256_or_si256( 113| 1.04k| forbidden_bytemask, 114| 1.04k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_d800)); 115| | 116| 1.04k| const __m256i dup_even = _mm256_setr_epi16( 117| 1.04k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 118| 1.04k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 119| | 120| | /* In this branch we handle three cases: 121| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 122| | single UFT-8 byte 123| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - two 124| | UTF-8 bytes 125| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 126| | three UTF-8 bytes 127| | 128| | We expand the input word (16-bit) into two code units (32-bit), thus 129| | we have room for four bytes. However, we need five distinct bit 130| | layouts. Note that the last byte in cases #2 and #3 is the same. 131| | 132| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 133| | in register t2. 134| | 135| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 136| | either byte 1 for case #2 or byte 2 for case #3. Note that they 137| | differ by exactly one bit. 138| | 139| | Finally from these two code units we build proper UTF-8 sequence, taking 140| | into account the case (i.e, the number of bytes to write). 141| | */ 142| | /** 143| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 144| | * t2 => [0ccc|cccc] [10cc|cccc] 145| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 146| | */ 147| 1.04k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 148| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 149| 1.04k| const __m256i t0 = _mm256_shuffle_epi8(in_16, dup_even); 150| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 151| 1.04k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 147| 1.04k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 152| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 153| 1.04k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 147| 1.04k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 154| | 155| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 156| 1.04k| const __m256i s0 = _mm256_srli_epi16(in_16, 4); 157| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 158| 1.04k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 147| 1.04k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 159| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 160| 1.04k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 147| 1.04k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 161| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 162| 1.04k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 147| 1.04k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 163| 1.04k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 164| 1.04k| simdutf_vec(0b0100000000000000)); ------------------ | | 147| 1.04k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 165| 1.04k| const __m256i s4 = _mm256_xor_si256(s3, m0); 166| 1.04k|#undef simdutf_vec 167| | 168| | // 4. expand code units 16-bit => 32-bit 169| 1.04k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 170| 1.04k| const __m256i out1 = _mm256_unpackhi_epi16(t2, s4); 171| | 172| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 173| 1.04k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 174| 1.04k| (one_or_two_bytes_bitmask & 0xaaaaaaaa); 175| | // Due to the wider registers, the following path is less likely to be 176| | // useful. 177| | /*if(mask == 0) { 178| | // We only have three-byte code units. Use fast path. 179| | const __m256i shuffle = 180| | _mm256_setr_epi8(2,3,1,6,7,5,10,11,9,14,15,13,-1,-1,-1,-1, 181| | 2,3,1,6,7,5,10,11,9,14,15,13,-1,-1,-1,-1); const __m256i utf8_0 = 182| | _mm256_shuffle_epi8(out0, shuffle); const __m256i utf8_1 = 183| | _mm256_shuffle_epi8(out1, shuffle); 184| | _mm_storeu_si128((__m128i*)utf8_output, _mm256_castsi256_si128(utf8_0)); 185| | utf8_output += 12; 186| | _mm_storeu_si128((__m128i*)utf8_output, _mm256_castsi256_si128(utf8_1)); 187| | utf8_output += 12; 188| | _mm_storeu_si128((__m128i*)utf8_output, 189| | _mm256_extractf128_si256(utf8_0,1)); utf8_output += 12; 190| | _mm_storeu_si128((__m128i*)utf8_output, 191| | _mm256_extractf128_si256(utf8_1,1)); utf8_output += 12; buf += 16; 192| | continue; 193| | }*/ 194| 1.04k| const uint8_t mask0 = uint8_t(mask); 195| 1.04k| const uint8_t *row0 = 196| 1.04k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 197| 1.04k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 198| 1.04k| const __m128i utf8_0 = 199| 1.04k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 200| | 201| 1.04k| const uint8_t mask1 = static_cast(mask >> 8); 202| 1.04k| const uint8_t *row1 = 203| 1.04k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 204| 1.04k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 205| 1.04k| const __m128i utf8_1 = 206| 1.04k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 207| | 208| 1.04k| const uint8_t mask2 = static_cast(mask >> 16); 209| 1.04k| const uint8_t *row2 = 210| 1.04k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 211| 1.04k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 212| 1.04k| const __m128i utf8_2 = 213| 1.04k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 214| | 215| 1.04k| const uint8_t mask3 = static_cast(mask >> 24); 216| 1.04k| const uint8_t *row3 = 217| 1.04k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 218| 1.04k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 219| 1.04k| const __m128i utf8_3 = 220| 1.04k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 221| | 222| 1.04k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 223| 1.04k| utf8_output += row0[0]; 224| 1.04k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 225| 1.04k| utf8_output += row1[0]; 226| 1.04k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 227| 1.04k| utf8_output += row2[0]; 228| 1.04k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 229| 1.04k| utf8_output += row3[0]; 230| 1.04k| buf += 16; 231| 1.78k| } else { 232| | // case: at least one 32-bit word is larger than 0xFFFF <=> it will 233| | // produce four UTF-8 bytes. Let us do a scalar fallback. It may seem 234| | // wasteful to use scalar code, but being efficient with SIMD may require 235| | // large, non-trivial tables? 236| 1.78k| size_t forward = 15; 237| 1.78k| size_t k = 0; 238| 1.78k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (238:11): [True: 0, False: 1.78k] ------------------ 239| 0| forward = size_t(end - buf - 1); 240| 0| } 241| 26.2k| for (; k < forward; k++) { ------------------ | Branch (241:14): [True: 24.7k, False: 1.57k] ------------------ 242| 24.7k| uint32_t word = buf[k]; 243| 24.7k| if ((word & 0xFFFFFF80) == 0) { // 1-byte (ASCII) ------------------ | Branch (243:13): [True: 16.4k, False: 8.26k] ------------------ 244| 16.4k| *utf8_output++ = char(word); 245| 16.4k| } else if ((word & 0xFFFFF800) == 0) { // 2-byte ------------------ | Branch (245:20): [True: 1.35k, False: 6.91k] ------------------ 246| 1.35k| *utf8_output++ = char((word >> 6) | 0b11000000); 247| 1.35k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 248| 6.91k| } else if ((word & 0xFFFF0000) == 0) { // 3-byte ------------------ | Branch (248:20): [True: 2.78k, False: 4.12k] ------------------ 249| 2.78k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (249:15): [True: 868, False: 1.91k] | Branch (249:33): [True: 34, False: 834] ------------------ 250| 34| return std::make_pair(nullptr, utf8_output); 251| 34| } 252| 2.75k| *utf8_output++ = char((word >> 12) | 0b11100000); 253| 2.75k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 254| 2.75k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 255| 4.12k| } else { // 4-byte 256| 4.12k| if (word > 0x10FFFF) { ------------------ | Branch (256:15): [True: 180, False: 3.94k] ------------------ 257| 180| return std::make_pair(nullptr, utf8_output); 258| 180| } 259| 3.94k| *utf8_output++ = char((word >> 18) | 0b11110000); 260| 3.94k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 261| 3.94k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 262| 3.94k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 263| 3.94k| } 264| 24.7k| } 265| 1.57k| buf += k; 266| 1.57k| } 267| 2.83k| } // while 268| | 269| | // check for invalid input 270| 650| const __m256i v_10ffff = _mm256_set1_epi32((uint32_t)0x10ffff); 271| 650| if (static_cast(_mm256_movemask_epi8(_mm256_cmpeq_epi32( ------------------ | Branch (271:7): [True: 24, False: 626] ------------------ 272| 650| _mm256_max_epu32(running_max, v_10ffff), v_10ffff))) != 0xffffffff) { 273| 24| return std::make_pair(nullptr, utf8_output); 274| 24| } 275| | 276| 626| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (276:7): [True: 54, False: 572] ------------------ 277| 54| return std::make_pair(nullptr, utf8_output); 278| 54| } 279| | 280| 572| return std::make_pair(buf, utf8_output); 281| 626|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_138avx2_convert_utf32_to_utf8_with_errorsEPKDimPc: 285| 832| char *utf8_output) { 286| 832| const char32_t *end = buf + len; 287| 832| const char32_t *start = buf; 288| | 289| 832| const __m256i v_0000 = _mm256_setzero_si256(); 290| 832| const __m256i v_ffff0000 = _mm256_set1_epi32((uint32_t)0xffff0000); 291| 832| const __m256i v_ff80 = _mm256_set1_epi16((uint16_t)0xff80); 292| 832| const __m256i v_f800 = _mm256_set1_epi16((uint16_t)0xf800); 293| 832| const __m256i v_c080 = _mm256_set1_epi16((uint16_t)0xc080); 294| 832| const __m256i v_7fffffff = _mm256_set1_epi32((uint32_t)0x7fffffff); 295| 832| const __m256i v_10ffff = _mm256_set1_epi32((uint32_t)0x10ffff); 296| | 297| 832| const size_t safety_margin = 298| 832| 12; // to avoid overruns, see issue 299| | // https://github.com/simdutf/simdutf/issues/92 300| | 301| 24.1k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (301:10): [True: 23.5k, False: 620] ------------------ 302| 23.5k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 303| 23.5k| __m256i nextin = _mm256_loadu_si256((__m256i *)buf + 1); 304| | // Check for too large input 305| 23.5k| const __m256i max_input = 306| 23.5k| _mm256_max_epu32(_mm256_max_epu32(in, nextin), v_10ffff); 307| 23.5k| if (static_cast(_mm256_movemask_epi8( ------------------ | Branch (307:9): [True: 144, False: 23.3k] ------------------ 308| 23.5k| _mm256_cmpeq_epi32(max_input, v_10ffff))) != 0xffffffff) { 309| 144| return std::make_pair(result(error_code::TOO_LARGE, buf - start), 310| 144| utf8_output); 311| 144| } 312| | 313| | // Pack 32-bit UTF-32 code units to 16-bit UTF-16 code units with unsigned 314| | // saturation 315| 23.3k| __m256i in_16 = _mm256_packus_epi32(_mm256_and_si256(in, v_7fffffff), 316| 23.3k| _mm256_and_si256(nextin, v_7fffffff)); 317| 23.3k| in_16 = _mm256_permute4x64_epi64(in_16, 0b11011000); 318| | 319| | // Try to apply UTF-16 => UTF-8 routine on 256 bits 320| | // (haswell/avx2_convert_utf16_to_utf8.cpp) 321| | 322| 23.3k| if (_mm256_testz_si256(in_16, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (322:9): [True: 20.3k, False: 3.02k] ------------------ 323| | // 1. pack the bytes 324| 20.3k| const __m128i utf8_packed = _mm_packus_epi16( 325| 20.3k| _mm256_castsi256_si128(in_16), _mm256_extractf128_si256(in_16, 1)); 326| | // 2. store (16 bytes) 327| 20.3k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 328| | // 3. adjust pointers 329| 20.3k| buf += 16; 330| 20.3k| utf8_output += 16; 331| 20.3k| continue; // we are done for this round! 332| 20.3k| } 333| | // no bits set above 7th bit 334| 3.02k| const __m256i one_byte_bytemask = 335| 3.02k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_ff80), v_0000); 336| 3.02k| const uint32_t one_byte_bitmask = 337| 3.02k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 338| | 339| | // no bits set above 11th bit 340| 3.02k| const __m256i one_or_two_bytes_bytemask = 341| 3.02k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_0000); 342| 3.02k| const uint32_t one_or_two_bytes_bitmask = 343| 3.02k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 344| 3.02k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (344:9): [True: 1.53k, False: 1.48k] ------------------ 345| | // 1. prepare 2-byte values 346| | // input 16-bit word : [0000|0aaa|aabb|bbbb] x 8 347| | // expected output : [110a|aaaa|10bb|bbbb] x 8 348| 1.53k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 349| 1.53k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 350| | 351| | // t0 = [000a|aaaa|bbbb|bb00] 352| 1.53k| const __m256i t0 = _mm256_slli_epi16(in_16, 2); 353| | // t1 = [000a|aaaa|0000|0000] 354| 1.53k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 355| | // t2 = [0000|0000|00bb|bbbb] 356| 1.53k| const __m256i t2 = _mm256_and_si256(in_16, v_003f); 357| | // t3 = [000a|aaaa|00bb|bbbb] 358| 1.53k| const __m256i t3 = _mm256_or_si256(t1, t2); 359| | // t4 = [110a|aaaa|10bb|bbbb] 360| 1.53k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 361| | 362| | // 2. merge ASCII and 2-byte codewords 363| 1.53k| const __m256i utf8_unpacked = 364| 1.53k| _mm256_blendv_epi8(t4, in_16, one_byte_bytemask); 365| | 366| | // 3. prepare bitmask for 8-bit lookup 367| 1.53k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 368| 1.53k| const uint32_t M1 = M0 >> 7; 369| 1.53k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 370| | // 4. pack the bytes 371| | 372| 1.53k| const uint8_t *row = 373| 1.53k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 374| 1.53k| const uint8_t *row_2 = 375| 1.53k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 376| 1.53k| 16)][0]; 377| | 378| 1.53k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 379| 1.53k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 380| | 381| 1.53k| const __m256i utf8_packed = _mm256_shuffle_epi8( 382| 1.53k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 1.53k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 1.53k| _mm256_castsi128_si256(xmm2), 2) ------------------ 383| | // 5. store bytes 384| 1.53k| _mm_storeu_si128((__m128i *)utf8_output, 385| 1.53k| _mm256_castsi256_si128(utf8_packed)); 386| 1.53k| utf8_output += row[0]; 387| 1.53k| _mm_storeu_si128((__m128i *)utf8_output, 388| 1.53k| _mm256_extractf128_si256(utf8_packed, 1)); 389| 1.53k| utf8_output += row_2[0]; 390| | 391| | // 6. adjust pointers 392| 1.53k| buf += 16; 393| 1.53k| continue; 394| 1.53k| } 395| | // Must check for overflow in packing 396| 1.48k| const __m256i saturation_bytemask = _mm256_cmpeq_epi32( 397| 1.48k| _mm256_and_si256(_mm256_or_si256(in, nextin), v_ffff0000), v_0000); 398| 1.48k| const uint32_t saturation_bitmask = 399| 1.48k| static_cast(_mm256_movemask_epi8(saturation_bytemask)); 400| 1.48k| if (saturation_bitmask == 0xffffffff) { ------------------ | Branch (400:9): [True: 650, False: 838] ------------------ 401| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 402| | 403| | // Check for illegal surrogate code units 404| 650| const __m256i v_d800 = _mm256_set1_epi16((uint16_t)0xd800); 405| 650| const __m256i forbidden_bytemask = 406| 650| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_d800); 407| 650| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != ------------------ | Branch (407:11): [True: 34, False: 616] ------------------ 408| 650| 0x0) { 409| 34| return std::make_pair(result(error_code::SURROGATE, buf - start), 410| 34| utf8_output); 411| 34| } 412| | 413| 616| const __m256i dup_even = _mm256_setr_epi16( 414| 616| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 415| 616| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 416| | 417| | /* In this branch we handle three cases: 418| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 419| | single UFT-8 byte 420| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - two 421| | UTF-8 bytes 422| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 423| | three UTF-8 bytes 424| | 425| | We expand the input word (16-bit) into two code units (32-bit), thus 426| | we have room for four bytes. However, we need five distinct bit 427| | layouts. Note that the last byte in cases #2 and #3 is the same. 428| | 429| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 430| | in register t2. 431| | 432| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 433| | either byte 1 for case #2 or byte 2 for case #3. Note that they 434| | differ by exactly one bit. 435| | 436| | Finally from these two code units we build proper UTF-8 sequence, taking 437| | into account the case (i.e, the number of bytes to write). 438| | */ 439| | /** 440| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 441| | * t2 => [0ccc|cccc] [10cc|cccc] 442| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 443| | */ 444| 616|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 445| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 446| 616| const __m256i t0 = _mm256_shuffle_epi8(in_16, dup_even); 447| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 448| 616| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 444| 616|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 449| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 450| 616| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 444| 616|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 451| | 452| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 453| 616| const __m256i s0 = _mm256_srli_epi16(in_16, 4); 454| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 455| 616| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 444| 616|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 456| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 457| 616| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 444| 616|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 458| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 459| 616| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 444| 616|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 460| 616| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 461| 616| simdutf_vec(0b0100000000000000)); ------------------ | | 444| 616|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 462| 616| const __m256i s4 = _mm256_xor_si256(s3, m0); 463| 616|#undef simdutf_vec 464| | 465| | // 4. expand code units 16-bit => 32-bit 466| 616| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 467| 616| const __m256i out1 = _mm256_unpackhi_epi16(t2, s4); 468| | 469| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 470| 616| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 471| 616| (one_or_two_bytes_bitmask & 0xaaaaaaaa); 472| | // Due to the wider registers, the following path is less likely to be 473| | // useful. 474| | /*if(mask == 0) { 475| | // We only have three-byte code units. Use fast path. 476| | const __m256i shuffle = 477| | _mm256_setr_epi8(2,3,1,6,7,5,10,11,9,14,15,13,-1,-1,-1,-1, 478| | 2,3,1,6,7,5,10,11,9,14,15,13,-1,-1,-1,-1); const __m256i utf8_0 = 479| | _mm256_shuffle_epi8(out0, shuffle); const __m256i utf8_1 = 480| | _mm256_shuffle_epi8(out1, shuffle); 481| | _mm_storeu_si128((__m128i*)utf8_output, _mm256_castsi256_si128(utf8_0)); 482| | utf8_output += 12; 483| | _mm_storeu_si128((__m128i*)utf8_output, _mm256_castsi256_si128(utf8_1)); 484| | utf8_output += 12; 485| | _mm_storeu_si128((__m128i*)utf8_output, 486| | _mm256_extractf128_si256(utf8_0,1)); utf8_output += 12; 487| | _mm_storeu_si128((__m128i*)utf8_output, 488| | _mm256_extractf128_si256(utf8_1,1)); utf8_output += 12; buf += 16; 489| | continue; 490| | }*/ 491| 616| const uint8_t mask0 = uint8_t(mask); 492| 616| const uint8_t *row0 = 493| 616| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 494| 616| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 495| 616| const __m128i utf8_0 = 496| 616| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 497| | 498| 616| const uint8_t mask1 = static_cast(mask >> 8); 499| 616| const uint8_t *row1 = 500| 616| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 501| 616| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 502| 616| const __m128i utf8_1 = 503| 616| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 504| | 505| 616| const uint8_t mask2 = static_cast(mask >> 16); 506| 616| const uint8_t *row2 = 507| 616| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 508| 616| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 509| 616| const __m128i utf8_2 = 510| 616| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 511| | 512| 616| const uint8_t mask3 = static_cast(mask >> 24); 513| 616| const uint8_t *row3 = 514| 616| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 515| 616| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 516| 616| const __m128i utf8_3 = 517| 616| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 518| | 519| 616| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 520| 616| utf8_output += row0[0]; 521| 616| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 522| 616| utf8_output += row1[0]; 523| 616| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 524| 616| utf8_output += row2[0]; 525| 616| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 526| 616| utf8_output += row3[0]; 527| 616| buf += 16; 528| 838| } else { 529| | // case: at least one 32-bit word is larger than 0xFFFF <=> it will 530| | // produce four UTF-8 bytes. Let us do a scalar fallback. It may seem 531| | // wasteful to use scalar code, but being efficient with SIMD may require 532| | // large, non-trivial tables? 533| 838| size_t forward = 15; 534| 838| size_t k = 0; 535| 838| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (535:11): [True: 0, False: 838] ------------------ 536| 0| forward = size_t(end - buf - 1); 537| 0| } 538| 13.1k| for (; k < forward; k++) { ------------------ | Branch (538:14): [True: 12.3k, False: 804] ------------------ 539| 12.3k| uint32_t word = buf[k]; 540| 12.3k| if ((word & 0xFFFFFF80) == 0) { // 1-byte (ASCII) ------------------ | Branch (540:13): [True: 4.71k, False: 7.59k] ------------------ 541| 4.71k| *utf8_output++ = char(word); 542| 7.59k| } else if ((word & 0xFFFFF800) == 0) { // 2-byte ------------------ | Branch (542:20): [True: 1.09k, False: 6.50k] ------------------ 543| 1.09k| *utf8_output++ = char((word >> 6) | 0b11000000); 544| 1.09k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 545| 6.50k| } else if ((word & 0xFFFF0000) == 0) { // 3-byte ------------------ | Branch (545:20): [True: 3.42k, False: 3.08k] ------------------ 546| 3.42k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (546:15): [True: 1.93k, False: 1.49k] | Branch (546:33): [True: 34, False: 1.89k] ------------------ 547| 34| return std::make_pair( 548| 34| result(error_code::SURROGATE, buf - start + k), utf8_output); 549| 34| } 550| 3.38k| *utf8_output++ = char((word >> 12) | 0b11100000); 551| 3.38k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 552| 3.38k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 553| 3.38k| } else { // 4-byte 554| 3.08k| if (word > 0x10FFFF) { ------------------ | Branch (554:15): [True: 0, False: 3.08k] ------------------ 555| 0| return std::make_pair( 556| 0| result(error_code::TOO_LARGE, buf - start + k), utf8_output); 557| 0| } 558| 3.08k| *utf8_output++ = char((word >> 18) | 0b11110000); 559| 3.08k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 560| 3.08k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 561| 3.08k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 562| 3.08k| } 563| 12.3k| } 564| 804| buf += k; 565| 804| } 566| 1.48k| } // while 567| | 568| 620| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 569| 832|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_129convert_masked_utf8_to_latin1EPKcmRPc: 9| 588k| char *&latin1_output) { 10| | // we use an approach where we try to process up to 12 input bytes. 11| | // Why 12 input bytes and not 16? Because we are concerned with the size of 12| | // the lookup tables. Also 12 is nicely divisible by two and three. 13| | // 14| | // 15| | // Optimization note: our main path below is load-latency dependent. Thus it 16| | // is maybe beneficial to have fast paths that depend on branch prediction but 17| | // have less latency. This results in more instructions but, potentially, also 18| | // higher speeds. 19| | // 20| 588k| const __m128i in = _mm_loadu_si128((__m128i *)input); 21| | 22| 588k| const uint16_t input_utf8_end_of_code_point_mask = 23| 588k| utf8_end_of_code_point_mask & 24| 588k| 0xfff; // we are only processing 12 bytes in case it is not all ASCII 25| | 26| 588k| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (26:7): [True: 3.61k, False: 585k] ------------------ 27| | // We process the data in chunks of 12 bytes. 28| 3.61k| _mm_storeu_si128(reinterpret_cast<__m128i *>(latin1_output), in); 29| 3.61k| latin1_output += 12; // We wrote 12 characters. 30| 3.61k| return 12; // We consumed 1 bytes. 31| 3.61k| } 32| | /// We do not have a fast path available, so we fallback. 33| 585k| const uint8_t idx = 34| 585k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 35| 585k| const uint8_t consumed = 36| 585k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 37| | // this indicates an invalid input: 38| 585k| if (idx >= 64) { ------------------ | Branch (38:7): [True: 30.3k, False: 554k] ------------------ 39| 30.3k| return consumed; 40| 30.3k| } 41| | // Here we should have (idx < 64), if not, there is a bug in the validation or 42| | // elsewhere. SIX (6) input code-code units this is a relatively easy scenario 43| | // we process SIX (6) input code-code units. The max length in bytes of six 44| | // code code units spanning between 1 and 2 bytes each is 12 bytes. On 45| | // processors where pdep/pext is fast, we might be able to use a small lookup 46| | // table. 47| 554k| const __m128i sh = 48| 554k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 49| 554k| const __m128i perm = _mm_shuffle_epi8(in, sh); 50| 554k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 51| 554k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 52| 554k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 53| 554k| const __m128i latin1_packed = _mm_packus_epi16(composed, composed); 54| | // writing 8 bytes even though we only care about the first 6 bytes. 55| | // performance note: it would be faster to use _mm_storeu_si128, we should 56| | // investigate. 57| 554k| _mm_storel_epi64((__m128i *)latin1_output, latin1_packed); 58| 554k| latin1_output += 6; // We wrote 6 bytes. 59| 554k| return consumed; 60| 585k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE0EEEmPKcmRPDs: 10| 3.17M| char16_t *&utf16_output) { 11| | // we use an approach where we try to process up to 12 input bytes. 12| | // Why 12 input bytes and not 16? Because we are concerned with the size of 13| | // the lookup tables. Also 12 is nicely divisible by two and three. 14| | // 15| | // 16| | // Optimization note: our main path below is load-latency dependent. Thus it 17| | // is maybe beneficial to have fast paths that depend on branch prediction but 18| | // have less latency. This results in more instructions but, potentially, also 19| | // higher speeds. 20| | // 21| | // We first try a few fast paths. 22| 3.17M| const __m128i swap = 23| 3.17M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 3.17M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 3.17M| const uint16_t input_utf8_end_of_code_point_mask = 26| 3.17M| utf8_end_of_code_point_mask & 0xfff; 27| 3.17M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 10.8k, False: 3.16M] ------------------ 28| | // We process the data in chunks of 12 bytes. 29| 10.8k| __m256i ascii = _mm256_cvtepu8_epi16(in); 30| 10.8k| if (big_endian) { ------------------ | Branch (30:9): [Folded, False: 10.8k] ------------------ 31| 0| const __m256i swap256 = _mm256_setr_epi8( 32| 0| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 33| 0| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 34| 0| ascii = _mm256_shuffle_epi8(ascii, swap256); 35| 0| } 36| 10.8k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output), ascii); 37| 10.8k| utf16_output += 12; // We wrote 12 16-bit characters. 38| 10.8k| return 12; // We consumed 12 bytes. 39| 10.8k| } 40| 3.16M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (40:7): [True: 15.9k, False: 3.15M] ------------------ 41| | // We want to take 8 2-byte UTF-8 code units and turn them into 8 2-byte 42| | // UTF-16 code units. There is probably a more efficient sequence, but the 43| | // following might do. 44| 15.9k| const __m128i sh = 45| 15.9k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 46| 15.9k| const __m128i perm = _mm_shuffle_epi8(in, sh); 47| 15.9k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 48| 15.9k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 49| 15.9k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 50| 15.9k| if (big_endian) ------------------ | Branch (50:9): [Folded, False: 15.9k] ------------------ 51| 0| composed = _mm_shuffle_epi8(composed, swap); 52| 15.9k| _mm_storeu_si128((__m128i *)utf16_output, composed); 53| 15.9k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 54| 15.9k| return 16; 55| 15.9k| } 56| 3.15M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (56:7): [True: 2.28k, False: 3.14M] ------------------ 57| | // We want to take 4 3-byte UTF-8 code units and turn them into 4 2-byte 58| | // UTF-16 code units. There is probably a more efficient sequence, but the 59| | // following might do. 60| 2.28k| const __m128i sh = 61| 2.28k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 62| 2.28k| const __m128i perm = _mm_shuffle_epi8(in, sh); 63| 2.28k| const __m128i ascii = 64| 2.28k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 65| 2.28k| const __m128i middlebyte = 66| 2.28k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 67| 2.28k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 68| 2.28k| const __m128i highbyte = 69| 2.28k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 70| 2.28k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 71| 2.28k| const __m128i composed = 72| 2.28k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 73| 2.28k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 74| 2.28k| if (big_endian) ------------------ | Branch (74:9): [Folded, False: 2.28k] ------------------ 75| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 76| 2.28k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 77| 2.28k| utf16_output += 4; 78| 2.28k| return 12; 79| 2.28k| } 80| | 81| 3.14M| const uint8_t idx = simdutf::tables::utf8_to_utf16::utf8bigindex 82| 3.14M| [input_utf8_end_of_code_point_mask][0]; 83| 3.14M| const uint8_t consumed = simdutf::tables::utf8_to_utf16::utf8bigindex 84| 3.14M| [input_utf8_end_of_code_point_mask][1]; 85| 3.14M| if (idx < 64) { ------------------ | Branch (85:7): [True: 2.64M, False: 500k] ------------------ 86| | // SIX (6) input code-code units 87| | // this is a relatively easy scenario 88| | // we process SIX (6) input code-code units. The max length in bytes of six 89| | // code code units spanning between 1 and 2 bytes each is 12 bytes. On 90| | // processors where pdep/pext is fast, we might be able to use a small 91| | // lookup table. 92| 2.64M| const __m128i sh = _mm_loadu_si128( 93| 2.64M| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 94| 2.64M| const __m128i perm = _mm_shuffle_epi8(in, sh); 95| 2.64M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 96| 2.64M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 97| 2.64M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 98| 2.64M| if (big_endian) ------------------ | Branch (98:9): [Folded, False: 2.64M] ------------------ 99| 0| composed = _mm_shuffle_epi8(composed, swap); 100| 2.64M| _mm_storeu_si128((__m128i *)utf16_output, composed); 101| 2.64M| utf16_output += 6; // We wrote 12 bytes, 6 code points. There is a potential 102| | // overflow of 4 bytes. 103| 2.64M| } else if (idx < 145) { ------------------ | Branch (103:14): [True: 434k, False: 66.2k] ------------------ 104| | // FOUR (4) input code-code units 105| 434k| const __m128i sh = _mm_loadu_si128( 106| 434k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 107| 434k| const __m128i perm = _mm_shuffle_epi8(in, sh); 108| 434k| const __m128i ascii = 109| 434k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 110| 434k| const __m128i middlebyte = 111| 434k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 112| 434k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 113| 434k| const __m128i highbyte = 114| 434k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 115| 434k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 116| 434k| const __m128i composed = 117| 434k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 118| 434k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 119| 434k| if (big_endian) ------------------ | Branch (119:9): [Folded, False: 434k] ------------------ 120| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 121| 434k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 122| 434k| utf16_output += 4; // Here we overflow by 8 bytes. 123| 434k| } else if (idx < 209) { ------------------ | Branch (123:14): [True: 65.5k, False: 712] ------------------ 124| | // TWO (2) input code-code units 125| | ////////////// 126| | // There might be garbage inputs where a leading byte mascarades as a 127| | // four-byte leading byte (by being followed by 3 continuation byte), but is 128| | // not greater than 0xf0. This could trigger a buffer overflow if we only 129| | // counted leading bytes of the form 0xf0 as generating surrogate pairs, 130| | // without further UTF-8 validation. Thus we must be careful to ensure that 131| | // only leading bytes at least as large as 0xf0 generate surrogate pairs. We 132| | // do as at the cost of an extra mask. 133| | ///////////// 134| 65.5k| const __m128i sh = _mm_loadu_si128( 135| 65.5k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 136| 65.5k| const __m128i perm = _mm_shuffle_epi8(in, sh); 137| 65.5k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 138| 65.5k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 139| 65.5k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 140| 65.5k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 141| | // correct for spurious high bit 142| 65.5k| const __m128i correct = 143| 65.5k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 144| 65.5k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 145| 65.5k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 146| | // We deliberately carry the leading four bits in highbyte if they are 147| | // present, we remove them later when computing hightenbits. 148| 65.5k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 149| 65.5k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 150| | // When we need to generate a surrogate pair (leading byte > 0xF0), then 151| | // the corresponding 32-bit value in 'composed' will be greater than 152| | // > (0xff00000>>6) or > 0x3c00000. This can be used later to identify the 153| | // location of the surrogate pairs. 154| 65.5k| const __m128i composed = 155| 65.5k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 156| 65.5k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 157| 65.5k| const __m128i composedminus = 158| 65.5k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 159| 65.5k| const __m128i lowtenbits = 160| 65.5k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 161| | // Notice the 0x3ff mask: 162| 65.5k| const __m128i hightenbits = 163| 65.5k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 164| 65.5k| const __m128i lowtenbitsadd = 165| 65.5k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 166| 65.5k| const __m128i hightenbitsadd = 167| 65.5k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 168| 65.5k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 169| 65.5k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 170| 65.5k| uint32_t basic_buffer[4]; 171| 65.5k| uint32_t basic_buffer_swap[4]; 172| 65.5k| if (big_endian) { ------------------ | Branch (172:9): [Folded, False: 65.5k] ------------------ 173| 0| _mm_storeu_si128((__m128i *)basic_buffer_swap, 174| 0| _mm_shuffle_epi8(composed, swap)); 175| 0| surrogates = _mm_shuffle_epi8(surrogates, swap); 176| 0| } 177| 65.5k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 178| 65.5k| uint32_t surrogate_buffer[4]; 179| 65.5k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 180| 262k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (180:24): [True: 196k, False: 65.5k] ------------------ 181| 196k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (181:11): [True: 54.1k, False: 142k] ------------------ 182| 54.1k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 183| 54.1k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 184| 54.1k| utf16_output += 2; 185| 142k| } else { 186| 142k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (186:27): [Folded, False: 142k] ------------------ 187| 142k| : uint16_t(basic_buffer[i]); 188| 142k| utf16_output++; 189| 142k| } 190| 196k| } 191| 65.5k| } else { 192| | // here we know that there is an error but we do not handle errors 193| 712| } 194| 3.14M| return consumed; 195| 3.15M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE1EEEmPKcmRPDs: 10| 3.00M| char16_t *&utf16_output) { 11| | // we use an approach where we try to process up to 12 input bytes. 12| | // Why 12 input bytes and not 16? Because we are concerned with the size of 13| | // the lookup tables. Also 12 is nicely divisible by two and three. 14| | // 15| | // 16| | // Optimization note: our main path below is load-latency dependent. Thus it 17| | // is maybe beneficial to have fast paths that depend on branch prediction but 18| | // have less latency. This results in more instructions but, potentially, also 19| | // higher speeds. 20| | // 21| | // We first try a few fast paths. 22| 3.00M| const __m128i swap = 23| 3.00M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 3.00M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 3.00M| const uint16_t input_utf8_end_of_code_point_mask = 26| 3.00M| utf8_end_of_code_point_mask & 0xfff; 27| 3.00M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 9.13k, False: 2.99M] ------------------ 28| | // We process the data in chunks of 12 bytes. 29| 9.13k| __m256i ascii = _mm256_cvtepu8_epi16(in); 30| 9.13k| if (big_endian) { ------------------ | Branch (30:9): [True: 9.13k, Folded] ------------------ 31| 9.13k| const __m256i swap256 = _mm256_setr_epi8( 32| 9.13k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 33| 9.13k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 34| 9.13k| ascii = _mm256_shuffle_epi8(ascii, swap256); 35| 9.13k| } 36| 9.13k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output), ascii); 37| 9.13k| utf16_output += 12; // We wrote 12 16-bit characters. 38| 9.13k| return 12; // We consumed 12 bytes. 39| 9.13k| } 40| 2.99M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (40:7): [True: 23.5k, False: 2.96M] ------------------ 41| | // We want to take 8 2-byte UTF-8 code units and turn them into 8 2-byte 42| | // UTF-16 code units. There is probably a more efficient sequence, but the 43| | // following might do. 44| 23.5k| const __m128i sh = 45| 23.5k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 46| 23.5k| const __m128i perm = _mm_shuffle_epi8(in, sh); 47| 23.5k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 48| 23.5k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 49| 23.5k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 50| 23.5k| if (big_endian) ------------------ | Branch (50:9): [True: 23.5k, Folded] ------------------ 51| 23.5k| composed = _mm_shuffle_epi8(composed, swap); 52| 23.5k| _mm_storeu_si128((__m128i *)utf16_output, composed); 53| 23.5k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 54| 23.5k| return 16; 55| 23.5k| } 56| 2.96M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (56:7): [True: 5.17k, False: 2.96M] ------------------ 57| | // We want to take 4 3-byte UTF-8 code units and turn them into 4 2-byte 58| | // UTF-16 code units. There is probably a more efficient sequence, but the 59| | // following might do. 60| 5.17k| const __m128i sh = 61| 5.17k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 62| 5.17k| const __m128i perm = _mm_shuffle_epi8(in, sh); 63| 5.17k| const __m128i ascii = 64| 5.17k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 65| 5.17k| const __m128i middlebyte = 66| 5.17k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 67| 5.17k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 68| 5.17k| const __m128i highbyte = 69| 5.17k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 70| 5.17k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 71| 5.17k| const __m128i composed = 72| 5.17k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 73| 5.17k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 74| 5.17k| if (big_endian) ------------------ | Branch (74:9): [True: 5.17k, Folded] ------------------ 75| 5.17k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 76| 5.17k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 77| 5.17k| utf16_output += 4; 78| 5.17k| return 12; 79| 5.17k| } 80| | 81| 2.96M| const uint8_t idx = simdutf::tables::utf8_to_utf16::utf8bigindex 82| 2.96M| [input_utf8_end_of_code_point_mask][0]; 83| 2.96M| const uint8_t consumed = simdutf::tables::utf8_to_utf16::utf8bigindex 84| 2.96M| [input_utf8_end_of_code_point_mask][1]; 85| 2.96M| if (idx < 64) { ------------------ | Branch (85:7): [True: 2.67M, False: 286k] ------------------ 86| | // SIX (6) input code-code units 87| | // this is a relatively easy scenario 88| | // we process SIX (6) input code-code units. The max length in bytes of six 89| | // code code units spanning between 1 and 2 bytes each is 12 bytes. On 90| | // processors where pdep/pext is fast, we might be able to use a small 91| | // lookup table. 92| 2.67M| const __m128i sh = _mm_loadu_si128( 93| 2.67M| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 94| 2.67M| const __m128i perm = _mm_shuffle_epi8(in, sh); 95| 2.67M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 96| 2.67M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 97| 2.67M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 98| 2.67M| if (big_endian) ------------------ | Branch (98:9): [True: 2.67M, Folded] ------------------ 99| 2.67M| composed = _mm_shuffle_epi8(composed, swap); 100| 2.67M| _mm_storeu_si128((__m128i *)utf16_output, composed); 101| 2.67M| utf16_output += 6; // We wrote 12 bytes, 6 code points. There is a potential 102| | // overflow of 4 bytes. 103| 2.67M| } else if (idx < 145) { ------------------ | Branch (103:14): [True: 221k, False: 65.3k] ------------------ 104| | // FOUR (4) input code-code units 105| 221k| const __m128i sh = _mm_loadu_si128( 106| 221k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 107| 221k| const __m128i perm = _mm_shuffle_epi8(in, sh); 108| 221k| const __m128i ascii = 109| 221k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 110| 221k| const __m128i middlebyte = 111| 221k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 112| 221k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 113| 221k| const __m128i highbyte = 114| 221k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 115| 221k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 116| 221k| const __m128i composed = 117| 221k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 118| 221k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 119| 221k| if (big_endian) ------------------ | Branch (119:9): [True: 221k, Folded] ------------------ 120| 221k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 121| 221k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 122| 221k| utf16_output += 4; // Here we overflow by 8 bytes. 123| 221k| } else if (idx < 209) { ------------------ | Branch (123:14): [True: 64.0k, False: 1.31k] ------------------ 124| | // TWO (2) input code-code units 125| | ////////////// 126| | // There might be garbage inputs where a leading byte mascarades as a 127| | // four-byte leading byte (by being followed by 3 continuation byte), but is 128| | // not greater than 0xf0. This could trigger a buffer overflow if we only 129| | // counted leading bytes of the form 0xf0 as generating surrogate pairs, 130| | // without further UTF-8 validation. Thus we must be careful to ensure that 131| | // only leading bytes at least as large as 0xf0 generate surrogate pairs. We 132| | // do as at the cost of an extra mask. 133| | ///////////// 134| 64.0k| const __m128i sh = _mm_loadu_si128( 135| 64.0k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 136| 64.0k| const __m128i perm = _mm_shuffle_epi8(in, sh); 137| 64.0k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 138| 64.0k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 139| 64.0k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 140| 64.0k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 141| | // correct for spurious high bit 142| 64.0k| const __m128i correct = 143| 64.0k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 144| 64.0k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 145| 64.0k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 146| | // We deliberately carry the leading four bits in highbyte if they are 147| | // present, we remove them later when computing hightenbits. 