_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| 9.24k|extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size) { 640| 9.24k| static const auto fptrs = populate_functions(); 641| 9.24k| 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| 9.24k| if (size < 4) { ------------------ | Branch (646:7): [True: 3, False: 9.24k] ------------------ 647| 3| return 0; 648| 3| } 649| | 650| 9.24k| constexpr auto actionmask = std::bit_ceil(Ncases) - 1; 651| 9.24k| const auto action = data[0] & actionmask; 652| 9.24k| data += 4; 653| 9.24k| size -= 4; 654| | 655| 9.24k| if (action >= Ncases) { ------------------ | Branch (655:7): [True: 1, False: 9.24k] ------------------ 656| 1| return 0; 657| 1| } 658| | 659| 9.24k| if constexpr (use_separate_allocation) { 660| | // this is better at exercising null input and catch buffer underflows 661| 9.24k| const std::vector separate{data, data + size}; 662| 9.24k| fptrs[action](std::span(separate)); 663| | } else { 664| | std::span chardata{(const char*)data, size}; 665| | fptrs[action](chardata); 666| | } 667| | 668| 9.24k| return 0; 669| 9.24k|} conversion.cpp:_ZZ18populate_functionsvENK3$_0clENSt3__14spanIKcLm18446744073709551615EEE: 554| 99| +[](std::span chardata) { \ 555| 99| const auto c = \ 556| 99| Conversion{ \ 559| 99| &I::lenfunc, &I::conversionfunc, \ 560| 99| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 99|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 99| ::std::void_t(); \ | | 1180| 99| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 99| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 99| constexpr auto _size = _name.size(); \ | | 1183| 99| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 99| return _nameof; }() ------------------ 561| 99| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 99|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 99| ::std::void_t(); \ | | 1180| 99| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 99| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 99| constexpr auto _size = _name.size(); \ | | 1183| 99| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 99| return _nameof; }() ------------------ 562| 99| c.fuzz(chardata); \ 563| 99| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 209| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 209| const FromSpan from{reinterpret_cast(chardata.data()), 178| 209| chardata.size() / sizeof(FromType)}; 179| | 180| 209| static const bool do_print_testcase = 181| 209| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 209| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 209] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 209| do { 189| | // step 0 - is the input valid? 190| 209| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 209| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 209] | 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| 209| From == UtfEncodings::UTF8) { 198| 209| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 209] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 209| } 201| | 202| | // step 2 - what is the required size of the output? 203| 209| const auto [output_length, length_agree] = 204| 209| calculate_length(from, inputisvalid); 205| 209| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 209] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 209| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 83, False: 126] | Branch (208:28): [True: 13, False: 70] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 13| return; 211| 13| } 212| | 213| | // step 3 - run the conversion 214| 196| const auto [written, outputs_agree] = 215| 196| do_conversion(from, output_length, inputisvalid); 216| 196| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 196] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 196| return; 221| 196| } 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| 209| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 209| validation_result verify_valid_input(FromSpan src) const { 231| 209| validation_result ret{}; 232| | 233| 209| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 209| const auto implementations = get_supported_implementations(); 235| 209| std::vector results; 236| 209| results.reserve(implementations.size()); 237| | 238| 627| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 627, False: 209] ------------------ 239| 627| results.push_back( 240| 627| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 627| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 627| const bool validation2 = 245| 627| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 627| src.data(), src.size()); 247| 627| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 627] ------------------ 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| 627| } 258| | 259| 209| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 209| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 209] ------------------ 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| 209| } else { 273| 209| ret.implementations_agree = true; 274| 209| } 275| 209| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 209| return r.error == simdutf::SUCCESS; 277| 209| }); 278| 209| return ret; 279| 209| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 418| 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| 461| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 461| return r.error == simdutf::SUCCESS; 277| 461| }); _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 209| bool count_the_input(FromSpan src) const { 290| 209| const auto implementations = get_supported_implementations(); 291| 209| std::vector results; 292| 209| results.reserve(implementations.size()); 293| | 294| 627| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 627, False: 209] ------------------ 295| 627| std::size_t ret; 296| 627| if constexpr (From == UtfEncodings::UTF16BE) { 297| 627| 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| 627| results.push_back(ret); 304| 627| } 305| 209| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 209| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 209] ------------------ 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| 209| return true; 321| 209| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 418| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 209| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 209| length_result ret{}; 344| | 345| 209| const auto implementations = get_supported_implementations(); 346| 209| std::vector results; 347| 209| results.reserve(implementations.size()); 348| | 349| 627| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 627, False: 209] ------------------ 350| 627| const auto len = invoke_lengthcalc(impl, src); 351| 627| results.push_back(len); 352| 627| ret.length.push_back(len); 353| 627| } 354| | 355| 209| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 209| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 209] ------------------ 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| 209| } else { 375| 209| ret.implementations_agree = true; 376| 209| } 377| 209| return ret; 378| 209| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 627| FromSpan src) const { 331| 627| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 627| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 418| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 196| const bool inputisvalid) const { 383| 196| conversion_result ret{}; 384| | 385| 196| const auto implementations = get_supported_implementations(); 386| | 387| 196| std::vector> results; 388| 196| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 196| std::vector> outputbuffers; 393| 196| outputbuffers.reserve(implementations.size()); 394| 784| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 588, False: 196] ------------------ 395| 588| auto impl = implementations[i]; 396| 588| const ToType canary1{42}; 397| 588| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 588| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 588| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 588| const auto success = [](const ConversionResult& r) -> bool { 402| 588| if constexpr (std::is_same_v) { 403| 588| return r != 0; 404| 588| } else { 405| 588| return r.error == simdutf::error_code::SUCCESS; 406| 588| } 407| 588| }(implret1); 408| 588| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 588| 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| 588| const ToType canary2{25}; 414| 588| const auto outputbuffer_first_run = outputbuffer; 415| 588| std::ranges::fill(outputbuffer, canary2); 416| 588| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 588| src.size(), outputbuffer.data()); 418| | 419| 588| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 588] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 588| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 378, False: 210] | 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| 588| } 441| 588| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 366, False: 222] ------------------ 442| 588| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 196| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 