148| 64.0k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 149| 64.0k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 150| | // When we need to generate a surrogate pair (leading byte > 0xF0), then 151| | // the corresponding 32-bit value in 'composed' will be greater than 152| | // > (0xff00000>>6) or > 0x3c00000. This can be used later to identify the 153| | // location of the surrogate pairs. 154| 64.0k| const __m128i composed = 155| 64.0k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 156| 64.0k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 157| 64.0k| const __m128i composedminus = 158| 64.0k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 159| 64.0k| const __m128i lowtenbits = 160| 64.0k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 161| | // Notice the 0x3ff mask: 162| 64.0k| const __m128i hightenbits = 163| 64.0k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 164| 64.0k| const __m128i lowtenbitsadd = 165| 64.0k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 166| 64.0k| const __m128i hightenbitsadd = 167| 64.0k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 168| 64.0k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 169| 64.0k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 170| 64.0k| uint32_t basic_buffer[4]; 171| 64.0k| uint32_t basic_buffer_swap[4]; 172| 64.0k| if (big_endian) { ------------------ | Branch (172:9): [True: 64.0k, Folded] ------------------ 173| 64.0k| _mm_storeu_si128((__m128i *)basic_buffer_swap, 174| 64.0k| _mm_shuffle_epi8(composed, swap)); 175| 64.0k| surrogates = _mm_shuffle_epi8(surrogates, swap); 176| 64.0k| } 177| 64.0k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 178| 64.0k| uint32_t surrogate_buffer[4]; 179| 64.0k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 180| 256k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (180:24): [True: 192k, False: 64.0k] ------------------ 181| 192k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (181:11): [True: 53.9k, False: 138k] ------------------ 182| 53.9k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 183| 53.9k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 184| 53.9k| utf16_output += 2; 185| 138k| } else { 186| 138k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (186:27): [True: 138k, Folded] ------------------ 187| 138k| : uint16_t(basic_buffer[i]); 188| 138k| utf16_output++; 189| 138k| } 190| 192k| } 191| 64.0k| } else { 192| | // here we know that there is an error but we do not handle errors 193| 1.31k| } 194| 2.96M| return consumed; 195| 2.96M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128convert_masked_utf8_to_utf32EPKcmRPDi: 9| 2.58M| char32_t *&utf32_output) { 10| | // we use an approach where we try to process up to 12 input bytes. 11| | // Why 12 input bytes and not 16? Because we are concerned with the size of 12| | // the lookup tables. Also 12 is nicely divisible by two and three. 13| | // 14| | // 15| | // Optimization note: our main path below is load-latency dependent. Thus it 16| | // is maybe beneficial to have fast paths that depend on branch prediction but 17| | // have less latency. This results in more instructions but, potentially, also 18| | // higher speeds. 19| | // 20| | // We first try a few fast paths. 21| 2.58M| const __m128i in = _mm_loadu_si128((__m128i *)input); 22| 2.58M| const uint16_t input_utf8_end_of_code_point_mask = 23| 2.58M| utf8_end_of_code_point_mask & 0xfff; 24| 2.58M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (24:7): [True: 7.57k, False: 2.57M] ------------------ 25| | // We process the data in chunks of 12 bytes. 26| 7.57k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 27| 7.57k| _mm256_cvtepu8_epi32(in)); 28| 7.57k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output + 8), 29| 7.57k| _mm256_cvtepu8_epi32(_mm_srli_si128(in, 8))); 30| 7.57k| utf32_output += 12; // We wrote 12 32-bit characters. 31| 7.57k| return 12; // We consumed 12 bytes. 32| 7.57k| } 33| 2.57M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (33:7): [True: 17.0k, False: 2.56M] ------------------ 34| | // We want to take 8 2-byte UTF-8 code units and turn them into 8 4-byte 35| | // UTF-32 code units. There is probably a more efficient sequence, but the 36| | // following might do. 37| 17.0k| const __m128i sh = 38| 17.0k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 39| 17.0k| const __m128i perm = _mm_shuffle_epi8(in, sh); 40| 17.0k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 41| 17.0k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 42| 17.0k| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 43| 17.0k| _mm256_storeu_si256((__m256i *)utf32_output, 44| 17.0k| _mm256_cvtepu16_epi32(composed)); 45| 17.0k| utf32_output += 8; // We wrote 16 bytes, 8 code points. 46| 17.0k| return 16; 47| 17.0k| } 48| 2.56M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (48:7): [True: 2.91k, False: 2.55M] ------------------ 49| | // We want to take 4 3-byte UTF-8 code units and turn them into 4 4-byte 50| | // UTF-32 code units. There is probably a more efficient sequence, but the 51| | // following might do. 52| 2.91k| const __m128i sh = 53| 2.91k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 54| 2.91k| const __m128i perm = _mm_shuffle_epi8(in, sh); 55| 2.91k| const __m128i ascii = 56| 2.91k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 57| 2.91k| const __m128i middlebyte = 58| 2.91k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 59| 2.91k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 60| 2.91k| const __m128i highbyte = 61| 2.91k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 62| 2.91k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 63| 2.91k| const __m128i composed = 64| 2.91k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 65| 2.91k| _mm_storeu_si128((__m128i *)utf32_output, composed); 66| 2.91k| utf32_output += 4; 67| 2.91k| return 12; 68| 2.91k| } 69| | /// We do not have a fast path available, so we fallback. 70| | 71| 2.55M| const uint8_t idx = 72| 2.55M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 73| 2.55M| const uint8_t consumed = 74| 2.55M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 75| 2.55M| if (idx < 64) { ------------------ | Branch (75:7): [True: 2.25M, False: 304k] ------------------ 76| | // SIX (6) input code-code units 77| | // this is a relatively easy scenario 78| | // we process SIX (6) input code-code units. The max length in bytes of six 79| | // code code units spanning between 1 and 2 bytes each is 12 bytes. On 80| | // processors where pdep/pext is fast, we might be able to use a small 81| | // lookup table. 82| 2.25M| const __m128i sh = 83| 2.25M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 84| 2.25M| const __m128i perm = _mm_shuffle_epi8(in, sh); 85| 2.25M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 86| 2.25M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 87| 2.25M| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 88| 2.25M| _mm256_storeu_si256((__m256i *)utf32_output, 89| 2.25M| _mm256_cvtepu16_epi32(composed)); 90| 2.25M| utf32_output += 6; // We wrote 24 bytes, 6 code points. There is a potential 91| | // overflow of 32 - 24 = 8 bytes. 92| 2.25M| } else if (idx < 145) { ------------------ | Branch (92:14): [True: 253k, False: 51.8k] ------------------ 93| | // FOUR (4) input code-code units 94| 253k| const __m128i sh = 95| 253k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 96| 253k| const __m128i perm = _mm_shuffle_epi8(in, sh); 97| 253k| const __m128i ascii = 98| 253k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 99| 253k| const __m128i middlebyte = 100| 253k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 101| 253k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 102| 253k| const __m128i highbyte = 103| 253k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 104| 253k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 105| 253k| const __m128i composed = 106| 253k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 107| 253k| _mm_storeu_si128((__m128i *)utf32_output, composed); 108| 253k| utf32_output += 4; 109| 253k| } else if (idx < 209) { ------------------ | Branch (109:14): [True: 51.1k, False: 668] ------------------ 110| | // TWO (2) input code-code units 111| 51.1k| const __m128i sh = 112| 51.1k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 113| 51.1k| const __m128i perm = _mm_shuffle_epi8(in, sh); 114| 51.1k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 115| 51.1k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 116| 51.1k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 117| 51.1k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 118| | // correct for spurious high bit 119| 51.1k| const __m128i correct = 120| 51.1k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 121| 51.1k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 122| 51.1k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 123| 51.1k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0x07000000)); 124| 51.1k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 125| 51.1k| const __m128i composed = 126| 51.1k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 127| 51.1k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 128| 51.1k| _mm_storeu_si128((__m128i *)utf32_output, composed); 129| 51.1k| utf32_output += 130| 51.1k| 3; // We wrote 3 * 4 bytes, there is a potential overflow of 4 bytes. 131| 51.1k| } else { 132| | // here we know that there is an error but we do not handle errors 133| 668| } 134| 2.55M| return consumed; 135| 2.56M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf16_gather_high_bytesILNS_10endiannessE0EEENS1_4simd5simd8IhEERKNS5_6simd16ItEESB_: 3| 2.54M| const simd16 &in1) { 4| 2.54M| if (big_endian) { ------------------ | Branch (4:7): [Folded, False: 2.54M] ------------------ 5| | // we want lower bytes 6| 0| const auto mask = simd16(0x00ff); 7| 0| const auto t0 = in0 & mask; 8| 0| const auto t1 = in1 & mask; 9| | 10| 0| return simd16::pack(t0, t1); 11| 2.54M| } else { 12| 2.54M| const auto t0 = in0.shr<8>(); 13| 2.54M| const auto t1 = in1.shr<8>(); 14| | 15| 2.54M| return simd16::pack(t0, t1); 16| 2.54M| } 17| 2.54M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf16_gather_high_bytesILNS_10endiannessE1EEENS1_4simd5simd8IhEERKNS5_6simd16ItEESB_: 3| 2.61M| const simd16 &in1) { 4| 2.61M| if (big_endian) { ------------------ | Branch (4:7): [True: 2.61M, Folded] ------------------ 5| | // we want lower bytes 6| 2.61M| const auto mask = simd16(0x00ff); 7| 2.61M| const auto t0 = in0 & mask; 8| 2.61M| const auto t1 = in1 & mask; 9| | 10| 2.61M| return simd16::pack(t0, t1); 11| 2.61M| } else { 12| 0| const auto t0 = in0.shr<8>(); 13| 0| const auto t1 = in1.shr<8>(); 14| | 15| 0| return simd16::pack(t0, t1); 16| 0| } 17| 2.61M|} _ZNK7simdutf7haswell14implementation13validate_utf8EPKcm: 266| 3.48k|implementation::validate_utf8(const char *buf, size_t len) const noexcept { 267| 3.48k| return haswell::utf8_validation::generic_validate_utf8(buf, len); 268| 3.48k|} _ZNK7simdutf7haswell14implementation25validate_utf8_with_errorsEPKcm: 273| 3.48k| const char *buf, size_t len) const noexcept { 274| 3.48k| return haswell::utf8_validation::generic_validate_utf8_with_errors(buf, len); 275| 3.48k|} _ZNK7simdutf7haswell14implementation16validate_utf16leEPKDsm: 312| 1.35k| size_t len) const noexcept { 313| 1.35k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.35k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 17, False: 1.33k] | | ------------------ ------------------ 314| | // empty input is valid UTF-16. protect the implementation from 315| | // handling nullptr 316| 17| return true; 317| 17| } 318| 1.33k| const auto res = 319| 1.33k| haswell::utf16::validate_utf16_with_errors(buf, len); 320| 1.33k| if (res.is_err()) { ------------------ | Branch (320:7): [True: 88, False: 1.25k] ------------------ 321| 88| return false; 322| 88| } 323| | 324| 1.25k| if (res.count == len) { ------------------ | Branch (324:7): [True: 0, False: 1.25k] ------------------ 325| 0| return true; 326| 0| } 327| | 328| 1.25k| return scalar::utf16::validate(buf + res.count, 329| 1.25k| len - res.count); 330| 1.25k|} _ZNK7simdutf7haswell14implementation16validate_utf16beEPKDsm: 336| 1.33k| size_t len) const noexcept { 337| 1.33k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.33k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 1.32k] | | ------------------ ------------------ 338| | // empty input is valid UTF-16. protect the implementation from 339| | // handling nullptr 340| 16| return true; 341| 16| } 342| 1.32k| const auto res = 343| 1.32k| haswell::utf16::validate_utf16_with_errors(buf, len); 344| 1.32k| if (res.is_err()) { ------------------ | Branch (344:7): [True: 102, False: 1.22k] ------------------ 345| 102| return false; 346| 102| } 347| | 348| 1.22k| if (res.count == len) { ------------------ | Branch (348:7): [True: 0, False: 1.22k] ------------------ 349| 0| return true; 350| 0| } 351| | 352| 1.22k| return scalar::utf16::validate(buf + res.count, 353| 1.22k| len - res.count); 354| 1.22k|} _ZNK7simdutf7haswell14implementation28validate_utf16le_with_errorsEPKDsm: 357| 1.35k| const char16_t *buf, size_t len) const noexcept { 358| | 359| 1.35k| const result res = 360| 1.35k| haswell::utf16::validate_utf16_with_errors(buf, len); 361| 1.35k| if (res.count != len) { ------------------ | Branch (361:7): [True: 1.33k, False: 17] ------------------ 362| 1.33k| const result scalar_res = 363| 1.33k| scalar::utf16::validate_with_errors( 364| 1.33k| buf + res.count, len - res.count); 365| 1.33k| return result(scalar_res.error, res.count + scalar_res.count); 366| 1.33k| } else { 367| 17| return res; 368| 17| } 369| 1.35k|} _ZNK7simdutf7haswell14implementation28validate_utf16be_with_errorsEPKDsm: 372| 1.33k| const char16_t *buf, size_t len) const noexcept { 373| 1.33k| const result res = 374| 1.33k| haswell::utf16::validate_utf16_with_errors(buf, len); 375| 1.33k| if (res.count != len) { ------------------ | Branch (375:7): [True: 1.32k, False: 16] ------------------ 376| 1.32k| const result scalar_res = 377| 1.32k| scalar::utf16::validate_with_errors(buf + res.count, 378| 1.32k| len - res.count); 379| 1.32k| return result(scalar_res.error, res.count + scalar_res.count); 380| 1.32k| } else { 381| 16| return res; 382| 16| } 383| 1.33k|} _ZNK7simdutf7haswell14implementation14validate_utf32EPKDim: 398| 2.63k|implementation::validate_utf32(const char32_t *buf, size_t len) const noexcept { 399| 2.63k| return utf32::validate(buf, len); 400| 2.63k|} _ZNK7simdutf7haswell14implementation26validate_utf32_with_errorsEPKDim: 405| 2.63k| const char32_t *buf, size_t len) const noexcept { 406| 2.63k| return utf32::validate_with_errors(buf, len); 407| 2.63k|} _ZNK7simdutf7haswell14implementation22convert_latin1_to_utf8EPKcmPc: 412| 604| const char *buf, size_t len, char *utf8_output) const noexcept { 413| 604| std::pair ret = 414| 604| avx2_convert_latin1_to_utf8(buf, len, utf8_output); 415| 604| size_t converted_chars = ret.second - utf8_output; 416| | 417| 604| if (ret.first != buf + len) { ------------------ | Branch (417:7): [True: 600, False: 4] ------------------ 418| 600| const size_t scalar_converted_chars = scalar::latin1_to_utf8::convert( 419| 600| ret.first, len - (ret.first - buf), ret.second); 420| 600| converted_chars += scalar_converted_chars; 421| 600| } 422| | 423| 604| return converted_chars; 424| 604|} _ZNK7simdutf7haswell14implementation25convert_latin1_to_utf16leEPKcmPDs: 429| 76| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 430| 76| std::pair ret = 431| 76| avx2_convert_latin1_to_utf16(buf, len, utf16_output); 432| 76| if (ret.first == nullptr) { ------------------ | Branch (432:7): [True: 4, False: 72] ------------------ 433| 4| return 0; 434| 4| } 435| 72| size_t converted_chars = ret.second - utf16_output; 436| 72| if (ret.first != buf + len) { ------------------ | Branch (436:7): [True: 58, False: 14] ------------------ 437| 58| const size_t scalar_converted_chars = 438| 58| scalar::latin1_to_utf16::convert( 439| 58| ret.first, len - (ret.first - buf), ret.second); 440| 58| if (scalar_converted_chars == 0) { ------------------ | Branch (440:9): [True: 0, False: 58] ------------------ 441| 0| return 0; 442| 0| } 443| 58| converted_chars += scalar_converted_chars; 444| 58| } 445| 72| return converted_chars; 446| 72|} _ZNK7simdutf7haswell14implementation25convert_latin1_to_utf16beEPKcmPDs: 449| 96| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 450| 96| std::pair ret = 451| 96| avx2_convert_latin1_to_utf16(buf, len, utf16_output); 452| 96| if (ret.first == nullptr) { ------------------ | Branch (452:7): [True: 4, False: 92] ------------------ 453| 4| return 0; 454| 4| } 455| 92| size_t converted_chars = ret.second - utf16_output; 456| 92| if (ret.first != buf + len) { ------------------ | Branch (456:7): [True: 78, False: 14] ------------------ 457| 78| const size_t scalar_converted_chars = 458| 78| scalar::latin1_to_utf16::convert( 459| 78| ret.first, len - (ret.first - buf), ret.second); 460| 78| if (scalar_converted_chars == 0) { ------------------ | Branch (460:9): [True: 0, False: 78] ------------------ 461| 0| return 0; 462| 0| } 463| 78| converted_chars += scalar_converted_chars; 464| 78| } 465| 92| return converted_chars; 466| 92|} _ZNK7simdutf7haswell14implementation23convert_latin1_to_utf32EPKcmPDi: 471| 124| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 472| 124| std::pair ret = 473| 124| avx2_convert_latin1_to_utf32(buf, len, utf32_output); 474| 124| if (ret.first == nullptr) { ------------------ | Branch (474:7): [True: 4, False: 120] ------------------ 475| 4| return 0; 476| 4| } 477| 120| size_t converted_chars = ret.second - utf32_output; 478| 120| if (ret.first != buf + len) { ------------------ | Branch (478:7): [True: 74, False: 46] ------------------ 479| 74| const size_t scalar_converted_chars = scalar::latin1_to_utf32::convert( 480| 74| ret.first, len - (ret.first - buf), ret.second); 481| 74| if (scalar_converted_chars == 0) { ------------------ | Branch (481:9): [True: 0, False: 74] ------------------ 482| 0| return 0; 483| 0| } 484| 74| converted_chars += scalar_converted_chars; 485| 74| } 486| 120| return converted_chars; 487| 120|} _ZNK7simdutf7haswell14implementation22convert_utf8_to_latin1EPKcmPc: 492| 586| const char *buf, size_t len, char *latin1_output) const noexcept { 493| 586| utf8_to_latin1::validating_transcoder converter; 494| 586| return converter.convert(buf, len, latin1_output); 495| 586|} _ZNK7simdutf7haswell14implementation34convert_utf8_to_latin1_with_errorsEPKcmPc: 498| 512| const char *buf, size_t len, char *latin1_output) const noexcept { 499| 512| utf8_to_latin1::validating_transcoder converter; 500| 512| return converter.convert_with_errors(buf, len, latin1_output); 501| 512|} _ZNK7simdutf7haswell14implementation23convert_utf8_to_utf16leEPKcmPDs: 511| 776| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 512| 776| utf8_to_utf16::validating_transcoder converter; 513| 776| return converter.convert(buf, len, utf16_output); 514| 776|} _ZNK7simdutf7haswell14implementation23convert_utf8_to_utf16beEPKcmPDs: 517| 794| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 518| 794| utf8_to_utf16::validating_transcoder converter; 519| 794| return converter.convert(buf, len, utf16_output); 520| 794|} _ZNK7simdutf7haswell14implementation35convert_utf8_to_utf16le_with_errorsEPKcmPDs: 523| 610| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 524| 610| utf8_to_utf16::validating_transcoder converter; 525| 610| return converter.convert_with_errors(buf, len, 526| 610| utf16_output); 527| 610|} _ZNK7simdutf7haswell14implementation35convert_utf8_to_utf16be_with_errorsEPKcmPDs: 530| 706| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 531| 706| utf8_to_utf16::validating_transcoder converter; 532| 706| return converter.convert_with_errors(buf, len, utf16_output); 533| 706|} _ZNK7simdutf7haswell14implementation29convert_valid_utf8_to_utf16leEPKcmPDs: 536| 450| const char *input, size_t size, char16_t *utf16_output) const noexcept { 537| 450| return utf8_to_utf16::convert_valid(input, size, 538| 450| utf16_output); 539| 450|} _ZNK7simdutf7haswell14implementation29convert_valid_utf8_to_utf16beEPKcmPDs: 542| 510| const char *input, size_t size, char16_t *utf16_output) const noexcept { 543| 510| return utf8_to_utf16::convert_valid(input, size, 544| 510| utf16_output); 545| 510|} _ZNK7simdutf7haswell14implementation21convert_utf8_to_utf32EPKcmPDi: 550| 758| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 551| 758| utf8_to_utf32::validating_transcoder converter; 552| 758| return converter.convert(buf, len, utf32_output); 553| 758|} _ZNK7simdutf7haswell14implementation33convert_utf8_to_utf32_with_errorsEPKcmPDi: 556| 746| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 557| 746| utf8_to_utf32::validating_transcoder converter; 558| 746| return converter.convert_with_errors(buf, len, utf32_output); 559| 746|} _ZNK7simdutf7haswell14implementation27convert_valid_utf8_to_utf32EPKcmPDi: 562| 464| const char *input, size_t size, char32_t *utf32_output) const noexcept { 563| 464| return utf8_to_utf32::convert_valid(input, size, utf32_output); 564| 464|} _ZNK7simdutf7haswell14implementation25convert_utf16le_to_latin1EPKDsmPc: 569| 124| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 570| 124| std::pair ret = 571| 124| haswell::avx2_convert_utf16_to_latin1(buf, len, 572| 124| latin1_output); 573| 124| if (ret.first == nullptr) { ------------------ | Branch (573:7): [True: 26, False: 98] ------------------ 574| 26| return 0; 575| 26| } 576| 98| size_t saved_bytes = ret.second - latin1_output; 577| 98| if (ret.first != buf + len) { ------------------ | Branch (577:7): [True: 84, False: 14] ------------------ 578| 84| const size_t scalar_saved_bytes = 579| 84| scalar::utf16_to_latin1::convert( 580| 84| ret.first, len - (ret.first - buf), ret.second); 581| 84| if (scalar_saved_bytes == 0) { ------------------ | Branch (581:9): [True: 56, False: 28] ------------------ 582| 56| return 0; 583| 56| } 584| 28| saved_bytes += scalar_saved_bytes; 585| 28| } 586| 42| return saved_bytes; 587| 98|} _ZNK7simdutf7haswell14implementation25convert_utf16be_to_latin1EPKDsmPc: 590| 114| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 591| 114| std::pair ret = 592| 114| haswell::avx2_convert_utf16_to_latin1(buf, len, 593| 114| latin1_output); 594| 114| if (ret.first == nullptr) { ------------------ | Branch (594:7): [True: 28, False: 86] ------------------ 595| 28| return 0; 596| 28| } 597| 86| size_t saved_bytes = ret.second - latin1_output; 598| 86| if (ret.first != buf + len) { ------------------ | Branch (598:7): [True: 74, False: 12] ------------------ 599| 74| const size_t scalar_saved_bytes = 600| 74| scalar::utf16_to_latin1::convert( 601| 74| ret.first, len - (ret.first - buf), ret.second); 602| 74| if (scalar_saved_bytes == 0) { ------------------ | Branch (602:9): [True: 38, False: 36] ------------------ 603| 38| return 0; 604| 38| } 605| 36| saved_bytes += scalar_saved_bytes; 606| 36| } 607| 48| return saved_bytes; 608| 86|} _ZNK7simdutf7haswell14implementation37convert_utf16le_to_latin1_with_errorsEPKDsmPc: 612| 278| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 613| 278| std::pair ret = 614| 278| avx2_convert_utf16_to_latin1_with_errors( 615| 278| buf, len, latin1_output); 616| 278| if (ret.first.error) { ------------------ | Branch (616:7): [True: 116, False: 162] ------------------ 617| 116| return ret.first; 618| 116| } // Can return directly since scalar fallback already found correct 619| | // ret.first.count 620| 162| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (620:7): [True: 124, False: 38] ------------------ 621| 124| result scalar_res = 622| 124| scalar::utf16_to_latin1::convert_with_errors( 623| 124| buf + ret.first.count, len - ret.first.count, ret.second); 624| 124| if (scalar_res.error) { ------------------ | Branch (624:9): [True: 68, False: 56] ------------------ 625| 68| scalar_res.count += ret.first.count; 626| 68| return scalar_res; 627| 68| } else { 628| 56| ret.second += scalar_res.count; 629| 56| } 630| 124| } 631| 94| ret.first.count = 632| 94| ret.second - 633| 94| latin1_output; // Set count to the number of 8-bit code units written 634| 94| return ret.first; 635| 162|} _ZNK7simdutf7haswell14implementation37convert_utf16be_to_latin1_with_errorsEPKDsmPc: 639| 266| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 640| 266| std::pair ret = 641| 266| avx2_convert_utf16_to_latin1_with_errors(buf, len, 642| 266| latin1_output); 643| 266| if (ret.first.error) { ------------------ | Branch (643:7): [True: 114, False: 152] ------------------ 644| 114| return ret.first; 645| 114| } // Can return directly since scalar fallback already found correct 646| | // ret.first.count 647| 152| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (647:7): [True: 118, False: 34] ------------------ 648| 118| result scalar_res = 649| 118| scalar::utf16_to_latin1::convert_with_errors( 650| 118| buf + ret.first.count, len - ret.first.count, ret.second); 651| 118| if (scalar_res.error) { ------------------ | Branch (651:9): [True: 60, False: 58] ------------------ 652| 60| scalar_res.count += ret.first.count; 653| 60| return scalar_res; 654| 60| } else { 655| 58| ret.second += scalar_res.count; 656| 58| } 657| 118| } 658| 92| ret.first.count = 659| 92| ret.second - 660| 92| latin1_output; // Set count to the number of 8-bit code units written 661| 92| return ret.first; 662| 152|} _ZNK7simdutf7haswell14implementation23convert_utf16le_to_utf8EPKDsmPc: 679| 842| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 680| 842| std::pair ret = 681| 842| haswell::avx2_convert_utf16_to_utf8(buf, len, 682| 842| utf8_output); 683| 842| if (ret.first == nullptr) { ------------------ | Branch (683:7): [True: 124, False: 718] ------------------ 684| 124| return 0; 685| 124| } 686| 718| size_t saved_bytes = ret.second - utf8_output; 687| 718| if (ret.first != buf + len) { ------------------ | Branch (687:7): [True: 716, False: 2] ------------------ 688| 716| const size_t scalar_saved_bytes = 689| 716| scalar::utf16_to_utf8::convert( 690| 716| ret.first, len - (ret.first - buf), ret.second); 691| 716| if (scalar_saved_bytes == 0) { ------------------ | Branch (691:9): [True: 80, False: 636] ------------------ 692| 80| return 0; 693| 80| } 694| 636| saved_bytes += scalar_saved_bytes; 695| 636| } 696| 638| return saved_bytes; 697| 718|} _ZNK7simdutf7haswell14implementation23convert_utf16be_to_utf8EPKDsmPc: 700| 798| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 701| 798| std::pair ret = 702| 798| haswell::avx2_convert_utf16_to_utf8(buf, len, 703| 798| utf8_output); 704| 798| if (ret.first == nullptr) { ------------------ | Branch (704:7): [True: 110, False: 688] ------------------ 705| 110| return 0; 706| 110| } 707| 688| size_t saved_bytes = ret.second - utf8_output; 708| 688| if (ret.first != buf + len) { ------------------ | Branch (708:7): [True: 686, False: 2] ------------------ 709| 686| const size_t scalar_saved_bytes = 710| 686| scalar::utf16_to_utf8::convert( 711| 686| ret.first, len - (ret.first - buf), ret.second); 712| 686| if (scalar_saved_bytes == 0) { ------------------ | Branch (712:9): [True: 54, False: 632] ------------------ 713| 54| return 0; 714| 54| } 715| 632| saved_bytes += scalar_saved_bytes; 716| 632| } 717| 634| return saved_bytes; 718| 688|} _ZNK7simdutf7haswell14implementation35convert_utf16le_to_utf8_with_errorsEPKDsmPc: 721| 642| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 722| | // ret.first.count is always the position in the buffer, not the number of 723| | // code units written even if finished 724| 642| std::pair ret = 725| 642| haswell::avx2_convert_utf16_to_utf8_with_errors( 726| 642| buf, len, utf8_output); 727| 642| if (ret.first.error) { ------------------ | Branch (727:7): [True: 158, False: 484] ------------------ 728| 158| return ret.first; 729| 158| } // Can return directly since scalar fallback already found correct 730| | // ret.first.count 731| 484| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (731:7): [True: 480, False: 4] ------------------ 732| 480| result scalar_res = 733| 480| scalar::utf16_to_utf8::convert_with_errors( 734| 480| buf + ret.first.count, len - ret.first.count, ret.second); 735| 480| if (scalar_res.error) { ------------------ | Branch (735:9): [True: 82, False: 398] ------------------ 736| 82| scalar_res.count += ret.first.count; 737| 82| return scalar_res; 738| 398| } else { 739| 398| ret.second += scalar_res.count; 740| 398| } 741| 480| } 742| 402| ret.first.count = 743| 402| ret.second - 744| 402| utf8_output; // Set count to the number of 8-bit code units written 745| 402| return ret.first; 746| 484|} _ZNK7simdutf7haswell14implementation35convert_utf16be_to_utf8_with_errorsEPKDsmPc: 749| 654| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 750| | // ret.first.count is always the position in the buffer, not the number of 751| | // code units written even if finished 752| 654| std::pair ret = 753| 654| haswell::avx2_convert_utf16_to_utf8_with_errors( 754| 654| buf, len, utf8_output); 755| 654| if (ret.first.error) { ------------------ | Branch (755:7): [True: 180, False: 474] ------------------ 756| 180| return ret.first; 757| 180| } // Can return directly since scalar fallback already found correct 758| | // ret.first.count 759| 474| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (759:7): [True: 470, False: 4] ------------------ 760| 470| result scalar_res = 761| 470| scalar::utf16_to_utf8::convert_with_errors( 762| 470| buf + ret.first.count, len - ret.first.count, ret.second); 763| 470| if (scalar_res.error) { ------------------ | Branch (763:9): [True: 92, False: 378] ------------------ 764| 92| scalar_res.count += ret.first.count; 765| 92| return scalar_res; 766| 378| } else { 767| 378| ret.second += scalar_res.count; 768| 378| } 769| 470| } 770| 382| ret.first.count = 771| 382| ret.second - 772| 382| utf8_output; // Set count to the number of 8-bit code units written 773| 382| return ret.first; 774| 474|} _ZNK7simdutf7haswell14implementation29convert_valid_utf16le_to_utf8EPKDsmPc: 777| 406| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 778| 406| return convert_utf16le_to_utf8(buf, len, utf8_output); 779| 406|} _ZNK7simdutf7haswell14implementation29convert_valid_utf16be_to_utf8EPKDsmPc: 782| 452| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 783| 452| return convert_utf16be_to_utf8(buf, len, utf8_output); 784| 452|} _ZNK7simdutf7haswell14implementation21convert_utf32_to_utf8EPKDimPc: 789| 864| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 790| 864| std::pair ret = 791| 864| avx2_convert_utf32_to_utf8(buf, len, utf8_output); 792| 864| if (ret.first == nullptr) { ------------------ | Branch (792:7): [True: 298, False: 566] ------------------ 793| 298| return 0; 794| 298| } 795| 566| size_t saved_bytes = ret.second - utf8_output; 796| 566| if (ret.first != buf + len) { ------------------ | Branch (796:7): [True: 564, False: 2] ------------------ 797| 564| const size_t scalar_saved_bytes = scalar::utf32_to_utf8::convert( 798| 564| ret.first, len - (ret.first - buf), ret.second); 799| 564| if (scalar_saved_bytes == 0) { ------------------ | Branch (799:9): [True: 256, False: 308] ------------------ 800| 256| return 0; 801| 256| } 802| 308| saved_bytes += scalar_saved_bytes; 803| 308| } 804| 310| return saved_bytes; 805| 566|} _ZNK7simdutf7haswell14implementation23convert_utf32_to_latin1EPKDimPc: 810| 228| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 811| 228| std::pair ret = 812| 228| avx2_convert_utf32_to_latin1(buf, len, latin1_output); 813| 228| if (ret.first == nullptr) { ------------------ | Branch (813:7): [True: 50, False: 178] ------------------ 814| 50| return 0; 815| 50| } 816| 178| size_t saved_bytes = ret.second - latin1_output; 817| 178| if (ret.first != buf + len) { ------------------ | Branch (817:7): [True: 166, False: 12] ------------------ 818| 166| const size_t scalar_saved_bytes = scalar::utf32_to_latin1::convert( 819| 166| ret.first, len - (ret.first - buf), ret.second); 820| 166| if (scalar_saved_bytes == 0) { ------------------ | Branch (820:9): [True: 124, False: 42] ------------------ 821| 124| return 0; 822| 124| } 823| 42| saved_bytes += scalar_saved_bytes; 824| 42| } 825| 54| return saved_bytes; 826| 178|} _ZNK7simdutf7haswell14implementation35convert_utf32_to_latin1_with_errorsEPKDimPc: 829| 436| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 830| | // ret.first.count is always the position in the buffer, not the number of 831| | // code units written even if finished 832| 436| std::pair ret = 833| 436| avx2_convert_utf32_to_latin1_with_errors(buf, len, latin1_output); 834| 436| if (ret.first.count != len) { ------------------ | Branch (834:7): [True: 418, False: 18] ------------------ 835| 418| result scalar_res = scalar::utf32_to_latin1::convert_with_errors( 836| 418| buf + ret.first.count, len - ret.first.count, ret.second); 837| 418| if (scalar_res.error) { ------------------ | Branch (837:9): [True: 394, False: 24] ------------------ 838| 394| scalar_res.count += ret.first.count; 839| 394| return scalar_res; 840| 394| } else { 841| 24| ret.second += scalar_res.count; 842| 24| } 843| 418| } 844| 42| ret.first.count = 845| 42| ret.second - 846| 42| latin1_output; // Set count to the number of 8-bit code units written 847| 42| return ret.first; 848| 436|} _ZNK7simdutf7haswell14implementation33convert_utf32_to_utf8_with_errorsEPKDimPc: 858| 832| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 859| | // ret.first.count is always the position in the buffer, not the number of 860| | // code units written even if finished 861| 832| std::pair ret = 862| 832| haswell::avx2_convert_utf32_to_utf8_with_errors(buf, len, utf8_output); 863| 832| if (ret.first.count != len) { ------------------ | Branch (863:7): [True: 828, False: 4] ------------------ 864| 828| result scalar_res = scalar::utf32_to_utf8::convert_with_errors( 865| 828| buf + ret.first.count, len - ret.first.count, ret.second); 866| 828| if (scalar_res.error) { ------------------ | Branch (866:9): [True: 618, False: 210] ------------------ 867| 618| scalar_res.count += ret.first.count; 868| 618| return scalar_res; 869| 618| } else { 870| 210| ret.second += scalar_res.count; 871| 210| } 872| 828| } 873| 214| ret.first.count = 874| 214| ret.second - 875| 214| utf8_output; // Set count to the number of 8-bit code units written 876| 214| return ret.first; 877| 832|} _ZNK7simdutf7haswell14implementation24convert_utf16le_to_utf32EPKDsmPDi: 882| 454| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 883| 454| std::pair ret = 884| 454| haswell::avx2_convert_utf16_to_utf32(buf, len, 885| 454| utf32_output); 886| 454| if (ret.first == nullptr) { ------------------ | Branch (886:7): [True: 124, False: 330] ------------------ 887| 124| return 0; 888| 124| } 889| 330| size_t saved_bytes = ret.second - utf32_output; 890| 330| if (ret.first != buf + len) { ------------------ | Branch (890:7): [True: 274, False: 56] ------------------ 891| 274| const size_t scalar_saved_bytes = 892| 274| scalar::utf16_to_utf32::convert( 893| 274| ret.first, len - (ret.first - buf), ret.second); 894| 274| if (scalar_saved_bytes == 0) { ------------------ | Branch (894:9): [True: 82, False: 192] ------------------ 895| 82| return 0; 896| 82| } 897| 192| saved_bytes += scalar_saved_bytes; 898| 192| } 899| 248| return saved_bytes; 900| 330|} _ZNK7simdutf7haswell14implementation24convert_utf16be_to_utf32EPKDsmPDi: 903| 422| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 904| 422| std::pair ret = 905| 422| haswell::avx2_convert_utf16_to_utf32(buf, len, 906| 422| utf32_output); 907| 422| if (ret.first == nullptr) { ------------------ | Branch (907:7): [True: 88, False: 334] ------------------ 908| 88| return 0; 909| 88| } 910| 334| size_t saved_bytes = ret.second - utf32_output; 911| 334| if (ret.first != buf + len) { ------------------ | Branch (911:7): [True: 290, False: 44] ------------------ 912| 290| const size_t scalar_saved_bytes = 913| 290| scalar::utf16_to_utf32::convert( 914| 290| ret.first, len - (ret.first - buf), ret.second); 915| 290| if (scalar_saved_bytes == 0) { ------------------ | Branch (915:9): [True: 88, False: 202] ------------------ 916| 88| return 0; 917| 88| } 918| 202| saved_bytes += scalar_saved_bytes; 919| 202| } 920| 246| return saved_bytes; 921| 334|} _ZNK7simdutf7haswell14implementation36convert_utf16le_to_utf32_with_errorsEPKDsmPDi: 924| 334| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 925| | // ret.first.count is always the position in the buffer, not the number of 926| | // code units written even if finished 927| 334| std::pair ret = 928| 334| haswell::avx2_convert_utf16_to_utf32_with_errors( 929| 334| buf, len, utf32_output); 930| 334| if (ret.first.error) { ------------------ | Branch (930:7): [True: 108, False: 226] ------------------ 931| 108| return ret.first; 932| 108| } // Can return directly since scalar fallback already found correct 933| | // ret.first.count 934| 226| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (934:7): [True: 188, False: 38] ------------------ 935| 188| result scalar_res = 936| 188| scalar::utf16_to_utf32::convert_with_errors( 937| 188| buf + ret.first.count, len - ret.first.count, ret.second); 938| 188| if (scalar_res.error) { ------------------ | Branch (938:9): [True: 72, False: 116] ------------------ 939| 72| scalar_res.count += ret.first.count; 940| 72| return scalar_res; 941| 116| } else { 942| 116| ret.second += scalar_res.count; 943| 116| } 944| 188| } 945| 154| ret.first.count = 946| 154| ret.second - 947| 154| utf32_output; // Set count to the number of 8-bit code units written 948| 154| return ret.first; 949| 226|} _ZNK7simdutf7haswell14implementation36convert_utf16be_to_utf32_with_errorsEPKDsmPDi: 952| 346| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 953| | // ret.first.count is always the position in the buffer, not the number of 954| | // code units written even if finished 955| 346| std::pair ret = 956| 346| haswell::avx2_convert_utf16_to_utf32_with_errors( 957| 346| buf, len, utf32_output); 958| 346| if (ret.first.error) { ------------------ | Branch (958:7): [True: 92, False: 254] ------------------ 959| 92| return ret.first; 960| 92| } // Can return directly since scalar fallback already found correct 961| | // ret.first.count 962| 254| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (962:7): [True: 228, False: 26] ------------------ 963| 228| result scalar_res = 964| 228| scalar::utf16_to_utf32::convert_with_errors( 965| 228| buf + ret.first.count, len - ret.first.count, ret.second); 966| 228| if (scalar_res.error) { ------------------ | Branch (966:9): [True: 92, False: 136] ------------------ 967| 92| scalar_res.count += ret.first.count; 968| 92| return scalar_res; 969| 136| } else { 970| 136| ret.second += scalar_res.count; 971| 136| } 972| 228| } 973| 162| ret.first.count = 974| 162| ret.second - 975| 162| utf32_output; // Set count to the number of 8-bit code units written 976| 162| return ret.first; 977| 254|} _ZNK7simdutf7haswell14implementation27convert_valid_utf32_to_utf8EPKDimPc: 982| 216| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 983| 216| return convert_utf32_to_utf8(buf, len, utf8_output); 984| 216|} _ZNK7simdutf7haswell14implementation24convert_utf32_to_utf16leEPKDimPDs: 989| 822| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 990| 822| std::pair ret = 991| 822| avx2_convert_utf32_to_utf16(buf, len, utf16_output); 992| 822| if (ret.first == nullptr) { ------------------ | Branch (992:7): [True: 206, False: 616] ------------------ 993| 206| return 0; 994| 206| } 995| 616| size_t saved_bytes = ret.second - utf16_output; 996| 616| if (ret.first != buf + len) { ------------------ | Branch (996:7): [True: 612, False: 4] ------------------ 997| 612| const size_t scalar_saved_bytes = 998| 612| scalar::utf32_to_utf16::convert( 999| 612| ret.first, len - (ret.first - buf), ret.second); 1000| 612| if (scalar_saved_bytes == 0) { ------------------ | Branch (1000:9): [True: 188, False: 424] ------------------ 1001| 188| return 0; 1002| 188| } 1003| 424| saved_bytes += scalar_saved_bytes; 1004| 424| } 1005| 428| return saved_bytes; 1006| 616|} _ZNK7simdutf7haswell14implementation24convert_utf32_to_utf16beEPKDimPDs: 1009| 674| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1010| 674| std::pair ret = 1011| 674| avx2_convert_utf32_to_utf16(buf, len, utf16_output); 1012| 674| if (ret.first == nullptr) { ------------------ | Branch (1012:7): [True: 224, False: 450] ------------------ 1013| 224| return 0; 1014| 224| } 1015| 450| size_t saved_bytes = ret.second - utf16_output; 1016| 450| if (ret.first != buf + len) { ------------------ | Branch (1016:7): [True: 448, False: 2] ------------------ 1017| 448| const size_t scalar_saved_bytes = 1018| 448| scalar::utf32_to_utf16::convert( 1019| 448| ret.first, len - (ret.first - buf), ret.second); 1020| 448| if (scalar_saved_bytes == 0) { ------------------ | Branch (1020:9): [True: 196, False: 252] ------------------ 1021| 196| return 0; 1022| 196| } 1023| 252| saved_bytes += scalar_saved_bytes; 1024| 252| } 1025| 254| return saved_bytes; 1026| 450|} _ZNK7simdutf7haswell14implementation36convert_utf32_to_utf16le_with_errorsEPKDimPDs: 1029| 674| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1030| | // ret.first.count is always the position in the buffer, not the number of 1031| | // code units written even if finished 1032| 674| std::pair ret = 1033| 674| haswell::avx2_convert_utf32_to_utf16_with_errors( 1034| 674| buf, len, utf16_output); 1035| 674| if (ret.first.count != len) { ------------------ | Branch (1035:7): [True: 670, False: 4] ------------------ 1036| 670| result scalar_res = 1037| 670| scalar::utf32_to_utf16::convert_with_errors( 1038| 670| buf + ret.first.count, len - ret.first.count, ret.second); 1039| 670| if (scalar_res.error) { ------------------ | Branch (1039:9): [True: 504, False: 166] ------------------ 1040| 504| scalar_res.count += ret.first.count; 1041| 504| return scalar_res; 1042| 504| } else { 1043| 166| ret.second += scalar_res.count; 1044| 166| } 1045| 670| } 1046| 170| ret.first.count = 1047| 170| ret.second - 1048| 170| utf16_output; // Set count to the number of 8-bit code units written 1049| 170| return ret.first; 1050| 674|} _ZNK7simdutf7haswell14implementation36convert_utf32_to_utf16be_with_errorsEPKDimPDs: 1053| 694| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1054| | // ret.first.count is always the position in the buffer, not the number of 1055| | // code units written even if finished 1056| 694| std::pair ret = 1057| 694| haswell::avx2_convert_utf32_to_utf16_with_errors( 1058| 694| buf, len, utf16_output); 1059| 694| if (ret.first.count != len) { ------------------ | Branch (1059:7): [True: 690, False: 4] ------------------ 1060| 690| result scalar_res = 1061| 690| scalar::utf32_to_utf16::convert_with_errors( 1062| 690| buf + ret.first.count, len - ret.first.count, ret.second); 1063| 690| if (scalar_res.error) { ------------------ | Branch (1063:9): [True: 512, False: 178] ------------------ 1064| 512| scalar_res.count += ret.first.count; 1065| 512| return scalar_res; 1066| 512| } else { 1067| 178| ret.second += scalar_res.count; 1068| 178| } 1069| 690| } 1070| 182| ret.first.count = 1071| 182| ret.second - 1072| 182| utf16_output; // Set count to the number of 8-bit code units written 1073| 182| return ret.first; 1074| 694|} _ZNK7simdutf7haswell14implementation30convert_valid_utf32_to_utf16leEPKDimPDs: 1077| 280| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1078| 280| return convert_utf32_to_utf16le(buf, len, utf16_output); 1079| 280|} _ZNK7simdutf7haswell14implementation30convert_valid_utf32_to_utf16beEPKDimPDs: 1082| 134| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1083| 134| return convert_utf32_to_utf16be(buf, len, utf16_output); 1084| 134|} _ZNK7simdutf7haswell14implementation30convert_valid_utf16le_to_utf32EPKDsmPDi: 1087| 128| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1088| 128| return convert_utf16le_to_utf32(buf, len, utf32_output); 1089| 128|} _ZNK7simdutf7haswell14implementation30convert_valid_utf16be_to_utf32EPKDsmPDi: 1092| 142| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1093| 142| return convert_utf16be_to_utf32(buf, len, utf32_output); 1094| 142|} _ZNK7simdutf7haswell14implementation13count_utf16leEPKDsm: 1105| 1.35k| const char16_t *input, size_t length) const noexcept { 1106| 1.35k| return utf16::count_code_points(input, length); 1107| 1.35k|} _ZNK7simdutf7haswell14implementation13count_utf16beEPKDsm: 1110| 1.33k| const char16_t *input, size_t length) const noexcept { 1111| 1.33k| return utf16::count_code_points(input, length); 1112| 1.33k|} _ZNK7simdutf7haswell14implementation10count_utf8EPKcm: 1117| 4.03k|implementation::count_utf8(const char *in, size_t size) const noexcept { 1118| 4.03k| return utf8::count_code_points_bytemask(in, size); 1119| 4.03k|} _ZNK7simdutf7haswell14implementation23latin1_length_from_utf8EPKcm: 1124| 549| const char *buf, size_t len) const noexcept { 1125| 549| return count_utf8(buf, len); 1126| 549|} _ZNK7simdutf7haswell14implementation24utf8_length_from_utf16leEPKDsm: 1131| 754| const char16_t *input, size_t length) const noexcept { 1132| 754| return utf16::utf8_length_from_utf16_bytemask(input, 1133| 754| length); 1134| 754|} _ZNK7simdutf7haswell14implementation24utf8_length_from_utf16beEPKDsm: 1137| 748| const char16_t *input, size_t length) const noexcept { 1138| 748| return utf16::utf8_length_from_utf16_bytemask(input, length); 1139| 748|} _ZNK7simdutf7haswell14implementation25utf32_length_from_utf16leEPKDsm: 1144| 400| const char16_t *input, size_t length) const noexcept { 1145| 400| return utf16::utf32_length_from_utf16(input, length); 1146| 400|} _ZNK7simdutf7haswell14implementation25utf32_length_from_utf16beEPKDsm: 1149| 400| const char16_t *input, size_t length) const noexcept { 1150| 400| return utf16::utf32_length_from_utf16(input, length); 1151| 400|} _ZNK7simdutf7haswell14implementation22utf16_length_from_utf8EPKcm: 1156| 1.94k| const char *input, size_t length) const noexcept { 1157| 1.94k| return utf8::utf16_length_from_utf8_bytemask(input, length); 1158| 1.94k|} _ZNK7simdutf7haswell14implementation23utf8_length_from_latin1EPKcm: 1203| 302| const char *input, size_t len) const noexcept { 1204| 302| const uint8_t *data = reinterpret_cast(input); 1205| 302| size_t answer = len / sizeof(__m256i) * sizeof(__m256i); 1206| 302| size_t i = 0; 1207| 302| if (answer >= 2048) { // long strings optimization ------------------ | Branch (1207:7): [True: 139, False: 163] ------------------ 1208| 139| __m256i four_64bits = _mm256_setzero_si256(); 1209| 3.51k| while (i + sizeof(__m256i) <= len) { ------------------ | Branch (1209:12): [True: 3.37k, False: 139] ------------------ 1210| 3.37k| __m256i runner = _mm256_setzero_si256(); 1211| | // We can do up to 255 loops without overflow. 