70, False: 126] ------------------ 447| 210| for (auto& e : results) { ------------------ | Branch (447:20): [True: 210, False: 70] ------------------ 448| 210| e.outputhash.clear(); 449| 210| } 450| 70| } 451| | 452| 196| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 196| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 196] ------------------ 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| 196| } else { 474| 196| ret.implementations_agree = true; 475| 196| } 476| 196| return ret; 477| 196| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 588| const auto success = [](const ConversionResult& r) -> bool { 402| 588| if constexpr (std::is_same_v) { 403| 588| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 588| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 392| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_1clENSt3__14spanIKcLm18446744073709551615EEE: 554| 216| +[](std::span chardata) { \ 555| 216| const auto c = \ 556| 216| Conversion{ \ 559| 216| &I::lenfunc, &I::conversionfunc, \ 560| 216| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 216|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 216| ::std::void_t(); \ | | 1180| 216| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 216| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 216| constexpr auto _size = _name.size(); \ | | 1183| 216| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 216| return _nameof; }() ------------------ 561| 216| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 216|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 216| ::std::void_t(); \ | | 1180| 216| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 216| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 216| constexpr auto _size = _name.size(); \ | | 1183| 216| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 216| return _nameof; }() ------------------ 562| 216| c.fuzz(chardata); \ 563| 216| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 392| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 392| const FromSpan from{reinterpret_cast(chardata.data()), 178| 392| chardata.size() / sizeof(FromType)}; 179| | 180| 392| static const bool do_print_testcase = 181| 392| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 392| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 392] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 392| do { 189| | // step 0 - is the input valid? 190| 392| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 392| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 392] | 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| 392| From == UtfEncodings::UTF8) { 198| 392| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 392] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 392| } 201| | 202| | // step 2 - what is the required size of the output? 203| 392| const auto [output_length, length_agree] = 204| 392| calculate_length(from, inputisvalid); 205| 392| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 392] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 392| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 94, False: 298] | Branch (208:28): [True: 17, False: 77] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 17| return; 211| 17| } 212| | 213| | // step 3 - run the conversion 214| 375| const auto [written, outputs_agree] = 215| 375| do_conversion(from, output_length, inputisvalid); 216| 375| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 375] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 375| return; 221| 375| } 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| 392| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 392| validation_result verify_valid_input(FromSpan src) const { 231| 392| validation_result ret{}; 232| | 233| 392| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 392| const auto implementations = get_supported_implementations(); 235| 392| std::vector results; 236| 392| results.reserve(implementations.size()); 237| | 238| 1.17k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.17k, False: 392] ------------------ 239| 1.17k| results.push_back( 240| 1.17k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.17k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.17k| const bool validation2 = 245| 1.17k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.17k| src.data(), src.size()); 247| 1.17k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.17k] ------------------ 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| 1.17k| } 258| | 259| 392| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 392| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 392] ------------------ 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| 392| } else { 273| 392| ret.implementations_agree = true; 274| 392| } 275| 392| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 392| return r.error == simdutf::SUCCESS; 277| 392| }); 278| 392| return ret; 279| 392| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 784| 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| 988| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 988| return r.error == simdutf::SUCCESS; 277| 988| }); _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 392| bool count_the_input(FromSpan src) const { 290| 392| const auto implementations = get_supported_implementations(); 291| 392| std::vector results; 292| 392| results.reserve(implementations.size()); 293| | 294| 1.17k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.17k, False: 392] ------------------ 295| 1.17k| std::size_t ret; 296| 1.17k| if constexpr (From == UtfEncodings::UTF16BE) { 297| 1.17k| 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| 1.17k| results.push_back(ret); 304| 1.17k| } 305| 392| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 392| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 392] ------------------ 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| 392| return true; 321| 392| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 784| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 392| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 392| length_result ret{}; 344| | 345| 392| const auto implementations = get_supported_implementations(); 346| 392| std::vector results; 347| 392| results.reserve(implementations.size()); 348| | 349| 1.17k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.17k, False: 392] ------------------ 350| 1.17k| const auto len = invoke_lengthcalc(impl, src); 351| 1.17k| results.push_back(len); 352| 1.17k| ret.length.push_back(len); 353| 1.17k| } 354| | 355| 392| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 392| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 392] ------------------ 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| 392| } else { 375| 392| ret.implementations_agree = true; 376| 392| } 377| 392| return ret; 378| 392| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.17k| FromSpan src) const { 331| 1.17k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.17k| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 784| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 375| const bool inputisvalid) const { 383| 375| conversion_result ret{}; 384| | 385| 375| const auto implementations = get_supported_implementations(); 386| | 387| 375| std::vector> results; 388| 375| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 375| std::vector> outputbuffers; 393| 375| outputbuffers.reserve(implementations.size()); 394| 1.50k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.12k, False: 375] ------------------ 395| 1.12k| auto impl = implementations[i]; 396| 1.12k| const ToType canary1{42}; 397| 1.12k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.12k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.12k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.12k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.12k| if constexpr (std::is_same_v) { 403| 1.12k| return r != 0; 404| 1.12k| } else { 405| 1.12k| return r.error == simdutf::error_code::SUCCESS; 406| 1.12k| } 407| 1.12k| }(implret1); 408| 1.12k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.12k| 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| 1.12k| const ToType canary2{25}; 414| 1.12k| const auto outputbuffer_first_run = outputbuffer; 415| 1.12k| std::ranges::fill(outputbuffer, canary2); 416| 1.12k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.12k| src.size(), outputbuffer.data()); 418| | 419| 1.12k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.12k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.12k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 894, False: 231] | Branch (423:29): [True: 885, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 885| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 885| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 885] ------------------ 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| 885| } 440| 1.12k| } 441| 1.12k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 885, False: 240] ------------------ 442| 1.12k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 375| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 77, False: 298] ------------------ 447| 231| for (auto& e : results) { ------------------ | Branch (447:20): [True: 231, False: 77] ------------------ 448| 231| e.outputhash.clear(); 449| 231| } 450| 77| } 451| | 452| 375| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 375| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 375] ------------------ 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| 375| } else { 474| 375| ret.implementations_agree = true; 475| 375| } 476| 375| return ret; 477| 375| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.12k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.12k| if constexpr (std::is_same_v) { 403| 1.12k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.12k| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 750| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_2clENSt3__14spanIKcLm18446744073709551615EEE: 554| 58| +[](std::span chardata) { \ 555| 58| const auto c = \ 556| 58| Conversion{ \ 559| 58| &I::lenfunc, &I::conversionfunc, \ 560| 58| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 58|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 58| ::std::void_t(); \ | | 1180| 58| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 58| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 58| constexpr auto _size = _name.size(); \ | | 1183| 58| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 58| return _nameof; }() ------------------ 561| 58| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 58|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 58| ::std::void_t(); \ | | 1180| 58| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 58| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 58| constexpr auto _size = _name.size(); \ | | 1183| 58| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 58| return _nameof; }() ------------------ 562| 58| c.fuzz(chardata); \ 563| 58| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 211| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 211| const FromSpan from{reinterpret_cast(chardata.data()), 178| 211| chardata.size() / sizeof(FromType)}; 179| | 180| 211| static const bool do_print_testcase = 181| 211| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 211| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 211] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 211| do { 189| | // step 0 - is the input valid? 190| 211| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 211| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 211] | 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| 211| From == UtfEncodings::UTF8) { 198| 211| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 211] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 211| } 201| | 202| | // step 2 - what is the required size of the output? 203| 211| const auto [output_length, length_agree] = 204| 211| calculate_length(from, inputisvalid); 205| 211| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 211] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 211| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 91, False: 120] | Branch (208:28): [True: 5, False: 86] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 5| return; 211| 5| } 212| | 213| | // step 3 - run the conversion 214| 206| const auto [written, outputs_agree] = 215| 206| do_conversion(from, output_length, inputisvalid); 216| 206| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 206] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 206| return; 221| 206| } 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| 211| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 211| validation_result verify_valid_input(FromSpan src) const { 231| 211| validation_result ret{}; 232| | 233| 211| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 211| const auto implementations = get_supported_implementations(); 235| 211| std::vector results; 236| 211| results.reserve(implementations.size()); 237| | 238| 633| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 633, False: 211] ------------------ 239| 633| results.push_back( 240| 633| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 633| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 633| const bool validation2 = 245| 633| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 633| src.data(), src.size()); 247| 633| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 633] ------------------ 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| 633| } 258| | 259| 211| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 211| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 211] ------------------ 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| 211| } else { 273| 211| ret.implementations_agree = true; 274| 211| } 275| 211| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 211| return r.error == simdutf::SUCCESS; 277| 211| }); 278| 211| return ret; 279| 211| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 422| 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| 451| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 451| return r.error == simdutf::SUCCESS; 277| 451| }); _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 211| bool count_the_input(FromSpan src) const { 290| 211| const auto implementations = get_supported_implementations(); 291| 211| std::vector results; 292| 211| results.reserve(implementations.size()); 293| | 294| 633| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 633, False: 211] ------------------ 295| 633| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 633| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 633| 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| 633| results.push_back(ret); 304| 633| } 305| 211| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 211| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 211] ------------------ 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| 211| return true; 321| 211| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 422| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 211| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 211| length_result ret{}; 344| | 345| 211| const auto implementations = get_supported_implementations(); 346| 211| std::vector results; 347| 211| results.reserve(implementations.size()); 348| | 349| 633| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 633, False: 211] ------------------ 350| 633| const auto len = invoke_lengthcalc(impl, src); 351| 633| results.push_back(len); 352| 633| ret.length.push_back(len); 353| 633| } 354| | 355| 211| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 211| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 211] ------------------ 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| 211| } else { 375| 211| ret.implementations_agree = true; 376| 211| } 377| 211| return ret; 378| 211| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 633| FromSpan src) const { 331| 633| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 633| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 422| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 206| const bool inputisvalid) const { 383| 206| conversion_result ret{}; 384| | 385| 206| const auto implementations = get_supported_implementations(); 386| | 387| 206| std::vector> results; 388| 206| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 206| std::vector> outputbuffers; 393| 206| outputbuffers.reserve(implementations.size()); 394| 824| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 618, False: 206] ------------------ 395| 618| auto impl = implementations[i]; 396| 618| const ToType canary1{42}; 397| 618| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 618| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 618| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 618| const auto success = [](const ConversionResult& r) -> bool { 402| 618| if constexpr (std::is_same_v) { 403| 618| return r != 0; 404| 618| } else { 405| 618| return r.error == simdutf::error_code::SUCCESS; 406| 618| } 407| 618| }(implret1); 408| 618| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 618| 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| 618| const ToType canary2{25}; 414| 618| const auto outputbuffer_first_run = outputbuffer; 415| 618| std::ranges::fill(outputbuffer, canary2); 416| 618| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 618| src.size(), outputbuffer.data()); 418| | 419| 618| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 618] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 618| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 360, False: 258] | Branch (423:29): [True: 348, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 348| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 348| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 348] ------------------ 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| 348| } 440| 618| } 441| 618| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 348, False: 270] ------------------ 442| 618| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 206| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 86, False: 120] ------------------ 447| 258| for (auto& e : results) { ------------------ | Branch (447:20): [True: 258, False: 86] ------------------ 448| 258| e.outputhash.clear(); 449| 258| } 450| 86| } 451| | 452| 206| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 206| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 206] ------------------ 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| 206| } else { 474| 206| ret.implementations_agree = true; 475| 206| } 476| 206| return ret; 477| 206| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 618| const auto success = [](const ConversionResult& r) -> bool { 402| 618| if constexpr (std::is_same_v) { 403| 618| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 618| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 412| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_3clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_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| 444| From == UtfEncodings::UTF8) { 198| 444| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 444] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 444| } 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: 117, False: 327] | Branch (208:28): [True: 16, False: 101] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 16| return; 211| 16| } 212| | 213| | // step 3 - run the conversion 214| 428| const auto [written, outputs_agree] = 215| 428| do_conversion(from, output_length, inputisvalid); 216| 428| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 428] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 428| return; 221| 428| } 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| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_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| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_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; }; _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 1.09k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.09k| return r.error == simdutf::SUCCESS; 277| 1.09k| }); _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 444| bool count_the_input(FromSpan src) const { 290| 444| const auto implementations = get_supported_implementations(); 291| 444| std::vector results; 292| 444| results.reserve(implementations.size()); 293| | 294| 1.33k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.33k, False: 444] ------------------ 295| 1.33k| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 1.33k| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 1.33k| 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| 1.33k| results.push_back(ret); 304| 1.33k| } 305| 444| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 444| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 444] ------------------ 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| 444| return true; 321| 444| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 888| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_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| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_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| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 888| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 428| const bool inputisvalid) const { 383| 428| conversion_result ret{}; 384| | 385| 428| const auto implementations = get_supported_implementations(); 386| | 387| 428| std::vector> results; 388| 428| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 428| std::vector> outputbuffers; 393| 428| outputbuffers.reserve(implementations.size()); 394| 1.71k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.28k, False: 428] ------------------ 395| 1.28k| auto impl = implementations[i]; 396| 1.28k| const ToType canary1{42}; 397| 1.28k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.28k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.28k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.28k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.28k| if constexpr (std::is_same_v) { 403| 1.28k| return r != 0; 404| 1.28k| } else { 405| 1.28k| return r.error == simdutf::error_code::SUCCESS; 406| 1.28k| } 407| 1.28k| }(implret1); 408| 1.28k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.28k| 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| 1.28k| const ToType canary2{25}; 414| 1.28k| const auto outputbuffer_first_run = outputbuffer; 415| 1.28k| std::ranges::fill(outputbuffer, canary2); 416| 1.28k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.28k| src.size(), outputbuffer.data()); 418| | 419| 1.28k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.28k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.28k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 981, False: 303] | Branch (423:29): [True: 969, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 969| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 969| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 969] ------------------ 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| 969| } 440| 1.28k| } 441| 1.28k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 969, False: 315] ------------------ 442| 1.28k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 428| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 101, False: 327] ------------------ 447| 303| for (auto& e : results) { ------------------ | Branch (447:20): [True: 303, False: 101] ------------------ 448| 303| e.outputhash.clear(); 449| 303| } 450| 101| } 451| | 452| 428| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 428| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 428] ------------------ 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| 428| } else { 474| 428| ret.implementations_agree = true; 475| 428| } 476| 428| return ret; 477| 428| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.28k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.28k| if constexpr (std::is_same_v) { 403| 1.28k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.28k| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 856| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_4clENSt3__14spanIKcLm18446744073709551615EEE: 554| 66| +[](std::span chardata) { \ 555| 66| const auto c = \ 556| 66| Conversion{ \ 559| 66| &I::lenfunc, &I::conversionfunc, \ 560| 66| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 66|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 66| ::std::void_t(); \ | | 1180| 66| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 66| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 66| constexpr auto _size = _name.size(); \ | | 1183| 66| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 66| return _nameof; }() ------------------ 561| 66| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 66|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 66| ::std::void_t(); \ | | 1180| 66| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 66| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 66| constexpr auto _size = _name.size(); \ | | 1183| 66| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 66| return _nameof; }() ------------------ 562| 66| c.fuzz(chardata); \ 563| 66| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 333| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 333| const FromSpan from{reinterpret_cast(chardata.data()), 178| 333| chardata.size() / sizeof(FromType)}; 179| | 180| 333| static const bool do_print_testcase = 181| 333| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 333| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 333] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 333| do { 189| | // step 0 - is the input valid? 190| 333| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 333| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 333] | 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| 333| const auto [output_length, length_agree] = 204| 333| calculate_length(from, inputisvalid); 205| 333| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 333] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 333| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 190, False: 143] | Branch (208:28): [True: 2, False: 188] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 2| return; 211| 2| } 212| | 213| | // step 3 - run the conversion 214| 331| const auto [written, outputs_agree] = 215| 331| do_conversion(from, output_length, inputisvalid); 216| 331| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 331] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 331| return; 221| 331| } 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| 333| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 333| validation_result verify_valid_input(FromSpan src) const { 231| 333| validation_result ret{}; 232| | 233| 333| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 333| const auto implementations = get_supported_implementations(); 235| 333| std::vector results; 236| 333| results.reserve(implementations.size()); 237| | 238| 999| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 999, False: 333] ------------------ 239| 999| results.push_back( 240| 999| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 999| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 999| const bool validation2 = 245| 999| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 999| src.data(), src.size()); 247| 999| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 999] ------------------ 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| 999| } 258| | 259| 333| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 333| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 333] ------------------ 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| 333| } else { 273| 333| ret.implementations_agree = true; 274| 333| } 275| 333| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 333| return r.error == simdutf::SUCCESS; 277| 333| }); 278| 333| return ret; 279| 333| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 666| 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| 619| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 619| return r.error == simdutf::SUCCESS; 277| 619| }); _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 333| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 333| length_result ret{}; 344| | 345| 333| const auto implementations = get_supported_implementations(); 346| 333| std::vector results; 347| 333| results.reserve(implementations.size()); 348| | 349| 999| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 999, False: 333] ------------------ 350| 999| const auto len = invoke_lengthcalc(impl, src); 351| 999| results.push_back(len); 352| 999| ret.length.push_back(len); 353| 999| } 354| | 355| 333| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 333| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 333] ------------------ 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| 333| } else { 375| 333| ret.implementations_agree = true; 376| 333| } 377| 333| return ret; 378| 333| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 999| FromSpan src) const { 331| 999| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 999| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 666| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 331| const bool inputisvalid) const { 383| 331| conversion_result ret{}; 384| | 385| 331| const auto implementations = get_supported_implementations(); 386| | 387| 331| std::vector> results; 388| 331| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 331| std::vector> outputbuffers; 393| 331| outputbuffers.reserve(implementations.size()); 394| 1.32k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 993, False: 331] ------------------ 395| 993| auto impl = implementations[i]; 396| 993| const ToType canary1{42}; 397| 993| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 993| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 993| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 993| const auto success = [](const ConversionResult& r) -> bool { 402| 993| if constexpr (std::is_same_v) { 403| 993| return r != 0; 404| 993| } else { 405| 993| return r.error == simdutf::error_code::SUCCESS; 406| 993| } 407| 993| }(implret1); 408| 993| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 993| 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| 993| const ToType canary2{25}; 414| 993| const auto outputbuffer_first_run = outputbuffer; 415| 993| std::ranges::fill(outputbuffer, canary2); 416| 993| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 993| src.size(), outputbuffer.data()); 418| | 419| 993| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 993] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 993| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 429, False: 564] | Branch (423:29): [True: 417, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 417| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 417| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 417] ------------------ 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| 417| } 440| 993| } 441| 993| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 417, False: 576] ------------------ 442| 993| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 331| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 188, False: 143] ------------------ 447| 564| for (auto& e : results) { ------------------ | Branch (447:20): [True: 564, False: 188] ------------------ 448| 564| e.outputhash.clear(); 449| 564| } 450| 188| } 451| | 452| 331| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 331| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 331] ------------------ 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| 331| } else { 474| 331| ret.implementations_agree = true; 475| 331| } 476| 331| return ret; 477| 331| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 993| const auto success = [](const ConversionResult& r) -> bool { 402| 993| if constexpr (std::is_same_v) { 403| 993| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 993| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 662| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_5clENSt3__14spanIKcLm18446744073709551615EEE: 554| 131| +[](std::span chardata) { \ 555| 131| const auto c = \ 556| 131| Conversion{ \ 559| 131| &I::lenfunc, &I::conversionfunc, \ 560| 131| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 131|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 131| ::std::void_t(); \ | | 1180| 131| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 131| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 131| constexpr auto _size = _name.size(); \ | | 1183| 131| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 131| return _nameof; }() ------------------ 561| 131| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 131|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 131| ::std::void_t(); \ | | 1180| 131| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 131| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 131| constexpr auto _size = _name.size(); \ | | 1183| 131| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 131| return _nameof; }() ------------------ 562| 131| c.fuzz(chardata); \ 563| 131| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 394| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 394| const FromSpan from{reinterpret_cast(chardata.data()), 178| 394| chardata.size() / sizeof(FromType)}; 179| | 180| 394| static const bool do_print_testcase = 181| 394| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 394| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 394] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 394| do { 189| | // step 0 - is the input valid? 190| 394| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 394| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 394] | 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| 394| const auto [output_length, length_agree] = 204| 394| calculate_length(from, inputisvalid); 205| 394| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 394] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 394| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 193, False: 201] | Branch (208:28): [True: 7, False: 186] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 7| return; 211| 7| } 212| | 213| | // step 3 - run the conversion 214| 387| const auto [written, outputs_agree] = 215| 387| do_conversion(from, output_length, inputisvalid); 216| 387| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 387] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 387| return; 221| 387| } 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| 394| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 394| validation_result verify_valid_input(FromSpan src) const { 231| 394| validation_result ret{}; 232| | 233| 394| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 394| const auto implementations = get_supported_implementations(); 235| 394| std::vector results; 236| 394| results.reserve(implementations.size()); 237| | 238| 1.18k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.18k, False: 394] ------------------ 239| 1.18k| results.push_back( 240| 1.18k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.18k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.18k| const bool validation2 = 245| 1.18k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.18k| src.data(), src.size()); 247| 1.18k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.18k] ------------------ 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| 1.18k| } 258| | 259| 394| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 394| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 394] ------------------ 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| 394| } else { 273| 394| ret.implementations_agree = true; 274| 394| } 275| 394| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 394| return r.error == simdutf::SUCCESS; 277| 394| }); 278| 394| return ret; 279| 394| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 788| 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| 796| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 796| return r.error == simdutf::SUCCESS; 277| 796| }); _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 394| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 394| length_result