1212| 3.37k| size_t iterations = (len - i) / sizeof(__m256i); 1213| 3.37k| if (iterations > 255) { ------------------ | Branch (1213:11): [True: 3.23k, False: 139] ------------------ 1214| 3.23k| iterations = 255; 1215| 3.23k| } 1216| 3.37k| size_t max_i = i + iterations * sizeof(__m256i) - sizeof(__m256i); 1217| 210k| for (; i + 4 * sizeof(__m256i) <= max_i; i += 4 * sizeof(__m256i)) { ------------------ | Branch (1217:14): [True: 206k, False: 3.37k] ------------------ 1218| 206k| __m256i input1 = _mm256_loadu_si256((const __m256i *)(data + i)); 1219| 206k| __m256i input2 = 1220| 206k| _mm256_loadu_si256((const __m256i *)(data + i + sizeof(__m256i))); 1221| 206k| __m256i input3 = _mm256_loadu_si256( 1222| 206k| (const __m256i *)(data + i + 2 * sizeof(__m256i))); 1223| 206k| __m256i input4 = _mm256_loadu_si256( 1224| 206k| (const __m256i *)(data + i + 3 * sizeof(__m256i))); 1225| 206k| __m256i input12 = 1226| 206k| _mm256_add_epi8(_mm256_cmpgt_epi8(_mm256_setzero_si256(), input1), 1227| 206k| _mm256_cmpgt_epi8(_mm256_setzero_si256(), input2)); 1228| 206k| __m256i input23 = 1229| 206k| _mm256_add_epi8(_mm256_cmpgt_epi8(_mm256_setzero_si256(), input3), 1230| 206k| _mm256_cmpgt_epi8(_mm256_setzero_si256(), input4)); 1231| 206k| __m256i input1234 = _mm256_add_epi8(input12, input23); 1232| 206k| runner = _mm256_sub_epi8(runner, input1234); 1233| 206k| } 1234| 13.4k| for (; i <= max_i; i += sizeof(__m256i)) { ------------------ | Branch (1234:14): [True: 10.0k, False: 3.37k] ------------------ 1235| 10.0k| __m256i input_256_chunk = 1236| 10.0k| _mm256_loadu_si256((const __m256i *)(data + i)); 1237| 10.0k| runner = _mm256_sub_epi8( 1238| 10.0k| runner, _mm256_cmpgt_epi8(_mm256_setzero_si256(), input_256_chunk)); 1239| 10.0k| } 1240| 3.37k| four_64bits = _mm256_add_epi64( 1241| 3.37k| four_64bits, _mm256_sad_epu8(runner, _mm256_setzero_si256())); 1242| 3.37k| } 1243| 139| answer += _mm256_extract_epi64(four_64bits, 0) + 1244| 139| _mm256_extract_epi64(four_64bits, 1) + 1245| 139| _mm256_extract_epi64(four_64bits, 2) + 1246| 139| _mm256_extract_epi64(four_64bits, 3); 1247| 163| } else if (answer > 0) { ------------------ | Branch (1247:14): [True: 80, False: 83] ------------------ 1248| 1.40k| for (; i + sizeof(__m256i) <= len; i += sizeof(__m256i)) { ------------------ | Branch (1248:12): [True: 1.32k, False: 80] ------------------ 1249| 1.32k| __m256i latin = _mm256_loadu_si256((const __m256i *)(data + i)); 1250| 1.32k| uint32_t non_ascii = _mm256_movemask_epi8(latin); 1251| 1.32k| answer += count_ones(non_ascii); 1252| 1.32k| } 1253| 80| } 1254| 302| return answer + scalar::latin1::utf8_length_from_latin1( 1255| 302| reinterpret_cast(data + i), len - i); 1256| 302|} _ZNK7simdutf7haswell14implementation22utf8_length_from_utf32EPKDim: 1261| 860| const char32_t *input, size_t length) const noexcept { 1262| 860| return utf32::utf8_length_from_utf32(input, length); 1263| 860|} _ZNK7simdutf7haswell14implementation23utf16_length_from_utf32EPKDim: 1268| 1.44k| const char32_t *input, size_t length) const noexcept { 1269| 1.44k| const __m256i v_00000000 = _mm256_setzero_si256(); 1270| 1.44k| const __m256i v_ffff0000 = _mm256_set1_epi32((uint32_t)0xffff0000); 1271| 1.44k| size_t pos = 0; 1272| 1.44k| size_t count = 0; 1273| 582k| for (; pos + 8 <= length; pos += 8) { ------------------ | Branch (1273:10): [True: 581k, False: 1.44k] ------------------ 1274| 581k| __m256i in = _mm256_loadu_si256((__m256i *)(input + pos)); 1275| 581k| const __m256i surrogate_bytemask = 1276| 581k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_ffff0000), v_00000000); 1277| 581k| const uint32_t surrogate_bitmask = 1278| 581k| static_cast(_mm256_movemask_epi8(surrogate_bytemask)); 1279| 581k| size_t surrogate_count = (32 - count_ones(surrogate_bitmask)) / 4; 1280| 581k| count += 8 + surrogate_count; 1281| 581k| } 1282| 1.44k| return count + 1283| 1.44k| scalar::utf32::utf16_length_from_utf32(input + pos, length - pos); 1284| 1.44k|} _ZNK7simdutf7haswell14implementation22utf32_length_from_utf8EPKcm: 1289| 994| const char *input, size_t length) const noexcept { 1290| 994| return utf8::count_code_points(input, length); 1291| 994|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_18is_asciiERKNS1_4simd8simd8x64IhEE: 13| 4.87M|simdutf_really_inline bool is_ascii(const simd8x64 &input) { 14| 4.87M| return input.reduce_or().is_ascii(); 15| 4.87M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_124must_be_2_3_continuationENS1_4simd5simd8IhEES4_: 22| 3.00M| const simd8 prev3) { 23| 3.00M| simd8 is_third_byte = 24| 3.00M| prev2.saturating_sub(0xe0u - 0x80); // Only 111_____ will be > 0x80 25| 3.00M| simd8 is_fourth_byte = 26| 3.00M| prev3.saturating_sub(0xf0u - 0x80); // Only 1111____ will be > 0x80 27| 3.00M| return simd8(is_third_byte | is_fourth_byte); 28| 3.00M|} _ZNK7simdutf14implementation27supported_by_runtime_systemEv: 64| 8|bool implementation::supported_by_runtime_system() const { 65| 8| uint32_t required_instruction_sets = this->required_instruction_sets(); 66| 8| uint32_t supported_instruction_sets = 67| 8| internal::detect_supported_architectures(); 68| 8| return ((supported_instruction_sets & required_instruction_sets) == 69| 8| required_instruction_sets); 70| 8|} _ZNK7simdutf8internal29available_implementation_list5beginEv: 1503| 1|available_implementation_list::begin() const noexcept { 1504| 1| return internal::get_available_implementation_pointers().begin(); 1505| 1|} _ZNK7simdutf8internal29available_implementation_list3endEv: 1507| 1|available_implementation_list::end() const noexcept { 1508| 1| return internal::get_available_implementation_pointers().end(); 1509| 1|} _ZN7simdutf29get_available_implementationsEv: 1558| 1|get_available_implementations() { 1559| 1|#if !SIMDUTF_USE_STATIC_INITIALIZATION 1560| 1| static const internal::available_implementation_list 1561| 1| available_implementations_instance{}; 1562| 1|#endif 1563| 1| return available_implementations_instance; 1564| 1|} simdutf.cpp:_ZN7simdutf8internalL37get_available_implementation_pointersEv: 933| 2|get_available_implementation_pointers() { 934| 2|#if !SIMDUTF_USE_STATIC_INITIALIZATION 935| 2| static const std::initializer_list 936| 2| available_implementation_pointers{ 937| 2| #if SIMDUTF_IMPLEMENTATION_ICELAKE 938| 2| get_icelake_singleton(), 939| 2| #endif 940| 2| #if SIMDUTF_IMPLEMENTATION_HASWELL 941| 2| get_haswell_singleton(), 942| 2| #endif 943| 2| #if SIMDUTF_IMPLEMENTATION_WESTMERE 944| 2| get_westmere_singleton(), 945| 2| #endif 946| | #if SIMDUTF_IMPLEMENTATION_ARM64 947| | get_arm64_singleton(), 948| | #endif 949| | #if SIMDUTF_IMPLEMENTATION_PPC64 950| | get_ppc64_singleton(), 951| | #endif 952| | #if SIMDUTF_IMPLEMENTATION_RVV 953| | get_rvv_singleton(), 954| | #endif 955| | #if SIMDUTF_IMPLEMENTATION_LASX 956| | get_lasx_singleton(), 957| | #endif 958| | #if SIMDUTF_IMPLEMENTATION_LSX 959| | get_lsx_singleton(), 960| | #endif 961| 2| #if SIMDUTF_IMPLEMENTATION_FALLBACK 962| 2| get_fallback_singleton(), 963| 2| #endif 964| 2| }; 965| 2|#endif 966| 2| return available_implementation_pointers; 967| 2|} simdutf.cpp:_ZN7simdutf8internalL21get_icelake_singletonEv: 157| 1|static const icelake::implementation *get_icelake_singleton() { 158| 1| #if !SIMDUTF_USE_STATIC_INITIALIZATION 159| 1| static const icelake::implementation icelake_singleton{}; 160| 1| #endif 161| 1| return &icelake_singleton; 162| 1|} simdutf.cpp:_ZN7simdutf8internalL21get_haswell_singletonEv: 168| 1|static const haswell::implementation *get_haswell_singleton() { 169| 1| #if !SIMDUTF_USE_STATIC_INITIALIZATION 170| 1| static const haswell::implementation haswell_singleton{}; 171| 1| #endif 172| 1| return &haswell_singleton; 173| 1|} simdutf.cpp:_ZN7simdutf8internalL22get_westmere_singletonEv: 179| 1|static const westmere::implementation *get_westmere_singleton() { 180| 1| #if !SIMDUTF_USE_STATIC_INITIALIZATION 181| 1| static const westmere::implementation westmere_singleton{}; 182| 1| #endif 183| 1| return &westmere_singleton; 184| 1|} simdutf.cpp:_ZN7simdutf8internalL22get_fallback_singletonEv: 245| 1|static const fallback::implementation *get_fallback_singleton() { 246| 1| #if !SIMDUTF_USE_STATIC_INITIALIZATION 247| 1| static const fallback::implementation fallback_singleton{}; 248| 1| #endif 249| 1| return &fallback_singleton; 250| 1|} _ZN7simdutf8fallback14implementationC2Ev: 16| 1| : simdutf::implementation("fallback", "Generic fallback implementation", 17| 1| 0) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_110count_onesEm: 14| 3.52M|simdutf_really_inline long long int count_ones(uint64_t input_num) { 15| | return _popcnt64(input_num); 16| 3.52M|} _ZN7simdutf7haswell14implementationC2Ev: 16| 1| : simdutf::implementation("haswell", "Intel/AMD AVX2", 17| 1| internal::instruction_set::AVX2 | 18| 1| internal::instruction_set::BMI1 | 19| 1| internal::instruction_set::BMI2) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2Ev: 191| 20.9k| simdutf_really_inline simd8() : base8_numeric() {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEC2Ev: 124| 20.9k| simdutf_really_inline base8_numeric() : base8() {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2Ev: 71| 20.9k| simdutf_really_inline base8() : base>() {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2Ev: 14| 20.9k| simdutf_really_inline base() : value{__m256i()} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IhE9reduce_orEv: 294| 4.87M| simdutf_really_inline simd8 reduce_or() const { 295| 4.87M| return this->chunks[0] | this->chunks[1]; 296| 4.87M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEorES5_: 50| 15.0M| simdutf_really_inline Child operator|(const Child other) const { 51| 15.0M| return _mm256_or_si256(*this, other); 52| 15.0M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2EDv4_x: 193| 157M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEC2EDv4_x: 126| 157M| : base8(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2EDv4_x: 73| 157M| simdutf_really_inline base8(const __m256i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2EDv4_x: 17| 157M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE8is_asciiEv: 234| 4.87M| simdutf_really_inline bool is_ascii() const { 235| 4.87M| return _mm256_movemask_epi8(*this) == 0; 236| 4.87M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEoRES5_: 59| 7.13M| simdutf_really_inline Child &operator|=(const Child other) { 60| 7.13M| auto this_cast = static_cast(this); 61| 7.13M| *this_cast = *this_cast | other; 62| 7.13M| return *this_cast; 63| 7.13M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi1EEENS4_IhEES8_: 83| 3.13M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 84| | return _mm256_alignr_epi8( 85| 3.13M| *this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N); 86| 3.13M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE3shrILi4EEES4_v: 245| 6.27M| template simdutf_really_inline simd8 shr() const { 246| 6.27M| return simd8(_mm256_srli_epi16(*this, N)) & uint8_t(0xFFu >> N); 247| 6.27M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_: 157| 9.41M| L replace15) const { 158| 9.41M| return lookup_16(simd8::repeat_16( 159| 9.41M| replace0, replace1, replace2, replace3, replace4, replace5, replace6, 160| 9.41M| replace7, replace8, replace9, replace10, replace11, replace12, 161| 9.41M| replace13, replace14, replace15)); 162| 9.41M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES8_: 148| 9.41M| simdutf_really_inline simd8 lookup_16(simd8 lookup_table) const { 149| 9.41M| return _mm256_shuffle_epi8(lookup_table, *this); 150| 9.41M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE9repeat_16Ehhhhhhhhhhhhhhhh: 118| 9.41M| T v14, T v15) { 119| 9.41M| return simd8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, 120| 9.41M| v14, v15, v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, 121| 9.41M| v12, v13, v14, v15); 122| 9.41M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2Ehhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh: 207| 9.41M| : simd8(_mm256_setr_epi8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, 208| 9.41M| v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, 209| 9.41M| v22, v23, v24, v25, v26, v27, v28, v29, v30, 210| 9.41M| v31)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2Eh: 195| 22.1M| simdutf_really_inline simd8(uint8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi2EEENS4_IhEES8_: 83| 3.00M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 84| | return _mm256_alignr_epi8( 85| 3.00M| *this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N); 86| 3.00M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi3EEENS4_IhEES8_: 83| 3.00M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 84| | return _mm256_alignr_epi8( 85| 3.00M| *this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N); 86| 3.00M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE14saturating_subES4_: 214| 6.89M| saturating_sub(const simd8 other) const { 215| 6.89M| return _mm256_subs_epu8(*this, other); 216| 6.89M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IbEC2EDv4_x: 95| 24.6M| simdutf_really_inline simd8(const __m256i _value) : base8(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IbNS2_5simd8IbEEEC2EDv4_x: 73| 24.6M| simdutf_really_inline base8(const __m256i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEC2EDv4_x: 17| 24.6M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEcvRKDv4_xEv: 19| 15.8M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEeoES5_: 56| 3.00M| simdutf_really_inline Child operator^(const Child other) const { 57| 3.00M| return _mm256_xor_si256(*this, other); 58| 3.00M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2EPKh: 197| 881k| simdutf_really_inline simd8(const uint8_t values[32]) : simd8(load(values)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE4loadEPKh: 111| 10.6M| static simdutf_really_inline simd8 load(const T values[32]) { 112| 10.6M| return _mm256_loadu_si256(reinterpret_cast(values)); 113| 10.6M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE7gt_bitsES4_: 225| 881k| gt_bits(const simd8 other) const { 226| 881k| return this->saturating_sub(other); 227| 881k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEC2EDv4_x: 17| 71.8M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEcvRKDv4_xEv: 19| 102M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IbE10to_bitmaskEv: 99| 8.81M| simdutf_really_inline uint32_t to_bitmask() const { 100| 8.81M| return uint32_t(_mm256_movemask_epi8(value)); 101| 8.81M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE21any_bits_set_anywhereEv: 241| 2.30M| simdutf_really_inline bool any_bits_set_anywhere() const { 242| 2.30M| return !bits_not_set_anywhere(); 243| 2.30M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE21bits_not_set_anywhereEv: 237| 2.30M| simdutf_really_inline bool bits_not_set_anywhere() const { 238| 2.30M| return _mm256_testz_si256(*this, *this); 239| 2.30M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaEC2EPKa: 274| 6.04M| : chunks{simd8::load(ptr), 275| 6.04M| simd8::load(ptr + sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaE4loadEPKa: 111| 21.8M| static simdutf_really_inline simd8 load(const T values[32]) { 112| 21.8M| return _mm256_loadu_si256(reinterpret_cast(values)); 113| 21.8M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEC2EDv4_x: 169| 38.7M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaEC2EDv4_x: 126| 38.7M| : base8(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IaNS2_5simd8IbEEEC2EDv4_x: 73| 38.7M| simdutf_really_inline base8(const __m256i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEC2EDv4_x: 17| 38.7M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE8is_asciiEv: 298| 5.29M| simdutf_really_inline bool is_ascii() const { 299| 5.29M| return this->reduce_or().is_ascii(); 300| 5.29M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE9reduce_orEv: 294| 5.29M| simdutf_really_inline simd8 reduce_or() const { 295| 5.29M| return this->chunks[0] | this->chunks[1]; 296| 5.29M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEorES5_: 50| 5.29M| simdutf_really_inline Child operator|(const Child other) const { 51| 5.29M| return _mm256_or_si256(*this, other); 52| 5.29M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEcvRKDv4_xEv: 19| 62.6M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaE8is_asciiEv: 177| 5.29M| simdutf_really_inline bool is_ascii() const { 178| 5.29M| return _mm256_movemask_epi8(*this) == 0; 179| 5.29M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE5storeEPa: 277| 903k| simdutf_really_inline void store(T *ptr) const { 278| 903k| this->chunks[0].store(ptr + sizeof(simd8) * 0 / sizeof(T)); 279| 903k| this->chunks[1].store(ptr + sizeof(simd8) * 1 / sizeof(T)); 280| 903k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaE5storeEPa: 129| 1.80M| simdutf_really_inline void store(T dst[32]) const { 130| 1.80M| return _mm256_storeu_si256(reinterpret_cast<__m256i *>(dst), *this); 131| 1.80M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEcvNS3_IhEEEv: 256| 2.06M|simdutf_really_inline simd8::operator simd8() const { 257| 2.06M| return this->value; 258| 2.06M|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE2ltEa: 325| 1.05M| simdutf_really_inline uint64_t lt(const T m) const { 326| 1.05M| const simd8 mask = simd8::splat(m); 327| 1.05M| return simd8x64(this->chunks[0] < mask, this->chunks[1] < mask) 328| 1.05M| .to_bitmask(); 329| 1.05M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaE5splatEa: 105| 11.6M| static simdutf_really_inline simd8 splat(T _value) { 106| 11.6M| return _mm256_set1_epi8(_value); 107| 11.6M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEltES4_: 184| 2.11M| simdutf_really_inline simd8 operator<(const simd8 other) const { 185| 2.11M| return _mm256_cmpgt_epi8(other, *this); 186| 2.11M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IbEC2ENS2_5simd8IbEES6_: 272| 1.80M| : chunks{chunk0, chunk1} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IbE10to_bitmaskEv: 282| 1.80M| simdutf_really_inline uint64_t to_bitmask() const { 283| 1.80M| uint64_t r_lo = uint32_t(this->chunks[0].to_bitmask()); 284| 1.80M| uint64_t r_hi = this->chunks[1].to_bitmask(); 285| 1.80M| return r_lo | (r_hi << 32); 286| 1.80M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf32EPDi: 310| 1.19M| simdutf_really_inline void store_ascii_as_utf32(char32_t *ptr) const { 311| 1.19M| this->chunks[0].store_ascii_as_utf32(ptr + sizeof(simd8) * 0); 312| 1.19M| this->chunks[1].store_ascii_as_utf32(ptr + sizeof(simd8) * 1); 313| 1.19M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf32EPDi: 36| 2.39M| simdutf_really_inline void store_ascii_as_utf32(char32_t *ptr) const { 37| 2.39M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), 38| 2.39M| _mm256_cvtepu8_epi32(_mm256_castsi256_si128(*this))); 39| 2.39M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 8), 40| 2.39M| _mm256_cvtepu8_epi32(_mm256_castsi256_si128( 41| 2.39M| _mm256_srli_si256(*this, 8)))); 42| 2.39M| _mm256_storeu_si256( 43| 2.39M| reinterpret_cast<__m256i *>(ptr + 16), 44| 2.39M| _mm256_cvtepu8_epi32(_mm256_extractf128_si256(*this, 1))); 45| 2.39M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 24), 46| | _mm256_cvtepu8_epi32(_mm_srli_si128( 47| 2.39M| _mm256_extractf128_si256(*this, 1), 8))); 48| 2.39M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE4zeroEv: 108| 106k| static simdutf_really_inline simd8 zero() { 109| 106k| return _mm256_setzero_si256(); 110| 106k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEgtES4_: 181| 11.3M| simdutf_really_inline simd8 operator>(const simd8 other) const { 182| 11.3M| return _mm256_cmpgt_epi8(*this, other); 183| 11.3M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEC2Ea: 172| 9.79M| simdutf_really_inline simd8(int8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEmIENS2_5simd8IhEE: 137| 12.8M| simdutf_really_inline simd8 &operator-=(const simd8 other) { 138| 12.8M| *this = *this - other; 139| 12.8M| return *static_cast *>(this); 140| 12.8M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEmiENS2_5simd8IhEE: 134| 12.8M| simdutf_really_inline simd8 operator-(const simd8 other) const { 135| 12.8M| return _mm256_sub_epi8(*this, other); 136| 12.8M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd10sum_8bytesENS2_5simd8IhEE: 353| 50.3k|simdutf_really_inline simd64 sum_8bytes(const simd8 v) { 354| 50.3k| return _mm256_sad_epu8(v.value, simd8::zero()); 355| 50.3k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE9sum_bytesEv: 249| 2.77k| simdutf_really_inline uint64_t sum_bytes() const { 250| 2.77k| const auto tmp = _mm256_sad_epu8(value, _mm256_setzero_si256()); 251| | 252| 2.77k| return _mm256_extract_epi64(tmp, 0) + _mm256_extract_epi64(tmp, 1) + 253| 2.77k| _mm256_extract_epi64(tmp, 2) + _mm256_extract_epi64(tmp, 3); 254| 2.77k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEgeES4_: 229| 3.04M| operator>=(const simd8 other) const { 230| 3.04M| return other.min_val(*this) == other; 231| 3.04M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE7min_valES4_: 220| 3.04M| min_val(const simd8 other) const { 221| 3.04M| return _mm256_min_epu8(other, *this); 222| 3.04M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE2gtEa: 331| 752k| simdutf_really_inline uint64_t gt(const T m) const { 332| 752k| const simd8 mask = simd8::splat(m); 333| 752k| return simd8x64(this->chunks[0] > mask, this->chunks[1] > mask) 334| 752k| .to_bitmask(); 335| 752k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEC2EDv4_x: 17| 15.1M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEcvRKDv4_xEv: 19| 15.1M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE5splatEh: 105| 22.1M| static simdutf_really_inline simd8 splat(T _value) { 106| 22.1M| return _mm256_set1_epi8(_value); 107| 22.1M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEcvRKDv4_xEv: 19| 210M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdeqENS2_5simd8IhEES4_: 76| 8.23M| const simd8 rhs) { 77| 8.23M| return _mm256_cmpeq_epi8(lhs, rhs); 78| 8.23M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEanES5_: 53| 23.8M| simdutf_really_inline Child operator&(const Child other) const { 54| 23.8M| return _mm256_and_si256(*this, other); 55| 23.8M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IhEC2EPKh: 274| 4.87M| : chunks{simd8::load(ptr), 275| 4.87M| simd8::load(ptr + sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEanES5_: 53| 11.2M| simdutf_really_inline Child operator&(const Child other) const { 54| 11.2M| return _mm256_and_si256(*this, other); 55| 11.2M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 303| 1.00M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 304| 1.00M| this->chunks[0].template store_ascii_as_utf16(ptr + 305| 1.00M| sizeof(simd8) * 0); 306| 1.00M| this->chunks[1].template store_ascii_as_utf16(ptr + 307| 1.00M| sizeof(simd8) * 1); 308| 1.00M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 22| 2.00M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 23| 2.00M| __m256i first = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(*this)); 24| 2.00M| __m256i second = _mm256_cvtepu8_epi16(_mm256_extractf128_si256(*this, 1)); 25| 2.00M| if (big_endian) { ------------------ | Branch (25:9): [Folded, False: 2.00M] ------------------ 26| 0| const __m256i swap = _mm256_setr_epi8( 27| 0| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 28| 0| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 29| 0| first = _mm256_shuffle_epi8(first, swap); 30| 0| second = _mm256_shuffle_epi8(second, swap); 31| 0| } 32| 2.00M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), first); 33| 2.00M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 16), second); 34| 2.00M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 303| 1.13M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 304| 1.13M| this->chunks[0].template store_ascii_as_utf16(ptr + 305| 1.13M| sizeof(simd8) * 0); 306| 1.13M| this->chunks[1].template store_ascii_as_utf16(ptr + 307| 1.13M| sizeof(simd8) * 1); 308| 1.13M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 22| 2.26M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 23| 2.26M| __m256i first = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(*this)); 24| 2.26M| __m256i second = _mm256_cvtepu8_epi16(_mm256_extractf128_si256(*this, 1)); 25| 2.26M| if (big_endian) { ------------------ | Branch (25:9): [True: 2.26M, Folded] ------------------ 26| 2.26M| const __m256i swap = _mm256_setr_epi8( 27| 2.26M| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 28| 2.26M| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 29| 2.26M| first = _mm256_shuffle_epi8(first, swap); 30| 2.26M| second = _mm256_shuffle_epi8(second, swap); 31| 2.26M| } 32| 2.26M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), first); 33| 2.26M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 16), second); 34| 2.26M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEorES5_: 50| 4.38M| simdutf_really_inline Child operator|(const Child other) const { 51| 4.38M| return _mm256_or_si256(*this, other); 52| 4.38M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEC2EDv4_x: 97| 71.8M| : base16_numeric(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItEC2EDv4_x: 73| 71.8M| : base16(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6base16ItNS2_6simd16IbEEEC2EDv4_x: 20| 71.8M| : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE4packERKS4_S6_: 149| 5.15M| const simd16 &v1) { 150| | // Note: the AVX2 variant of pack operates on 128-bit lanes, thus 151| | // we have to shuffle lanes in order to produce bytes in the 152| | // correct order. 153| | 154| | // get the 0th lanes 155| 5.15M| const __m128i lo_0 = _mm256_extracti128_si256(v0, 0); 156| 5.15M| const __m128i lo_1 = _mm256_extracti128_si256(v1, 0); 157| | 158| | // get the 1st lanes 159| 5.15M| const __m128i hi_0 = _mm256_extracti128_si256(v0, 1); 160| 5.15M| const __m128i hi_1 = _mm256_extracti128_si256(v1, 1); 161| | 162| | // build new vectors (shuffle lanes) 163| 5.15M| const __m256i t0 = _mm256_set_m128i(lo_1, lo_0); ------------------ | | 4| 5.15M| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 5.15M| _mm256_castsi128_si256(xmm2), 2) ------------------ 164| 5.15M| const __m256i t1 = _mm256_set_m128i(hi_1, hi_0); ------------------ | | 4| 5.15M| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 5.15M| _mm256_castsi128_si256(xmm2), 2) ------------------ 165| | 166| | // pack code units in linear order from v0 and v1 167| 5.15M| return _mm256_packus_epi16(t0, t1); 168| 5.15M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd9simd16x32ItEC2EPKt: 202| 2.19M| : chunks{simd16::load(ptr), 203| 2.19M| simd16::load(ptr + sizeof(simd16) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItE4loadEPKt: 66| 16.7M| static simdutf_really_inline simd16 load(const T values[8]) { 67| 16.7M| return _mm256_loadu_si256(reinterpret_cast(values)); 68| 16.7M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd9simd16x32ItE10swap_bytesEv: 229| 1.03M| simdutf_really_inline void swap_bytes() { 230| 1.03M| this->chunks[0] = this->chunks[0].swap_bytes(); 231| 1.03M| this->chunks[1] = this->chunks[1].swap_bytes(); 232| 1.03M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE10swap_bytesEv: 139| 3.02M| simdutf_really_inline simd16 swap_bytes() const { 140| 3.02M| const __m256i swap = _mm256_setr_epi8( 141| 3.02M| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 142| 3.02M| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 143| 3.02M| return _mm256_shuffle_epi8(*this, swap); 144| 3.02M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItE5splatEt: 58| 15.0M| static simdutf_really_inline simd16 splat(T _value) { 59| 15.0M| return _mm256_set1_epi16(_value); 60| 15.0M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdeqENS2_6simd16ItEES4_: 26| 10.7M| const simd16 rhs) { 27| 10.7M| return _mm256_cmpeq_epi16(lhs, rhs); 28| 10.7M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16IbEC2EDv4_x: 45| 15.1M| simdutf_really_inline simd16(const __m256i _value) : base16(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6base16IbNS2_6simd16IbEEEC2EDv4_x: 20| 15.1M| : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd9simd16x32IbEC2ENS2_6simd16IbEES6_: 200| 2.19M| : chunks{chunk0, chunk1} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd9simd16x32IbE10to_bitmaskEv: 210| 2.19M| simdutf_really_inline uint64_t to_bitmask() const { 211| 2.19M| uint64_t r_lo = uint32_t(this->chunks[0].to_bitmask()); 212| 2.19M| uint64_t r_hi = this->chunks[1].to_bitmask(); 213| 2.19M| return r_lo | (r_hi << 32); 214| 2.19M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16IbE10to_bitmaskEv: 50| 4.38M| simdutf_really_inline bitmask_type to_bitmask() const { 51| 4.38M| return _mm256_movemask_epi8(*this); 52| 4.38M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE3shrILi8EEES4_v: 134| 5.08M| template simdutf_really_inline simd16 shr() const { 135| 5.08M| return simd16(_mm256_srli_epi16(*this, N)); 136| 5.08M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEleES4_: 117| 4.38M| operator<=(const simd16 other) const { 118| 4.38M| return other.max_val(*this) == other; 119| 4.38M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE7max_valES4_: 108| 4.38M| max_val(const simd16 other) const { 109| 4.38M| return _mm256_max_epu16(*this, other); 110| 4.38M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEC2EPKDs: 104| 10.3M| : simd16(load(reinterpret_cast(values))) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEC2Et: 100| 10.6M| simdutf_really_inline simd16(uint16_t _value) : simd16(splat(_value)) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd9simd16x32ItE12not_in_rangeEtt: 249| 2.19M| simdutf_really_inline uint64_t not_in_range(const T low, const T high) const { 250| 2.19M| const simd16 mask_low = simd16::splat(static_cast(low - 1)); 251| 2.19M| const simd16 mask_high = simd16::splat(static_cast(high + 1)); 252| 2.19M| return simd16x32( 253| 2.19M| (this->chunks[0] >= mask_high) | (this->chunks[0] <= mask_low), 254| 2.19M| (this->chunks[1] >= mask_high) | (this->chunks[1] <= mask_low)) 255| 2.19M| .to_bitmask(); 256| 2.19M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEgeES4_: 121| 4.38M| operator>=(const simd16 other) const { 122| 4.38M| return other.min_val(*this) == other; 123| 4.38M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE7min_valES4_: 112| 4.38M| min_val(const simd16 other) const { 113| 4.38M| return _mm256_min_epu16(*this, other); 114| 4.38M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItE4zeroEv: 62| 1.53k| static simdutf_really_inline simd16 zero() { 63| 1.53k| return _mm256_setzero_si256(); 64| 1.53k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd3minENS2_6simd16ItEES4_: 259| 4.00M|simd16 min(const simd16 a, simd16 b) { 260| 4.00M| return _mm256_min_epu16(a.value, b.value); 261| 4.00M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItEpLENS2_6simd16ItEE: 87| 6.01M| simdutf_really_inline simd16 &operator+=(const simd16 other) { 88| 6.01M| *this = *this + other; 89| 6.01M| return *static_cast *>(this); 90| 6.01M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItEplENS2_6simd16ItEE: 84| 6.01M| simdutf_really_inline simd16 operator+(const simd16 other) const { 85| 6.01M| return _mm256_add_epi16(*this, other); 86| 6.01M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE3sumEv: 170| 1.53k| simdutf_really_inline uint64_t sum() const { 171| 1.53k| const auto lo_u16 = _mm256_and_si256(value, _mm256_set1_epi32(0x0000ffff)); 172| 1.53k| const auto hi_u16 = _mm256_srli_epi32(value, 16); 173| 1.53k| const auto sum_u32 = _mm256_add_epi32(lo_u16, hi_u16); 174| | 175| 1.53k| const auto lo_u32 = 176| 1.53k| _mm256_and_si256(sum_u32, _mm256_set1_epi64x(0xffffffff)); 177| 1.53k| const auto hi_u32 = _mm256_srli_epi64(sum_u32, 32); 178| 1.53k| const auto sum_u64 = _mm256_add_epi64(lo_u32, hi_u32); 179| | 180| 1.53k| return uint64_t(_mm256_extract_epi64(sum_u64, 0)) + 181| 1.53k| uint64_t(_mm256_extract_epi64(sum_u64, 1)) + 182| 1.53k| uint64_t(_mm256_extract_epi64(sum_u64, 2)) + 183| | uint64_t(_mm256_extract_epi64(sum_u64, 3)); 184| 1.53k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE5splatEj: 46| 19.1k| simdutf_really_inline static simd32 splat(uint32_t v) { 47| 19.1k| return _mm256_set1_epi32(v); 48| 19.1k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjEC2EDv4_x: 9| 7.09M| simdutf_really_inline simd32(const __m256i v) : value(v) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE4zeroEv: 42| 5.84k| simdutf_really_inline static simd32 zero() { 43| 5.84k| return _mm256_setzero_si256(); 44| 5.84k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjEC2IDiEEPKT_: 13| 1.71M| : value(_mm256_loadu_si256(reinterpret_cast(ptr))) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE10swap_bytesEv: 27| 1.36M| simdutf_really_inline simd32 swap_bytes() const { 28| 1.36M| const __m256i shuffle = 29| 1.36M| _mm256_setr_epi8(3, 2, 1, 0, 7, 6, 5, 4, 8, 9, 10, 11, 15, 14, 13, 12, 30| 1.36M| 3, 2, 1, 0, 7, 6, 5, 4, 8, 9, 10, 11, 15, 14, 13, 12); 31| | 32| 1.36M| return _mm256_shuffle_epi8(value, shuffle); 33| 1.36M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd3maxENS2_6simd32IjEES4_: 80| 2.23M| const simd32 b) { 81| 2.23M| return _mm256_max_epu32(a.value, b.value); 82| 2.23M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdplENS2_6simd32IjEES4_: 90| 1.11M| const simd32 b) { 91| 1.11M| return _mm256_add_epi32(a.value, b.value); 92| 1.11M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdgtENS2_6simd32IjEES4_: 109| 4.25k| const simd32 b) { 110| 4.25k| return !(b >= a); 111| 4.25k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdntENS2_6simd32IbEE: 104| 4.25k|simdutf_really_inline simd32 operator!(const simd32 v) { 105| 4.25k| return _mm256_xor_si256(v.value, _mm256_set1_epi8(-1)); 106| 4.25k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IbEC2EDv4_x: 59| 755k| simdutf_really_inline simd32(const __m256i v) : value(v) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdgeENS2_6simd32IjEES4_: 100| 253k| const simd32 b) { 101| 253k| return _mm256_cmpeq_epi32(_mm256_max_epu32(a.value, b.value), a.value); 102| 253k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd32IbE3anyEv: 61| 253k| simdutf_really_inline bool any() const { 62| 253k| return _mm256_movemask_epi8(value) != 0; 63| 253k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdeqENS2_6simd32IjEES4_: 95| 248k| const simd32 b) { 96| 248k| return _mm256_cmpeq_epi32(a.value, b.value); 97| 248k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdanENS2_6simd32IjEES4_: 85| 1.29M| const simd32 a) { 86| 1.29M| return _mm256_and_si256(a.value, b.value); 87| 1.29M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdorIbEENS2_6simd32IT_EES6_S6_: 70| 248k| const simd32 b) { 71| 248k| return _mm256_or_si256(a.value, b.value); 72| 248k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd3minENS2_6simd32IjEES4_: 75| 1.04M| const simd32 a) { 76| 1.04M| return _mm256_min_epu32(a.value, b.value); 77| 1.04M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjEpLES4_: 36| 1.04M| simdutf_really_inline simd32 &operator+=(const simd32 other) { 37| 1.04M| value = _mm256_add_epi32(value, other.value); 38| 1.04M| return *this; 39| 1.04M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE3sumEv: 15| 623| simdutf_really_inline uint64_t sum() const { 16| 623| const __m256i mask = _mm256_set1_epi64x(0xffffffff); 17| 623| const __m256i t0 = _mm256_and_si256(value, mask); 18| 623| const __m256i t1 = _mm256_srli_epi64(value, 32); 19| 623| const __m256i t2 = _mm256_add_epi64(t0, t1); 20| | 21| 623| return uint64_t(_mm256_extract_epi64(t2, 0)) + 22| 623| uint64_t(_mm256_extract_epi64(t2, 1)) + 23| 623| uint64_t(_mm256_extract_epi64(t2, 2)) + 24| | uint64_t(_mm256_extract_epi64(t2, 3)); 25| 623| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd64ImE4zeroEv: 27| 5.98k| simdutf_really_inline static simd64 zero() { 28| 5.98k| return _mm256_setzero_si256(); 29| 5.98k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd64ImEC2EDv4_x: 9| 56.3k| simdutf_really_inline simd64(const __m256i v) : value(v) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd64ImEpLES4_: 21| 50.3k| simdutf_really_inline simd64 &operator+=(const simd64 other) { 22| 50.3k| value = _mm256_add_epi64(value, other.value); 23| 50.3k| return *this; 24| 50.3k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd64ImE3sumEv: 15| 5.98k| simdutf_really_inline uint64_t sum() const { 16| 5.98k| return _mm256_extract_epi64(value, 0) + _mm256_extract_epi64(value, 1) + 17| 5.98k| _mm256_extract_epi64(value, 2) + _mm256_extract_epi64(value, 3); 18| 5.98k| } _ZN7simdutf7icelake14implementationC2Ev: 16| 1| : simdutf::implementation( 17| 1| "icelake", 18| 1| "Intel AVX512 (AVX-512BW, AVX-512CD, AVX-512VL, AVX-512VBMI2 " 19| 1| "extensions)", 20| 1| internal::instruction_set::AVX2 | internal::instruction_set::BMI1 | 21| 1| internal::instruction_set::BMI2 | 22| 1| internal::instruction_set::AVX512BW | 23| 1| internal::instruction_set::AVX512CD | 24| 1| internal::instruction_set::AVX512VL | 25| 1| internal::instruction_set::AVX512VBMI2 | 26| 1| internal::instruction_set::AVX512VPOPCNTDQ) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_110count_onesEm: 14| 4.12M|simdutf_really_inline long long int count_ones(uint64_t input_num) { 15| | return _popcnt64(input_num); 16| 4.12M|} _ZN7simdutf8westmere14implementationC2Ev: 18| 1| : simdutf::implementation("westmere", "Intel/AMD SSE4.2", 19| 1| internal::instruction_set::SSE42) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2Ev: 183| 20.9k| simdutf_really_inline simd8() : base8_numeric() {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEC2Ev: 118| 20.9k| simdutf_really_inline base8_numeric() : base8() {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2Ev: 68| 20.9k| simdutf_really_inline base8() : base>() {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2Ev: 13| 20.9k| simdutf_really_inline base() : value{__m128i()} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IhE9reduce_orEv: 290| 4.87M| simdutf_really_inline simd8 reduce_or() const { 291| 4.87M| return (this->chunks[0] | this->chunks[1]) | 292| 4.87M| (this->chunks[2] | this->chunks[3]); 293| 4.87M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEorES5_: 42| 30.9M| simdutf_really_inline Child operator|(const Child other) const { 43| 30.9M| return _mm_or_si128(*this, other); 44| 30.9M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2EDv2_x: 185| 304M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEC2EDv2_x: 120| 304M| : base8(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2EDv2_x: 69| 304M| simdutf_really_inline base8(const __m128i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2EDv2_x: 16| 304M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE8is_asciiEv: 228| 4.87M| simdutf_really_inline bool is_ascii() const { 229| 4.87M| return _mm_movemask_epi8(*this) == 0; 230| 4.87M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEoRES5_: 51| 10.2M| simdutf_really_inline Child &operator|=(const Child other) { 52| 10.2M| auto this_cast = static_cast(this); 53| 10.2M| *this_cast = *this_cast | other; 54| 10.2M| return *this_cast; 55| 10.2M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi1EEENS4_IhEES8_: 79| 6.27M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 80| | return _mm_alignr_epi8(*this, prev_chunk, 16 - N); 81| 6.27M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE3shrILi4EEES4_v: 238| 12.5M| template simdutf_really_inline simd8 shr() const { 239| 12.5M| return simd8(_mm_srli_epi16(*this, N)) & uint8_t(0xFFu >> N); 240| 12.5M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_: 151| 18.8M| L replace15) const { 152| 18.8M| return lookup_16(simd8::repeat_16( 153| 18.8M| replace0, replace1, replace2, replace3, replace4, replace5, replace6, 154| 18.8M| replace7, replace8, replace9, replace10, replace11, replace12, 155| 18.8M| replace13, replace14, replace15)); 156| 18.8M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES8_: 142| 18.8M| simdutf_really_inline simd8 lookup_16(simd8 lookup_table) const { 143| 18.8M| return _mm_shuffle_epi8(lookup_table, *this); 144| 18.8M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE9repeat_16Ehhhhhhhhhhhhhhhh: 113| 18.8M| T v14, T v15) { 114| 18.8M| return simd8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, 115| 18.8M| v14, v15); 116| 18.8M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2Ehhhhhhhhhhhhhhhh: 196| 18.8M| : simd8(_mm_setr_epi8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, 197| 18.8M| v12, v13, v14, v15)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2Eh: 188| 43.6M| simdutf_really_inline simd8(uint8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi2EEENS4_IhEES8_: 79| 6.01M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 80| | return _mm_alignr_epi8(*this, prev_chunk, 16 - N); 81| 6.01M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi3EEENS4_IhEES8_: 79| 6.01M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 80| | return _mm_alignr_epi8(*this, prev_chunk, 16 - N); 81| 6.01M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE14saturating_subES4_: 201| 12.9M| saturating_sub(const simd8 other) const { 202| 12.9M| return _mm_subs_epu8(*this, other); 203| 12.9M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IbEC2EDv2_x: 91| 44.1M| simdutf_really_inline simd8(const __m128i _value) : base8(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IbNS2_5simd8IbEEEC2EDv2_x: 69| 44.1M| simdutf_really_inline base8(const __m128i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEC2EDv2_x: 16| 44.1M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEcvRKDv2_xEv: 18| 44.1M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEeoES5_: 48| 6.01M| simdutf_really_inline Child operator^(const Child