ret{}; 344| | 345| 394| const auto implementations = get_supported_implementations(); 346| 394| std::vector results; 347| 394| results.reserve(implementations.size()); 348| | 349| 1.18k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.18k, False: 394] ------------------ 350| 1.18k| const auto len = invoke_lengthcalc(impl, src); 351| 1.18k| results.push_back(len); 352| 1.18k| ret.length.push_back(len); 353| 1.18k| } 354| | 355| 394| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 394| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 394] ------------------ 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| 394| } else { 375| 394| ret.implementations_agree = true; 376| 394| } 377| 394| return ret; 378| 394| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.18k| FromSpan src) const { 331| 1.18k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.18k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 788| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 387| const bool inputisvalid) const { 383| 387| conversion_result ret{}; 384| | 385| 387| const auto implementations = get_supported_implementations(); 386| | 387| 387| std::vector> results; 388| 387| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 387| std::vector> outputbuffers; 393| 387| outputbuffers.reserve(implementations.size()); 394| 1.54k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.16k, False: 387] ------------------ 395| 1.16k| auto impl = implementations[i]; 396| 1.16k| const ToType canary1{42}; 397| 1.16k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.16k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.16k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.16k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.16k| if constexpr (std::is_same_v) { 403| 1.16k| return r != 0; 404| 1.16k| } else { 405| 1.16k| return r.error == simdutf::error_code::SUCCESS; 406| 1.16k| } 407| 1.16k| }(implret1); 408| 1.16k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.16k| 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| 1.16k| const ToType canary2{25}; 414| 1.16k| const auto outputbuffer_first_run = outputbuffer; 415| 1.16k| std::ranges::fill(outputbuffer, canary2); 416| 1.16k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.16k| src.size(), outputbuffer.data()); 418| | 419| 1.16k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.16k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.16k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 603, False: 558] | Branch (423:29): [True: 588, False: 15] ------------------ 424| | // only care about the output if the input is valid 425| 588| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 588| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 588] ------------------ 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| 588| } 440| 1.16k| } 441| 1.16k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 588, False: 573] ------------------ 442| 1.16k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 387| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 186, False: 201] ------------------ 447| 558| for (auto& e : results) { ------------------ | Branch (447:20): [True: 558, False: 186] ------------------ 448| 558| e.outputhash.clear(); 449| 558| } 450| 186| } 451| | 452| 387| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 387| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 387] ------------------ 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| 387| } else { 474| 387| ret.implementations_agree = true; 475| 387| } 476| 387| return ret; 477| 387| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.16k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.16k| if constexpr (std::is_same_v) { 403| 1.16k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.16k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 774| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_6clENSt3__14spanIKcLm18446744073709551615EEE: 554| 115| +[](std::span chardata) { \ 555| 115| const auto c = \ 556| 115| Conversion{ \ 559| 115| &I::lenfunc, &I::conversionfunc, \ 560| 115| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 115|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 115| ::std::void_t(); \ | | 1180| 115| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 115| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 115| constexpr auto _size = _name.size(); \ | | 1183| 115| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 115| return _nameof; }() ------------------ 561| 115| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 115|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 115| ::std::void_t(); \ | | 1180| 115| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 115| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 115| constexpr auto _size = _name.size(); \ | | 1183| 115| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 115| return _nameof; }() ------------------ 562| 115| c.fuzz(chardata); \ 563| 115| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 470| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 470| const FromSpan from{reinterpret_cast(chardata.data()), 178| 470| chardata.size() / sizeof(FromType)}; 179| | 180| 470| static const bool do_print_testcase = 181| 470| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 470| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 470] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 470| do { 189| | // step 0 - is the input valid? 190| 470| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 470| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 470] | 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| 470| const auto [output_length, length_agree] = 204| 470| calculate_length(from, inputisvalid); 205| 470| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 470] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 470| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 311, False: 159] | Branch (208:28): [True: 12, False: 299] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 12| return; 211| 12| } 212| | 213| | // step 3 - run the conversion 214| 458| const auto [written, outputs_agree] = 215| 458| do_conversion(from, output_length, inputisvalid); 216| 458| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 458] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 458| return; 221| 458| } 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| 470| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 470| validation_result verify_valid_input(FromSpan src) const { 231| 470| validation_result ret{}; 232| | 233| 470| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 470| const auto implementations = get_supported_implementations(); 235| 470| std::vector results; 236| 470| results.reserve(implementations.size()); 237| | 238| 1.41k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.41k, False: 470] ------------------ 239| 1.41k| results.push_back( 240| 1.41k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.41k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.41k| const bool validation2 = 245| 1.41k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.41k| src.data(), src.size()); 247| 1.41k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.41k] ------------------ 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| 1.41k| } 258| | 259| 470| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 470| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 470] ------------------ 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| 470| } else { 273| 470| ret.implementations_agree = true; 274| 470| } 275| 470| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 470| return r.error == simdutf::SUCCESS; 277| 470| }); 278| 470| return ret; 279| 470| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 940| 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| 788| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 788| return r.error == simdutf::SUCCESS; 277| 788| }); _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 470| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 470| length_result ret{}; 344| | 345| 470| const auto implementations = get_supported_implementations(); 346| 470| std::vector results; 347| 470| results.reserve(implementations.size()); 348| | 349| 1.41k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.41k, False: 470] ------------------ 350| 1.41k| const auto len = invoke_lengthcalc(impl, src); 351| 1.41k| results.push_back(len); 352| 1.41k| ret.length.push_back(len); 353| 1.41k| } 354| | 355| 470| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 470| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 470] ------------------ 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| 470| } else { 375| 470| ret.implementations_agree = true; 376| 470| } 377| 470| return ret; 378| 470| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.41k| FromSpan src) const { 331| 1.41k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.41k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 940| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 458| const bool inputisvalid) const { 383| 458| conversion_result