other) const { 49| 6.01M| return _mm_xor_si128(*this, other); 50| 6.01M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2EPKh: 190| 881k| simdutf_really_inline simd8(const uint8_t *values) : simd8(load(values)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE4loadEPKh: 106| 20.3M| static simdutf_really_inline simd8 load(const T values[16]) { 107| 20.3M| return _mm_loadu_si128(reinterpret_cast(values)); 108| 20.3M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE7gt_bitsES4_: 212| 881k| gt_bits(const simd8 other) const { 213| 881k| return this->saturating_sub(other); 214| 881k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEC2EDv2_x: 16| 116M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEcvRKDv2_xEv: 18| 158M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IbE10to_bitmaskEv: 95| 12.4M| simdutf_really_inline int to_bitmask() const { 96| 12.4M| return _mm_movemask_epi8(*this); 97| 12.4M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE21any_bits_set_anywhereEv: 235| 2.30M| simdutf_really_inline bool any_bits_set_anywhere() const { 236| 2.30M| return !bits_not_set_anywhere(); 237| 2.30M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE21bits_not_set_anywhereEv: 232| 2.30M| simdutf_really_inline bool bits_not_set_anywhere() const { 233| 2.30M| return _mm_testz_si128(*this, *this); 234| 2.30M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaEC2EPKa: 270| 6.04M| : chunks{simd8::load(ptr), 271| 6.04M| simd8::load(ptr + sizeof(simd8) / sizeof(T)), 272| 6.04M| simd8::load(ptr + 2 * sizeof(simd8) / sizeof(T)), 273| 6.04M| simd8::load(ptr + 3 * sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaE4loadEPKa: 106| 43.7M| static simdutf_really_inline simd8 load(const T values[16]) { 107| 43.7M| return _mm_loadu_si128(reinterpret_cast(values)); 108| 43.7M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEC2EDv2_x: 163| 81.0M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaEC2EDv2_x: 120| 81.0M| : base8(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IaNS2_5simd8IbEEEC2EDv2_x: 69| 81.0M| simdutf_really_inline base8(const __m128i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEC2EDv2_x: 16| 81.0M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE8is_asciiEv: 295| 5.29M| simdutf_really_inline bool is_ascii() const { 296| 5.29M| return this->reduce_or().is_ascii(); 297| 5.29M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE9reduce_orEv: 290| 5.29M| simdutf_really_inline simd8 reduce_or() const { 291| 5.29M| return (this->chunks[0] | this->chunks[1]) | 292| 5.29M| (this->chunks[2] | this->chunks[3]); 293| 5.29M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEorES5_: 42| 15.8M| simdutf_really_inline Child operator|(const Child other) const { 43| 15.8M| return _mm_or_si128(*this, other); 44| 15.8M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEcvRKDv2_xEv: 18| 130M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaE8is_asciiEv: 168| 5.29M| simdutf_really_inline bool is_ascii() const { 169| 5.29M| return _mm_movemask_epi8(*this) == 0; 170| 5.29M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE5storeEPa: 275| 903k| simdutf_really_inline void store(T *ptr) const { 276| 903k| this->chunks[0].store(ptr + sizeof(simd8) * 0 / sizeof(T)); 277| 903k| this->chunks[1].store(ptr + sizeof(simd8) * 1 / sizeof(T)); 278| 903k| this->chunks[2].store(ptr + sizeof(simd8) * 2 / sizeof(T)); 279| 903k| this->chunks[3].store(ptr + sizeof(simd8) * 3 / sizeof(T)); 280| 903k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaE5storeEPa: 123| 3.61M| simdutf_really_inline void store(T dst[16]) const { 124| 3.61M| return _mm_storeu_si128(reinterpret_cast<__m128i *>(dst), *this); 125| 3.61M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEcvNS3_IhEEEv: 251| 4.80M|simdutf_really_inline simd8::operator simd8() const { 252| 4.80M| return this->value; 253| 4.80M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE2ltEa: 326| 1.05M| simdutf_really_inline uint64_t lt(const T m) const { 327| 1.05M| const simd8 mask = simd8::splat(m); 328| 1.05M| return simd8x64(this->chunks[0] < mask, this->chunks[1] < mask, 329| 1.05M| this->chunks[2] < mask, this->chunks[3] < mask) 330| 1.05M| .to_bitmask(); 331| 1.05M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaE5splatEa: 102| 21.4M| static simdutf_really_inline simd8 splat(T _value) { 103| 21.4M| return _mm_set1_epi8(_value); 104| 21.4M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEltES4_: 176| 4.22M| simdutf_really_inline simd8 operator<(const simd8 other) const { 177| 4.22M| return _mm_cmpgt_epi8(other, *this); 178| 4.22M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IbEC2ENS2_5simd8IbEES6_S6_S6_: 268| 1.80M| : chunks{chunk0, chunk1, chunk2, chunk3} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IbE10to_bitmaskEv: 318| 1.80M| simdutf_really_inline uint64_t to_bitmask() const { 319| 1.80M| uint64_t r0 = uint32_t(this->chunks[0].to_bitmask()); 320| 1.80M| uint64_t r1 = this->chunks[1].to_bitmask(); 321| 1.80M| uint64_t r2 = this->chunks[2].to_bitmask(); 322| 1.80M| uint64_t r3 = this->chunks[3].to_bitmask(); 323| 1.80M| return r0 | (r1 << 16) | (r2 << 32) | (r3 << 48); 324| 1.80M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf32EPDi: 311| 1.19M| simdutf_really_inline void store_ascii_as_utf32(char32_t *ptr) const { 312| 1.19M| this->chunks[0].store_ascii_as_utf32(ptr + sizeof(simd8) * 0); 313| 1.19M| this->chunks[1].store_ascii_as_utf32(ptr + sizeof(simd8) * 1); 314| 1.19M| this->chunks[2].store_ascii_as_utf32(ptr + sizeof(simd8) * 2); 315| 1.19M| this->chunks[3].store_ascii_as_utf32(ptr + sizeof(simd8) * 3); 316| 1.19M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf32EPDi: 32| 4.78M| simdutf_really_inline void store_ascii_as_utf32(char32_t *p) const { 33| 4.78M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p), _mm_cvtepu8_epi32(*this)); 34| 4.78M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 4), 35| 4.78M| _mm_cvtepu8_epi32(_mm_srli_si128(*this, 4))); 36| 4.78M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 8), 37| 4.78M| _mm_cvtepu8_epi32(_mm_srli_si128(*this, 8))); 38| 4.78M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 12), 39| | _mm_cvtepu8_epi32(_mm_srli_si128(*this, 12))); 40| 4.78M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE4zeroEv: 105| 208k| static simdutf_really_inline simd8 zero() { return _mm_setzero_si128(); } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEgtES4_: 173| 22.6M| simdutf_really_inline simd8 operator>(const simd8 other) const { 174| 22.6M| return _mm_cmpgt_epi8(*this, other); 175| 22.6M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEC2Ea: 165| 19.6M| simdutf_really_inline simd8(int8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEmIENS2_5simd8IhEE: 134| 25.6M| simdutf_really_inline simd8 &operator-=(const simd8 other) { 135| 25.6M| *this = *this - other; 136| 25.6M| return *static_cast *>(this); 137| 25.6M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEmiENS2_5simd8IhEE: 131| 25.6M| simdutf_really_inline simd8 operator-(const simd8 other) const { 132| 25.6M| return _mm_sub_epi8(*this, other); 133| 25.6M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd10sum_8bytesENS2_5simd8IhEE: 366| 101k|simdutf_really_inline simd64 sum_8bytes(const simd8 v) { 367| 101k| return _mm_sad_epu8(v.value, simd8::zero()); 368| 101k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE9sum_bytesEv: 245| 3.84k| simdutf_really_inline uint64_t sum_bytes() const { 246| 3.84k| const auto tmp = _mm_sad_epu8(value, _mm_setzero_si128()); 247| 3.84k| return _mm_extract_epi64(tmp, 0) + _mm_extract_epi64(tmp, 1); 248| 3.84k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEgeES4_: 217| 6.08M| operator>=(const simd8 other) const { 218| 6.08M| return other.min_val(*this) == other; 219| 6.08M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE7min_valES4_: 207| 6.08M| min_val(const simd8 other) const { 208| 6.08M| return _mm_min_epu8(*this, other); 209| 6.08M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE2gtEa: 333| 752k| simdutf_really_inline uint64_t gt(const T m) const { 334| 752k| const simd8 mask = simd8::splat(m); 335| 752k| return simd8x64(this->chunks[0] > mask, this->chunks[1] > mask, 336| 752k| this->chunks[2] > mask, this->chunks[3] > mask) 337| 752k| .to_bitmask(); 338| 752k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEC2EDv2_x: 16| 30.3M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEcvRKDv2_xEv: 18| 30.3M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE5splatEh: 102| 43.6M| static simdutf_really_inline simd8 splat(T _value) { 103| 43.6M| return _mm_set1_epi8(_value); 104| 43.6M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEcvRKDv2_xEv: 18| 373M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_5simd8IhEES4_: 72| 11.2M| const simd8 rhs) { 73| 11.2M| return _mm_cmpeq_epi8(lhs, rhs); 74| 11.2M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEanES5_: 45| 42.5M| simdutf_really_inline Child operator&(const Child other) const { 46| 42.5M| return _mm_and_si128(*this, other); 47| 42.5M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IhEC2EPKh: 270| 4.87M| : chunks{simd8::load(ptr), 271| 4.87M| simd8::load(ptr + sizeof(simd8) / sizeof(T)), 272| 4.87M| simd8::load(ptr + 2 * sizeof(simd8) / sizeof(T)), 273| 4.87M| simd8::load(ptr + 3 * sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEanES5_: 45| 17.2M| simdutf_really_inline Child operator&(const Child other) const { 46| 17.2M| return _mm_and_si128(*this, other); 47| 17.2M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 300| 1.00M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 301| 1.00M| this->chunks[0].template store_ascii_as_utf16(ptr + 302| 1.00M| sizeof(simd8) * 0); 303| 1.00M| this->chunks[1].template store_ascii_as_utf16(ptr + 304| 1.00M| sizeof(simd8) * 1); 305| 1.00M| this->chunks[2].template store_ascii_as_utf16(ptr + 306| 1.00M| sizeof(simd8) * 2); 307| 1.00M| this->chunks[3].template store_ascii_as_utf16(ptr + 308| 1.00M| sizeof(simd8) * 3); 309| 1.00M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 20| 4.00M| simdutf_really_inline void store_ascii_as_utf16(char16_t *p) const { 21| 4.00M| __m128i first = _mm_cvtepu8_epi16(*this); 22| 4.00M| __m128i second = _mm_cvtepu8_epi16(_mm_srli_si128(*this, 8)); 23| 4.00M| if (big_endian) { ------------------ | Branch (23:9): [Folded, False: 4.00M] ------------------ 24| 0| const __m128i swap = 25| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 26| 0| first = _mm_shuffle_epi8(first, swap); 27| 0| second = _mm_shuffle_epi8(second, swap); 28| 0| } 29| 4.00M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p), first); 30| 4.00M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 8), second); 31| 4.00M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 300| 1.13M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 301| 1.13M| this->chunks[0].template store_ascii_as_utf16(ptr + 302| 1.13M| sizeof(simd8) * 0); 303| 1.13M| this->chunks[1].template store_ascii_as_utf16(ptr + 304| 1.13M| sizeof(simd8) * 1); 305| 1.13M| this->chunks[2].template store_ascii_as_utf16(ptr + 306| 1.13M| sizeof(simd8) * 2); 307| 1.13M| this->chunks[3].template store_ascii_as_utf16(ptr + 308| 1.13M| sizeof(simd8) * 3); 309| 1.13M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 20| 4.53M| simdutf_really_inline void store_ascii_as_utf16(char16_t *p) const { 21| 4.53M| __m128i first = _mm_cvtepu8_epi16(*this); 22| 4.53M| __m128i second = _mm_cvtepu8_epi16(_mm_srli_si128(*this, 8)); 23| 4.53M| if (big_endian) { ------------------ | Branch (23:9): [True: 4.53M, Folded] ------------------ 24| 4.53M| const __m128i swap = 25| 4.53M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 26| 4.53M| first = _mm_shuffle_epi8(first, swap); 27| 4.53M| second = _mm_shuffle_epi8(second, swap); 28| 4.53M| } 29| 4.53M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p), first); 30| 4.53M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 8), second); 31| 4.53M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd12as_vector_u8ENS2_6simd32IjEE: 370| 13.8k|simdutf_really_inline simd8 as_vector_u8(const simd32 v) { 371| 13.8k| return simd8(v.value); 372| 13.8k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEorES5_: 42| 8.77M| simdutf_really_inline Child operator|(const Child other) const { 43| 8.77M| return _mm_or_si128(*this, other); 44| 8.77M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEC2EDv2_x: 79| 116M| : base16_numeric(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItEC2EDv2_x: 54| 116M| : base16(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6base16ItNS2_6simd16IbEEEC2EDv2_x: 8| 116M| : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE4packERKS4_S6_: 130| 5.15M| const simd16 &v1) { 131| 5.15M| return _mm_packus_epi16(v0, v1); 132| 5.15M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd9simd16x32ItEC2EPKt: 164| 2.19M| : chunks{simd16::load(ptr), 165| 2.19M| simd16::load(ptr + sizeof(simd16) / sizeof(T)), 166| 2.19M| simd16::load(ptr + 2 * sizeof(simd16) / sizeof(T)), 167| 2.19M| simd16::load(ptr + 3 * sizeof(simd16) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItE4loadEPKt: 47| 23.1M| static simdutf_really_inline simd16 load(const T values[8]) { 48| 23.1M| return _mm_loadu_si128(reinterpret_cast(values)); 49| 23.1M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd9simd16x32ItE10swap_bytesEv: 200| 1.03M| simdutf_really_inline void swap_bytes() { 201| 1.03M| this->chunks[0] = this->chunks[0].swap_bytes(); 202| 1.03M| this->chunks[1] = this->chunks[1].swap_bytes(); 203| 1.03M| this->chunks[2] = this->chunks[2].swap_bytes(); 204| 1.03M| this->chunks[3] = this->chunks[3].swap_bytes(); 205| 1.03M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE10swap_bytesEv: 121| 6.05M| simdutf_really_inline simd16 swap_bytes() const { 122| 6.05M| const __m128i swap = 123| 6.05M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 124| 6.05M| return _mm_shuffle_epi8(*this, swap); 125| 6.05M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItE5splatEt: 41| 23.0M| static simdutf_really_inline simd16 splat(T _value) { 42| 23.0M| return _mm_set1_epi16(_value); 43| 23.0M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_6simd16ItEES4_: 11| 21.5M| const simd16 rhs) { 12| 21.5M| return _mm_cmpeq_epi16(lhs, rhs); 13| 21.5M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16IbEC2EDv2_x: 28| 30.3M| simdutf_really_inline simd16(const __m128i _value) : base16(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6base16IbNS2_6simd16IbEEEC2EDv2_x: 8| 30.3M| : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd9simd16x32IbEC2ENS2_6simd16IbEES6_S6_S6_: 162| 2.19M| : chunks{chunk0, chunk1, chunk2, chunk3} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd9simd16x32IbE10to_bitmaskEv: 192| 2.19M| simdutf_really_inline uint64_t to_bitmask() const { 193| 2.19M| uint64_t r0 = uint32_t(this->chunks[0].to_bitmask()); 194| 2.19M| uint64_t r1 = this->chunks[1].to_bitmask(); 195| 2.19M| uint64_t r2 = this->chunks[2].to_bitmask(); 196| 2.19M| uint64_t r3 = this->chunks[3].to_bitmask(); 197| 2.19M| return r0 | (r1 << 16) | (r2 << 32) | (r3 << 48); 198| 2.19M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16IbE10to_bitmaskEv: 33| 8.77M| simdutf_really_inline int to_bitmask() const { 34| 8.77M| return _mm_movemask_epi8(*this); 35| 8.77M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEgeES4_: 104| 8.77M| operator>=(const simd16 other) const { 105| 8.77M| return other.min_val(*this) == other; 106| 8.77M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE7min_valES4_: 95| 8.77M| min_val(const simd16 other) const { 96| 8.77M| return _mm_min_epu16(*this, other); 97| 8.77M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE3shrILi8EEES4_v: 116| 5.08M| template simdutf_really_inline simd16 shr() const { 117| 5.08M| return simd16(_mm_srli_epi16(*this, N)); 118| 5.08M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEleES4_: 100| 8.77M| operator<=(const simd16 other) const { 101| 8.77M| return other.max_val(*this) == other; 102| 8.77M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE7max_valES4_: 90| 8.77M| max_val(const simd16 other) const { 91| 8.77M| return _mm_max_epu16(*this, other); 92| 8.77M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEC2EPKDs: 86| 10.3M| : simd16(load(reinterpret_cast(values))) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEC2Et: 82| 18.6M| simdutf_really_inline simd16(uint16_t _value) : simd16(splat(_value)) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd9simd16x32ItE12not_in_rangeEtt: 228| 2.19M| simdutf_really_inline uint64_t not_in_range(const T low, const T high) const { 229| 2.19M| const simd16 mask_low = simd16::splat(static_cast(low - 1)); 230| 2.19M| const simd16 mask_high = simd16::splat(static_cast(high + 1)); 231| 2.19M| return simd16x32( 232| 2.19M| (this->chunks[0] >= mask_high) | (this->chunks[0] <= mask_low), 233| 2.19M| (this->chunks[1] >= mask_high) | (this->chunks[1] <= mask_low), 234| 2.19M| (this->chunks[2] >= mask_high) | (this->chunks[2] <= mask_low), 235| 2.19M| (this->chunks[3] >= mask_high) | (this->chunks[3] <= mask_low)) 236| 2.19M| .to_bitmask(); 237| 2.19M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItE4zeroEv: 45| 1.59k| static simdutf_really_inline simd16 zero() { return _mm_setzero_si128(); } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd3minENS2_6simd16ItEES4_: 240| 8.02M|simd16 min(const simd16 a, simd16 b) { 241| 8.02M| return _mm_min_epu16(a.value, b.value); 242| 8.02M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItEpLENS2_6simd16ItEE: 68| 12.0M| simdutf_really_inline simd16 &operator+=(const simd16 other) { 69| 12.0M| *this = *this + other; 70| 12.0M| return *static_cast *>(this); 71| 12.0M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItEplENS2_6simd16ItEE: 65| 12.0M| simdutf_really_inline simd16 operator+(const simd16 other) const { 66| 12.0M| return _mm_add_epi16(*this, other); 67| 12.0M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE3sumEv: 134| 1.59k| simdutf_really_inline uint64_t sum() const { 135| 1.59k| const auto lo_u16 = _mm_and_si128(value, _mm_set1_epi32(0x0000ffff)); 136| 1.59k| const auto hi_u16 = _mm_srli_epi32(value, 16); 137| 1.59k| const auto sum_u32 = _mm_add_epi32(lo_u16, hi_u16); 138| | 139| 1.59k| const auto lo_u32 = _mm_and_si128(sum_u32, _mm_set1_epi64x(0xffffffff)); 140| 1.59k| const auto hi_u32 = _mm_srli_epi64(sum_u32, 32); 141| 1.59k| const auto sum_u64 = _mm_add_epi64(lo_u32, hi_u32); 142| | 143| 1.59k| return uint64_t(_mm_extract_epi64(sum_u64, 0)) + 144| | uint64_t(_mm_extract_epi64(sum_u64, 1)); 145| 1.59k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE5splatEj: 56| 40.4k| simdutf_really_inline static simd32 splat(uint32_t v) { 57| 40.4k| return _mm_set1_epi32(v); 58| 40.4k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjEC2EDv2_x: 9| 14.3M| simdutf_really_inline simd32(const __m128i v) : value(v) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE4zeroEv: 52| 6.12k| simdutf_really_inline static simd32 zero() { 53| 6.12k| return _mm_setzero_si128(); 54| 6.12k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjEC2IDiEEPKT_: 13| 3.43M| : value(_mm_loadu_si128(reinterpret_cast(ptr))) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE10swap_bytesEv: 22| 2.73M| simdutf_really_inline simd32 swap_bytes() const { 23| 2.73M| const __m128i shuffle = 24| 2.73M| _mm_setr_epi8(3, 2, 1, 0, 7, 6, 5, 4, 8, 9, 10, 11, 15, 14, 13, 12); 25| | 26| 2.73M| return _mm_shuffle_epi8(value, shuffle); 27| 2.73M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd3maxENS2_6simd32IjEES4_: 94| 4.48M| const simd32 b) { 95| 4.48M| return _mm_max_epu32(a.value, b.value); 96| 4.48M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdplENS2_6simd32IjEES4_: 119| 2.25M| const simd32 b) { 120| 2.25M| return _mm_add_epi32(a.value, b.value); 121| 2.25M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdgtENS2_6simd32IjEES4_: 143| 25.3k| const simd32 b) { 144| 25.3k| return !(b >= a); 145| 25.3k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdntENS2_6simd32IbEE: 138| 25.3k|simdutf_really_inline simd32 operator!(const simd32 v) { 139| 25.3k| return _mm_xor_si128(v.value, _mm_set1_epi8(-1)); 140| 25.3k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IbEC2EDv2_x: 69| 1.57M| simdutf_really_inline simd32(const __m128i v) : value(v) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdgeENS2_6simd32IjEES4_: 134| 522k| const simd32 b) { 135| 522k| return _mm_cmpeq_epi32(_mm_max_epu32(a.value, b.value), a.value); 136| 522k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IbE3anyEv: 71| 508k| simdutf_really_inline bool any() const { 72| 508k| return _mm_movemask_epi8(value) != 0; 73| 508k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_6simd32IjEES4_: 129| 497k| const simd32 b) { 130| 497k| return _mm_cmpeq_epi32(a.value, b.value); 131| 497k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdanENS2_6simd32IjEES4_: 104| 2.59M| const simd32 b) { 105| 2.59M| return _mm_and_si128(a.value, b.value); 106| 2.59M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdorIbEENS2_6simd32IT_EES6_S6_: 84| 511k| const simd32 b) { 85| 511k| return _mm_or_si128(a.value, b.value); 86| 511k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd3minENS2_6simd32IjEES4_: 89| 2.09M| const simd32 b) { 90| 2.09M| return _mm_min_epu32(a.value, b.value); 91| 2.09M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjEpLES4_: 46| 2.09M| simdutf_really_inline simd32 &operator+=(const simd32 other) { 47| 2.09M| value = _mm_add_epi32(value, other.value); 48| 2.09M| return *this; 49| 2.09M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE3sumEv: 15| 895| simdutf_really_inline uint64_t sum() const { 16| 895| return uint64_t(_mm_extract_epi32(value, 0)) + 17| 895| uint64_t(_mm_extract_epi32(value, 1)) + 18| 895| uint64_t(_mm_extract_epi32(value, 2)) + 19| | uint64_t(_mm_extract_epi32(value, 3)); 20| 895| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_6simd32IjEEj: 99| 13.8k| uint32_t b) { 100| 13.8k| return _mm_cmpeq_epi32(a.value, _mm_set1_epi32(b)); 101| 13.8k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdanENS2_6simd32IjEEj: 109| 41.6k| uint32_t b) { 110| 41.6k| return _mm_and_si128(a.value, _mm_set1_epi32(b)); 111| 41.6k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IbE15to_4bit_bitmaskEv: 75| 13.8k| simdutf_really_inline uint8_t to_4bit_bitmask() const { 76| 13.8k| return uint8_t(_mm_movemask_ps(_mm_castsi128_ps(value))); 77| 13.8k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdmiENS2_6simd32IjEEj: 124| 13.8k| uint32_t b) { 125| 13.8k| return _mm_sub_epi32(a.value, _mm_set1_epi32(b)); 126| 13.8k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE3shrILi10EEES4_v: 29| 13.8k| template simdutf_really_inline simd32 shr() const { 30| 13.8k| return _mm_srli_epi32(value, N); 31| 13.8k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE3shlILi16EEES4_v: 33| 13.8k| template simdutf_really_inline simd32 shl() const { 34| 13.8k| return _mm_slli_epi32(value, N); 35| 13.8k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdorENS2_6simd32IjEEj: 114| 13.8k| uint32_t b) { 115| 13.8k| return _mm_or_si128(a.value, _mm_set1_epi32(b)); 116| 13.8k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdorIjEENS2_6simd32IT_EES6_S6_: 84| 13.8k| const simd32 b) { 85| 13.8k| return _mm_or_si128(a.value, b.value); 86| 13.8k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6selectENS2_6simd32IbEENS3_IjEES5_: 149| 13.8k| const simd32 v_false) { 150| 13.8k| return _mm_blendv_epi8(v_false.value, v_true.value, cond.value); 151| 13.8k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd64ImE4zeroEv: 26| 5.98k| simdutf_really_inline static simd64 zero() { 27| 5.98k| return _mm_setzero_si128(); 28| 5.98k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd64ImEC2EDv2_x: 9| 107k| simdutf_really_inline simd64(const __m128i v) : value(v) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd64ImEpLES4_: 20| 101k| simdutf_really_inline simd64 &operator+=(const simd64 other) { 21| 101k| value = _mm_add_epi64(value, other.value); 22| 101k| return *this; 23| 101k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd64ImE3sumEv: 15| 5.98k| simdutf_really_inline uint64_t sum() const { 16| 5.98k| return _mm_extract_epi64(value, 0) + _mm_extract_epi64(value, 1); 17| 5.98k| } _ZNK7simdutf8westmere14implementation13validate_utf8EPKcm: 299| 3.48k|implementation::validate_utf8(const char *buf, size_t len) const noexcept { 300| 3.48k| return westmere::utf8_validation::generic_validate_utf8(buf, len); 301| 3.48k|} _ZNK7simdutf8westmere14implementation25validate_utf8_with_errorsEPKcm: 306| 3.48k| const char *buf, size_t len) const noexcept { 307| 3.48k| return westmere::utf8_validation::generic_validate_utf8_with_errors(buf, len); 308| 3.48k|} _ZNK7simdutf8westmere14implementation16validate_utf16leEPKDsm: 345| 1.35k| size_t len) const noexcept { 346| 1.35k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.35k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 17, False: 1.33k] | | ------------------ ------------------ 347| | // empty input is valid UTF-16. protect the implementation from 348| | // handling nullptr 349| 17| return true; 350| 17| } 351| 1.33k| const auto res = 352| 1.33k| westmere::utf16::validate_utf16_with_errors(buf, len); 353| 1.33k| if (res.is_err()) { ------------------ | Branch (353:7): [True: 139, False: 1.19k] ------------------ 354| 139| return false; 355| 139| } 356| | 357| 1.19k| if (res.count == len) ------------------ | Branch (357:7): [True: 0, False: 1.19k] ------------------ 358| 0| return true; 359| | 360| 1.19k| return scalar::utf16::validate(buf + res.count, 361| 1.19k| len - res.count); 362| 1.19k|} _ZNK7simdutf8westmere14implementation16validate_utf16beEPKDsm: 368| 1.33k| size_t len) const noexcept { 369| 1.33k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.33k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 1.32k] | | ------------------ ------------------ 370| | // empty input is valid UTF-16. protect the implementation from 371| | // handling nullptr 372| 16| return true; 373| 16| } 374| 1.32k| const auto res = 375| 1.32k| westmere::utf16::validate_utf16_with_errors(buf, len); 376| 1.32k| if (res.is_err()) { ------------------ | Branch (376:7): [True: 155, False: 1.16k] ------------------ 377| 155| return false; 378| 155| } 379| | 380| 1.16k| if (res.count == len) ------------------ | Branch (380:7): [True: 0, False: 1.16k] ------------------ 381| 0| return true; 382| | 383| 1.16k| return scalar::utf16::validate(buf + res.count, 384| 1.16k| len - res.count); 385| 1.16k|} _ZNK7simdutf8westmere14implementation28validate_utf16le_with_errorsEPKDsm: 388| 1.35k| const char16_t *buf, size_t len) const noexcept { 389| 1.35k| const result res = 390| 1.35k| westmere::utf16::validate_utf16_with_errors(buf, len); 391| 1.35k| if (res.count != len) { ------------------ | Branch (391:7): [True: 1.33k, False: 17] ------------------ 392| 1.33k| const result scalar_res = 393| 1.33k| scalar::utf16::validate_with_errors( 394| 1.33k| buf + res.count, len - res.count); 395| 1.33k| return result(scalar_res.error, res.count + scalar_res.count); 396| 1.33k| } else { 397| 17| return res; 398| 17| } 399| 1.35k|} _ZNK7simdutf8westmere14implementation28validate_utf16be_with_errorsEPKDsm: 402| 1.33k| const char16_t *buf, size_t len) const noexcept { 403| 1.33k| const result res = 404| 1.33k| westmere::utf16::validate_utf16_with_errors(buf, len); 405| 1.33k| if (res.count != len) { ------------------ | Branch (405:7): [True: 1.32k, False: 16] ------------------ 406| 1.32k| result scalar_res = scalar::utf16::validate_with_errors( 407| 1.32k| buf + res.count, len - res.count); 408| 1.32k| return result(scalar_res.error, res.count + scalar_res.count); 409| 1.32k| } else { 410| 16| return res; 411| 16| } 412| 1.33k|} _ZNK7simdutf8westmere14implementation14validate_utf32EPKDim: 427| 2.63k|implementation::validate_utf32(const char32_t *buf, size_t len) const noexcept { 428| 2.63k| return utf32::validate(buf, len); 429| 2.63k|} _ZNK7simdutf8westmere14implementation26validate_utf32_with_errorsEPKDim: 434| 2.63k| const char32_t *buf, size_t len) const noexcept { 435| 2.63k| return utf32::validate_with_errors(buf, len); 436| 2.63k|} _ZNK7simdutf8westmere14implementation22convert_latin1_to_utf8EPKcmPc: 441| 604| const char *buf, size_t len, char *utf8_output) const noexcept { 442| | 443| 604| std::pair ret = 444| 604| sse_convert_latin1_to_utf8(buf, len, utf8_output); 445| 604| size_t converted_chars = ret.second - utf8_output; 446| | 447| 604| if (ret.first != buf + len) { ------------------ | Branch (447:7): [True: 512, False: 92] ------------------ 448| 512| const size_t scalar_converted_chars = scalar::latin1_to_utf8::convert( 449| 512| ret.first, len - (ret.first - buf), ret.second); 450| 512| converted_chars += scalar_converted_chars; 451| 512| } 452| | 453| 604| return converted_chars; 454| 604|} _ZNK7simdutf8westmere14implementation25convert_latin1_to_utf16leEPKcmPDs: 459| 76| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 460| 76| std::pair ret = 461| 76| sse_convert_latin1_to_utf16(buf, len, utf16_output); 462| 76| if (ret.first == nullptr) { ------------------ | Branch (462:7): [True: 4, False: 72] ------------------ 463| 4| return 0; 464| 4| } 465| 72| size_t converted_chars = ret.second - utf16_output; 466| 72| if (ret.first != buf + len) { ------------------ | Branch (466:7): [True: 58, False: 14] ------------------ 467| 58| const size_t scalar_converted_chars = 468| 58| scalar::latin1_to_utf16::convert( 469| 58| ret.first, len - (ret.first - buf), ret.second); 470| 58| if (scalar_converted_chars == 0) { ------------------ | Branch (470:9): [True: 0, False: 58] ------------------ 471| 0| return 0; 472| 0| } 473| 58| converted_chars += scalar_converted_chars; 474| 58| } 475| 72| return converted_chars; 476| 72|} _ZNK7simdutf8westmere14implementation25convert_latin1_to_utf16beEPKcmPDs: 479| 96| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 480| 96| std::pair ret = 481| 96| sse_convert_latin1_to_utf16(buf, len, utf16_output); 482| 96| if (ret.first == nullptr) { ------------------ | Branch (482:7): [True: 4, False: 92] ------------------ 483| 4| return 0; 484| 4| } 485| 92| size_t converted_chars = ret.second - utf16_output; 486| 92| if (ret.first != buf + len) { ------------------ | Branch (486:7): [True: 78, False: 14] ------------------ 487| 78| const size_t scalar_converted_chars = 488| 78| scalar::latin1_to_utf16::convert( 489| 78| ret.first, len - (ret.first - buf), ret.second); 490| 78| if (scalar_converted_chars == 0) { ------------------ | Branch (490:9): [True: 0, False: 78] ------------------ 491| 0| return 0; 492| 0| } 493| 78| converted_chars += scalar_converted_chars; 494| 78| } 495| 92| return converted_chars; 496| 92|} _ZNK7simdutf8westmere14implementation23convert_latin1_to_utf32EPKcmPDi: 501| 124| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 502| 124| std::pair ret = 503| 124| sse_convert_latin1_to_utf32(buf, len, utf32_output); 504| 124| if (ret.first == nullptr) { ------------------ | Branch (504:7): [True: 4, False: 120] ------------------ 505| 4| return 0; 506| 4| } 507| 120| size_t converted_chars = ret.second - utf32_output; 508| 120| if (ret.first != buf + len) { ------------------ | Branch (508:7): [True: 106, False: 14] ------------------ 509| 106| const size_t scalar_converted_chars = scalar::latin1_to_utf32::convert( 510| 106| ret.first, len - (ret.first - buf), ret.second); 511| 106| if (scalar_converted_chars == 0) { ------------------ | Branch (511:9): [True: 0, False: 106] ------------------ 512| 0| return 0; 513| 0| } 514| 106| converted_chars += scalar_converted_chars; 515| 106| } 516| 120| return converted_chars; 517| 120|} _ZNK7simdutf8westmere14implementation22convert_utf8_to_latin1EPKcmPc: 522| 586| const char *buf, size_t len, char *latin1_output) const noexcept { 523| 586| utf8_to_latin1::validating_transcoder converter; 524| 586| return converter.convert(buf, len, latin1_output); 525| 586|} _ZNK7simdutf8westmere14implementation34convert_utf8_to_latin1_with_errorsEPKcmPc: 528| 512| const char *buf, size_t len, char *latin1_output) const noexcept { 529| 512| utf8_to_latin1::validating_transcoder converter; 530| 512| return converter.convert_with_errors(buf, len, latin1_output); 531| 512|} _ZNK7simdutf8westmere14implementation23convert_utf8_to_utf16leEPKcmPDs: 541| 776| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 542| 776| utf8_to_utf16::validating_transcoder converter; 543| 776| return converter.convert(buf, len, utf16_output); 544| 776|} _ZNK7simdutf8westmere14implementation23convert_utf8_to_utf16beEPKcmPDs: 547| 794| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 548| 794| utf8_to_utf16::validating_transcoder converter; 549| 794| return converter.convert(buf, len, utf16_output); 550| 794|} _ZNK7simdutf8westmere14implementation35convert_utf8_to_utf16le_with_errorsEPKcmPDs: 553| 610| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 554| 610| utf8_to_utf16::validating_transcoder converter; 555| 610| return converter.convert_with_errors(buf, len, 556| 610| utf16_output); 557| 610|} _ZNK7simdutf8westmere14implementation35convert_utf8_to_utf16be_with_errorsEPKcmPDs: 560| 706| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 561| 706| utf8_to_utf16::validating_transcoder converter; 562| 706| return converter.convert_with_errors(buf, len, utf16_output); 563| 706|} _ZNK7simdutf8westmere14implementation29convert_valid_utf8_to_utf16leEPKcmPDs: 566| 450| const char *input, size_t size, char16_t *utf16_output) const noexcept { 567| 450| return utf8_to_utf16::convert_valid(input, size, 568| 450| utf16_output); 569| 450|} _ZNK7simdutf8westmere14implementation29convert_valid_utf8_to_utf16beEPKcmPDs: 572| 510| const char *input, size_t size, char16_t *utf16_output) const noexcept { 573| 510| return utf8_to_utf16::convert_valid(input, size, 574| 510| utf16_output); 575| 510|} _ZNK7simdutf8westmere14implementation21convert_utf8_to_utf32EPKcmPDi: 580| 758| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 581| 758| utf8_to_utf32::validating_transcoder converter; 582| 758| return converter.convert(buf, len, utf32_output); 583| 758|} _ZNK7simdutf8westmere14implementation33convert_utf8_to_utf32_with_errorsEPKcmPDi: 586| 746| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 587| 746| utf8_to_utf32::validating_transcoder converter; 588| 746| return converter.convert_with_errors(buf, len, utf32_output); 589| 746|} _ZNK7simdutf8westmere14implementation27convert_valid_utf8_to_utf32EPKcmPDi: 592| 464| const char *input, size_t size, char32_t *utf32_output) const noexcept { 593| 464| return utf8_to_utf32::convert_valid(input, size, utf32_output); 594| 464|} _ZNK7simdutf8westmere14implementation25convert_utf16le_to_latin1EPKDsmPc: 599| 124| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 600| 124| std::pair ret = 601| 124| sse_convert_utf16_to_latin1(buf, len, latin1_output); 602| 124| if (ret.first == nullptr) { ------------------ | Branch (602:7): [True: 42, False: 82] ------------------ 603| 42| return 0; 604| 42| } 605| 82| size_t saved_bytes = ret.second - latin1_output; 606| | 607| 82| if (ret.first != buf + len) { ------------------ | Branch (607:7): [True: 64, False: 18] ------------------ 608| 64| const size_t scalar_saved_bytes = 609| 64| scalar::utf16_to_latin1::convert( 610| 64| ret.first, len - (ret.first - buf), ret.second); 611| 64| if (scalar_saved_bytes == 0) { ------------------ | Branch (611:9): [True: 40, False: 24] ------------------ 612| 40| return 0; 613| 40| } 614| 24| saved_bytes += scalar_saved_bytes; 615| 24| } 616| 42| return saved_bytes; 617| 82|} _ZNK7simdutf8westmere14implementation25convert_utf16be_to_latin1EPKDsmPc: 620| 114| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 621| 114| std::pair ret = 622| 114| sse_convert_utf16_to_latin1(buf, len, latin1_output); 623| 114| if (ret.first == nullptr) { ------------------ | Branch (623:7): [True: 42, False: 72] ------------------ 624| 42| return 0; 625| 42| } 626| 72| size_t saved_bytes = ret.second - latin1_output; 627| | 628| 72| if (ret.first != buf + len) { ------------------ | Branch (628:7): [True: 56, False: 16] ------------------ 629| 56| const size_t scalar_saved_bytes = 630| 56| scalar::utf16_to_latin1::convert( 631| 56| ret.first, len - (ret.first - buf), ret.second); 632| 56| if (scalar_saved_bytes == 0) { ------------------ | Branch (632:9): [True: 24, False: 32] ------------------ 633| 24| return 0; 634| 24| } 635| 32| saved_bytes += scalar_saved_bytes; 636| 32| } 637| 48| return saved_bytes; 638| 72|} _ZNK7simdutf8westmere14implementation37convert_utf16le_to_latin1_with_errorsEPKDsmPc: 642| 278| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 643| 278| std::pair ret = 644| 278| sse_convert_utf16_to_latin1_with_errors( 645| 278| buf, len, latin1_output); 646| 278| if (ret.first.error) { ------------------ | Branch (646:7): [True: 158, False: 120] ------------------ 647| 158| return ret.first; 648| 158| } // Can return directly since scalar fallback already found correct 649| | // ret.first.count 650| 120| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (650:7): [True: 64, False: 56] ------------------ 651| 64| result scalar_res = 652| 64| scalar::utf16_to_latin1::convert_with_errors( 653| 64| buf + ret.first.count, len - ret.first.count, ret.second); 654| 64| if (scalar_res.error) { ------------------ | Branch (654:9): [True: 26, False: 38] ------------------ 655| 26| scalar_res.count += ret.first.count; 656| 26| return scalar_res; 657| 38| } else { 658| 38| ret.second += scalar_res.count; 659| 38| } 660| 64| } 661| 94| ret.first.count = 662| 94| ret.second - 663| 94| latin1_output; // Set count to the number of 8-bit code units written 664| 94| return ret.first; 665| 120|} _ZNK7simdutf8westmere14implementation37convert_utf16be_to_latin1_with_errorsEPKDsmPc: 669| 266| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 670| 266| std::pair ret = 671| 266| sse_convert_utf16_to_latin1_with_errors(buf, len, 672| 266| latin1_output); 673| 266| if (ret.first.error) { ------------------ | Branch (673:7): [True: 142, False: 124] ------------------ 674| 142| return ret.first; 675| 142| } // Can return directly since scalar fallback already found correct 676| | // ret.first.count 677| 124| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (677:7): [True: 80, False: 44] ------------------ 678| 80| result scalar_res = 679| 80| scalar::utf16_to_latin1::convert_with_errors( 680| 80| buf + ret.first.count, len - ret.first.count, ret.second); 681| 80| if (scalar_res.error) { ------------------ | Branch (681:9): [True: 32, False: 48] ------------------ 682| 32| scalar_res.count += ret.first.count; 683| 32| return scalar_res; 684| 48| } else { 685| 48| ret.second += scalar_res.count; 686| 48| } 687| 80| } 688| 92| ret.first.count = 689| 92| ret.second - 690| 92| latin1_output; // Set count to the number of 8-bit code units written 691| 92| return ret.first; 692| 124|} _ZNK7simdutf8westmere14implementation23convert_utf16le_to_utf8EPKDsmPc: 709| 842| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 710| 842| std::pair ret = 711| 842| sse_convert_utf16_to_utf8(buf, len, utf8_output); 712| 842| if (ret.first == nullptr) { ------------------ | Branch (712:7): [True: 112, False: 730] ------------------ 713| 112| return 0; 714| 112| } 715| 730| size_t saved_bytes = ret.second - utf8_output; 716| 730| if (ret.first != buf + len) { ------------------ | Branch (716:7): [True: 728, False: 2] ------------------ 717| 728| const size_t scalar_saved_bytes = 718| 728| scalar::utf16_to_utf8::convert( 719| 728| ret.first, len - (ret.first - buf), ret.second); 720| 728| if (scalar_saved_bytes == 0) { ------------------ | Branch (720:9): [True: 92, False: 636] ------------------ 721| 92| return 0; 722| 92| } 723| 636| saved_bytes += scalar_saved_bytes; 724| 636| } 725| 638| return saved_bytes; 726| 730|} _ZNK7simdutf8westmere14implementation23convert_utf16be_to_utf8EPKDsmPc: 729| 798| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 730| 798| std::pair ret = 731| 798| sse_convert_utf16_to_utf8(buf, len, utf8_output); 732| 798| if (ret.first == nullptr) { ------------------ | Branch (732:7): [True: 94, False: 704] ------------------ 733| 94| return 0; 734| 94| } 735| 704| size_t saved_bytes = ret.second - utf8_output; 736| 704| if (ret.first != buf + len) { ------------------ | Branch (736:7): [True: 702, False: 2] ------------------ 737| 702| const size_t scalar_saved_bytes = 738| 702| scalar::utf16_to_utf8::convert( 739| 702| ret.first, len - (ret.first - buf), ret.second); 740| 702| if (scalar_saved_bytes == 0) { ------------------ | Branch (740:9): [True: 70, False: 632] ------------------ 741| 70| return 0; 742| 70| } 743| 632| saved_bytes += scalar_saved_bytes; 744| 632| } 745| 634| return saved_bytes; 746| 704|} _ZNK7simdutf8westmere14implementation35convert_utf16le_to_utf8_with_errorsEPKDsmPc: 749| 642| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 750| | // ret.first.count is always the position in the buffer, not the number of 751| | // code units written even if finished 752| 642| std::pair ret = 753| 642| westmere::sse_convert_utf16_to_utf8_with_errors( 754| 642| buf, len, utf8_output); 755| 642| if (ret.first.error) { ------------------ | Branch (755:7): [True: 146, False: 496] ------------------ 756| 146| return ret.first; 757| 146| } // Can return directly since scalar fallback already found correct 758| | // ret.first.count 759| 496| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (759:7): [True: 492, False: 4] ------------------ 760| 492| result scalar_res = 761| 492| scalar::utf16_to_utf8::convert_with_errors( 762| 492| buf + ret.first.count, len - ret.first.count, ret.second); 763| 492| if (scalar_res.error) { ------------------ | Branch (763:9): [True: 94, False: 398] ------------------ 764| 94| scalar_res.count += ret.first.count; 765| 94| return scalar_res; 766| 398| } else { 767| 398| ret.second += scalar_res.count; 768| 398| } 769| 492| } 770| 402| ret.first.count = 771| 402| ret.second - 772| 402| utf8_output; // Set count to the number of 8-bit code units written 773| 402| return ret.first; 774| 496|} _ZNK7simdutf8westmere14implementation35convert_utf16be_to_utf8_with_errorsEPKDsmPc: 777| 654| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 778| | // ret.first.count is always the position in the buffer, not the number of 779| | // code units written even if finished 780| 654| std::pair ret = 781| 654| westmere::sse_convert_utf16_to_utf8_with_errors( 782| 654| buf, len, utf8_output); 783| 654| if (ret.first.error) { ------------------ | Branch (783:7): [True: 158, False: 496] ------------------ 784| 158| return ret.first; 785| 158| } // Can return directly since scalar fallback already found correct 786| | // ret.first.count 787| 496| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (787:7): [True: 492, False: 4] ------------------ 788| 492| result scalar_res = 789| 492| scalar::utf16_to_utf8::convert_with_errors( 790| 492| buf + ret.first.count, len - ret.first.count, ret.second); 791| 492| if (scalar_res.error) { ------------------ | Branch (791:9): [True: 114, False: 378] ------------------ 792| 114| scalar_res.count += ret.first.count; 793| 114| return scalar_res; 794| 378| } else { 795| 378| ret.second += scalar_res.count; 796| 378| } 797| 492| } 798| 382| ret.first.count = 799| 382| ret.second - 800| 382| utf8_output; // Set count to the number of 8-bit code units written 801| 382| return ret.first; 802| 496|} _ZNK7simdutf8westmere14implementation29convert_valid_utf16le_to_utf8EPKDsmPc: 805| 406| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 806| 406| return convert_utf16le_to_utf8(buf, len, utf8_output); 807| 406|} _ZNK7simdutf8westmere14implementation29convert_valid_utf16be_to_utf8EPKDsmPc: 810| 452| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 811| 452| return convert_utf16be_to_utf8(buf, len, utf8_output); 812| 452|} _ZNK7simdutf8westmere14implementation23convert_utf32_to_latin1EPKDimPc: 817| 228| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 818| 228| std::pair ret = 819| 228| sse_convert_utf32_to_latin1(buf, len, latin1_output); 820| 228| if (ret.first == nullptr) { ------------------ | Branch (820:7): [True: 64, False: 164] ------------------ 821| 64| return 0; 822| 64| } 823| 164| size_t saved_bytes = ret.second - latin1_output; 824| | // if (ret.first != buf + len) { 825| 164| if (ret.first < buf + len) { ------------------ | Branch (825:7): [True: 150, False: 14] ------------------ 826| 150| const size_t scalar_saved_bytes = scalar::utf32_to_latin1::convert( 827| 150| ret.first, len - (ret.first - buf), ret.second); 828| 150| if (scalar_saved_bytes == 0) { ------------------ | Branch (828:9): [True: 110, False: 40] ------------------ 829| 110| return 0; 830| 110| } 831| 40| saved_bytes += scalar_saved_bytes; 832| 40| } 833| 54| return saved_bytes; 834| 164|} _ZNK7simdutf8westmere14implementation35convert_utf32_to_latin1_with_errorsEPKDimPc: 837| 436| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 838| | // ret.first.count is always the position in the buffer, not the number of 839| | // code units written even if finished 840| 436| std::pair ret = 841| 436| westmere::sse_convert_utf32_to_latin1_with_errors(buf, len, 842| 436| latin1_output); 843| 436| if (ret.first.count != len) { ------------------ | Branch (843:7): [True: 416, False: 20] ------------------ 844| 416| result scalar_res = scalar::utf32_to_latin1::convert_with_errors( 845| 416| buf + ret.first.count, len - ret.first.count, ret.second); 846| 416| if (scalar_res.error) { ------------------ | Branch (846:9): [True: 394, False: 22] ------------------ 847| 394| scalar_res.count += ret.first.count; 848| 394| return scalar_res; 849| 394| } else { 850| 22| ret.second += scalar_res.count; 851| 22| } 852| 416| } 853| 42| ret.first.count = 854| 42| ret.second - 855| 42| latin1_output; // Set count to the number of 8-bit code units written 856| 42| return ret.first; 857| 436|} _ZNK7simdutf8westmere14implementation21convert_utf32_to_utf8EPKDimPc: 868| 864| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 869| 864| std::pair ret = 870| 864| sse_convert_utf32_to_utf8(buf, len, utf8_output); 871| 864| if (ret.first == nullptr) { ------------------ | Branch (871:7): [True: 262, False: 602] ------------------ 872| 262| return 0; 873| 262| } 874| 602| size_t saved_bytes = ret.second - utf8_output; 875| 602| if (ret.first != buf + len) { ------------------ | Branch (875:7): [True: 600, False: 2] ------------------ 876| 600| const size_t scalar_saved_bytes = scalar::utf32_to_utf8::convert( 877| 600| ret.first, len - (ret.first - buf), ret.second); 878| 600| if (scalar_saved_bytes == 0) { ------------------ | Branch (878:9): [True: 292, False: 308] ------------------ 879| 292| return 0; 880| 292| } 881| 308| saved_bytes += scalar_saved_bytes; 882| 308| } 883| 310| return saved_bytes; 884| 602|} _ZNK7simdutf8westmere14implementation33convert_utf32_to_utf8_with_errorsEPKDimPc: 887| 832| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 888| | // ret.first.count is always the position in the buffer, not the number of 889| | // code units written even if finished 890| 832| std::pair ret = 891| 832| westmere::sse_convert_utf32_to_utf8_with_errors(buf, len, utf8_output); 892| 832| if (ret.first.count != len) { ------------------ | Branch (892:7): [True: 828, False: 4] ------------------ 893| 828| result scalar_res = scalar::utf32_to_utf8::convert_with_errors( 894| 828| buf + ret.first.count, len - ret.first.count, ret.second); 895| 828| if (scalar_res.error) { ------------------ | Branch (895:9): [True: 618, False: 210] ------------------ 896| 618| scalar_res.count += ret.first.count; 897| 618| return scalar_res; 898| 618| } else { 899| 210| ret.second += scalar_res.count; 900| 210| } 901| 828| } 902| 214| ret.first.count = 903| 214| ret.second - 904| 214| utf8_output; // Set count to the number of 8-bit code units written 905| 214| return ret.first; 906| 832|} _ZNK7simdutf8westmere14implementation24convert_utf16le_to_utf32EPKDsmPDi: 911| 454| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 912| 454| std::pair ret = 913| 454| sse_convert_utf16_to_utf32(buf, len, utf32_output); 914| 454| if (ret.first == nullptr) { ------------------ | Branch (914:7): [True: 142, False: 312] ------------------ 915| 142| return 0; 916| 142| } 917| 312| size_t saved_bytes = ret.second - utf32_output; 918| 312| if (ret.first != buf + len) { ------------------ | Branch (918:7): [True: 240, False: 72] ------------------ 919| 240| const size_t scalar_saved_bytes = 920| 240| scalar::utf16_to_utf32::convert( 921| 240| ret.first, len - (ret.first - buf), ret.second); 922| 240| if (scalar_saved_bytes == 0) { ------------------ | Branch (922:9): [True: 64, False: 176] ------------------ 923| 64| return 0; 924| 64| } 925| 176| saved_bytes += scalar_saved_bytes; 926| 176| } 927| 248| return saved_bytes; 928| 312|} _ZNK7simdutf8westmere14implementation24convert_utf16be_to_utf32EPKDsmPDi: 931| 422| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 932| 422| std::pair ret = 933| 422| sse_convert_utf16_to_utf32(buf, len, utf32_output); 934| 422| if (ret.first == nullptr) { ------------------ | Branch (934:7): [True: 114, False: 308] ------------------ 935| 114| return 0; 936| 114| } 937| 308| size_t saved_bytes = ret.second - utf32_output; 938| 308| if (ret.first != buf + len) { ------------------ | Branch (938:7): [True: 250, False: 58] ------------------ 939| 250| const size_t scalar_saved_bytes = 940| 250| scalar::utf16_to_utf32::convert( 941| 250| ret.first, len - (ret.first - buf), ret.second); 942| 250| if (scalar_saved_bytes == 0) { ------------------ | Branch (942:9): [True: 62, False: 188] ------------------ 943| 62| return 0; 944| 62| } 945| 188| saved_bytes += scalar_saved_bytes; 946| 188| } 947| 246| return saved_bytes; 948| 308|} _ZNK7simdutf8westmere14implementation36convert_utf16le_to_utf32_with_errorsEPKDsmPDi: 951| 334| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 952| | // ret.first.count is always the position in the buffer, not the number of 953| | // code units written even if finished 954| 334| std::pair ret = 955| 334| westmere::sse_convert_utf16_to_utf32_with_errors( 956| 334| buf, len, utf32_output); 957| 334| if (ret.first.error) { ------------------ | Branch (957:7): [True: 120, False: 214] ------------------ 958| 120| return ret.first; 959| 120| } // Can return directly since scalar fallback already found correct 960| | // ret.first.count 961| 214| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (961:7): [True: 168, False: 46] ------------------ 962| 168| result scalar_res = 963| 168| scalar::utf16_to_utf32::convert_with_errors( 964| 168| buf + ret.first.count, len - ret.first.count, ret.second); 965| 168| if (scalar_res.error) { ------------------ | Branch (965:9): [True: 60, False: 108] ------------------ 966| 60| scalar_res.count += ret.first.count; 967| 60| return scalar_res; 968| 108| } else { 969| 108| ret.second += scalar_res.count; 970| 108| } 971| 168| } 972| 154| ret.first.count = 973| 154| ret.second - 974| 154| utf32_output; // Set count to the number of 8-bit code units written 975| 154| return ret.first; 976| 214|} _ZNK7simdutf8westmere14implementation36convert_utf16be_to_utf32_with_errorsEPKDsmPDi: 979| 346| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 980| | // ret.first.count is always the position in the buffer, not the number of 981| | // code units written even if finished 982| 346| std::pair ret = 983| 346| westmere::sse_convert_utf16_to_utf32_with_errors( 984| 346| buf, len, utf32_output); 985| 346| if (ret.first.error) { ------------------ | Branch (985:7): [True: 116, False: 230] ------------------ 986| 116| return ret.first; 987| 116| } // Can return directly since scalar fallback already found correct 988| | // ret.first.count 989| 230| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (989:7): [True: 184, False: 46] ------------------ 990| 184| result scalar_res = 991| 184| scalar::utf16_to_utf32::convert_with_errors( 992| 184| buf + ret.first.count, len - ret.first.count, ret.second); 993| 184| if (scalar_res.error) { ------------------ | Branch (993:9): [True: 68, False: 116] ------------------ 994| 68| scalar_res.count += ret.first.count; 995| 68| return scalar_res; 996| 116| } else { 997| 116| ret.second += scalar_res.count; 998| 116| } 999| 184| } 1000| 162| ret.first.count = 1001| 162| ret.second - 1002| 162| utf32_output; // Set count to the number of 8-bit code units written 1003| 162| return ret.first; 1004| 230|} _ZNK7simdutf8westmere14implementation27convert_valid_utf32_to_utf8EPKDimPc: 1009| 216| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 1010| 216| return convert_utf32_to_utf8(buf, len, utf8_output); 1011| 216|} _ZNK7simdutf8westmere14implementation24convert_utf32_to_utf16leEPKDimPDs: 1016| 822| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1017| 822| std::pair ret = 1018| 822| sse_convert_utf32_to_utf16(buf, len, utf16_output); 1019| 822| if (ret.first == nullptr) { ------------------ | Branch (1019:7): [True: 60, False: 762] ------------------ 1020| 60| return 0; 1021| 60| } 1022| 762| size_t saved_bytes = ret.second - utf16_output; 1023| 762| if (ret.first != buf + len) { ------------------ | Branch (1023:7): [True: 758, False: 4] ------------------ 1024| 758| const size_t scalar_saved_bytes = 1025| 758| scalar::utf32_to_utf16::convert( 1026| 758| ret.first, len - (ret.first - buf), ret.second); 1027| 758| if (scalar_saved_bytes == 0) { ------------------ | Branch (1027:9): [True: 334, False: 424] ------------------ 1028| 334| return 0; 1029| 334| } 1030| 424| saved_bytes += scalar_saved_bytes; 1031| 424| } 1032| 428| return saved_bytes; 1033| 762|} _ZNK7simdutf8westmere14implementation24convert_utf32_to_utf16beEPKDimPDs: 1036| 674| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1037| 674| std::pair ret = 1038| 674| sse_convert_utf32_to_utf16(buf, len, utf16_output); 1039| 674| if (ret.first == nullptr) { ------------------ | Branch (1039:7): [True: 94, False: 580] ------------------ 1040| 94| return 0; 1041| 94| } 1042| 580| size_t saved_bytes = ret.second - utf16_output; 1043| 580| if (ret.first != buf + len) { ------------------ | Branch (1043:7): [True: 578, False: 2] ------------------ 1044| 578| const size_t scalar_saved_bytes = 1045| 578| scalar::utf32_to_utf16::convert( 1046| 578| ret.first, len - (ret.first - buf), ret.second); 1047| 578| if (scalar_saved_bytes == 0) { ------------------ | Branch (1047:9): [True: 326, False: 252] ------------------ 1048| 326| return 0; 1049| 326| } 1050| 252| saved_bytes += scalar_saved_bytes; 1051| 252| } 1052| 254| return saved_bytes; 1053| 580|} _ZNK7simdutf8westmere14implementation36convert_utf32_to_utf16le_with_errorsEPKDimPDs: 1056| 674| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1057| | // ret.first.count is always the position in the buffer, not the number of 1058| | // code units written even if finished 1059| 674| std::pair ret = 1060| 674| westmere::sse_convert_utf32_to_utf16_with_errors( 1061| 674| buf, len, utf16_output); 1062| 674| if (ret.first.count != len) { ------------------ | Branch (1062:7): [True: 652, False: 22] ------------------ 1063| 652| result scalar_res = 1064| 652| scalar::utf32_to_utf16::convert_with_errors( 1065| 652| buf + ret.first.count, len - ret.first.count, ret.second); 1066| 652| if (scalar_res.error) { ------------------ | Branch (1066:9): [True: 504, False: 148] ------------------ 1067| 504| scalar_res.count += ret.first.count; 1068| 504| return scalar_res; 1069| 504| } else { 1070| 148| ret.second += scalar_res.count; 1071| 148| } 1072| 652| } 1073| 170| ret.first.count = 1074| 170| ret.second - 1075| 170| utf16_output; // Set count to the number of 8-bit code units written 1076| 170| return ret.first; 1077| 674|} _ZNK7simdutf8westmere14implementation36convert_utf32_to_utf16be_with_errorsEPKDimPDs: 1080| 694| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1081| | // ret.first.count is always the position in the buffer, not the number of 1082| | // code units written even if finished 1083| 694| std::pair ret = 1084| 694| westmere::sse_convert_utf32_to_utf16_with_errors( 1085| 694| buf, len, utf16_output); 1086| 694| if (ret.first.count != len) { ------------------ | Branch (1086:7): [True: 670, False: 24] ------------------ 1087| 670| result scalar_res = 1088| 670| scalar::utf32_to_utf16::convert_with_errors( 1089| 670| buf + ret.first.count, len - ret.first.count, ret.second); 1090| 670| if (scalar_res.error) { ------------------ | Branch (1090:9): [True: 512, False: 158] ------------------ 1091| 512| scalar_res.count += ret.first.count; 1092| 512| return scalar_res; 1093| 512| } else { 1094| 158| ret.second += scalar_res.count; 1095| 158| } 1096| 670| } 1097| 182| ret.first.count = 1098| 182| ret.second - 1099| 182| utf16_output; // Set count to the number of 8-bit code units written 1100| 182| return ret.first; 1101| 694|} _ZNK7simdutf8westmere14implementation30convert_valid_utf32_to_utf16leEPKDimPDs: 1104| 280| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1105| 280| return convert_utf32_to_utf16le(buf, len, utf16_output); 1106| 280|} _ZNK7simdutf8westmere14implementation30convert_valid_utf32_to_utf16beEPKDimPDs: 1109| 134| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1110| 134| return convert_utf32_to_utf16be(buf, len, utf16_output); 1111| 134|} _ZNK7simdutf8westmere14implementation30convert_valid_utf16le_to_utf32EPKDsmPDi: 1114| 128| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1115| 128| return convert_utf16le_to_utf32(buf, len, utf32_output); 1116| 128|} _ZNK7simdutf8westmere14implementation30convert_valid_utf16be_to_utf32EPKDsmPDi: 1119| 142| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1120| 142| return convert_utf16be_to_utf32(buf, len, utf32_output); 1121| 142|} _ZNK7simdutf8westmere14implementation13count_utf16leEPKDsm: 1132| 1.35k| const char16_t *input, size_t length) const noexcept { 1133| 1.35k| return utf16::count_code_points(input, length); 1134| 1.35k|} _ZNK7simdutf8westmere14implementation13count_utf16beEPKDsm: 1137| 1.33k| const char16_t *input, size_t length) const noexcept { 1138| 1.33k| return utf16::count_code_points(input, length); 1139| 1.33k|} _ZNK7simdutf8westmere14implementation10count_utf8EPKcm: 1144| 4.03k|implementation::count_utf8(const char *input, size_t length) const noexcept { 1145| 4.03k| return utf8::count_code_points_bytemask(input, length); 1146| 4.03k|} _ZNK7simdutf8westmere14implementation23latin1_length_from_utf8EPKcm: 1151| 549| const char *buf, size_t len) const noexcept { 1152| 549| return count_utf8(buf, len); 1153| 549|} _ZNK7simdutf8westmere14implementation24utf8_length_from_utf16leEPKDsm: 1158| 754| const char16_t *input, size_t length) const noexcept { 1159| 754| return utf16::utf8_length_from_utf16_bytemask(input, 1160| 754| length); 1161| 754|} _ZNK7simdutf8westmere14implementation24utf8_length_from_utf16beEPKDsm: 1164| 748| const char16_t *input, size_t length) const noexcept { 1165| 748| return utf16::utf8_length_from_utf16_bytemask(input, length); 1166| 748|} _ZNK7simdutf8westmere14implementation23utf8_length_from_latin1EPKcm: 1171| 302| const char *input, size_t len) const noexcept { 1172| 302| const uint8_t *str = reinterpret_cast(input); 1173| 302| size_t answer = len / sizeof(__m128i) * sizeof(__m128i); 1174| 302| size_t i = 0; 1175| 302| if (answer >= 2048) { // long strings optimization ------------------ | Branch (1175:7): [True: 139, False: 163] ------------------ 1176| 139| __m128i two_64bits = _mm_setzero_si128(); 1177| 6.80k| while (i + sizeof(__m128i) <= len) { ------------------ | Branch (1177:12): [True: 6.67k, False: 139] ------------------ 1178| 6.67k| __m128i runner = _mm_setzero_si128(); 1179| 6.67k| size_t iterations = (len - i) / sizeof(__m128i); 1180| 6.67k| if (iterations > 255) { ------------------ | Branch (1180:11): [True: 6.53k, False: 139] ------------------ 1181| 6.53k| iterations = 255; 1182| 6.53k| } 1183| 6.67k| size_t max_i = i + iterations * sizeof(__m128i) - sizeof(__m128i); 1184| 420k| for (; i + 4 * sizeof(__m128i) <= max_i; i += 4 * sizeof(__m128i)) { ------------------ | Branch (1184:14): [True: 413k, False: 6.67k] ------------------ 1185| 413k| __m128i input1 = _mm_loadu_si128((const __m128i *)(str + i)); 1186| 413k| __m128i input2 = 1187| 413k| _mm_loadu_si128((const __m128i *)(str + i + sizeof(__m128i))); 1188| 413k| __m128i input3 = 1189| 413k| _mm_loadu_si128((const __m128i *)(str + i + 2 * sizeof(__m128i))); 1190| 413k| __m128i input4 = 1191| 413k| _mm_loadu_si128((const __m128i *)(str + i + 3 * sizeof(__m128i))); 1192| 413k| __m128i input12 = 1193| 413k| _mm_add_epi8(_mm_cmpgt_epi8(_mm_setzero_si128(), input1), 1194| 413k| _mm_cmpgt_epi8(_mm_setzero_si128(), input2)); 1195| 413k| __m128i input34 = 1196| 413k| _mm_add_epi8(_mm_cmpgt_epi8(_mm_setzero_si128(), input3), 1197| 413k| _mm_cmpgt_epi8(_mm_setzero_si128(), input4)); 1198| 413k| __m128i input1234 = _mm_add_epi8(input12, input34); 1199| 413k| runner = _mm_sub_epi8(runner, input1234); 1200| 413k| } 1201| 26.5k| for (; i <= max_i; i += sizeof(__m128i)) { ------------------ | Branch (1201:14): [True: 19.9k, False: 6.67k] ------------------ 1202| 19.9k| __m128i more_input = _mm_loadu_si128((const __m128i *)(str + i)); 1203| 19.9k| runner = _mm_sub_epi8(runner, 1204| 19.9k| _mm_cmpgt_epi8(_mm_setzero_si128(), more_input)); 1205| 19.9k| } 1206| 6.67k| two_64bits = 1207| 6.67k| _mm_add_epi64(two_64bits, _mm_sad_epu8(runner, _mm_setzero_si128())); 1208| 6.67k| } 1209| 139| answer += 1210| 139| _mm_extract_epi64(two_64bits, 0) + _mm_extract_epi64(two_64bits, 1); 1211| 163| } else if (answer > 0) { // short string optimization ------------------ | Branch (1211:14): [True: 105, False: 58] ------------------ 1212| 1.43k| for (; i + 2 * sizeof(__m128i) <= len; i += 2 * sizeof(__m128i)) { ------------------ | Branch (1212:12): [True: 1.32k, False: 105] ------------------ 1213| 1.32k| __m128i latin = _mm_loadu_si128((const __m128i *)(input + i)); 1214| 1.32k| uint16_t non_ascii = (uint16_t)_mm_movemask_epi8(latin); 1215| 1.32k| answer += count_ones(non_ascii); 1216| 1.32k| latin = _mm_loadu_si128((const __m128i *)(input + i) + 1); 1217| 1.32k| non_ascii = (uint16_t)_mm_movemask_epi8(latin); 1218| 1.32k| answer += count_ones(non_ascii); 1219| 1.32k| } 1220| 156| for (; i + sizeof(__m128i) <= len; i += sizeof(__m128i)) { ------------------ | Branch (1220:12): [True: 51, False: 105] ------------------ 1221| 51| __m128i latin = _mm_loadu_si128((const __m128i *)(input + i)); 1222| 51| uint16_t non_ascii = (uint16_t)_mm_movemask_epi8(latin); 1223| 51| answer += count_ones(non_ascii); 1224| 51| } 1225| 105| } 1226| 302| return answer + scalar::latin1::utf8_length_from_latin1( 1227| 302| reinterpret_cast(str + i), len - i); 1228| 302|} _ZNK7simdutf8westmere14implementation25utf32_length_from_utf16leEPKDsm: 1233| 400| const char16_t *input, size_t length) const noexcept { 1234| 400| return utf16::utf32_length_from_utf16(input, length); 1235| 400|} _ZNK7simdutf8westmere14implementation25utf32_length_from_utf16beEPKDsm: 1238| 400| const char16_t *input, size_t length) const noexcept { 1239| 400| return utf16::utf32_length_from_utf16(input, length); 1240| 400|} _ZNK7simdutf8westmere14implementation22utf16_length_from_utf8EPKcm: 1245| 1.94k| const char *input, size_t length) const noexcept { 1246| 1.94k| return utf8::utf16_length_from_utf8_bytemask(input, length); 1247| 1.94k|} _ZNK7simdutf8westmere14implementation22utf8_length_from_utf32EPKDim: 1292| 860| const char32_t *input, size_t length) const noexcept { 1293| 860| return utf32::utf8_length_from_utf32(input, length); 1294| 860|} _ZNK7simdutf8westmere14implementation23utf16_length_from_utf32EPKDim: 1299| 1.44k| const char32_t *input, size_t length) const noexcept { 1300| 1.44k| const __m128i v_00000000 = _mm_setzero_si128(); 1301| 1.44k| const __m128i v_ffff0000 = _mm_set1_epi32((uint32_t)0xffff0000); 1302| 1.44k| size_t pos = 0; 1303| 1.44k| size_t count = 0; 1304| 1.16M| for (; pos + 4 <= length; pos += 4) { ------------------ | Branch (1304:10): [True: 1.16M, False: 1.44k] ------------------ 1305| 1.16M| __m128i in = _mm_loadu_si128((__m128i *)(input + pos)); 1306| 1.16M| const __m128i surrogate_bytemask = 1307| 1.16M| _mm_cmpeq_epi32(_mm_and_si128(in, v_ffff0000), v_00000000); 1308| 1.16M| const uint16_t surrogate_bitmask = 1309| 1.16M| static_cast(_mm_movemask_epi8(surrogate_bytemask)); 1310| 1.16M| size_t surrogate_count = (16 - count_ones(surrogate_bitmask)) / 4; 1311| 1.16M| count += 4 + surrogate_count; 1312| 1.16M| } 1313| 1.44k| return count + 1314| 1.44k| scalar::utf32::utf16_length_from_utf32(input + pos, length - pos); 1315| 1.44k|} _ZNK7simdutf8westmere14implementation22utf32_length_from_utf8EPKcm: 1320| 994| const char *input, size_t length) const noexcept { 1321| 994| return utf8::count_code_points(input, length); 1322| 994|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_18is_asciiERKNS1_4simd8simd8x64IhEE: 13| 4.87M|simdutf_really_inline bool is_ascii(const simd8x64 &input) { 14| 4.87M| return input.reduce_or().is_ascii(); 15| 4.87M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_124must_be_2_3_continuationENS1_4simd5simd8IhEES4_: 22| 6.01M| const simd8 prev3) { 23| 6.01M| simd8 is_third_byte = 24| 6.01M| prev2.saturating_sub(0xe0u - 0x80); // Only 111_____ will be >= 0x80 25| 6.01M| simd8 is_fourth_byte = 26| 6.01M| prev3.saturating_sub(0xf0u - 0x80); // Only 1111____ will be >= 0x80 27| 6.01M| return simd8(is_third_byte | is_fourth_byte); 28| 6.01M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_18internal8westmere26write_v_u16_11bits_to_utf8EDv2_xRPcS4_S4_: 57| 891k| const __m128i v_ff80) { 58| | // no bits set above 7th bit 59| 891k| const __m128i one_byte_bytemask = 60| 891k| _mm_cmpeq_epi16(_mm_and_si128(v_u16, v_ff80), v_0000); 61| 891k| const uint16_t one_byte_bitmask = 62| 891k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 63| | 64| 891k| write_v_u16_11bits_to_utf8(v_u16, utf8_output, one_byte_bytemask, 65| 891k| one_byte_bitmask); 66| 891k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_18internal8westmere26write_v_u16_11bits_to_utf8EDv2_xRPcS4_t: 10| 1.32M| const uint16_t one_byte_bitmask) { 11| | // 0b1100_0000_1000_0000 12| 1.32M| const __m128i v_c080 = _mm_set1_epi16((int16_t)0xc080); 13| | // 0b0001_1111_0000_0000 14| 1.32M| const __m128i v_1f00 = _mm_set1_epi16((int16_t)0x1f00); 15| | // 0b0000_0000_0011_1111 16| 1.32M| const __m128i v_003f = _mm_set1_epi16((int16_t)0x003f); 17| | 18| | // 1. prepare 2-byte values 19| | // input 16-bit word : [0000|0aaa|aabb|bbbb] x 8 20| | // expected output : [110a|aaaa|10bb|bbbb] x 8 21| | 22| | // t0 = [000a|aaaa|bbbb|bb00] 23| 1.32M| const __m128i t0 = _mm_slli_epi16(v_u16, 2); 24| | // t1 = [000a|aaaa|0000|0000] 25| 1.32M| const __m128i t1 = _mm_and_si128(t0, v_1f00); 26| | // t2 = [0000|0000|00bb|bbbb] 27| 1.32M| const __m128i t2 = _mm_and_si128(v_u16, v_003f); 28| | // t3 = [000a|aaaa|00bb|bbbb] 29| 1.32M| const __m128i t3 = _mm_or_si128(t1, t2); 30| | // t4 = [110a|aaaa|10bb|bbbb] 31| 1.32M| const __m128i t4 = _mm_or_si128(t3, v_c080); 32| | 33| | // 2. merge ASCII and 2-byte codewords 34| 1.32M| const __m128i utf8_unpacked = _mm_blendv_epi8(t4, v_u16, one_byte_bytemask); 35| | 36| | // 3. prepare bitmask for 8-bit lookup 37| | // one_byte_bitmask = hhggffeeddccbbaa -- the bits are doubled (h - MSB, a 38| | // - LSB) 39| 1.32M| const uint16_t m0 = one_byte_bitmask & 0x5555; // m0 = 0h0g0f0e0d0c0b0a 40| 1.32M| const uint16_t m1 = static_cast(m0 >> 7); // m1 = 00000000h0g0f0e0 41| 1.32M| const uint8_t m2 = static_cast((m0 | m1) & 0xff); // m2 = hdgcfbea 42| | // 4. pack the bytes 43| 1.32M| const uint8_t *row = 44| 1.32M| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[m2][0]; 45| 1.32M| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 46| 1.32M| const __m128i utf8_packed = _mm_shuffle_epi8(utf8_unpacked, shuffle); 47| | 48| | // 5. store bytes 49| 1.32M| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 50| | 51| | // 6. adjust pointers 52| 1.32M| utf8_output += row[0]; 53| 1.32M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_latin1_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 76| char16_t *utf16_output) { 5| 76| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| 93.4k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (6:22): [True: 93.3k, False: 76] ------------------ 7| | // Load 16 Latin1 characters into a 128-bit register 8| 93.3k| __m128i in = 9| 93.3k| _mm_loadu_si128(reinterpret_cast(&latin1_input[i])); 10| 93.3k| __m128i out1 = big_endian ? _mm_unpacklo_epi8(_mm_setzero_si128(), in) ------------------ | Branch (10:20): [Folded, False: 93.3k] ------------------ 11| 93.3k| : _mm_unpacklo_epi8(in, _mm_setzero_si128()); 12| 93.3k| __m128i out2 = big_endian ? _mm_unpackhi_epi8(_mm_setzero_si128(), in) ------------------ | Branch (12:20): [Folded, False: 93.3k] ------------------ 13| 93.3k| : _mm_unpackhi_epi8(in, _mm_setzero_si128()); 14| | // Zero extend each Latin1 character to 16-bit integers and store the 15| | // results back to memory 16| 93.3k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i]), out1); 17| 93.3k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i + 8]), out2); 18| 93.3k| } 19| | // return pointers pointing to where we left off 20| 76| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 21| 76|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_latin1_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 96| char16_t *utf16_output) { 5| 96| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| 190k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (6:22): [True: 190k, False: 96] ------------------ 7| | // Load 16 Latin1 characters into a 128-bit register 8| 190k| __m128i in = 9| 190k| _mm_loadu_si128(reinterpret_cast(&latin1_input[i])); 10| 190k| __m128i out1 = big_endian ? _mm_unpacklo_epi8(_mm_setzero_si128(), in) ------------------ | Branch (10:20): [True: 190k, Folded] ------------------ 11| 190k| : _mm_unpacklo_epi8(in, _mm_setzero_si128()); 12| 190k| __m128i out2 = big_endian ? _mm_unpackhi_epi8(_mm_setzero_si128(), in) ------------------ | Branch (12:20): [True: 190k, Folded] ------------------ 13| 190k| : _mm_unpackhi_epi8(in, _mm_setzero_si128()); 14| | // Zero extend each Latin1 character to 16-bit integers and store the 15| | // results back to memory 16| 190k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i]), out1); 17| 190k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i + 8]), out2); 18| 190k| } 19| | // return pointers pointing to where we left off 20| 96| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 21| 96|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_latin1_to_utf32EPKcmPDi: 3| 124| char32_t *utf32_output) { 4| 124| const char *end = buf + len; 5| | 6| 433k| while (end - buf >= 16) { ------------------ | Branch (6:10): [True: 433k, False: 124] ------------------ 7| | // Load 16 Latin1 characters (16 bytes) into a 128-bit register 8| 433k| __m128i in = _mm_loadu_si128((__m128i *)buf); 9| | 10| | // Shift input to process next 4 bytes 11| 433k| __m128i in_shifted1 = _mm_srli_si128(in, 4); 12| 433k| __m128i in_shifted2 = _mm_srli_si128(in, 8); 13| 433k| __m128i in_shifted3 = _mm_srli_si128(in, 12); 14| | 15| | // expand 8-bit to 32-bit unit 16| 433k| __m128i out1 = _mm_cvtepu8_epi32(in); 17| 433k| __m128i out2 = _mm_cvtepu8_epi32(in_shifted1); 18| 433k| __m128i out3 = _mm_cvtepu8_epi32(in_shifted2); 19| 433k| __m128i out4 = _mm_cvtepu8_epi32(in_shifted3); 20| | 21| 433k| _mm_storeu_si128((__m128i *)utf32_output, out1); 22| 433k| _mm_storeu_si128((__m128i *)(utf32_output + 4), out2); 23| 433k| _mm_storeu_si128((__m128i *)(utf32_output + 8), out3); 24| 433k| _mm_storeu_si128((__m128i *)(utf32_output + 12), out4); 25| | 26| 433k| utf32_output += 16; 27| 433k| buf += 16; 28| 433k| } 29| | 30| 124| return std::make_pair(buf, utf32_output); 31| 124|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_latin1_to_utf8EPKcmPc: 3| 604| const size_t latin_input_length, char *utf8_output) { 4| 604| const char *end = latin_input + latin_input_length; 5| | 6| 604| const __m128i v_0000 = _mm_setzero_si128(); 7| | // 0b1000_0000 8| 604| const __m128i v_80 = _mm_set1_epi8((uint8_t)0x80); 9| | // 0b1111_1111_1000_0000 10| 604| const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80); 11| | 12| 604| const __m128i latin_1_half_into_u16_byte_mask = 13| 604| _mm_setr_epi8(0, '\x80', 1, '\x80', 2, '\x80', 3, '\x80', 4, '\x80', 5, 14| 604| '\x80', 6, '\x80', 7, '\x80'); 15| | 16| 604| const __m128i latin_2_half_into_u16_byte_mask = 17| 604| _mm_setr_epi8(8, '\x80', 9, '\x80', 10, '\x80', 11, '\x80', 12, '\x80', 18| 604| 13, '\x80', 14, '\x80', 15, '\x80'); 19| | 20| | // each latin1 takes 1-2 utf8 bytes 21| | // slow path writes useful 8-15 bytes twice (eagerly writes 16 bytes and then 22| | // adjust the pointer) so the last write can exceed the utf8_output size by 23| | // 8-1 bytes by reserving 8 extra input bytes, we expect the output to have 24| | // 8-16 bytes free 25| 3.35M| while (end - latin_input >= 16 + 8) { ------------------ | Branch (25:10): [True: 3.35M, False: 604] ------------------ 26| | // Load 16 Latin1 characters (16 bytes) into a 128-bit register 27| 3.35M| __m128i v_latin = _mm_loadu_si128((__m128i *)latin_input); 28| | 29| 3.35M| if (_mm_testz_si128(v_latin, v_80)) { // ASCII fast path!!!! ------------------ | Branch (29:9): [True: 2.90M, False: 445k] ------------------ 30| 2.90M| _mm_storeu_si128((__m128i *)utf8_output, v_latin); 31| 2.90M| latin_input += 16; 32| 2.90M| utf8_output += 16; 33| 2.90M| continue; 34| 2.90M| } 35| | 36| | // assuming a/b are bytes and A/B are uint16 of the same value 37| | // aaaa_aaaa_bbbb_bbbb -> AAAA_AAAA 38| 445k| __m128i v_u16_latin_1_half = 39| 445k| _mm_shuffle_epi8(v_latin, latin_1_half_into_u16_byte_mask); 40| | // aaaa_aaaa_bbbb_bbbb -> BBBB_BBBB 41| 445k| __m128i v_u16_latin_2_half = 42| 445k| _mm_shuffle_epi8(v_latin, latin_2_half_into_u16_byte_mask); 43| | 44| 445k| internal::westmere::write_v_u16_11bits_to_utf8(v_u16_latin_1_half, 45| 445k| utf8_output, v_0000, v_ff80); 46| 445k| internal::westmere::write_v_u16_11bits_to_utf8(v_u16_latin_2_half, 47| 445k| utf8_output, v_0000, v_ff80); 48| 445k| latin_input += 16; 49| 445k| } 50| | 51| 604| if (end - latin_input >= 16) { ------------------ | Branch (51:7): [True: 312, False: 292] ------------------ 52| | // Load 16 Latin1 characters (16 bytes) into a 128-bit register 53| 312| __m128i v_latin = _mm_loadu_si128((__m128i *)latin_input); 54| | 55| 312| if (_mm_testz_si128(v_latin, v_80)) { // ASCII fast path!!!! ------------------ | Branch (55:9): [True: 120, False: 192] ------------------ 56| 120| _mm_storeu_si128((__m128i *)utf8_output, v_latin); 57| 120| latin_input += 16; 58| 120| utf8_output += 16; 59| 192| } else { 60| | // assuming a/b are bytes and A/B are uint16 of the same value 61| | // aaaa_aaaa_bbbb_bbbb -> AAAA_AAAA 62| 192| __m128i v_u16_latin_1_half = 63| 192| _mm_shuffle_epi8(v_latin, latin_1_half_into_u16_byte_mask); 64| 192| internal::westmere::write_v_u16_11bits_to_utf8( 65| 192| v_u16_latin_1_half, utf8_output, v_0000, v_ff80); 66| 192| latin_input += 8; 67| 192| } 68| 312| } 69| | 70| 604| return std::make_pair(latin_input, utf8_output); 71| 604|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_utf16_to_latin1ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 124| char *latin1_output) { 5| 124| const char16_t *end = buf + len; 6| 1.12k| while (end - buf >= 8) { ------------------ | Branch (6:10): [True: 1.03k, False: 86] ------------------ 7| | // Load 8 UTF-16 characters into 128-bit SSE register 8| 1.03k| __m128i in = _mm_loadu_si128(reinterpret_cast(buf)); 9| | 10| | if constexpr (!match_system(big_endian)) { 11| | const __m128i swap = 12| | _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 13| | in = _mm_shuffle_epi8(in, swap); 14| | } 15| | 16| 1.03k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 17| 1.03k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (17:9): [True: 998, False: 38] ------------------ 18| | // Pack 16-bit characters into 8-bit and store in latin1_output 19| 998| __m128i latin1_packed = _mm_packus_epi16(in, in); 20| 998| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 21| 998| latin1_packed); 22| | // Adjust pointers for next iteration 23| 998| buf += 8; 24| 998| latin1_output += 8; 25| 998| } else { 26| 38| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 27| 38| } 28| 1.03k| } // while 29| 86| return std::make_pair(buf, latin1_output); 30| 124|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_utf16_to_latin1ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 114| char *latin1_output) { 5| 114| const char16_t *end = buf + len; 6| 3.12k| while (end - buf >= 8) { ------------------ | Branch (6:10): [True: 3.04k, False: 76] ------------------ 7| | // Load 8 UTF-16 characters into 128-bit SSE register 8| 3.04k| __m128i in = _mm_loadu_si128(reinterpret_cast(buf)); 9| | 10| 3.04k| if constexpr (!match_system(big_endian)) { 11| 3.04k| const __m128i swap = 12| 3.04k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 13| 3.04k| in = _mm_shuffle_epi8(in, swap); 14| 3.04k| } 15| | 16| 3.04k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 17| 3.04k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (17:9): [True: 3.01k, False: 38] ------------------ 18| | // Pack 16-bit characters into 8-bit and store in latin1_output 19| 3.01k| __m128i latin1_packed = _mm_packus_epi16(in, in); 20| 3.01k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 21| 3.01k| latin1_packed); 22| | // Adjust pointers for next iteration 23| 3.01k| buf += 8; 24| 3.01k| latin1_output += 8; 25| 3.01k| } else { 26| 38| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 27| 38| } 28| 3.04k| } // while 29| 76| return std::make_pair(buf, latin1_output); 30| 114|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_139sse_convert_utf16_to_latin1_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 35| 278| char *latin1_output) { 36| 278| const char16_t *start = buf; 37| 278| const char16_t *end = buf + len; 38| 820k| while (end - buf >= 8) { ------------------ | Branch (38:10): [True: 820k, False: 120] ------------------ 39| 820k| __m128i in = _mm_loadu_si128(reinterpret_cast(buf)); 40| | 41| | if constexpr (!match_system(big_endian)) { 42| | const __m128i swap = 43| | _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 44| | in = _mm_shuffle_epi8(in, swap); 45| | } 46| | 47| 820k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 48| 820k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (48:9): [True: 820k, False: 158] ------------------ 49| 820k| __m128i latin1_packed = _mm_packus_epi16(in, in); 50| 820k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 51| 820k| latin1_packed); 52| 820k| buf += 8; 53| 820k| latin1_output += 8; 54| 820k| } else { 55| | // Fallback to scalar code for handling errors 56| 514| for (int k = 0; k < 8; k++) { ------------------ | Branch (56:23): [True: 514, False: 0] ------------------ 57| 514| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 58| 514| if (word <= 0xff) { ------------------ | Branch (58:13): [True: 356, False: 158] ------------------ 59| 356| *latin1_output++ = char(word); 60| 356| } else { 61| 158| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 62| 158| latin1_output); 63| 158| } 64| 514| } 65| 0| buf += 8; 66| 0| } 67| 820k| } // while 68| 120| return std::make_pair(result(error_code::SUCCESS, buf - start), 69| 120| latin1_output); 70| 278|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_139sse_convert_utf16_to_latin1_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 35| 266| char *latin1_output) { 36| 266| const char16_t *start = buf; 37| 266| const char16_t *end = buf + len; 38| 1.04M| while (end - buf >= 8) { ------------------ | Branch (38:10): [True: 1.04M, False: 124] ------------------ 39| 1.04M| __m128i in = _mm_loadu_si128(reinterpret_cast(buf)); 40| | 41| 1.04M| if constexpr (!match_system(big_endian)) { 42| 1.04M| const __m128i swap = 43| 1.04M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 44| 1.04M| in = _mm_shuffle_epi8(in, swap); 45| 1.04M| } 46| | 47| 1.04M| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 48| 1.04M| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (48:9): [True: 1.04M, False: 142] ------------------ 49| 1.04M| __m128i latin1_packed = _mm_packus_epi16(in, in); 50| 1.04M| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 51| 1.04M| latin1_packed); 52| 1.04M| buf += 8; 53| 1.04M| latin1_output += 8; 54| 1.04M| } else { 55| | // Fallback to scalar code for handling errors 56| 454| for (int k = 0; k < 8; k++) { ------------------ | Branch (56:23): [True: 454, False: 0] ------------------ 57| 454| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 58| 454| if (word <= 0xff) { ------------------ | Branch (58:13): [True: 312, False: 142] ------------------ 59| 312| *latin1_output++ = char(word); 60| 312| } else { 61| 142| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 62| 142| latin1_output); 63| 142| } 64| 454| } 65| 0| buf += 8; 66| 0| } 67| 1.04M| } // while 68| 124| return std::make_pair(result(error_code::SUCCESS, buf - start), 69| 124| latin1_output); 70| 266|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf16_to_utf32ILNS_10endiannessE0EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 454| char32_t *utf32_output) { 58| 454| const char16_t *end = buf + len; 59| | 60| 454| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 61| 454| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 62| | 63| 542k| while (end - buf >= 8) { ------------------ | Branch (63:10): [True: 542k, False: 318] ------------------ 64| 542k| __m128i in = _mm_loadu_si128((__m128i *)buf); 65| | 66| 542k| if (big_endian) { ------------------ | Branch (66:9): [Folded, False: 542k] ------------------ 67| 0| const __m128i swap = 68| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 69| 0| in = _mm_shuffle_epi8(in, swap); 70| 0| } 71| | 72| | // 1. Check if there are any surrogate word in the input chunk. 73| | // We have also deal with situation when there is a surrogate word 74| | // at the end of a chunk. 75| 542k| const __m128i surrogates_bytemask = 76| 542k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_d800); 77| | 78| | // bitmask = 0x0000 if there are no surrogates 79| | // = 0xc000 if the last word is a surrogate 80| 542k| const uint16_t surrogates_bitmask = 81| 542k| static_cast(_mm_movemask_epi8(surrogates_bytemask)); 82| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 83| | // However, it is likely an uncommon occurrence. 84| 542k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (84:9): [True: 536k, False: 5.51k] ------------------ 85| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 86| 536k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 87| 536k| _mm_cvtepu16_epi32(in)); 88| 536k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 89| 536k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 90| 536k| utf32_output += 8; 91| 536k| buf += 8; 92| | // surrogate pair(s) in a register 93| 536k| } else { 94| | // Let us do a scalar fallback. 95| | // It may seem wasteful to use scalar code, but being efficient with SIMD 96| | // in the presence of surrogate pairs may require non-trivial tables. 97| 5.51k| size_t forward = 15; 98| 5.51k| size_t k = 0; 99| 5.51k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (99:11): [True: 70, False: 5.44k] ------------------ 100| 70| forward = size_t(end - buf - 1); 101| 70| } 102| 80.1k| for (; k < forward; k++) { ------------------ | Branch (102:14): [True: 74.7k, False: 5.38k] ------------------ 103| 74.7k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 104| 74.7k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (104:13): [True: 67.9k, False: 6.73k] ------------------ 105| 67.9k| *utf32_output++ = char32_t(word); 106| 67.9k| } else { 107| | // must be a surrogate pair 108| 6.73k| uint16_t diff = uint16_t(word - 0xD800); 109| 6.73k| uint16_t next_word = 110| 6.73k| scalar::utf16::swap_if_needed(buf[k + 1]); 111| 6.73k| k++; 112| 6.73k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 113| 6.73k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (113:15): [True: 136, False: 6.59k] ------------------ 114| 136| return std::make_pair(nullptr, utf32_output); 115| 136| } 116| 6.59k| uint32_t value = (diff << 10) + diff2 + 0x10000; 117| 6.59k| *utf32_output++ = char32_t(value); 118| 6.59k| } 119| 74.7k| } 120| 5.38k| buf += k; 121| 5.38k| } 122| 542k| } // while 123| 318| return std::make_pair(buf, utf32_output); 124| 454|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf16_to_utf32ILNS_10endiannessE1EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 422| char32_t *utf32_output) { 58| 422| const char16_t *end = buf + len; 59| | 60| 422| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 61| 422| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 62| | 63| 737k| while (end - buf >= 8) { ------------------ | Branch (63:10): [True: 737k, False: 314] ------------------ 64| 737k| __m128i in = _mm_loadu_si128((__m128i *)buf); 65| | 66| 737k| if (big_endian) { ------------------ | Branch (66:9): [True: 737k, Folded] ------------------ 67| 737k| const __m128i swap = 68| 737k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 69| 737k| in = _mm_shuffle_epi8(in, swap); 70| 737k| } 71| | 72| | // 1. Check if there are any surrogate word in the input chunk. 73| | // We have also deal with situation when there is a surrogate word 74| | // at the end of a chunk. 75| 737k| const __m128i surrogates_bytemask = 76| 737k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_d800); 77| | 78| | // bitmask = 0x0000 if there are no surrogates 79| | // = 0xc000 if the last word is a surrogate 80| 737k| const uint16_t surrogates_bitmask = 81| 737k| static_cast(_mm_movemask_epi8(surrogates_bytemask)); 82| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 83| | // However, it is likely an uncommon occurrence. 84| 737k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (84:9): [True: 735k, False: 1.48k] ------------------ 85| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 86| 735k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 87| 735k| _mm_cvtepu16_epi32(in)); 88| 735k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 89| 735k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 90| 735k| utf32_output += 8; 91| 735k| buf += 8; 92| | // surrogate pair(s) in a register 93| 735k| } else { 94| | // Let us do a scalar fallback. 95| | // It may seem wasteful to use scalar code, but being efficient with SIMD 96| | // in the presence of surrogate pairs may require non-trivial tables. 97| 1.48k| size_t forward = 15; 98| 1.48k| size_t k = 0; 99| 1.48k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (99:11): [True: 76, False: 1.41k] ------------------ 100| 76| forward = size_t(end - buf - 1); 101| 76| } 102| 20.1k| for (; k < forward; k++) { ------------------ | Branch (102:14): [True: 18.7k, False: 1.38k] ------------------ 103| 18.7k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 104| 18.7k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (104:13): [True: 16.4k, False: 2.34k] ------------------ 105| 16.4k| *utf32_output++ = char32_t(word); 106| 16.4k| } else { 107| | // must be a surrogate pair 108| 2.34k| uint16_t diff = uint16_t(word - 0xD800); 109| 2.34k| uint16_t next_word = 110| 2.34k| scalar::utf16::swap_if_needed(buf[k + 1]); 111| 2.34k| k++; 112| 2.34k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 113| 2.34k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (113:15): [True: 108, False: 2.23k] ------------------ 114| 108| return std::make_pair(nullptr, utf32_output); 115| 108| } 116| 2.23k| uint32_t value = (diff << 10) + diff2 + 0x10000; 117| 2.23k| *utf32_output++ = char32_t(value); 118| 2.23k| } 119| 18.7k| } 120| 1.38k| buf += k; 121| 1.38k| } 122| 737k| } // while 123| 314| return std::make_pair(buf, utf32_output); 124| 422|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf16_to_utf32_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 136| 334| char32_t *utf32_output) { 137| 334| const char16_t *start = buf; 138| 334| const char16_t *end = buf + len; 139| | 140| 334| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 141| 334| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 142| | 143| 315k| while (end - buf >= 8) { ------------------ | Branch (143:10): [True: 315k, False: 214] ------------------ 144| 315k| __m128i in = _mm_loadu_si128((__m128i *)buf); 145| | 146| 315k| if (big_endian) { ------------------ | Branch (146:9): [Folded, False: 315k] ------------------ 147| 0| const __m128i swap = 148| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 149| 0| in = _mm_shuffle_epi8(in, swap); 150| 0| } 151| | 152| | // 1. Check if there are any surrogate word in the input chunk. 153| | // We have also deal with situation when there is a surrogate word 154| | // at the end of a chunk. 155| 315k| const __m128i surrogates_bytemask = 156| 315k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_d800); 157| | 158| | // bitmask = 0x0000 if there are no surrogates 159| | // = 0xc000 if the last word is a surrogate 160| 315k| const uint16_t surrogates_bitmask = 161| 315k| static_cast(_mm_movemask_epi8(surrogates_bytemask)); 162| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 163| | // However, it is likely an uncommon occurrence. 164| 315k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (164:9): [True: 310k, False: 5.06k] ------------------ 165| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 166| 310k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 167| 310k| _mm_cvtepu16_epi32(in)); 168| 310k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 169| 310k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 170| 310k| utf32_output += 8; 171| 310k| buf += 8; 172| | // surrogate pair(s) in a register 173| 310k| } else { 174| | // Let us do a scalar fallback. 175| | // It may seem wasteful to use scalar code, but being efficient with SIMD 176| | // in the presence of surrogate pairs may require non-trivial tables. 177| 5.06k| size_t forward = 15; 178| 5.06k| size_t k = 0; 179| 5.06k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (179:11): [True: 58, False: 5.00k] ------------------ 180| 58| forward = size_t(end - buf - 1); 181| 58| } 182| 70.6k| for (; k < forward; k++) { ------------------ | Branch (182:14): [True: 65.6k, False: 4.94k] ------------------ 183| 65.6k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 184| 65.6k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (184:13): [True: 56.1k, False: 9.50k] ------------------ 185| 56.1k| *utf32_output++ = char32_t(word); 186| 56.1k| } else { 187| | // must be a surrogate pair 188| 9.50k| uint16_t diff = uint16_t(word - 0xD800); 189| 9.50k| uint16_t next_word = 190| 9.50k| scalar::utf16::swap_if_needed(buf[k + 1]); 191| 9.50k| k++; 192| 9.50k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 193| 9.50k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (193:15): [True: 120, False: 9.38k] ------------------ 194| 120| return std::make_pair( 195| 120| result(error_code::SURROGATE, buf - start + k - 1), 196| 120| utf32_output); 197| 120| } 198| 9.38k| uint32_t value = (diff << 10) + diff2 + 0x10000; 199| 9.38k| *utf32_output++ = char32_t(value); 200| 9.38k| } 201| 65.6k| } 202| 4.94k| buf += k; 203| 4.94k| } 204| 315k| } // while 205| 214| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 206| 334|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf16_to_utf32_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 136| 346| char32_t *utf32_output) { 137| 346| const char16_t *start = buf; 138| 346| const char16_t *end = buf + len; 139| | 140| 346| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 141| 346| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 142| | 143| 297k| while (end - buf >= 8) { ------------------ | Branch (143:10): [True: 297k, False: 230] ------------------ 144| 297k| __m128i in = _mm_loadu_si128((__m128i *)buf); 145| | 146| 297k| if (big_endian) { ------------------ | Branch (146:9): [True: 297k, Folded] ------------------ 147| 297k| const __m128i swap = 148| 297k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 149| 297k| in = _mm_shuffle_epi8(in, swap); 150| 297k| } 151| | 152| | // 1. Check if there are any surrogate word in the input chunk. 153| | // We have also deal with situation when there is a surrogate word 154| | // at the end of a chunk. 155| 297k| const __m128i surrogates_bytemask = 156| 297k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_d800); 157| | 158| | // bitmask = 0x0000 if there are no surrogates 159| | // = 0xc000 if the last word is a surrogate 160| 297k| const uint16_t surrogates_bitmask = 161| 297k| static_cast(_mm_movemask_epi8(surrogates_bytemask)); 162| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 163| | // However, it is likely an uncommon occurrence. 164| 297k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (164:9): [True: 292k, False: 5.15k] ------------------ 165| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 166| 292k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 167| 292k| _mm_cvtepu16_epi32(in)); 168| 292k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 169| 292k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 170| 292k| utf32_output += 8; 171| 292k| buf += 8; 172| | // surrogate pair(s) in a register 173| 292k| } else { 174| | // Let us do a scalar fallback. 175| | // It may seem wasteful to use scalar code, but being efficient with SIMD 176| | // in the presence of surrogate pairs may require non-trivial tables. 177| 5.15k| size_t forward = 15; 178| 5.15k| size_t k = 0; 179| 5.15k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (179:11): [True: 82, False: 5.07k] ------------------ 180| 82| forward = size_t(end - buf - 1); 181| 82| } 182| 71.8k| for (; k < forward; k++) { ------------------ | Branch (182:14): [True: 66.7k, False: 5.04k] ------------------ 183| 66.7k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 184| 66.7k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (184:13): [True: 56.8k, False: 9.90k] ------------------ 185| 56.8k| *utf32_output++ = char32_t(word); 186| 56.8k| } else { 187| | // must be a surrogate pair 188| 9.90k| uint16_t diff = uint16_t(word - 0xD800); 189| 9.90k| uint16_t next_word = 190| 9.90k| scalar::utf16::swap_if_needed(buf[k + 1]); 191| 9.90k| k++; 192| 9.90k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 193| 9.90k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (193:15): [True: 116, False: 9.78k] ------------------ 194| 116| return std::make_pair( 195| 116| result(error_code::SURROGATE, buf - start + k - 1), 196| 116| utf32_output); 197| 116| } 198| 9.78k| uint32_t value = (diff << 10) + diff2 + 0x10000; 199| 9.78k| *utf32_output++ = char32_t(value); 200| 9.78k| } 201| 66.7k| } 202| 5.04k| buf += k; 203| 5.04k| } 204| 297k| } // while 205| 230| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 206| 346|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_125sse_convert_utf16_to_utf8ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 842|sse_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| | 58| 842| const char16_t *end = buf + len; 59| | 60| 842| const __m128i v_0000 = _mm_setzero_si128(); 61| 842| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 62| 842| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 63| 842| const size_t safety_margin = 64| 842| 12; // to avoid overruns, see issue 65| | // https://github.com/simdutf/simdutf/issues/92 66| | 67| 1.07M| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (67:10): [True: 1.07M, False: 736] ------------------ 68| 1.07M| __m128i in = _mm_loadu_si128((__m128i *)buf); 69| 1.07M| if (big_endian) { ------------------ | Branch (69:9): [Folded, False: 1.07M] ------------------ 70| 0| const __m128i swap = 71| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 72| 0| in = _mm_shuffle_epi8(in, swap); 73| 0| } 74| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 75| 1.07M| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 76| 1.07M| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 763k, False: 312k] ------------------ 77| 763k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 78| 763k| if (big_endian) { ------------------ | Branch (78:11): [Folded, False: 763k] ------------------ 79| 0| const __m128i swap = 80| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 81| 0| nextin = _mm_shuffle_epi8(nextin, swap); 82| 0| } 83| 763k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (83:11): [True: 43.3k, False: 719k] ------------------ 84| | // 1. pack the bytes 85| | // obviously suboptimal. 86| 43.3k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 87| | // 2. store (16 bytes) 88| 43.3k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 89| | // 3. adjust pointers 90| 43.3k| buf += 8; 91| 43.3k| utf8_output += 8; 92| 43.3k| in = nextin; 93| 719k| } else { 94| | // 1. pack the bytes 95| | // obviously suboptimal. 96| 719k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 97| | // 2. store (16 bytes) 98| 719k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 99| | // 3. adjust pointers 100| 719k| buf += 16; 101| 719k| utf8_output += 16; 102| 719k| continue; // we are done for this round! 103| 719k| } 104| 763k| } 105| | 106| | // no bits set above 7th bit 107| 355k| const __m128i one_byte_bytemask = 108| 355k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 109| 355k| const uint16_t one_byte_bitmask = 110| 355k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 111| | 112| | // no bits set above 11th bit 113| 355k| const __m128i one_or_two_bytes_bytemask = 114| 355k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 115| 355k| const uint16_t one_or_two_bytes_bitmask = 116| 355k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 117| | 118| 355k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (118:9): [True: 235k, False: 119k] ------------------ 119| 235k| internal::westmere::write_v_u16_11bits_to_utf8( 120| 235k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 121| 235k| buf += 8; 122| 235k| continue; 123| 235k| } 124| | 125| | // 1. Check if there are any surrogate word in the input chunk. 126| | // We have also deal with situation when there is a surrogate word 127| | // at the end of a chunk. 128| 119k| const __m128i surrogates_bytemask = 129| 119k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_d800); 130| | 131| | // bitmask = 0x0000 if there are no surrogates 132| | // = 0xc000 if the last word is a surrogate 133| 119k| const uint16_t surrogates_bitmask = 134| 119k| static_cast(_mm_movemask_epi8(surrogates_bytemask)); 135| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 136| | // However, it is likely an uncommon occurrence. 137| 119k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (137:9): [True: 113k, False: 6.91k] ------------------ 138| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 139| 113k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 140| 113k| 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 141| | 142| | /* In this branch we handle three cases: 143| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 144| | single UFT-8 byte 145| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - two 146| | UTF-8 bytes 147| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 148| | three UTF-8 bytes 149| | 150| | We expand the input word (16-bit) into two code units (32-bit), thus 151| | we have room for four bytes. However, we need five distinct bit 152| | layouts. Note that the last byte in cases #2 and #3 is the same. 153| | 154| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 155| | in register t2. 156| | 157| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 158| | either byte 1 for case #2 or byte 2 for case #3. Note that they 159| | differ by exactly one bit. 160| | 161| | Finally from these two code units we build proper UTF-8 sequence, taking 162| | into account the case (i.e, the number of bytes to write). 163| | */ 164| | /** 165| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 166| | * t2 => [0ccc|cccc] [10cc|cccc] 167| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 168| | */ 169| 113k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 170| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 171| 113k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 172| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 173| 113k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 169| 113k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 174| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 175| 113k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 169| 113k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 176| | 177| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 178| 113k| const __m128i s0 = _mm_srli_epi16(in, 4); 179| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 180| 113k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 169| 113k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 181| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 182| 113k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 169| 113k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 183| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 184| 113k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 169| 113k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 185| 113k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 186| 113k| simdutf_vec(0b0100000000000000)); ------------------ | | 169| 113k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 187| 113k| const __m128i s4 = _mm_xor_si128(s3, m0); 188| 113k|#undef simdutf_vec 189| | 190| | // 4. expand code units 16-bit => 32-bit 191| 113k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 192| 113k| const __m128i out1 = _mm_unpackhi_epi16(t2, s4); 193| | 194| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 195| 113k| const uint16_t mask = 196| 113k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 197| 113k| if (mask == 0) { ------------------ | Branch (197:11): [True: 20.3k, False: 92.6k] ------------------ 198| | // We only have three-byte code units. Use fast path. 199| 20.3k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 200| 20.3k| 15, 13, -1, -1, -1, -1); 201| 20.3k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 202| 20.3k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 203| 20.3k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 204| 20.3k| utf8_output += 12; 205| 20.3k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 206| 20.3k| utf8_output += 12; 207| 20.3k| buf += 8; 208| 20.3k| continue; 209| 20.3k| } 210| 92.6k| const uint8_t mask0 = uint8_t(mask); 211| | 212| 92.6k| const uint8_t *row0 = 213| 92.6k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 214| 92.6k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 215| 92.6k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 216| | 217| 92.6k| const uint8_t mask1 = static_cast(mask >> 8); 218| | 219| 92.6k| const uint8_t *row1 = 220| 92.6k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 221| 92.6k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 222| 92.6k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 223| | 224| 92.6k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 225| 92.6k| utf8_output += row0[0]; 226| 92.6k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 227| 92.6k| utf8_output += row1[0]; 228| | 229| 92.6k| buf += 8; 230| | // surrogate pair(s) in a register 231| 92.6k| } else { 232| | // Let us do a scalar fallback. 233| | // It may seem wasteful to use scalar code, but being efficient with SIMD 234| | // in the presence of surrogate pairs may require non-trivial tables. 235| 6.91k| size_t forward = 15; 236| 6.91k| size_t k = 0; 237| 6.91k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (237:11): [True: 0, False: 6.91k] ------------------ 238| 0| forward = size_t(end - buf - 1); 239| 0| } 240| 101k| for (; k < forward; k++) { ------------------ | Branch (240:14): [True: 95.0k, False: 6.81k] ------------------ 241| 95.0k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 242| 95.0k| if ((word & 0xFF80) == 0) { ------------------ | Branch (242:13): [True: 13.3k, False: 81.6k] ------------------ 243| 13.3k| *utf8_output++ = char(word); 244| 81.6k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (244:20): [True: 3.92k, False: 77.7k] ------------------ 245| 3.92k| *utf8_output++ = char((word >> 6) | 0b11000000); 246| 3.92k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 247| 77.7k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (247:20): [True: 69.8k, False: 7.92k] ------------------ 248| 69.8k| *utf8_output++ = char((word >> 12) | 0b11100000); 249| 69.8k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 250| 69.8k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 251| 69.8k| } else { 252| | // must be a surrogate pair 253| 7.92k| uint16_t diff = uint16_t(word - 0xD800); 254| 7.92k| uint16_t next_word = 255| 7.92k| scalar::utf16::swap_if_needed(buf[k + 1]); 256| 7.92k| k++; 257| 7.92k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 258| 7.92k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (258:15): [True: 106, False: 7.81k] ------------------ 259| 106| return std::make_pair(nullptr, utf8_output); 260| 106| } 261| 7.81k| uint32_t value = (diff << 10) + diff2 + 0x10000; 262| 7.81k| *utf8_output++ = char((value >> 18) | 0b11110000); 263| 7.81k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 264| 7.81k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 265| 7.81k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 266| 7.81k| } 267| 95.0k| } 268| 6.81k| buf += k; 269| 6.81k| } 270| 119k| } // while 271| | 272| 736| return std::make_pair(buf, utf8_output); 273| 842|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_125sse_convert_utf16_to_utf8ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 798|sse_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| | 58| 798| const char16_t *end = buf + len; 59| | 60| 798| const __m128i v_0000 = _mm_setzero_si128(); 61| 798| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 62| 798| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 63| 798| const size_t safety_margin = 64| 798| 12; // to avoid overruns, see issue 65| | // https://github.com/simdutf/simdutf/issues/92 66| | 67| 1.16M| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (67:10): [True: 1.16M, False: 708] ------------------ 68| 1.16M| __m128i in = _mm_loadu_si128((__m128i *)buf); 69| 1.16M| if (big_endian) { ------------------ | Branch (69:9): [True: 1.16M, Folded] ------------------ 70| 1.16M| const __m128i swap = 71| 1.16M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 72| 1.16M| in = _mm_shuffle_epi8(in, swap); 73| 1.16M| } 74| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 75| 1.16M| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 76| 1.16M| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 741k, False: 425k] ------------------ 77| 741k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 78| 741k| if (big_endian) { ------------------ | Branch (78:11): [True: 741k, Folded] ------------------ 79| 741k| const __m128i swap = 80| 741k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 81| 741k| nextin = _mm_shuffle_epi8(nextin, swap); 82| 741k| } 83| 741k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (83:11): [True: 36.6k, False: 705k] ------------------ 84| | // 1. pack the bytes 85| | // obviously suboptimal. 86| 36.6k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 87| | // 2. store (16 bytes) 88| 36.6k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 89| | // 3. adjust pointers 90| 36.6k| buf += 8; 91| 36.6k| utf8_output += 8; 92| 36.6k| in = nextin; 93| 705k| } else { 94| | // 1. pack the bytes 95| | // obviously suboptimal. 96| 705k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 97| | // 2. store (16 bytes) 98| 705k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 99| | // 3. adjust pointers 100| 705k| buf += 16; 101| 705k| utf8_output += 16; 102| 705k| continue; // we are done for this round! 103| 705k| } 104| 741k| } 105| | 106| | // no bits set above 7th bit 107| 462k| const __m128i one_byte_bytemask = 108| 462k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 109| 462k| const uint16_t one_byte_bitmask = 110| 462k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 111| | 112| | // no bits set above 11th bit 113| 462k| const __m128i one_or_two_bytes_bytemask = 114| 462k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 115| 462k| const uint16_t one_or_two_bytes_bitmask = 116| 462k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 117| | 118| 462k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (118:9): [True: 29.0k, False: 433k] ------------------ 119| 29.0k| internal::westmere::write_v_u16_11bits_to_utf8( 120| 29.0k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 121| 29.0k| buf += 8; 122| 29.0k| continue; 123| 29.0k| } 124| | 125| | // 1. Check if there are any surrogate word in the input chunk. 126| | // We have also deal with situation when there is a surrogate word 127| | // at the end of a chunk. 128| 433k| const __m128i surrogates_bytemask = 129| 433k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_d800); 130| | 131| | // bitmask = 0x0000 if there are no surrogates 132| | // = 0xc000 if the last word is a surrogate 133| 433k| const uint16_t surrogates_bitmask = 134| 433k| static_cast(_mm_movemask_epi8(surrogates_bytemask)); 135| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 136| | // However, it is likely an uncommon occurrence. 137| 433k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (137:9): [True: 429k, False: 3.99k] ------------------ 138| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 139| 429k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 140| 429k| 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 141| | 142| | /* In this branch we handle three cases: 143| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 144| | single UFT-8 byte 145| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - two 146| | UTF-8 bytes 147| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 148| | three UTF-8 bytes 149| | 150| | We expand the input word (16-bit) into two code units (32-bit), thus 151| | we have room for four bytes. However, we need five distinct bit 152| | layouts. Note that the last byte in cases #2 and #3 is the same. 153| | 154| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 155| | in register t2. 156| | 157| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 158| | either byte 1 for case #2 or byte 2 for case #3. Note that they 159| | differ by exactly one bit. 160| | 161| | Finally from these two code units we build proper UTF-8 sequence, taking 162| | into account the case (i.e, the number of bytes to write). 163| | */ 164| | /** 165| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 166| | * t2 => [0ccc|cccc] [10cc|cccc] 167| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 168| | */ 169| 429k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 170| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 171| 429k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 172| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 173| 429k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 169| 429k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 174| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 175| 429k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 169| 429k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 176| | 177| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 178| 429k| const __m128i s0 = _mm_srli_epi16(in, 4); 179| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 180| 429k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 169| 429k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 181| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 182| 429k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 169| 429k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 183| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 184| 429k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 169| 429k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 185| 429k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 186| 429k| simdutf_vec(0b0100000000000000)); ------------------ | | 169| 429k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 187| 429k| const __m128i s4 = _mm_xor_si128(s3, m0); 188| 429k|#undef simdutf_vec 189| | 190| | // 4. expand code units 16-bit => 32-bit 191| 429k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 192| 429k| const __m128i out1 = _mm_unpackhi_epi16(t2, s4); 193| | 194| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 195| 429k| const uint16_t mask = 196| 429k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 197| 429k| if (mask == 0) { ------------------ | Branch (197:11): [True: 257k, False: 171k] ------------------ 198| | // We only have three-byte code units. Use fast path. 199| 257k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 200| 257k| 15, 13, -1, -1, -1, -1); 201| 257k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 202| 257k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 203| 257k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 204| 257k| utf8_output += 12; 205| 257k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 206| 257k| utf8_output += 12; 207| 257k| buf += 8; 208| 257k| continue; 209| 257k| } 210| 171k| const uint8_t mask0 = uint8_t(mask); 211| | 212| 171k| const uint8_t *row0 = 213| 171k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 214| 171k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 215| 171k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 216| | 217| 171k| const uint8_t mask1 = static_cast(mask >> 8); 218| | 219| 171k| const uint8_t *row1 = 220| 171k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 221| 171k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 222| 171k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 223| | 224| 171k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 225| 171k| utf8_output += row0[0]; 226| 171k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 227| 171k| utf8_output += row1[0]; 228| | 229| 171k| buf += 8; 230| | // surrogate pair(s) in a register 231| 171k| } else { 232| | // Let us do a scalar fallback. 233| | // It may seem wasteful to use scalar code, but being efficient with SIMD 234| | // in the presence of surrogate pairs may require non-trivial tables. 235| 3.99k| size_t forward = 15; 236| 3.99k| size_t k = 0; 237| 3.99k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (237:11): [True: 0, False: 3.99k] ------------------ 238| 0| forward = size_t(end - buf - 1); 239| 0| } 240| 58.0k| for (; k < forward; k++) { ------------------ | Branch (240:14): [True: 54.1k, False: 3.90k] ------------------ 241| 54.1k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 242| 54.1k| if ((word & 0xFF80) == 0) { ------------------ | Branch (242:13): [True: 7.65k, False: 46.5k] ------------------ 243| 7.65k| *utf8_output++ = char(word); 244| 46.5k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (244:20): [True: 2.08k, False: 44.4k] ------------------ 245| 2.08k| *utf8_output++ = char((word >> 6) | 0b11000000); 246| 2.08k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 247| 44.4k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (247:20): [True: 39.5k, False: 4.92k] ------------------ 248| 39.5k| *utf8_output++ = char((word >> 12) | 0b11100000); 249| 39.5k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 250| 39.5k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 251| 39.5k| } else { 252| | // must be a surrogate pair 253| 4.92k| uint16_t diff = uint16_t(word - 0xD800); 254| 4.92k| uint16_t next_word = 255| 4.92k| scalar::utf16::swap_if_needed(buf[k + 1]); 256| 4.92k| k++; 257| 4.92k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 258| 4.92k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (258:15): [True: 90, False: 4.83k] ------------------ 259| 90| return std::make_pair(nullptr, utf8_output); 260| 90| } 261| 4.83k| uint32_t value = (diff << 10) + diff2 + 0x10000; 262| 4.83k| *utf8_output++ = char((value >> 18) | 0b11110000); 263| 4.83k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 264| 4.83k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 265| 4.83k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 266| 4.83k| } 267| 54.1k| } 268| 3.90k| buf += k; 269| 3.90k| } 270| 433k| } // while 271| | 272| 708| return std::make_pair(buf, utf8_output); 273| 798|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_137sse_convert_utf16_to_utf8_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 285| 642| char *utf8_output) { 286| 642| const char16_t *start = buf; 287| 642| const char16_t *end = buf + len; 288| | 289| 642| const __m128i v_0000 = _mm_setzero_si128(); 290| 642| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 291| 642| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 292| 642| const size_t safety_margin = 293| 642| 12; // to avoid overruns, see issue 294| | // https://github.com/simdutf/simdutf/issues/92 295| | 296| 965k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (296:10): [True: 964k, False: 496] ------------------ 297| 964k| __m128i in = _mm_loadu_si128((__m128i *)buf); 298| 964k| if (big_endian) { ------------------ | Branch (298:9): [Folded, False: 964k] ------------------ 299| 0| const __m128i swap = 300| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 301| 0| in = _mm_shuffle_epi8(in, swap); 302| 0| } 303| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 304| 964k| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 305| 964k| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (305:9): [True: 325k, False: 639k] ------------------ 306| 325k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 307| 325k| if (big_endian) { ------------------ | Branch (307:11): [Folded, False: 325k] ------------------ 308| 0| const __m128i swap = 309| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 310| 0| nextin = _mm_shuffle_epi8(nextin, swap); 311| 0| } 312| 325k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (312:11): [True: 12.2k, False: 313k] ------------------ 313| | // 1. pack the bytes 314| | // obviously suboptimal. 315| 12.2k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 316| | // 2. store (16 bytes) 317| 12.2k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 318| | // 3. adjust pointers 319| 12.2k| buf += 8; 320| 12.2k| utf8_output += 8; 321| 12.2k| in = nextin; 322| 313k| } else { 323| | // 1. pack the bytes 324| | // obviously suboptimal. 325| 313k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 326| | // 2. store (16 bytes) 327| 313k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 328| | // 3. adjust pointers 329| 313k| buf += 16; 330| 313k| utf8_output += 16; 331| 313k| continue; // we are done for this round! 332| 313k| } 333| 325k| } 334| | 335| | // no bits set above 7th bit 336| 651k| const __m128i one_byte_bytemask = 337| 651k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 338| 651k| const uint16_t one_byte_bitmask = 339| 651k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 340| | 341| | // no bits set above 11th bit 342| 651k| const __m128i one_or_two_bytes_bytemask = 343| 651k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 344| 651k| const uint16_t one_or_two_bytes_bitmask = 345| 651k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 346| | 347| 651k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (347:9): [True: 123k, False: 528k] ------------------ 348| 123k| internal::westmere::write_v_u16_11bits_to_utf8( 349| 123k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 350| 123k| buf += 8; 351| 123k| continue; 352| 123k| } 353| | 354| | // 1. Check if there are any surrogate word in the input chunk. 355| | // We have also deal with situation when there is a surrogate word 356| | // at the end of a chunk. 357| 528k| const __m128i surrogates_bytemask = 358| 528k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_d800); 359| | 360| | // bitmask = 0x0000 if there are no surrogates 361| | // = 0xc000 if the last word is a surrogate 362| 528k| const uint16_t surrogates_bitmask = 363| 528k| static_cast(_mm_movemask_epi8(surrogates_bytemask)); 364| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 365| | // However, it is likely an uncommon occurrence. 366| 528k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (366:9): [True: 525k, False: 2.60k] ------------------ 367| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 368| 525k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 369| 525k| 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 370| | 371| | /* In this branch we handle three cases: 372| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 373| | single UFT-8 byte 374| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - two 375| | UTF-8 bytes 376| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 377| | three UTF-8 bytes 378| | 379| | We expand the input word (16-bit) into two code units (32-bit), thus 380| | we have room for four bytes. However, we need five distinct bit 381| | layouts. Note that the last byte in cases #2 and #3 is the same. 382| | 383| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 384| | in register t2. 385| | 386| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 387| | either byte 1 for case #2 or byte 2 for case #3. Note that they 388| | differ by exactly one bit. 389| | 390| | Finally from these two code units we build proper UTF-8 sequence, taking 391| | into account the case (i.e, the number of bytes to write). 392| | */ 393| | /** 394| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 395| | * t2 => [0ccc|cccc] [10cc|cccc] 396| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 397| | */ 398| 525k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 399| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 400| 525k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 401| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 402| 525k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 398| 525k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 403| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 404| 525k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 398| 525k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 405| | 406| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 407| 525k| const __m128i s0 = _mm_srli_epi16(in, 4); 408| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 409| 525k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 398| 525k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 410| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 411| 525k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 398| 525k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 412| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 413| 525k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 398| 525k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 414| 525k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 415| 525k| simdutf_vec(0b0100000000000000)); ------------------ | | 398| 525k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 416| 525k| const __m128i s4 = _mm_xor_si128(s3, m0); 417| 525k|#undef simdutf_vec 418| | 419| | // 4. expand code units 16-bit => 32-bit 420| 525k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 421| 525k| const __m128i out1 = _mm_unpackhi_epi16(t2, s4); 422| | 423| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 424| 525k| const uint16_t mask = 425| 525k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 426| 525k| if (mask == 0) { ------------------ | Branch (426:11): [True: 446k, False: 79.1k] ------------------ 427| | // We only have three-byte code units. Use fast path. 428| 446k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 429| 446k| 15, 13, -1, -1, -1, -1); 430| 446k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 431| 446k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 432| 446k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 433| 446k| utf8_output += 12; 434| 446k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 435| 446k| utf8_output += 12; 436| 446k| buf += 8; 437| 446k| continue; 438| 446k| } 439| 79.1k| const uint8_t mask0 = uint8_t(mask); 440| | 441| 79.1k| const uint8_t *row0 = 442| 79.1k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 443| 79.1k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 444| 79.1k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 445| | 446| 79.1k| const uint8_t mask1 = static_cast(mask >> 8); 447| | 448| 79.1k| const uint8_t *row1 = 449| 79.1k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 450| 79.1k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 451| 79.1k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 452| | 453| 79.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 454| 79.1k| utf8_output += row0[0]; 455| 79.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 456| 79.1k| utf8_output += row1[0]; 457| | 458| 79.1k| buf += 8; 459| | // surrogate pair(s) in a register 460| 79.1k| } else { 461| | // Let us do a scalar fallback. 462| | // It may seem wasteful to use scalar code, but being efficient with SIMD 463| | // in the presence of surrogate pairs may require non-trivial tables. 464| 2.60k| size_t forward = 15; 465| 2.60k| size_t k = 0; 466| 2.60k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (466:11): [True: 0, False: 2.60k] ------------------ 467| 0| forward = size_t(end - buf - 1); 468| 0| } 469| 37.1k| for (; k < forward; k++) { ------------------ | Branch (469:14): [True: 34.6k, False: 2.46k] ------------------ 470| 34.6k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 471| 34.6k| if ((word & 0xFF80) == 0) { ------------------ | Branch (471:13): [True: 5.97k, False: 28.6k] ------------------ 472| 5.97k| *utf8_output++ = char(word); 473| 28.6k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (473:20): [True: 1.66k, False: 27.0k] ------------------ 474| 1.66k| *utf8_output++ = char((word >> 6) | 0b11000000); 475| 1.66k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 476| 27.0k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (476:20): [True: 23.9k, False: 3.04k] ------------------ 477| 23.9k| *utf8_output++ = char((word >> 12) | 0b11100000); 478| 23.9k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 479| 23.9k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 480| 23.9k| } else { 481| | // must be a surrogate pair 482| 3.04k| uint16_t diff = uint16_t(word - 0xD800); 483| 3.04k| uint16_t next_word = 484| 3.04k| scalar::utf16::swap_if_needed(buf[k + 1]); 485| 3.04k| k++; 486| 3.04k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 487| 3.04k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (487:15): [True: 146, False: 2.89k] ------------------ 488| 146| return std::make_pair( 489| 146| result(error_code::SURROGATE, buf - start + k - 1), 490| 146| utf8_output); 491| 146| } 492| 2.89k| uint32_t value = (diff << 10) + diff2 + 0x10000; 493| 2.89k| *utf8_output++ = char((value >> 18) | 0b11110000); 494| 2.89k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 495| 2.89k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 496| 2.89k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 497| 2.89k| } 498| 34.6k| } 499| 2.46k| buf += k; 500| 2.46k| } 501| 528k| } // while 502| | 503| 496| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 504| 642|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_137sse_convert_utf16_to_utf8_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 285| 654| char *utf8_output) { 286| 654| const char16_t *start = buf; 287| 654| const char16_t *end = buf + len; 288| | 289| 654| const __m128i v_0000 = _mm_setzero_si128(); 290| 654| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 291| 654| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 292| 654| const size_t safety_margin = 293| 654| 12; // to avoid overruns, see issue 294| | // https://github.com/simdutf/simdutf/issues/92 295| | 296| 559k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (296:10): [True: 559k, False: 496] ------------------ 297| 559k| __m128i in = _mm_loadu_si128((__m128i *)buf); 298| 559k| if (big_endian) { ------------------ | Branch (298:9): [True: 559k, Folded] ------------------ 299| 559k| const __m128i swap = 300| 559k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 301| 559k| in = _mm_shuffle_epi8(in, swap); 302| 559k| } 303| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 304| 559k| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 305| 559k| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (305:9): [True: 406k, False: 153k] ------------------ 306| 406k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 307| 406k| if (big_endian) { ------------------ | Branch (307:11): [True: 406k, Folded] ------------------ 308| 406k| const __m128i swap = 309| 406k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 310| 406k| nextin = _mm_shuffle_epi8(nextin, swap); 311| 406k| } 312| 406k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (312:11): [True: 33.2k, False: 372k] ------------------ 313| | // 1. pack the bytes 314| | // obviously suboptimal. 315| 33.2k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 316| | // 2. store (16 bytes) 317| 33.2k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 318| | // 3. adjust pointers 319| 33.2k| buf += 8; 320| 33.2k| utf8_output += 8; 321| 33.2k| in = nextin; 322| 372k| } else { 323| | // 1. pack the bytes 324| | // obviously suboptimal. 325| 372k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 326| | // 2. store (16 bytes) 327| 372k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 328| | // 3. adjust pointers 329| 372k| buf += 16; 330| 372k| utf8_output += 16; 331| 372k| continue; // we are done for this round! 332| 372k| } 333| 406k| } 334| | 335| | // no bits set above 7th bit 336| 186k| const __m128i one_byte_bytemask = 337| 186k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 338| 186k| const uint16_t one_byte_bitmask = 339| 186k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 340| | 341| | // no bits set above 11th bit 342| 186k| const __m128i one_or_two_bytes_bytemask = 343| 186k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 344| 186k| const uint16_t one_or_two_bytes_bitmask = 345| 186k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 346| | 347| 186k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (347:9): [True: 42.5k, False: 143k] ------------------ 348| 42.5k| internal::westmere::write_v_u16_11bits_to_utf8( 349| 42.5k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 350| 42.5k| buf += 8; 351| 42.5k| continue; 352| 42.5k| } 353| | 354| | // 1. Check if there are any surrogate word in the input chunk. 355| | // We have also deal with situation when there is a surrogate word 356| | // at the end of a chunk. 357| 143k| const __m128i surrogates_bytemask = 358| 143k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_d800); 359| | 360| | // bitmask = 0x0000 if there are no surrogates 361| | // = 0xc000 if the last word is a surrogate 362| 143k| const uint16_t surrogates_bitmask = 363| 143k| static_cast(_mm_movemask_epi8(surrogates_bytemask)); 364| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 365| | // However, it is likely an uncommon occurrence. 366| 143k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (366:9): [True: 136k, False: 7.41k] ------------------ 367| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 368| 136k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 369| 136k| 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 370| | 371| | /* In this branch we handle three cases: 372| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 373| | single UFT-8 byte 374| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - two 375| | UTF-8 bytes 376| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 377| | three UTF-8 bytes 378| | 379| | We expand the input word (16-bit) into two code units (32-bit), thus 380| | we have room for four bytes. However, we need five distinct bit 381| | layouts. Note that the last byte in cases #2 and #3 is the same. 382| | 383| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 384| | in register t2. 385| | 386| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 387| | either byte 1 for case #2 or byte 2 for case #3. Note that they 388| | differ by exactly one bit. 389| | 390| | Finally from these two code units we build proper UTF-8 sequence, taking 391| | into account the case (i.e, the number of bytes to write). 392| | */ 393| | /** 394| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 395| | * t2 => [0ccc|cccc] [10cc|cccc] 396| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 397| | */ 398| 136k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 399| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 400| 136k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 401| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 402| 136k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 398| 136k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 403| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 404| 136k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 398| 136k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 405| | 406| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 407| 136k| const __m128i s0 = _mm_srli_epi16(in, 4); 408| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 409| 136k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 398| 136k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 410| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 411| 136k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 398| 136k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 412| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 413| 136k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 398| 136k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 414| 136k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 415| 136k| simdutf_vec(0b0100000000000000)); ------------------ | | 398| 136k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 416| 136k| const __m128i s4 = _mm_xor_si128(s3, m0); 417| 136k|#undef simdutf_vec 418| | 419| | // 4. expand code units 16-bit => 32-bit 420| 136k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 421| 136k| const __m128i out1 = _mm_unpackhi_epi16(t2, s4); 422| | 423| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 424| 136k| const uint16_t mask = 425| 136k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 426| 136k| if (mask == 0) { ------------------ | Branch (426:11): [True: 49.5k, False: 86.9k] ------------------ 427| | // We only have three-byte code units. Use fast path. 428| 49.5k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 429| 49.5k| 15, 13, -1, -1, -1, -1); 430| 49.5k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 431| 49.5k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 432| 49.5k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 433| 49.5k| utf8_output += 12; 434| 49.5k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 435| 49.5k| utf8_output += 12; 436| 49.5k| buf += 8; 437| 49.5k| continue; 438| 49.5k| } 439| 86.9k| const uint8_t mask0 = uint8_t(mask); 440| | 441| 86.9k| const uint8_t *row0 = 442| 86.9k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 443| 86.9k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 444| 86.9k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 445| | 446| 86.9k| const uint8_t mask1 = static_cast(mask >> 8); 447| | 448| 86.9k| const uint8_t *row1 = 449| 86.9k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 450| 86.9k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 451| 86.9k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 452| | 453| 86.9k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 454| 86.9k| utf8_output += row0[0]; 455| 86.9k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 456| 86.9k| utf8_output += row1[0]; 457| | 458| 86.9k| buf += 8; 459| | // surrogate pair(s) in a register 460| 86.9k| } else { 461| | // Let us do a scalar fallback. 462| | // It may seem wasteful to use scalar code, but being efficient with SIMD 463| | // in the presence of surrogate pairs may require non-trivial tables. 464| 7.41k| size_t forward = 15; 465| 7.41k| size_t k = 0; 466| 7.41k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (466:11): [True: 0, False: 7.41k] ------------------ 467| 0| forward = size_t(end - buf - 1); 468| 0| } 469| 108k| for (; k < forward; k++) { ------------------ | Branch (469:14): [True: 101k, False: 7.25k] ------------------ 470| 101k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 471| 101k| if ((word & 0xFF80) == 0) { ------------------ | Branch (471:13): [True: 34.4k, False: 66.7k] ------------------ 472| 34.4k| *utf8_output++ = char(word); 473| 66.7k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (473:20): [True: 4.94k, False: 61.7k] ------------------ 474| 4.94k| *utf8_output++ = char((word >> 6) | 0b11000000); 475| 4.94k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 476| 61.7k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (476:20): [True: 53.1k, False: 8.69k] ------------------ 477| 53.1k| *utf8_output++ = char((word >> 12) | 0b11100000); 478| 53.1k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 479| 53.1k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 480| 53.1k| } else { 481| | // must be a surrogate pair 482| 8.69k| uint16_t diff = uint16_t(word - 0xD800); 483| 8.69k| uint16_t next_word = 484| 8.69k| scalar::utf16::swap_if_needed(buf[k + 1]); 485| 8.69k| k++; 486| 8.69k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 487| 8.69k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (487:15): [True: 158, False: 8.54k] ------------------ 488| 158| return std::make_pair( 489| 158| result(error_code::SURROGATE, buf - start + k - 1), 490| 158| utf8_output); 491| 158| } 492| 8.54k| uint32_t value = (diff << 10) + diff2 + 0x10000; 493| 8.54k| *utf8_output++ = char((value >> 18) | 0b11110000); 494| 8.54k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 495| 8.54k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 496| 8.54k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 497| 8.54k| } 498| 101k| } 499| 7.25k| buf += k; 500| 7.25k| } 501| 143k| } // while 502| | 503| 496| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 504| 654|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_utf32_to_latin1EPKDimPc: 3| 228| char *latin1_output) { 4| 228| const size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 5| | 6| 228| __m128i high_bytes_mask = _mm_set1_epi32(0xFFFFFF00); 7| 228| __m128i shufmask = 8| 228| _mm_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 12, 8, 4, 0); 9| | 10| 33.1k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (10:22): [True: 33.0k, False: 168] ------------------ 11| 33.0k| __m128i in1 = _mm_loadu_si128((__m128i *)buf); 12| 33.0k| __m128i in2 = _mm_loadu_si128((__m128i *)(buf + 4)); 13| 33.0k| __m128i in3 = _mm_loadu_si128((__m128i *)(buf + 8)); 14| 33.0k| __m128i in4 = _mm_loadu_si128((__m128i *)(buf + 12)); 15| | 16| 33.0k| __m128i check_combined = _mm_or_si128(in1, in2); 17| 33.0k| check_combined = _mm_or_si128(check_combined, in3); 18| 33.0k| check_combined = _mm_or_si128(check_combined, in4); 19| | 20| 33.0k| if (!_mm_testz_si128(check_combined, high_bytes_mask)) { ------------------ | Branch (20:9): [True: 60, False: 32.9k] ------------------ 21| 60| return std::make_pair(nullptr, latin1_output); 22| 60| } 23| 32.9k| __m128i pack1 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in1, shufmask), 24| 32.9k| _mm_shuffle_epi8(in2, shufmask)); 25| 32.9k| __m128i pack2 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in3, shufmask), 26| 32.9k| _mm_shuffle_epi8(in4, shufmask)); 27| 32.9k| __m128i pack = _mm_unpacklo_epi64(pack1, pack2); 28| 32.9k| _mm_storeu_si128((__m128i *)latin1_output, pack); 29| 32.9k| latin1_output += 16; 30| 32.9k| buf += 16; 31| 32.9k| } 32| | 33| 168| return std::make_pair(buf, latin1_output); 34| 228|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_139sse_convert_utf32_to_latin1_with_errorsEPKDimPc: 38| 436| char *latin1_output) { 39| 436| const char32_t *start = buf; 40| 436| const size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 41| | 42| 436| __m128i high_bytes_mask = _mm_set1_epi32(0xFFFFFF00); 43| 436| __m128i shufmask = 44| 436| _mm_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 12, 8, 4, 0); 45| | 46| 33.7k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (46:22): [True: 33.4k, False: 234] ------------------ 47| 33.4k| __m128i in1 = _mm_loadu_si128((__m128i *)buf); 48| 33.4k| __m128i in2 = _mm_loadu_si128((__m128i *)(buf + 4)); 49| 33.4k| __m128i in3 = _mm_loadu_si128((__m128i *)(buf + 8)); 50| 33.4k| __m128i in4 = _mm_loadu_si128((__m128i *)(buf + 12)); 51| | 52| 33.4k| __m128i check_combined = _mm_or_si128(in1, in2); 53| 33.4k| check_combined = _mm_or_si128(check_combined, in3); 54| 33.4k| check_combined = _mm_or_si128(check_combined, in4); 55| | 56| 33.4k| if (!_mm_testz_si128(check_combined, high_bytes_mask)) { ------------------ | Branch (56:9): [True: 202, False: 33.2k] ------------------ 57| | // Fallback to scalar code for handling errors 58| 712| for (int k = 0; k < 16; k++) { ------------------ | Branch (58:23): [True: 712, False: 0] ------------------ 59| 712| char32_t codepoint = buf[k]; 60| 712| if (codepoint <= 0xff) { ------------------ | Branch (60:13): [True: 510, False: 202] ------------------ 61| 510| *latin1_output++ = char(codepoint); 62| 510| } else { 63| 202| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 64| 202| latin1_output); 65| 202| } 66| 712| } 67| 0| buf += 16; 68| 0| continue; 69| 202| } 70| 33.2k| __m128i pack1 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in1, shufmask), 71| 33.2k| _mm_shuffle_epi8(in2, shufmask)); 72| 33.2k| __m128i pack2 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in3, shufmask), 73| 33.2k| _mm_shuffle_epi8(in4, shufmask)); 74| 33.2k| __m128i pack = _mm_unpacklo_epi64(pack1, pack2); 75| 33.2k| _mm_storeu_si128((__m128i *)latin1_output, pack); 76| 33.2k| latin1_output += 16; 77| 33.2k| buf += 16; 78| 33.2k| } 79| | 80| 234| return std::make_pair(result(error_code::SUCCESS, buf - start), 81| 234| latin1_output); 82| 436|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf32_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKDiPDsEES7_mS8_: 58| 822| char16_t *utf16_output) { 59| | 60| 822| const char32_t *end = buf + len; 61| | 62| 822| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 63| 822| __m128i forbidden_bytemask = _mm_setzero_si128(); 64| | 65| 76.1k| while (end - buf >= 16 + 8) { ------------------ | Branch (65:10): [True: 75.3k, False: 778] ------------------ 66| 75.3k| const __m128i *ptr = reinterpret_cast(buf); 67| 75.3k| const __m128i in0 = _mm_loadu_si128(ptr + 0); 68| 75.3k| const __m128i in1 = _mm_loadu_si128(ptr + 1); 69| 75.3k| const __m128i in2 = _mm_loadu_si128(ptr + 2); 70| 75.3k| const __m128i in3 = _mm_loadu_si128(ptr + 3); 71| | 72| 75.3k| const __m128i combined = 73| 75.3k| _mm_or_si128(_mm_or_si128(in2, in3), _mm_or_si128(in0, in1)); 74| 75.3k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 75.3k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 73.3k, False: 2.03k] | | ------------------ ------------------ 75| | // No bits set above 16th, directly pack UTF-32 to UTF-16 76| 73.3k| __m128i utf16_packed0 = _mm_packus_epi32(in0, in1); 77| 73.3k| __m128i utf16_packed1 = _mm_packus_epi32(in2, in3); 78| | 79| 73.3k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 80| 73.3k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 81| 73.3k| forbidden_bytemask = _mm_or_si128( 82| 73.3k| forbidden_bytemask, 83| 73.3k| _mm_or_si128( 84| 73.3k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed0, v_f800), v_d800), 85| 73.3k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed1, v_f800), v_d800))); 86| | 87| 73.3k| if (big_endian) { ------------------ | Branch (87:11): [Folded, False: 73.3k] ------------------ 88| 0| const __m128i swap = 89| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 90| 0| utf16_packed0 = _mm_shuffle_epi8(utf16_packed0, swap); 91| 0| utf16_packed1 = _mm_shuffle_epi8(utf16_packed1, swap); 92| 0| } 93| | 94| 73.3k| _mm_storeu_si128((__m128i *)utf16_output + 0, utf16_packed0); 95| 73.3k| _mm_storeu_si128((__m128i *)utf16_output + 1, utf16_packed1); 96| 73.3k| utf16_output += 16; 97| 73.3k| buf += 16; 98| 73.3k| } else { 99| 2.03k| if (!validate_utf32(in0, in1) || !validate_utf32(in2, in3)) { ------------------ | Branch (99:11): [True: 26, False: 2.01k] | Branch (99:40): [True: 18, False: 1.99k] ------------------ 100| 44| return std::make_pair(nullptr, utf16_output); 101| 44| } 102| | 103| 1.99k| const auto ret0 = sse_expand_surrogate(in0); 104| 1.99k| _mm_storeu_si128((__m128i *)utf16_output, ret0.compressed); 105| 1.99k| utf16_output += ret0.u16count; 106| | 107| 1.99k| const auto ret1 = sse_expand_surrogate(in1); 108| 1.99k| _mm_storeu_si128((__m128i *)utf16_output, ret1.compressed); 109| 1.99k| utf16_output += ret1.u16count; 110| | 111| 1.99k| const auto ret2 = sse_expand_surrogate(in2); 112| 1.99k| _mm_storeu_si128((__m128i *)utf16_output, ret2.compressed); 113| 1.99k| utf16_output += ret2.u16count; 114| | 115| 1.99k| const auto ret3 = sse_expand_surrogate(in3); 116| 1.99k| _mm_storeu_si128((__m128i *)utf16_output, ret3.compressed); 117| 1.99k| utf16_output += ret3.u16count; 118| | 119| 1.99k| buf += 16; 120| 1.99k| } 121| 75.3k| } 122| | 123| | // check for invalid input 124| 778| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (124:7): [True: 12, False: 766] ------------------ 125| 12| return std::make_pair(nullptr, utf16_output); 126| 12| } 127| | 128| 766| return std::make_pair(buf, utf16_output); 129| 778|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114validate_utf32EDv2_xS2_: 37| 7.10k|simdutf_really_inline bool validate_utf32(const __m128i a, const __m128i b) { 38| 7.10k| using vector_u32 = simd32; 39| | 40| 7.10k| const auto in0 = vector_u32(a); 41| 7.10k| const auto in1 = vector_u32(b); 42| | 43| 7.10k| const auto standardmax = vector_u32::splat(0x10ffff); 44| 7.10k| const auto offset = vector_u32::splat(0xffff2000); 45| 7.10k| const auto standardoffsetmax = vector_u32::splat(0xfffff7ff); 46| | 47| 7.10k| const auto too_large = max(in0, in1) > standardmax; 48| 7.10k| const auto surrogate0 = (in0 + offset) > standardoffsetmax; 49| 7.10k| const auto surrogate1 = (in1 + offset) > standardoffsetmax; 50| | 51| 7.10k| const auto combined = too_large | surrogate0 | surrogate1; 52| 7.10k| return !combined.any(); 53| 7.10k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_120sse_expand_surrogateILNS_10endiannessE0EEENS1_18expansion_result_tEDv2_x: 9| 7.97k|expansion_result_t sse_expand_surrogate(const __m128i x) { 10| 7.97k| using vector_u32 = simd32; 11| 7.97k| using vector_u8 = simd8; 12| | 13| 7.97k| const auto in = vector_u32(x); 14| | 15| 7.97k| const auto non_surrogate_mask = (in & uint32_t(0xffff0000)) == uint32_t(0); 16| 7.97k| const auto mask = (~non_surrogate_mask.to_4bit_bitmask()) & 0xf; 17| | 18| 7.97k| const auto t0 = in - uint32_t(0x00010000); 19| 7.97k| const auto hi = t0.shr<10>() & uint32_t(0x000003ff); 20| 7.97k| const auto lo = t0.shl<16>() & uint32_t(0x03ff0000); 21| 7.97k| const auto surrogates = (lo | hi) | uint32_t(0xdc00d800); 22| | 23| 7.97k| const auto merged = as_vector_u8(select(non_surrogate_mask, in, surrogates)); 24| | 25| 7.97k| const auto shuffle = vector_u8::load( 26| 7.97k| (byte_order == endianness::LITTLE) ------------------ | Branch (26:7): [True: 7.97k, Folded] ------------------ 27| 7.97k| ? tables::utf32_to_utf16::pack_utf32_to_utf16le[mask] 28| 7.97k| : tables::utf32_to_utf16::pack_utf32_to_utf16be[mask]); 29| | 30| 7.97k| const size_t u16count = (4 + count_ones(mask)); 31| 7.97k| const auto compressed = shuffle.lookup_16(merged); 32| | 33| 7.97k| return {u16count, compressed}; 34| 7.97k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf32_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKDiPDsEES7_mS8_: 58| 674| char16_t *utf16_output) { 59| | 60| 674| const char32_t *end = buf + len; 61| | 62| 674| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 63| 674| __m128i forbidden_bytemask = _mm_setzero_si128(); 64| | 65| 9.12k| while (end - buf >= 16 + 8) { ------------------ | Branch (65:10): [True: 8.53k, False: 596] ------------------ 66| 8.53k| const __m128i *ptr = reinterpret_cast(buf); 67| 8.53k| const __m128i in0 = _mm_loadu_si128(ptr + 0); 68| 8.53k| const __m128i in1 = _mm_loadu_si128(ptr + 1); 69| 8.53k| const __m128i in2 = _mm_loadu_si128(ptr + 2); 70| 8.53k| const __m128i in3 = _mm_loadu_si128(ptr + 3); 71| | 72| 8.53k| const __m128i combined = 73| 8.53k| _mm_or_si128(_mm_or_si128(in2, in3), _mm_or_si128(in0, in1)); 74| 8.53k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 8.53k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 6.98k, False: 1.55k] | | ------------------ ------------------ 75| | // No bits set above 16th, directly pack UTF-32 to UTF-16 76| 6.98k| __m128i utf16_packed0 = _mm_packus_epi32(in0, in1); 77| 6.98k| __m128i utf16_packed1 = _mm_packus_epi32(in2, in3); 78| | 79| 6.98k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 80| 6.98k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 81| 6.98k| forbidden_bytemask = _mm_or_si128( 82| 6.98k| forbidden_bytemask, 83| 6.98k| _mm_or_si128( 84| 6.98k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed0, v_f800), v_d800), 85| 6.98k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed1, v_f800), v_d800))); 86| | 87| 6.98k| if (big_endian) { ------------------ | Branch (87:11): [True: 6.98k, Folded] ------------------ 88| 6.98k| const __m128i swap = 89| 6.98k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 90| 6.98k| utf16_packed0 = _mm_shuffle_epi8(utf16_packed0, swap); 91| 6.98k| utf16_packed1 = _mm_shuffle_epi8(utf16_packed1, swap); 92| 6.98k| } 93| | 94| 6.98k| _mm_storeu_si128((__m128i *)utf16_output + 0, utf16_packed0); 95| 6.98k| _mm_storeu_si128((__m128i *)utf16_output + 1, utf16_packed1); 96| 6.98k| utf16_output += 16; 97| 6.98k| buf += 16; 98| 6.98k| } else { 99| 1.55k| if (!validate_utf32(in0, in1) || !validate_utf32(in2, in3)) { ------------------ | Branch (99:11): [True: 54, False: 1.49k] | Branch (99:40): [True: 24, False: 1.47k] ------------------ 100| 78| return std::make_pair(nullptr, utf16_output); 101| 78| } 102| | 103| 1.47k| const auto ret0 = sse_expand_surrogate(in0); 104| 1.47k| _mm_storeu_si128((__m128i *)utf16_output, ret0.compressed); 105| 1.47k| utf16_output += ret0.u16count; 106| | 107| 1.47k| const auto ret1 = sse_expand_surrogate(in1); 108| 1.47k| _mm_storeu_si128((__m128i *)utf16_output, ret1.compressed); 109| 1.47k| utf16_output += ret1.u16count; 110| | 111| 1.47k| const auto ret2 = sse_expand_surrogate(in2); 112| 1.47k| _mm_storeu_si128((__m128i *)utf16_output, ret2.compressed); 113| 1.47k| utf16_output += ret2.u16count; 114| | 115| 1.47k| const auto ret3 = sse_expand_surrogate(in3); 116| 1.47k| _mm_storeu_si128((__m128i *)utf16_output, ret3.compressed); 117| 1.47k| utf16_output += ret3.u16count; 118| | 119| 1.47k| buf += 16; 120| 1.47k| } 121| 8.53k| } 122| | 123| | // check for invalid input 124| 596| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (124:7): [True: 12, False: 584] ------------------ 125| 12| return std::make_pair(nullptr, utf16_output); 126| 12| } 127| | 128| 584| return std::make_pair(buf, utf16_output); 129| 596|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_120sse_expand_surrogateILNS_10endiannessE1EEENS1_18expansion_result_tEDv2_x: 9| 5.89k|expansion_result_t sse_expand_surrogate(const __m128i x) { 10| 5.89k| using vector_u32 = simd32; 11| 5.89k| using vector_u8 = simd8; 12| | 13| 5.89k| const auto in = vector_u32(x); 14| | 15| 5.89k| const auto non_surrogate_mask = (in & uint32_t(0xffff0000)) == uint32_t(0); 16| 5.89k| const auto mask = (~non_surrogate_mask.to_4bit_bitmask()) & 0xf; 17| | 18| 5.89k| const auto t0 = in - uint32_t(0x00010000); 19| 5.89k| const auto hi = t0.shr<10>() & uint32_t(0x000003ff); 20| 5.89k| const auto lo = t0.shl<16>() & uint32_t(0x03ff0000); 21| 5.89k| const auto surrogates = (lo | hi) | uint32_t(0xdc00d800); 22| | 23| 5.89k| const auto merged = as_vector_u8(select(non_surrogate_mask, in, surrogates)); 24| | 25| 5.89k| const auto shuffle = vector_u8::load( 26| 5.89k| (byte_order == endianness::LITTLE) ------------------ | Branch (26:7): [Folded, False: 5.89k] ------------------ 27| 5.89k| ? tables::utf32_to_utf16::pack_utf32_to_utf16le[mask] 28| 5.89k| : tables::utf32_to_utf16::pack_utf32_to_utf16be[mask]); 29| | 30| 5.89k| const size_t u16count = (4 + count_ones(mask)); 31| 5.89k| const auto compressed = shuffle.lookup_16(merged); 32| | 33| 5.89k| return {u16count, compressed}; 34| 5.89k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf32_to_utf16_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 134| 674| char16_t *utf16_output) { 135| 674| const char32_t *start = buf; 136| 674| const char32_t *end = buf + len; 137| | 138| 674| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 139| | 140| 68.6k| while (end - buf >= 8) { ------------------ | Branch (140:10): [True: 68.3k, False: 316] ------------------ 141| 68.3k| const __m128i in = _mm_loadu_si128((__m128i *)buf); 142| 68.3k| const __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 143| | 144| 68.3k| const __m128i combined = _mm_or_si128(in, nextin); 145| 68.3k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 68.3k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 66.6k, False: 1.68k] | | ------------------ ------------------ 146| | // No bits set above 16th, directly pack UTF-32 to UTF-16 147| 66.6k| __m128i utf16_packed = _mm_packus_epi32(in, nextin); 148| | 149| 66.6k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 150| 66.6k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 151| 66.6k| const __m128i forbidden_bytemask = 152| 66.6k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed, v_f800), v_d800); 153| 66.6k| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (153:11): [True: 46, False: 66.5k] ------------------ 154| 46| return std::make_pair(result(error_code::SURROGATE, buf - start), 155| 46| utf16_output); 156| 46| } 157| | 158| 66.5k| if (big_endian) { ------------------ | Branch (158:11): [Folded, False: 66.5k] ------------------ 159| 0| const __m128i swap = 160| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 161| 0| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 162| 0| } 163| | 164| 66.5k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 165| 66.5k| utf16_output += 8; 166| 66.5k| buf += 8; 167| 66.5k| } else { 168| 1.68k| size_t forward = 7; 169| 1.68k| size_t k = 0; 170| 1.68k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (170:11): [True: 0, False: 1.68k] ------------------ 171| 0| forward = size_t(end - buf - 1); 172| 0| } 173| 11.8k| for (; k < forward; k++) { ------------------ | Branch (173:14): [True: 10.5k, False: 1.36k] ------------------ 174| 10.5k| uint32_t word = buf[k]; 175| 10.5k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (175:13): [True: 4.93k, False: 5.56k] ------------------ 176| | // will not generate a surrogate pair 177| 4.93k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (177:15): [True: 720, False: 4.21k] | Branch (177:33): [True: 14, False: 706] ------------------ 178| 14| return std::make_pair( 179| 14| result(error_code::SURROGATE, buf - start + k), utf16_output); 180| 14| } 181| 4.92k| *utf16_output++ = 182| 4.92k| big_endian ------------------ | Branch (182:15): [Folded, False: 4.92k] ------------------ 183| 4.92k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 184| 4.92k| : char16_t(word); 185| 5.56k| } else { 186| | // will generate a surrogate pair 187| 5.56k| if (word > 0x10FFFF) { ------------------ | Branch (187:15): [True: 298, False: 5.27k] ------------------ 188| 298| return std::make_pair( 189| 298| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 190| 298| } 191| 5.27k| word -= 0x10000; 192| 5.27k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 193| 5.27k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 194| 5.27k| if (big_endian) { ------------------ | Branch (194:15): [Folded, False: 5.27k] ------------------ 195| 0| high_surrogate = 196| 0| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 197| 0| low_surrogate = 198| 0| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 199| 0| } 200| 5.27k| *utf16_output++ = char16_t(high_surrogate); 201| 5.27k| *utf16_output++ = char16_t(low_surrogate); 202| 5.27k| } 203| 10.5k| } 204| 1.36k| buf += k; 205| 1.36k| } 206| 68.3k| } 207| | 208| 316| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 209| 674|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf32_to_utf16_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 134| 694| char16_t *utf16_output) { 135| 694| const char32_t *start = buf; 136| 694| const char32_t *end = buf + len; 137| | 138| 694| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 139| | 140| 4.82k| while (end - buf >= 8) { ------------------ | Branch (140:10): [True: 4.50k, False: 316] ------------------ 141| 4.50k| const __m128i in = _mm_loadu_si128((__m128i *)buf); 142| 4.50k| const __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 143| | 144| 4.50k| const __m128i combined = _mm_or_si128(in, nextin); 145| 4.50k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 4.50k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.35k, False: 1.15k] | | ------------------ ------------------ 146| | // No bits set above 16th, directly pack UTF-32 to UTF-16 147| 3.35k| __m128i utf16_packed = _mm_packus_epi32(in, nextin); 148| | 149| 3.35k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 150| 3.35k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 151| 3.35k| const __m128i forbidden_bytemask = 152| 3.35k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed, v_f800), v_d800); 153| 3.35k| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (153:11): [True: 22, False: 3.33k] ------------------ 154| 22| return std::make_pair(result(error_code::SURROGATE, buf - start), 155| 22| utf16_output); 156| 22| } 157| | 158| 3.33k| if (big_endian) { ------------------ | Branch (158:11): [True: 3.33k, Folded] ------------------ 159| 3.33k| const __m128i swap = 160| 3.33k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 161| 3.33k| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 162| 3.33k| } 163| | 164| 3.33k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 165| 3.33k| utf16_output += 8; 166| 3.33k| buf += 8; 167| 3.33k| } else { 168| 1.15k| size_t forward = 7; 169| 1.15k| size_t k = 0; 170| 1.15k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (170:11): [True: 0, False: 1.15k] ------------------ 171| 0| forward = size_t(end - buf - 1); 172| 0| } 173| 7.49k| for (; k < forward; k++) { ------------------ | Branch (173:14): [True: 6.69k, False: 796] ------------------ 174| 6.69k| uint32_t word = buf[k]; 175| 6.69k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (175:13): [True: 3.35k, False: 3.34k] ------------------ 176| | // will not generate a surrogate pair 177| 3.35k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (177:15): [True: 850, False: 2.50k] | Branch (177:33): [True: 40, False: 810] ------------------ 178| 40| return std::make_pair( 179| 40| result(error_code::SURROGATE, buf - start + k), utf16_output); 180| 40| } 181| 3.31k| *utf16_output++ = 182| 3.31k| big_endian ------------------ | Branch (182:15): [True: 3.31k, Folded] ------------------ 183| 3.31k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 184| 3.31k| : char16_t(word); 185| 3.34k| } else { 186| | // will generate a surrogate pair 187| 3.34k| if (word > 0x10FFFF) { ------------------ | Branch (187:15): [True: 316, False: 3.02k] ------------------ 188| 316| return std::make_pair( 189| 316| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 190| 316| } 191| 3.02k| word -= 0x10000; 192| 3.02k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 193| 3.02k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 194| 3.02k| if (big_endian) { ------------------ | Branch (194:15): [True: 3.02k, Folded] ------------------ 195| 3.02k| high_surrogate = 196| 3.02k| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 197| 3.02k| low_surrogate = 198| 3.02k| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 199| 3.02k| } 200| 3.02k| *utf16_output++ = char16_t(high_surrogate); 201| 3.02k| *utf16_output++ = char16_t(low_surrogate); 202| 3.02k| } 203| 6.69k| } 204| 796| buf += k; 205| 796| } 206| 4.50k| } 207| | 208| 316| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 209| 694|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_125sse_convert_utf32_to_utf8EPKDimPc: 2| 864|sse_convert_utf32_to_utf8(const char32_t *buf, size_t len, char *utf8_output) { 3| 864| const char32_t *end = buf + len; 4| | 5| 864| const __m128i v_0000 = _mm_setzero_si128(); //__m128 = 128 bits 6| 864| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); // 1111 1000 0000 7| | // 0000 8| 864| const __m128i v_c080 = _mm_set1_epi16((uint16_t)0xc080); // 1100 0000 1000 9| | // 0000 10| 864| const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80); // 1111 1111 1000 11| | // 0000 12| 864| const __m128i v_ffff0000 = _mm_set1_epi32( 13| 864| (uint32_t)0xffff0000); // 1111 1111 1111 1111 0000 0000 0000 0000 14| 864| const __m128i v_7fffffff = _mm_set1_epi32( 15| 864| (uint32_t)0x7fffffff); // 0111 1111 1111 1111 1111 1111 1111 1111 16| 864| __m128i running_max = _mm_setzero_si128(); 17| 864| __m128i forbidden_bytemask = _mm_setzero_si128(); 18| 864| const size_t safety_margin = 19| 864| 12; // to avoid overruns, see issue 20| | // https://github.com/simdutf/simdutf/issues/92 21| | 22| 40.7k| while (end - buf >= ------------------ | Branch (22:10): [True: 40.0k, False: 672] ------------------ 23| 40.7k| std::ptrdiff_t( 24| 40.7k| 16 + safety_margin)) { // buf is a char32_t pointer, each char32_t 25| | // has 4 bytes or 32 bits, thus buf + 16 * 26| | // char_32t = 512 bits = 64 bytes 27| | // We load two 16 bytes registers for a total of 32 bytes or 16 characters. 28| 40.0k| __m128i in = _mm_loadu_si128((__m128i *)buf); 29| 40.0k| __m128i nextin = _mm_loadu_si128( 30| 40.0k| (__m128i *)buf + 1); // These two values can hold only 8 UTF32 chars 31| 40.0k| running_max = _mm_max_epu32( 32| 40.0k| _mm_max_epu32(in, running_max), // take element-wise max char32_t from 33| | // in and running_max vector 34| 40.0k| nextin); // and take element-wise max element from nextin and 35| | // running_max vector 36| | 37| | // Pack 32-bit UTF-32 code units to 16-bit UTF-16 code units with unsigned 38| | // saturation 39| 40.0k| __m128i in_16 = _mm_packus_epi32( 40| 40.0k| _mm_and_si128(in, v_7fffffff), 41| 40.0k| _mm_and_si128( 42| 40.0k| nextin, 43| 40.0k| v_7fffffff)); // in this context pack the two __m128 into a single 44| | // By ensuring the highest bit is set to 0(&v_7fffffff), we are making sure 45| | // all values are interpreted as non-negative, or specifically, the values 46| | // are within the range of valid Unicode code points. remember : having 47| | // leading byte 0 means a positive number by the two complements system. 48| | // Unicode is well beneath the range where you'll start getting issues so 49| | // that's OK. 50| | 51| | // Try to apply UTF-16 => UTF-8 from ./sse_convert_utf16_to_utf8.cpp 52| | 53| | // Check for ASCII fast path 54| | 55| | // ASCII fast path!!!! 56| | // We eagerly load another 32 bytes, hoping that they will be ASCII too. 57| | // The intuition is that we try to collect 16 ASCII characters which 58| | // requires a total of 64 bytes of input. If we fail, we just pass thirdin 59| | // and fourthin as our new inputs. 60| 40.0k| if (_mm_testz_si128(in_16, v_ff80)) { // if the first two blocks are ASCII ------------------ | Branch (60:9): [True: 35.3k, False: 4.73k] ------------------ 61| 35.3k| __m128i thirdin = _mm_loadu_si128((__m128i *)buf + 2); 62| 35.3k| __m128i fourthin = _mm_loadu_si128((__m128i *)buf + 3); 63| 35.3k| running_max = _mm_max_epu32( 64| 35.3k| _mm_max_epu32(thirdin, running_max), 65| 35.3k| fourthin); // take the running max of all 4 vectors thus far 66| 35.3k| __m128i nextin_16 = _mm_packus_epi32( 67| 35.3k| _mm_and_si128(thirdin, v_7fffffff), 68| 35.3k| _mm_and_si128(fourthin, 69| 35.3k| v_7fffffff)); // pack into 1 vector, now you have two 70| 35.3k| if (!_mm_testz_si128( ------------------ | Branch (70:11): [True: 1.21k, False: 34.1k] ------------------ 71| 35.3k| nextin_16, 72| 35.3k| v_ff80)) { // checks if the second packed vector is ASCII, if not: 73| | // 1. pack the bytes 74| | // obviously suboptimal. 75| 1.21k| const __m128i utf8_packed = _mm_packus_epi16( 76| 1.21k| in_16, in_16); // creates two copy of in_16 in 1 vector 77| | // 2. store (16 bytes) 78| 1.21k| _mm_storeu_si128((__m128i *)utf8_output, 79| 1.21k| utf8_packed); // put them into the output 80| | // 3. adjust pointers 81| 1.21k| buf += 8; // the char32_t buffer pointer goes up 8 char32_t chars* 32 82| | // bits = 256 bits 83| 1.21k| utf8_output += 84| 1.21k| 8; // same with output, e.g. lift the first two blocks alone. 85| | // Proceed with next input 86| 1.21k| in_16 = nextin_16; 87| | // We need to update in and nextin because they are used later. 88| 1.21k| in = thirdin; 89| 1.21k| nextin = fourthin; 90| 34.1k| } else { 91| | // 1. pack the bytes 92| 34.1k| const __m128i utf8_packed = _mm_packus_epi16(in_16, nextin_16); 93| | // 2. store (16 bytes) 94| 34.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 95| | // 3. adjust pointers 96| 34.1k| buf += 16; 97| 34.1k| utf8_output += 16; 98| 34.1k| continue; // we are done for this round! 99| 34.1k| } 100| 35.3k| } 101| | 102| | // no bits set above 7th bit -- find out all the ASCII characters 103| 5.94k| const __m128i one_byte_bytemask = 104| 5.94k| _mm_cmpeq_epi16( // this takes four bytes at a time and compares: 105| 5.94k| _mm_and_si128(in_16, v_ff80), // the vector that get only the first 106| | // 9 bits of each 16-bit/2-byte units 107| 5.94k| v_0000 // 108| 5.94k| ); // they should be all zero if they are ASCII. E.g. ASCII in UTF32 is 109| | // of format 0000 0000 0000 0XXX XXXX 110| | // _mm_cmpeq_epi16 should now return a 1111 1111 1111 1111 for equals, and 111| | // 0000 0000 0000 0000 if not for each 16-bit/2-byte units 112| 5.94k| const uint16_t one_byte_bitmask = static_cast(_mm_movemask_epi8( 113| 5.94k| one_byte_bytemask)); // collect the MSB from previous vector and put 114| | // them into uint16_t mas 115| | 116| | // no bits set above 11th bit 117| 5.94k| const __m128i one_or_two_bytes_bytemask = 118| 5.94k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_0000); 119| 5.94k| const uint16_t one_or_two_bytes_bitmask = 120| 5.94k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 121| | 122| 5.94k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (122:9): [True: 2.75k, False: 3.19k] ------------------ 123| | // case: all code units either produce 1 or 2 UTF-8 bytes (at least one 124| | // produces 2 bytes) 125| | // 1. prepare 2-byte values 126| | // input 16-bit word : [0000|0aaa|aabb|bbbb] x 8 127| | // expected output : [110a|aaaa|10bb|bbbb] x 8 128| 2.75k| const __m128i v_1f00 = 129| 2.75k| _mm_set1_epi16((int16_t)0x1f00); // 0001 1111 0000 0000 130| 2.75k| const __m128i v_003f = 131| 2.75k| _mm_set1_epi16((int16_t)0x003f); // 0000 0000 0011 1111 132| | 133| | // t0 = [000a|aaaa|bbbb|bb00] 134| 2.75k| const __m128i t0 = _mm_slli_epi16(in_16, 2); // shift packed vector by two 135| | // t1 = [000a|aaaa|0000|0000] 136| 2.75k| const __m128i t1 = _mm_and_si128(t0, v_1f00); // potential first utf8 byte 137| | // t2 = [0000|0000|00bb|bbbb] 138| 2.75k| const __m128i t2 = 139| 2.75k| _mm_and_si128(in_16, v_003f); // potential second utf8 byte 140| | // t3 = [000a|aaaa|00bb|bbbb] 141| 2.75k| const __m128i t3 = 142| 2.75k| _mm_or_si128(t1, t2); // first and second potential utf8 byte together 143| | // t4 = [110a|aaaa|10bb|bbbb] 144| 2.75k| const __m128i t4 = _mm_or_si128( 145| 2.75k| t3, 146| 2.75k| v_c080); // t3 | 1100 0000 1000 0000 = full potential 2-byte utf8 unit 147| | 148| | // 2. merge ASCII and 2-byte codewords 149| 2.75k| const __m128i utf8_unpacked = 150| 2.75k| _mm_blendv_epi8(t4, in_16, one_byte_bytemask); 151| | 152| | // 3. prepare bitmask for 8-bit lookup 153| | // one_byte_bitmask = hhggffeeddccbbaa -- the bits are doubled (h - 154| | // MSB, a - LSB) 155| 2.75k| const uint16_t m0 = one_byte_bitmask & 0x5555; // m0 = 0h0g0f0e0d0c0b0a 156| 2.75k| const uint16_t m1 = 157| 2.75k| static_cast(m0 >> 7); // m1 = 00000000h0g0f0e0 158| 2.75k| const uint8_t m2 = 159| 2.75k| static_cast((m0 | m1) & 0xff); // m2 = hdgcfbea 160| | // 4. pack the bytes 161| 2.75k| const uint8_t *row = 162| 2.75k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[m2][0]; 163| 2.75k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 164| 2.75k| const __m128i utf8_packed = _mm_shuffle_epi8(utf8_unpacked, shuffle); 165| | 166| | // 5. store bytes 167| 2.75k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 168| | 169| | // 6. adjust pointers 170| 2.75k| buf += 8; 171| 2.75k| utf8_output += row[0]; 172| 2.75k| continue; 173| 2.75k| } 174| | 175| | // Check for overflow in packing 176| | 177| 3.19k| const __m128i saturation_bytemask = _mm_cmpeq_epi32( 178| 3.19k| _mm_and_si128(_mm_or_si128(in, nextin), v_ffff0000), v_0000); 179| 3.19k| const uint32_t saturation_bitmask = 180| 3.19k| static_cast(_mm_movemask_epi8(saturation_bytemask)); 181| 3.19k| if (saturation_bitmask == 0xffff) { ------------------ | Branch (181:9): [True: 1.54k, False: 1.64k] ------------------ 182| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 183| 1.54k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 184| 1.54k| forbidden_bytemask = 185| 1.54k| _mm_or_si128(forbidden_bytemask, 186| 1.54k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_d800)); 187| | 188| 1.54k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 189| 1.54k| 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 190| | 191| | /* In this branch we handle three cases: 192| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 193| | single UFT-8 byte 194| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - 195| | two UTF-8 bytes 196| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 197| | three UTF-8 bytes 198| | 199| | We expand the input word (16-bit) into two code units (32-bit), thus 200| | we have room for four bytes. However, we need five distinct bit 201| | layouts. Note that the last byte in cases #2 and #3 is the same. 202| | 203| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 204| | in register t2. 205| | 206| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 207| | either byte 1 for case #2 or byte 2 for case #3. Note that they 208| | differ by exactly one bit. 209| | 210| | Finally from these two code units we build proper UTF-8 sequence, taking 211| | into account the case (i.e, the number of bytes to write). 212| | */ 213| | /** 214| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 215| | * t2 => [0ccc|cccc] [10cc|cccc] 216| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 217| | */ 218| 1.54k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 219| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 220| 1.54k| const __m128i t0 = _mm_shuffle_epi8(in_16, dup_even); 221| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 222| 1.54k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 218| 1.54k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 223| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 224| 1.54k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 218| 1.54k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 225| | 226| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 227| 1.54k| const __m128i s0 = _mm_srli_epi16(in_16, 4); 228| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 229| 1.54k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 218| 1.54k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 230| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 231| 1.54k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 218| 1.54k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 232| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 233| 1.54k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 218| 1.54k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 234| 1.54k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 235| 1.54k| simdutf_vec(0b0100000000000000)); ------------------ | | 218| 1.54k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 236| 1.54k| const __m128i s4 = _mm_xor_si128(s3, m0); 237| 1.54k|#undef simdutf_vec 238| | 239| | // 4. expand code units 16-bit => 32-bit 240| 1.54k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 241| 1.54k| const __m128i out1 = _mm_unpackhi_epi16(t2, s4); 242| | 243| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 244| 1.54k| const uint16_t mask = 245| 1.54k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 246| 1.54k| if (mask == 0) { ------------------ | Branch (246:11): [True: 588, False: 960] ------------------ 247| | // We only have three-byte code units. Use fast path. 248| 588| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 249| 588| 15, 13, -1, -1, -1, -1); 250| 588| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 251| 588| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 252| 588| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 253| 588| utf8_output += 12; 254| 588| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 255| 588| utf8_output += 12; 256| 588| buf += 8; 257| 588| continue; 258| 588| } 259| 960| const uint8_t mask0 = uint8_t(mask); 260| | 261| 960| const uint8_t *row0 = 262| 960| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 263| 960| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 264| 960| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 265| | 266| 960| const uint8_t mask1 = static_cast(mask >> 8); 267| | 268| 960| const uint8_t *row1 = 269| 960| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 270| 960| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 271| 960| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 272| | 273| 960| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 274| 960| utf8_output += row0[0]; 275| 960| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 276| 960| utf8_output += row1[0]; 277| | 278| 960| buf += 8; 279| 1.64k| } else { 280| | // case: at least one 32-bit word produce a surrogate pair in UTF-16 <=> 281| | // will produce four UTF-8 bytes Let us do a scalar fallback. It may seem 282| | // wasteful to use scalar code, but being efficient with SIMD in the 283| | // presence of surrogate pairs may require non-trivial tables. 284| 1.64k| size_t forward = 15; 285| 1.64k| size_t k = 0; 286| 1.64k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (286:11): [True: 0, False: 1.64k] ------------------ 287| 0| forward = size_t(end - buf - 1); 288| 0| } 289| 24.0k| for (; k < forward; k++) { ------------------ | Branch (289:14): [True: 22.5k, False: 1.45k] ------------------ 290| 22.5k| uint32_t word = buf[k]; 291| 22.5k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (291:13): [True: 14.8k, False: 7.77k] ------------------ 292| 14.8k| *utf8_output++ = char(word); 293| 14.8k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (293:20): [True: 1.13k, False: 6.64k] ------------------ 294| 1.13k| *utf8_output++ = char((word >> 6) | 0b11000000); 295| 1.13k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 296| 6.64k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (296:20): [True: 2.55k, False: 4.09k] ------------------ 297| 2.55k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (297:15): [True: 712, False: 1.84k] | Branch (297:33): [True: 22, False: 690] ------------------ 298| 22| return std::make_pair(nullptr, utf8_output); 299| 22| } 300| 2.53k| *utf8_output++ = char((word >> 12) | 0b11100000); 301| 2.53k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 302| 2.53k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 303| 4.09k| } else { 304| 4.09k| if (word > 0x10FFFF) { ------------------ | Branch (304:15): [True: 170, False: 3.92k] ------------------ 305| 170| return std::make_pair(nullptr, utf8_output); 306| 170| } 307| 3.92k| *utf8_output++ = char((word >> 18) | 0b11110000); 308| 3.92k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 309| 3.92k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 310| 3.92k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 311| 3.92k| } 312| 22.5k| } 313| 1.45k| buf += k; 314| 1.45k| } 315| 3.19k| } // while 316| | 317| | // check for invalid input 318| 672| const __m128i v_10ffff = _mm_set1_epi32((uint32_t)0x10ffff); 319| 672| if (static_cast(_mm_movemask_epi8(_mm_cmpeq_epi32( ------------------ | Branch (319:7): [True: 16, False: 656] ------------------ 320| 672| _mm_max_epu32(running_max, v_10ffff), v_10ffff))) != 0xffff) { 321| 16| return std::make_pair(nullptr, utf8_output); 322| 16| } 323| | 324| 656| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (324:7): [True: 48, False: 608] ------------------ 325| 48| return std::make_pair(nullptr, utf8_output); 326| 48| } 327| | 328| 608| return std::make_pair(buf, utf8_output); 329| 656|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_137sse_convert_utf32_to_utf8_with_errorsEPKDimPc: 333| 832| char *utf8_output) { 334| 832| const char32_t *end = buf + len; 335| 832| const char32_t *start = buf; 336| | 337| 832| const __m128i v_0000 = _mm_setzero_si128(); 338| 832| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 339| 832| const __m128i v_c080 = _mm_set1_epi16((uint16_t)0xc080); 340| 832| const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80); 341| 832| const __m128i v_ffff0000 = _mm_set1_epi32((uint32_t)0xffff0000); 342| 832| const __m128i v_7fffffff = _mm_set1_epi32((uint32_t)0x7fffffff); 343| 832| const __m128i v_10ffff = _mm_set1_epi32((uint32_t)0x10ffff); 344| | 345| 832| const size_t safety_margin = 346| 832| 12; // to avoid overruns, see issue 347| | // https://github.com/simdutf/simdutf/issues/92 348| | 349| 46.7k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (349:10): [True: 46.0k, False: 658] ------------------ 350| | // We load two 16 bytes registers for a total of 32 bytes or 8 characters. 351| 46.0k| __m128i in = _mm_loadu_si128((__m128i *)buf); 352| 46.0k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 353| | // Check for too large input 354| 46.0k| __m128i max_input = _mm_max_epu32(_mm_max_epu32(in, nextin), v_10ffff); 355| 46.0k| if (static_cast(_mm_movemask_epi8( ------------------ | Branch (355:9): [True: 50, False: 45.9k] ------------------ 356| 46.0k| _mm_cmpeq_epi32(max_input, v_10ffff))) != 0xffff) { 357| 50| return std::make_pair(result(error_code::TOO_LARGE, buf - start), 358| 50| utf8_output); 359| 50| } 360| | 361| | // Pack 32-bit UTF-32 code units to 16-bit UTF-16 code units with unsigned 362| | // saturation 363| 45.9k| __m128i in_16 = _mm_packus_epi32(_mm_and_si128(in, v_7fffffff), 364| 45.9k| _mm_and_si128(nextin, v_7fffffff)); 365| | 366| | // Try to apply UTF-16 => UTF-8 from ./sse_convert_utf16_to_utf8.cpp 367| | 368| | // Check for ASCII fast path 369| 45.9k| if (_mm_testz_si128(in_16, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (369:9): [True: 41.1k, False: 4.88k] ------------------ 370| | // 1. pack the bytes 371| | // obviously suboptimal. 372| 41.1k| const __m128i utf8_packed = _mm_packus_epi16(in_16, in_16); 373| | // 2. store (16 bytes) 374| 41.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 375| | // 3. adjust pointers 376| 41.1k| buf += 8; 377| 41.1k| utf8_output += 8; 378| 41.1k| continue; 379| 41.1k| } 380| | 381| | // no bits set above 7th bit 382| 4.88k| const __m128i one_byte_bytemask = 383| 4.88k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_ff80), v_0000); 384| 4.88k| const uint16_t one_byte_bitmask = 385| 4.88k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 386| | 387| | // no bits set above 11th bit 388| 4.88k| const __m128i one_or_two_bytes_bytemask = 389| 4.88k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_0000); 390| 4.88k| const uint16_t one_or_two_bytes_bitmask = 391| 4.88k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 392| | 393| 4.88k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (393:9): [True: 2.88k, False: 1.99k] ------------------ 394| | // case: all code units either produce 1 or 2 UTF-8 bytes (at least one 395| | // produces 2 bytes) 396| | // 1. prepare 2-byte values 397| | // input 16-bit word : [0000|0aaa|aabb|bbbb] x 8 398| | // expected output : [110a|aaaa|10bb|bbbb] x 8 399| 2.88k| const __m128i v_1f00 = _mm_set1_epi16((int16_t)0x1f00); 400| 2.88k| const __m128i v_003f = _mm_set1_epi16((int16_t)0x003f); 401| | 402| | // t0 = [000a|aaaa|bbbb|bb00] 403| 2.88k| const __m128i t0 = _mm_slli_epi16(in_16, 2); 404| | // t1 = [000a|aaaa|0000|0000] 405| 2.88k| const __m128i t1 = _mm_and_si128(t0, v_1f00); 406| | // t2 = [0000|0000|00bb|bbbb] 407| 2.88k| const __m128i t2 = _mm_and_si128(in_16, v_003f); 408| | // t3 = [000a|aaaa|00bb|bbbb] 409| 2.88k| const __m128i t3 = _mm_or_si128(t1, t2); 410| | // t4 = [110a|aaaa|10bb|bbbb] 411| 2.88k| const __m128i t4 = _mm_or_si128(t3, v_c080); 412| | 413| | // 2. merge ASCII and 2-byte codewords 414| 2.88k| const __m128i utf8_unpacked = 415| 2.88k| _mm_blendv_epi8(t4, in_16, one_byte_bytemask); 416| | 417| | // 3. prepare bitmask for 8-bit lookup 418| | // one_byte_bitmask = hhggffeeddccbbaa -- the bits are doubled (h - 419| | // MSB, a - LSB) 420| 2.88k| const uint16_t m0 = one_byte_bitmask & 0x5555; // m0 = 0h0g0f0e0d0c0b0a 421| 2.88k| const uint16_t m1 = 422| 2.88k| static_cast(m0 >> 7); // m1 = 00000000h0g0f0e0 423| 2.88k| const uint8_t m2 = 424| 2.88k| static_cast((m0 | m1) & 0xff); // m2 = hdgcfbea 425| | // 4. pack the bytes 426| 2.88k| const uint8_t *row = 427| 2.88k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[m2][0]; 428| 2.88k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 429| 2.88k| const __m128i utf8_packed = _mm_shuffle_epi8(utf8_unpacked, shuffle); 430| | 431| | // 5. store bytes 432| 2.88k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 433| | 434| | // 6. adjust pointers 435| 2.88k| buf += 8; 436| 2.88k| utf8_output += row[0]; 437| 2.88k| continue; 438| 2.88k| } 439| | 440| | // Check for overflow in packing 441| 1.99k| const __m128i saturation_bytemask = _mm_cmpeq_epi32( 442| 1.99k| _mm_and_si128(_mm_or_si128(in, nextin), v_ffff0000), v_0000); 443| 1.99k| const uint32_t saturation_bitmask = 444| 1.99k| static_cast(_mm_movemask_epi8(saturation_bytemask)); 445| | 446| 1.99k| if (saturation_bitmask == 0xffff) { ------------------ | Branch (446:9): [True: 1.15k, False: 844] ------------------ 447| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 448| | 449| | // Check for illegal surrogate code units 450| 1.15k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 451| 1.15k| const __m128i forbidden_bytemask = 452| 1.15k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_d800); 453| 1.15k| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (453:11): [True: 18, False: 1.13k] ------------------ 454| 18| return std::make_pair(result(error_code::SURROGATE, buf - start), 455| 18| utf8_output); 456| 18| } 457| | 458| 1.13k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 459| 1.13k| 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 460| | 461| | /* In this branch we handle three cases: 462| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 463| | single UFT-8 byte 464| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - 465| | two UTF-8 bytes 466| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 467| | three UTF-8 bytes 468| | 469| | We expand the input word (16-bit) into two code units (32-bit), thus 470| | we have room for four bytes. However, we need five distinct bit 471| | layouts. Note that the last byte in cases #2 and #3 is the same. 472| | 473| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 474| | in register t2. 475| | 476| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 477| | either byte 1 for case #2 or byte 2 for case #3. Note that they 478| | differ by exactly one bit. 479| | 480| | Finally from these two code units we build proper UTF-8 sequence, taking 481| | into account the case (i.e, the number of bytes to write). 482| | */ 483| | /** 484| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 485| | * t2 => [0ccc|cccc] [10cc|cccc] 486| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 487| | */ 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 489| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 490| 1.13k| const __m128i t0 = _mm_shuffle_epi8(in_16, dup_even); 491| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 492| 1.13k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 493| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 494| 1.13k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 495| | 496| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 497| 1.13k| const __m128i s0 = _mm_srli_epi16(in_16, 4); 498| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 499| 1.13k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 500| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 501| 1.13k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 502| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 503| 1.13k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 504| 1.13k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 505| 1.13k| simdutf_vec(0b0100000000000000)); ------------------ | | 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 506| 1.13k| const __m128i s4 = _mm_xor_si128(s3, m0); 507| 1.13k|#undef simdutf_vec 508| | 509| | // 4. expand code units 16-bit => 32-bit 510| 1.13k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 511| 1.13k| const __m128i out1 = _mm_unpackhi_epi16(t2, s4); 512| | 513| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 514| 1.13k| const uint16_t mask = 515| 1.13k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 516| 1.13k| if (mask == 0) { ------------------ | Branch (516:11): [True: 576, False: 558] ------------------ 517| | // We only have three-byte code units. Use fast path. 518| 576| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 519| 576| 15, 13, -1, -1, -1, -1); 520| 576| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 521| 576| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 522| 576| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 523| 576| utf8_output += 12; 524| 576| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 525| 576| utf8_output += 12; 526| 576| buf += 8; 527| 576| continue; 528| 576| } 529| 558| const uint8_t mask0 = uint8_t(mask); 530| | 531| 558| const uint8_t *row0 = 532| 558| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 533| 558| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 534| 558| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 535| | 536| 558| const uint8_t mask1 = static_cast(mask >> 8); 537| | 538| 558| const uint8_t *row1 = 539| 558| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 540| 558| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 541| 558| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 542| | 543| 558| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 544| 558| utf8_output += row0[0]; 545| 558| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 546| 558| utf8_output += row1[0]; 547| | 548| 558| buf += 8; 549| 844| } else { 550| | // case: at least one 32-bit word produce a surrogate pair in UTF-16 <=> 551| | // will produce four UTF-8 bytes Let us do a scalar fallback. It may seem 552| | // wasteful to use scalar code, but being efficient with SIMD in the 553| | // presence of surrogate pairs may require non-trivial tables. 554| 844| size_t forward = 15; 555| 844| size_t k = 0; 556| 844| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (556:11): [True: 0, False: 844] ------------------ 557| 0| forward = size_t(end - buf - 1); 558| 0| } 559| 12.8k| for (; k < forward; k++) { ------------------ | Branch (559:14): [True: 12.0k, False: 738] ------------------ 560| 12.0k| uint32_t word = buf[k]; 561| 12.0k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (561:13): [True: 4.47k, False: 7.61k] ------------------ 562| 4.47k| *utf8_output++ = char(word); 563| 7.61k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (563:20): [True: 920, False: 6.69k] ------------------ 564| 920| *utf8_output++ = char((word >> 6) | 0b11000000); 565| 920| *utf8_output++ = char((word & 0b111111) | 0b10000000); 566| 6.69k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (566:20): [True: 3.39k, False: 3.30k] ------------------ 567| 3.39k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (567:15): [True: 1.93k, False: 1.45k] | Branch (567:33): [True: 38, False: 1.89k] ------------------ 568| 38| return std::make_pair( 569| 38| result(error_code::SURROGATE, buf - start + k), utf8_output); 570| 38| } 571| 3.35k| *utf8_output++ = char((word >> 12) | 0b11100000); 572| 3.35k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 573| 3.35k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 574| 3.35k| } else { 575| 3.30k| if (word > 0x10FFFF) { ------------------ | Branch (575:15): [True: 68, False: 3.23k] ------------------ 576| 68| return std::make_pair( 577| 68| result(error_code::TOO_LARGE, buf - start + k), utf8_output); 578| 68| } 579| 3.23k| *utf8_output++ = char((word >> 18) | 0b11110000); 580| 3.23k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 581| 3.23k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 582| 3.23k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 583| 3.23k| } 584| 12.0k| } 585| 738| buf += k; 586| 738| } 587| 1.99k| } // while 588| 658| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 589| 832|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_129convert_masked_utf8_to_latin1EPKcmRPc: 9| 588k| char *&latin1_output) { 10| | // we use an approach where we try to process up to 12 input bytes. 11| | // Why 12 input bytes and not 16? Because we are concerned with the size of 12| | // the lookup tables. Also 12 is nicely divisible by two and three. 13| | // 14| | // 15| | // Optimization note: our main path below is load-latency dependent. Thus it 16| | // is maybe beneficial to have fast paths that depend on branch prediction but 17| | // have less latency. This results in more instructions but, potentially, also 18| | // higher speeds. 19| | // 20| 588k| const __m128i in = _mm_loadu_si128((__m128i *)input); 21| 588k| const uint16_t input_utf8_end_of_code_point_mask = 22| 588k| utf8_end_of_code_point_mask & 23| 588k| 0xfff; // we are only processing 12 bytes in case it is not all ASCII 24| 588k| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (24:7): [True: 3.61k, False: 585k] ------------------ 25| | // We process the data in chunks of 12 bytes. 26| 3.61k| _mm_storeu_si128(reinterpret_cast<__m128i *>(latin1_output), in); 27| 3.61k| latin1_output += 12; // We wrote 12 characters. 28| 3.61k| return 12; // We consumed 12 bytes. 29| 3.61k| } 30| | /// We do not have a fast path available, so we fallback. 31| 585k| const uint8_t idx = 32| 585k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 33| 585k| const uint8_t consumed = 34| 585k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 35| | // this indicates an invalid input: 36| 585k| if (idx >= 64) { ------------------ | Branch (36:7): [True: 30.3k, False: 554k] ------------------ 37| 30.3k| return consumed; 38| 30.3k| } 39| | // Here we should have (idx < 64), if not, there is a bug in the validation or 40| | // elsewhere. SIX (6) input code-code units this is a relatively easy scenario 41| | // we process SIX (6) input code-code units. The max length in bytes of six 42| | // code code units spanning between 1 and 2 bytes each is 12 bytes. On 43| | // processors where pdep/pext is fast, we might be able to use a small lookup 44| | // table. 45| 554k| const __m128i sh = 46| 554k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 47| 554k| const __m128i perm = _mm_shuffle_epi8(in, sh); 48| 554k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 49| 554k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 50| 554k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 51| 554k| const __m128i latin1_packed = _mm_packus_epi16(composed, composed); 52| | // writing 8 bytes even though we only care about the first 6 bytes. 53| | // performance note: it would be faster to use _mm_storeu_si128, we should 54| | // investigate. 55| 554k| _mm_storel_epi64((__m128i *)latin1_output, latin1_packed); 56| 554k| latin1_output += 6; // We wrote 6 bytes. 57| 554k| return consumed; 58| 585k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE0EEEmPKcmRPDs: 10| 3.17M| char16_t *&utf16_output) { 11| | // we use an approach where we try to process up to 12 input bytes. 12| | // Why 12 input bytes and not 16? Because we are concerned with the size of 13| | // the lookup tables. Also 12 is nicely divisible by two and three. 14| | // 15| | // 16| | // Optimization note: our main path below is load-latency dependent. Thus it 17| | // is maybe beneficial to have fast paths that depend on branch prediction but 18| | // have less latency. This results in more instructions but, potentially, also 19| | // higher speeds. 20| | // 21| | // We first try a few fast paths. 22| 3.17M| const __m128i swap = 23| 3.17M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 3.17M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 3.17M| const uint16_t input_utf8_end_of_code_point_mask = 26| 3.17M| utf8_end_of_code_point_mask & 0xfff; 27| 3.17M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 10.8k, False: 3.16M] ------------------ 28| | // We process the data in chunks of 12 bytes. 29| | // Note: using 16 bytes is unsafe, see issue_ossfuzz_71218 30| 10.8k| __m128i ascii_first = _mm_cvtepu8_epi16(in); 31| 10.8k| __m128i ascii_second = _mm_cvtepu8_epi16(_mm_srli_si128(in, 8)); 32| 10.8k| if (big_endian) { ------------------ | Branch (32:9): [Folded, False: 10.8k] ------------------ 33| 0| ascii_first = _mm_shuffle_epi8(ascii_first, swap); 34| 0| ascii_second = _mm_shuffle_epi8(ascii_second, swap); 35| 0| } 36| 10.8k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf16_output), ascii_first); 37| 10.8k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf16_output + 8), 38| 10.8k| ascii_second); 39| 10.8k| utf16_output += 12; // We wrote 12 16-bit characters. 40| 10.8k| return 12; // We consumed 12 bytes. 41| 10.8k| } 42| 3.16M| if (((utf8_end_of_code_point_mask & 0xFFFF) == 0xaaaa)) { ------------------ | Branch (42:7): [True: 15.9k, False: 3.15M] ------------------ 43| | // We want to take 8 2-byte UTF-8 code units and turn them into 8 2-byte 44| | // UTF-16 code units. There is probably a more efficient sequence, but the 45| | // following might do. 46| 15.9k| const __m128i sh = 47| 15.9k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 48| 15.9k| const __m128i perm = _mm_shuffle_epi8(in, sh); 49| 15.9k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 50| 15.9k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 51| 15.9k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 52| 15.9k| if (big_endian) ------------------ | Branch (52:9): [Folded, False: 15.9k] ------------------ 53| 0| composed = _mm_shuffle_epi8(composed, swap); 54| 15.9k| _mm_storeu_si128((__m128i *)utf16_output, composed); 55| 15.9k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 56| 15.9k| return 16; 57| 15.9k| } 58| 3.15M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (58:7): [True: 2.28k, False: 3.14M] ------------------ 59| | // We want to take 4 3-byte UTF-8 code units and turn them into 4 2-byte 60| | // UTF-16 code units. There is probably a more efficient sequence, but the 61| | // following might do. 62| 2.28k| const __m128i sh = 63| 2.28k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 64| 2.28k| const __m128i perm = _mm_shuffle_epi8(in, sh); 65| 2.28k| const __m128i ascii = 66| 2.28k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 67| 2.28k| const __m128i middlebyte = 68| 2.28k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 69| 2.28k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 70| 2.28k| const __m128i highbyte = 71| 2.28k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 72| 2.28k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 73| 2.28k| const __m128i composed = 74| 2.28k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 75| 2.28k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 76| 2.28k| if (big_endian) ------------------ | Branch (76:9): [Folded, False: 2.28k] ------------------ 77| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 78| 2.28k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 79| 2.28k| utf16_output += 4; 80| 2.28k| return 12; 81| 2.28k| } 82| | /// We do not have a fast path available, so we fallback. 83| | 84| 3.14M| const uint8_t idx = 85| 3.14M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 86| 3.14M| const uint8_t consumed = 87| 3.14M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 88| 3.14M| if (idx < 64) { ------------------ | Branch (88:7): [True: 2.64M, False: 500k] ------------------ 89| | // SIX (6) input code-code units 90| | // this is a relatively easy scenario 91| | // we process SIX (6) input code-code units. The max length in bytes of six 92| | // code code units spanning between 1 and 2 bytes each is 12 bytes. On 93| | // processors where pdep/pext is fast, we might be able to use a small 94| | // lookup table. 95| 2.64M| const __m128i sh = 96| 2.64M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 97| 2.64M| const __m128i perm = _mm_shuffle_epi8(in, sh); 98| 2.64M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 99| 2.64M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 100| 2.64M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 101| 2.64M| if (big_endian) ------------------ | Branch (101:9): [Folded, False: 2.64M] ------------------ 102| 0| composed = _mm_shuffle_epi8(composed, swap); 103| 2.64M| _mm_storeu_si128((__m128i *)utf16_output, composed); 104| 2.64M| utf16_output += 6; // We wrote 12 bytes, 6 code points. 105| 2.64M| } else if (idx < 145) { ------------------ | Branch (105:14): [True: 434k, False: 66.2k] ------------------ 106| | // FOUR (4) input code-code units 107| 434k| const __m128i sh = 108| 434k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 109| 434k| const __m128i perm = _mm_shuffle_epi8(in, sh); 110| 434k| const __m128i ascii = 111| 434k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 112| 434k| const __m128i middlebyte = 113| 434k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 114| 434k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 115| 434k| const __m128i highbyte = 116| 434k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 117| 434k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 118| 434k| const __m128i composed = 119| 434k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 120| 434k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 121| 434k| if (big_endian) ------------------ | Branch (121:9): [Folded, False: 434k] ------------------ 122| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 123| 434k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 124| 434k| utf16_output += 4; 125| 434k| } else if (idx < 209) { ------------------ | Branch (125:14): [True: 65.5k, False: 712] ------------------ 126| | // TWO (2) input code-code units 127| | ////////////// 128| | // There might be garbage inputs where a leading byte mascarades as a 129| | // four-byte leading byte (by being followed by 3 continuation byte), but is 130| | // not greater than 0xf0. This could trigger a buffer overflow if we only 131| | // counted leading bytes of the form 0xf0 as generating surrogate pairs, 132| | // without further UTF-8 validation. Thus we must be careful to ensure that 133| | // only leading bytes at least as large as 0xf0 generate surrogate pairs. We 134| | // do as at the cost of an extra mask. 135| | ///////////// 136| 65.5k| const __m128i sh = 137| 65.5k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 138| 65.5k| const __m128i perm = _mm_shuffle_epi8(in, sh); 139| 65.5k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 140| 65.5k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 141| 65.5k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 142| 65.5k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 143| | // correct for spurious high bit 144| 65.5k| const __m128i correct = 145| 65.5k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 146| 65.5k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 147| 65.5k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 148| | // We deliberately carry the leading four bits in highbyte if they are 149| | // present, we remove them later when computing hightenbits. 150| 65.5k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 151| 65.5k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 152| | // When we need to generate a surrogate pair (leading byte > 0xF0), then 153| | // the corresponding 32-bit value in 'composed' will be greater than 154| | // > (0xff00000>>6) or > 0x3c00000. This can be used later to identify the 155| | // location of the surrogate pairs. 156| 65.5k| const __m128i composed = 157| 65.5k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 158| 65.5k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 159| 65.5k| const __m128i composedminus = 160| 65.5k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 161| 65.5k| const __m128i lowtenbits = 162| 65.5k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 163| | // Notice the 0x3ff mask: 164| 65.5k| const __m128i hightenbits = 165| 65.5k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 166| 65.5k| const __m128i lowtenbitsadd = 167| 65.5k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 168| 65.5k| const __m128i hightenbitsadd = 169| 65.5k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 170| 65.5k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 171| 65.5k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 172| 65.5k| uint32_t basic_buffer[4]; 173| 65.5k| uint32_t basic_buffer_swap[4]; 174| 65.5k| if (big_endian) { ------------------ | Branch (174:9): [Folded, False: 65.5k] ------------------ 175| 0| _mm_storeu_si128((__m128i *)basic_buffer_swap, 176| 0| _mm_shuffle_epi8(composed, swap)); 177| 0| surrogates = _mm_shuffle_epi8(surrogates, swap); 178| 0| } 179| 65.5k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 180| 65.5k| uint32_t surrogate_buffer[4]; 181| 65.5k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 182| 262k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (182:24): [True: 196k, False: 65.5k] ------------------ 183| 196k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (183:11): [True: 54.1k, False: 142k] ------------------ 184| 54.1k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 185| 54.1k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 186| 54.1k| utf16_output += 2; 187| 142k| } else { 188| 142k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (188:27): [Folded, False: 142k] ------------------ 189| 142k| : uint16_t(basic_buffer[i]); 190| 142k| utf16_output++; 191| 142k| } 192| 196k| } 193| 65.5k| } else { 194| | // here we know that there is an error but we do not handle errors 195| 712| } 196| 3.14M| return consumed; 197| 3.15M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE1EEEmPKcmRPDs: 10| 3.00M| char16_t *&utf16_output) { 11| | // we use an approach where we try to process up to 12 input bytes. 12| | // Why 12 input bytes and not 16? Because we are concerned with the size of 13| | // the lookup tables. Also 12 is nicely divisible by two and three. 14| | // 15| | // 16| | // Optimization note: our main path below is load-latency dependent. Thus it 17| | // is maybe beneficial to have fast paths that depend on branch prediction but 18| | // have less latency. This results in more instructions but, potentially, also 19| | // higher speeds. 20| | // 21| | // We first try a few fast paths. 22| 3.00M| const __m128i swap = 23| 3.00M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 3.00M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 3.00M| const uint16_t input_utf8_end_of_code_point_mask = 26| 3.00M| utf8_end_of_code_point_mask & 0xfff; 27| 3.00M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 9.13k, False: 2.99M] ------------------ 28| | // We process the data in chunks of 12 bytes. 29| | // Note: using 16 bytes is unsafe, see issue_ossfuzz_71218 30| 9.13k| __m128i ascii_first = _mm_cvtepu8_epi16(in); 31| 9.13k| __m128i ascii_second = _mm_cvtepu8_epi16(_mm_srli_si128(in, 8)); 32| 9.13k| if (big_endian) { ------------------ | Branch (32:9): [True: 9.13k, Folded] ------------------ 33| 9.13k| ascii_first = _mm_shuffle_epi8(ascii_first, swap); 34| 9.13k| ascii_second = _mm_shuffle_epi8(ascii_second, swap); 35| 9.13k| } 36| 9.13k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf16_output), ascii_first); 37| 9.13k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf16_output + 8), 38| 9.13k| ascii_second); 39| 9.13k| utf16_output += 12; // We wrote 12 16-bit characters. 40| 9.13k| return 12; // We consumed 12 bytes. 41| 9.13k| } 42| 2.99M| if (((utf8_end_of_code_point_mask & 0xFFFF) == 0xaaaa)) { ------------------ | Branch (42:7): [True: 23.5k, False: 2.96M] ------------------ 43| | // We want to take 8 2-byte UTF-8 code units and turn them into 8 2-byte 44| | // UTF-16 code units. There is probably a more efficient sequence, but the 45| | // following might do. 46| 23.5k| const __m128i sh = 47| 23.5k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 48| 23.5k| const __m128i perm = _mm_shuffle_epi8(in, sh); 49| 23.5k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 50| 23.5k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 51| 23.5k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 52| 23.5k| if (big_endian) ------------------ | Branch (52:9): [True: 23.5k, Folded] ------------------ 53| 23.5k| composed = _mm_shuffle_epi8(composed, swap); 54| 23.5k| _mm_storeu_si128((__m128i *)utf16_output, composed); 55| 23.5k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 56| 23.5k| return 16; 57| 23.5k| } 58| 2.96M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (58:7): [True: 5.17k, False: 2.96M] ------------------ 59| | // We want to take 4 3-byte UTF-8 code units and turn them into 4 2-byte 60| | // UTF-16 code units. There is probably a more efficient sequence, but the 61| | // following might do. 62| 5.17k| const __m128i sh = 63| 5.17k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 64| 5.17k| const __m128i perm = _mm_shuffle_epi8(in, sh); 65| 5.17k| const __m128i ascii = 66| 5.17k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 67| 5.17k| const __m128i middlebyte = 68| 5.17k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 69| 5.17k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 70| 5.17k| const __m128i highbyte = 71| 5.17k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 72| 5.17k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 73| 5.17k| const __m128i composed = 74| 5.17k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 75| 5.17k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 76| 5.17k| if (big_endian) ------------------ | Branch (76:9): [True: 5.17k, Folded] ------------------ 77| 5.17k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 78| 5.17k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 79| 5.17k| utf16_output += 4; 80| 5.17k| return 12; 81| 5.17k| } 82| | /// We do not have a fast path available, so we fallback. 83| | 84| 2.96M| const uint8_t idx = 85| 2.96M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 86| 2.96M| const uint8_t consumed = 87| 2.96M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 88| 2.96M| if (idx < 64) { ------------------ | Branch (88:7): [True: 2.67M, False: 286k] ------------------ 89| | // SIX (6) input code-code units 90| | // this is a relatively easy scenario 91| | // we process SIX (6) input code-code units. The max length in bytes of six 92| | // code code units spanning between 1 and 2 bytes each is 12 bytes. On 93| | // processors where pdep/pext is fast, we might be able to use a small 94| | // lookup table. 95| 2.67M| const __m128i sh = 96| 2.67M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 97| 2.67M| const __m128i perm = _mm_shuffle_epi8(in, sh); 98| 2.67M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 99| 2.67M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 100| 2.67M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 101| 2.67M| if (big_endian) ------------------ | Branch (101:9): [True: 2.67M, Folded] ------------------ 102| 2.67M| composed = _mm_shuffle_epi8(composed, swap); 103| 2.67M| _mm_storeu_si128((__m128i *)utf16_output, composed); 104| 2.67M| utf16_output += 6; // We wrote 12 bytes, 6 code points. 105| 2.67M| } else if (idx < 145) { ------------------ | Branch (105:14): [True: 221k, False: 65.3k] ------------------ 106| | // FOUR (4) input code-code units 107| 221k| const __m128i sh = 108| 221k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 109| 221k| const __m128i perm = _mm_shuffle_epi8(in, sh); 110| 221k| const __m128i ascii = 111| 221k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 112| 221k| const __m128i middlebyte = 113| 221k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 114| 221k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 115| 221k| const __m128i highbyte = 116| 221k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 117| 221k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 118| 221k| const __m128i composed = 119| 221k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 120| 221k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 121| 221k| if (big_endian) ------------------ | Branch (121:9): [True: 221k, Folded] ------------------ 122| 221k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 123| 221k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 124| 221k| utf16_output += 4; 125| 221k| } else if (idx < 209) { ------------------ | Branch (125:14): [True: 64.0k, False: 1.31k] ------------------ 126| | // TWO (2) input code-code units 127| | ////////////// 128| | // There might be garbage inputs where a leading byte mascarades as a 129| | // four-byte leading byte (by being followed by 3 continuation byte), but is 130| | // not greater than 0xf0. This could trigger a buffer overflow if we only 131| | // counted leading bytes of the form 0xf0 as generating surrogate pairs, 132| | // without further UTF-8 validation. Thus we must be careful to ensure that 133| | // only leading bytes at least as large as 0xf0 generate surrogate pairs. We 134| | // do as at the cost of an extra mask. 135| | ///////////// 136| 64.0k| const __m128i sh = 137| 64.0k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 138| 64.0k| const __m128i perm = _mm_shuffle_epi8(in, sh); 139| 64.0k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 140| 64.0k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 141| 64.0k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 142| 64.0k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 143| | // correct for spurious high bit 144| 64.0k| const __m128i correct = 145| 64.0k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 146| 64.0k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 147| 64.0k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 148| | // We deliberately carry the leading four bits in highbyte if they are 149| | // present, we remove them later when computing hightenbits. 150| 64.0k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 151| 64.0k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 152| | // When we need to generate a surrogate pair (leading byte > 0xF0), then 153| | // the corresponding 32-bit value in 'composed' will be greater than 154| | // > (0xff00000>>6) or > 0x3c00000. This can be used later to identify the 155| | // location of the surrogate pairs. 156| 64.0k| const __m128i composed = 157| 64.0k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 158| 64.0k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 159| 64.0k| const __m128i composedminus = 160| 64.0k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 161| 64.0k| const __m128i lowtenbits = 162| 64.0k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 163| | // Notice the 0x3ff mask: 164| 64.0k| const __m128i hightenbits = 165| 64.0k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 166| 64.0k| const __m128i lowtenbitsadd = 167| 64.0k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 168| 64.0k| const __m128i hightenbitsadd = 169| 64.0k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 170| 64.0k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 171| 64.0k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 172| 64.0k| uint32_t basic_buffer[4]; 173| 64.0k| uint32_t basic_buffer_swap[4]; 174| 64.0k| if (big_endian) { ------------------ | Branch (174:9): [True: 64.0k, Folded] ------------------ 175| 64.0k| _mm_storeu_si128((__m128i *)basic_buffer_swap, 176| 64.0k| _mm_shuffle_epi8(composed, swap)); 177| 64.0k| surrogates = _mm_shuffle_epi8(surrogates, swap); 178| 64.0k| } 179| 64.0k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 180| 64.0k| uint32_t surrogate_buffer[4]; 181| 64.0k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 182| 256k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (182:24): [True: 192k, False: 64.0k] ------------------ 183| 192k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (183:11): [True: 53.9k, False: 138k] ------------------ 184| 53.9k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 185| 53.9k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 186| 53.9k| utf16_output += 2; 187| 138k| } else { 188| 138k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (188:27): [True: 138k, Folded] ------------------ 189| 138k| : uint16_t(basic_buffer[i]); 190| 138k| utf16_output++; 191| 138k| } 192| 192k| } 193| 64.0k| } else { 194| | // here we know that there is an error but we do not handle errors 195| 1.31k| } 196| 2.96M| return consumed; 197| 2.96M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_128convert_masked_utf8_to_utf32EPKcmRPDi: 9| 2.58M| char32_t *&utf32_output) { 10| | // we use an approach where we try to process up to 12 input bytes. 11| | // Why 12 input bytes and not 16? Because we are concerned with the size of 12| | // the lookup tables. Also 12 is nicely divisible by two and three. 13| | // 14| | // 15| | // Optimization note: our main path below is load-latency dependent. Thus it 16| | // is maybe beneficial to have fast paths that depend on branch prediction but 17| | // have less latency. This results in more instructions but, potentially, also 18| | // higher speeds. 19| | // 20| | // We first try a few fast paths. 21| 2.58M| const __m128i in = _mm_loadu_si128((__m128i *)input); 22| 2.58M| const uint16_t input_utf8_end_of_code_point_mask = 23| 2.58M| utf8_end_of_code_point_mask & 0xfff; 24| 2.58M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (24:7): [True: 7.57k, False: 2.57M] ------------------ 25| | // We process the data in chunks of 12 bytes. 26| 7.57k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 27| 7.57k| _mm_cvtepu8_epi32(in)); 28| 7.57k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 29| 7.57k| _mm_cvtepu8_epi32(_mm_srli_si128(in, 4))); 30| 7.57k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 8), 31| 7.57k| _mm_cvtepu8_epi32(_mm_srli_si128(in, 8))); 32| 7.57k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 12), 33| 7.57k| _mm_cvtepu8_epi32(_mm_srli_si128(in, 12))); 34| 7.57k| utf32_output += 12; // We wrote 12 32-bit characters. 35| 7.57k| return 12; // We consumed 12 bytes. 36| 7.57k| } 37| 2.57M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (37:7): [True: 17.0k, False: 2.56M] ------------------ 38| | // We want to take 8 2-byte UTF-8 code units and turn them into 8 4-byte 39| | // UTF-32 code units. There is probably a more efficient sequence, but the 40| | // following might do. 41| 17.0k| const __m128i sh = 42| 17.0k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 43| 17.0k| const __m128i perm = _mm_shuffle_epi8(in, sh); 44| 17.0k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 45| 17.0k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 46| 17.0k| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 47| 17.0k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 48| 17.0k| _mm_cvtepu16_epi32(composed)); 49| 17.0k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 50| 17.0k| _mm_cvtepu16_epi32(_mm_srli_si128(composed, 8))); 51| 17.0k| utf32_output += 8; // We wrote 32 bytes, 8 code points. 52| 17.0k| return 16; 53| 17.0k| } 54| 2.56M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (54:7): [True: 2.91k, False: 2.55M] ------------------ 55| | // We want to take 4 3-byte UTF-8 code units and turn them into 4 4-byte 56| | // UTF-32 code units. There is probably a more efficient sequence, but the 57| | // following might do. 58| 2.91k| const __m128i sh = 59| 2.91k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 60| 2.91k| const __m128i perm = _mm_shuffle_epi8(in, sh); 61| 2.91k| const __m128i ascii = 62| 2.91k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 63| 2.91k| const __m128i middlebyte = 64| 2.91k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 65| 2.91k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 66| 2.91k| const __m128i highbyte = 67| 2.91k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 68| 2.91k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 69| 2.91k| const __m128i composed = 70| 2.91k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 71| 2.91k| _mm_storeu_si128((__m128i *)utf32_output, composed); 72| 2.91k| utf32_output += 4; 73| 2.91k| return 12; 74| 2.91k| } 75| | /// We do not have a fast path available, so we fallback. 76| | 77| 2.55M| const uint8_t idx = 78| 2.55M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 79| 2.55M| const uint8_t consumed = 80| 2.55M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 81| 2.55M| if (idx < 64) { ------------------ | Branch (81:7): [True: 2.25M, False: 304k] ------------------ 82| | // SIX (6) input code-code units 83| | // this is a relatively easy scenario 84| | // we process SIX (6) input code-code units. The max length in bytes of six 85| | // code code units spanning between 1 and 2 bytes each is 12 bytes. On 86| | // processors where pdep/pext is fast, we might be able to use a small 87| | // lookup table. 88| 2.25M| const __m128i sh = 89| 2.25M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 90| 2.25M| const __m128i perm = _mm_shuffle_epi8(in, sh); 91| 2.25M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 92| 2.25M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 93| 2.25M| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 94| 2.25M| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 95| 2.25M| _mm_cvtepu16_epi32(composed)); 96| 2.25M| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 97| 2.25M| _mm_cvtepu16_epi32(_mm_srli_si128(composed, 8))); 98| 2.25M| utf32_output += 6; // We wrote 12 bytes, 6 code points. 99| 2.25M| } else if (idx < 145) { ------------------ | Branch (99:14): [True: 253k, False: 51.8k] ------------------ 100| | // FOUR (4) input code-code units 101| 253k| const __m128i sh = 102| 253k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 103| 253k| const __m128i perm = _mm_shuffle_epi8(in, sh); 104| 253k| const __m128i ascii = 105| 253k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 106| 253k| const __m128i middlebyte = 107| 253k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 108| 253k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 109| 253k| const __m128i highbyte = 110| 253k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 111| 253k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 112| 253k| const __m128i composed = 113| 253k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 114| 253k| _mm_storeu_si128((__m128i *)utf32_output, composed); 115| 253k| utf32_output += 4; 116| 253k| } else if (idx < 209) { ------------------ | Branch (116:14): [True: 51.1k, False: 668] ------------------ 117| | // TWO (2) input code-code units 118| 51.1k| const __m128i sh = 119| 51.1k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 120| 51.1k| const __m128i perm = _mm_shuffle_epi8(in, sh); 121| 51.1k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 122| 51.1k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 123| 51.1k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 124| 51.1k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 125| | // correct for spurious high bit 126| 51.1k| const __m128i correct = 127| 51.1k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 128| 51.1k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 129| 51.1k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 130| 51.1k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0x07000000)); 131| 51.1k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 132| 51.1k| const __m128i composed = 133| 51.1k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 134| 51.1k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 135| 51.1k| _mm_storeu_si128((__m128i *)utf32_output, composed); 136| 51.1k| utf32_output += 3; 137| 51.1k| } else { 138| | // here we know that there is an error but we do not handle errors 139| 668| } 140| 2.55M| return consumed; 141| 2.56M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_123utf16_gather_high_bytesILNS_10endiannessE0EEENS1_4simd5simd8IhEENS4_6simd16ItEES8_: 3| 2.54M| const simd16 in1) { 4| 2.54M| if (big_endian) { ------------------ | Branch (4:7): [Folded, False: 2.54M] ------------------ 5| | // we want lower bytes 6| 0| const auto mask = simd16(0x00ff); 7| 0| const auto t0 = in0 & mask; 8| 0| const auto t1 = in1 & mask; 9| | 10| 0| return simd16::pack(t0, t1); 11| 2.54M| } else { 12| 2.54M| const auto t0 = in0.shr<8>(); 13| 2.54M| const auto t1 = in1.shr<8>(); 14| | 15| 2.54M| return simd16::pack(t0, t1); 16| 2.54M| } 17| 2.54M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_123utf16_gather_high_bytesILNS_10endiannessE1EEENS1_4simd5simd8IhEENS4_6simd16ItEES8_: 3| 2.61M| const simd16 in1) { 4| 2.61M| if (big_endian) { ------------------ | Branch (4:7): [True: 2.61M, Folded] ------------------ 5| | // we want lower bytes 6| 2.61M| const auto mask = simd16(0x00ff); 7| 2.61M| const auto t0 = in0 & mask; 8| 2.61M| const auto t1 = in1 & mask; 9| | 10| 2.61M| return simd16::pack(t0, t1); 11| 2.61M| } else { 12| 0| const auto t0 = in0.shr<8>(); 13| 0| const auto t1 = in1.shr<8>(); 14| | 15| 0| return simd16::pack(t0, t1); 16| 0| } 17| 2.61M|}