ret{}; 384| | 385| 458| const auto implementations = get_supported_implementations(); 386| | 387| 458| std::vector> results; 388| 458| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 458| std::vector> outputbuffers; 393| 458| outputbuffers.reserve(implementations.size()); 394| 1.83k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.37k, False: 458] ------------------ 395| 1.37k| auto impl = implementations[i]; 396| 1.37k| const ToType canary1{42}; 397| 1.37k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.37k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.37k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.37k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.37k| if constexpr (std::is_same_v) { 403| 1.37k| return r != 0; 404| 1.37k| } else { 405| 1.37k| return r.error == simdutf::error_code::SUCCESS; 406| 1.37k| } 407| 1.37k| }(implret1); 408| 1.37k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.37k| 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| 1.37k| const ToType canary2{25}; 414| 1.37k| const auto outputbuffer_first_run = outputbuffer; 415| 1.37k| std::ranges::fill(outputbuffer, canary2); 416| 1.37k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.37k| src.size(), outputbuffer.data()); 418| | 419| 1.37k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.37k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.37k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 477, False: 897] | Branch (423:29): [True: 468, False: 9] ------------------ 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| 1.37k| } 441| 1.37k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 468, False: 906] ------------------ 442| 1.37k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 458| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 299, False: 159] ------------------ 447| 897| for (auto& e : results) { ------------------ | Branch (447:20): [True: 897, False: 299] ------------------ 448| 897| e.outputhash.clear(); 449| 897| } 450| 299| } 451| | 452| 458| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 458| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 458] ------------------ 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| 458| } else { 474| 458| ret.implementations_agree = true; 475| 458| } 476| 458| return ret; 477| 458| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.37k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.37k| if constexpr (std::is_same_v) { 403| 1.37k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.37k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 916| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_7clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 626| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 626| const FromSpan from{reinterpret_cast(chardata.data()), 178| 626| chardata.size() / sizeof(FromType)}; 179| | 180| 626| static const bool do_print_testcase = 181| 626| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 626| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 626] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 626| do { 189| | // step 0 - is the input valid? 190| 626| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 626| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 626] | 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| 626| From == UtfEncodings::UTF8) { 198| 626| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 626] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 626| } 201| | 202| | // step 2 - what is the required size of the output? 203| 626| const auto [output_length, length_agree] = 204| 626| calculate_length(from, inputisvalid); 205| 626| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 626] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 626| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 289, False: 337] | 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| 616| const auto [written, outputs_agree] = 215| 616| do_conversion(from, output_length, inputisvalid); 216| 616| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 616] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 616| return; 221| 616| } 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| 626| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 626| validation_result verify_valid_input(FromSpan src) const { 231| 626| validation_result ret{}; 232| | 233| 626| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 626| const auto implementations = get_supported_implementations(); 235| 626| std::vector results; 236| 626| results.reserve(implementations.size()); 237| | 238| 1.87k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.87k, False: 626] ------------------ 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| 626| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 626| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 626] ------------------ 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| 626| } else { 273| 626| ret.implementations_agree = true; 274| 626| } 275| 626| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 626| return r.error == simdutf::SUCCESS; 277| 626| }); 278| 626| return ret; 279| 626| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 1.25k| 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| 1.30k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.30k| return r.error == simdutf::SUCCESS; 277| 1.30k| }); _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 626| bool count_the_input(FromSpan src) const { 290| 626| const auto implementations = get_supported_implementations(); 291| 626| std::vector results; 292| 626| results.reserve(implementations.size()); 293| | 294| 1.87k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.87k, False: 626] ------------------ 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| 626| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 626| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 626] ------------------ 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| 626| return true; 321| 626| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 1.25k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 626| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 626| length_result ret{}; 344| | 345| 626| const auto implementations = get_supported_implementations(); 346| 626| std::vector results; 347| 626| results.reserve(implementations.size()); 348| | 349| 1.87k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.87k, False: 626] ------------------ 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| 626| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 626| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 626] ------------------ 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| 626| } else { 375| 626| ret.implementations_agree = true; 376| 626| } 377| 626| return ret; 378| 626| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_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_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 1.25k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 616| const bool inputisvalid) const { 383| 616| conversion_result ret{}; 384| | 385| 616| const auto implementations = get_supported_implementations(); 386| | 387| 616| std::vector> results; 388| 616| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 616| std::vector> outputbuffers; 393| 616| outputbuffers.reserve(implementations.size()); 394| 2.46k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.84k, False: 616] ------------------ 395| 1.84k| auto impl = implementations[i]; 396| 1.84k| const ToType canary1{42}; 397| 1.84k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.84k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.84k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.84k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.84k| if constexpr (std::is_same_v) { 403| 1.84k| return r != 0; 404| 1.84k| } else { 405| 1.84k| return r.error == simdutf::error_code::SUCCESS; 406| 1.84k| } 407| 1.84k| }(implret1); 408| 1.84k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.84k| 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| 1.84k| const ToType canary2{25}; 414| 1.84k| const auto outputbuffer_first_run = outputbuffer; 415| 1.84k| std::ranges::fill(outputbuffer, canary2); 416| 1.84k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.84k| src.size(), outputbuffer.data()); 418| | 419| 1.84k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.84k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.84k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 1.01k, False: 837] | Branch (423:29): [True: 1.00k, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 1.00k| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 1.00k| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 1.00k] ------------------ 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| 1.00k| } 440| 1.84k| } 441| 1.84k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 1.00k, False: 843] ------------------ 442| 1.84k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 616| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 279, False: 337] ------------------ 447| 837| for (auto& e : results) { ------------------ | Branch (447:20): [True: 837, False: 279] ------------------ 448| 837| e.outputhash.clear(); 449| 837| } 450| 279| } 451| | 452| 616| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 616| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 616] ------------------ 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| 616| } else { 474| 616| ret.implementations_agree = true; 475| 616| } 476| 616| return ret; 477| 616| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.84k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.84k| if constexpr (std::is_same_v) { 403| 1.84k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.84k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 1.23k| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_8clENSt3__14spanIKcLm18446744073709551615EEE: 554| 239| +[](std::span chardata) { \ 555| 239| const auto c = \ 556| 239| Conversion{ \ 559| 239| &I::lenfunc, &I::conversionfunc, \ 560| 239| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 239|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 239| ::std::void_t(); \ | | 1180| 239| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 239| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 239| constexpr auto _size = _name.size(); \ | | 1183| 239| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 239| return _nameof; }() ------------------ 561| 239| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 239|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 239| ::std::void_t(); \ | | 1180| 239| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 239| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 239| constexpr auto _size = _name.size(); \ | | 1183| 239| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 239| return _nameof; }() ------------------ 562| 239| c.fuzz(chardata); \ 563| 239| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 599| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 599| const FromSpan from{reinterpret_cast(chardata.data()), 178| 599| chardata.size() / sizeof(FromType)}; 179| | 180| 599| static const bool do_print_testcase = 181| 599| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 599| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 599] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 599| do { 189| | // step 0 - is the input valid? 190| 599| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 599| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 599] | 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| 599| From == UtfEncodings::UTF8) { 198| 599| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 599] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 599| } 201| | 202| | // step 2 - what is the required size of the output? 203| 599| const auto [output_length, length_agree] = 204| 599| calculate_length(from, inputisvalid); 205| 599| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 599] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 599| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 261, False: 338] | Branch (208:28): [True: 17, False: 244] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 17| return; 211| 17| } 212| | 213| | // step 3 - run the conversion 214| 582| const auto [written, outputs_agree] = 215| 582| do_conversion(from, output_length, inputisvalid); 216| 582| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 582] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 582| return; 221| 582| } 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| 599| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 599| validation_result verify_valid_input(FromSpan src) const { 231| 599| validation_result ret{}; 232| | 233| 599| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 599| const auto implementations = get_supported_implementations(); 235| 599| std::vector results; 236| 599| results.reserve(implementations.size()); 237| | 238| 1.79k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.79k, False: 599] ------------------ 239| 1.79k| results.push_back( 240| 1.79k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.79k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.79k| const bool validation2 = 245| 1.79k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.79k| src.data(), src.size()); 247| 1.79k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.79k] ------------------ 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| 1.79k| } 258| | 259| 599| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 599| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 599] ------------------ 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| 599| } else { 273| 599| ret.implementations_agree = true; 274| 599| } 275| 599| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 599| return r.error == simdutf::SUCCESS; 277| 599| }); 278| 599| return ret; 279| 599| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 1.19k| 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| 1.27k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.27k| return r.error == simdutf::SUCCESS; 277| 1.27k| }); _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 599| bool count_the_input(FromSpan src) const { 290| 599| const auto implementations = get_supported_implementations(); 291| 599| std::vector results; 292| 599| results.reserve(implementations.size()); 293| | 294| 1.79k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.79k, False: 599] ------------------ 295| 1.79k| 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| 1.79k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.79k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.79k| } 303| 1.79k| results.push_back(ret); 304| 1.79k| } 305| 599| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 599| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 599] ------------------ 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| 599| return true; 321| 599| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 1.19k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 599| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 599| length_result ret{}; 344| | 345| 599| const auto implementations = get_supported_implementations(); 346| 599| std::vector results; 347| 599| results.reserve(implementations.size()); 348| | 349| 1.79k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.79k, False: 599] ------------------ 350| 1.79k| const auto len = invoke_lengthcalc(impl, src); 351| 1.79k| results.push_back(len); 352| 1.79k| ret.length.push_back(len); 353| 1.79k| } 354| | 355| 599| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 599| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 599] ------------------ 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| 599| } else { 375| 599| ret.implementations_agree = true; 376| 599| } 377| 599| return ret; 378| 599| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.79k| FromSpan src) const { 331| 1.79k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.79k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 1.19k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 582| const bool inputisvalid) const { 383| 582| conversion_result ret{}; 384| | 385| 582| const auto implementations = get_supported_implementations(); 386| | 387| 582| std::vector> results; 388| 582| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 582| std::vector> outputbuffers; 393| 582| outputbuffers.reserve(implementations.size()); 394| 2.32k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.74k, False: 582] ------------------ 395| 1.74k| auto impl = implementations[i]; 396| 1.74k| const ToType canary1{42}; 397| 1.74k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.74k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.74k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.74k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.74k| if constexpr (std::is_same_v) { 403| 1.74k| return r != 0; 404| 1.74k| } else { 405| 1.74k| return r.error == simdutf::error_code::SUCCESS; 406| 1.74k| } 407| 1.74k| }(implret1); 408| 1.74k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.74k| 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| 1.74k| const ToType canary2{25}; 414| 1.74k| const auto outputbuffer_first_run = outputbuffer; 415| 1.74k| std::ranges::fill(outputbuffer, canary2); 416| 1.74k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.74k| src.size(), outputbuffer.data()); 418| | 419| 1.74k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.74k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.74k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 1.01k, False: 732] | Branch (423:29): [True: 1.00k, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 1.00k| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 1.00k| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 1.00k] ------------------ 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| 1.00k| } 440| 1.74k| } 441| 1.74k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 1.00k, False: 741] ------------------ 442| 1.74k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 582| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 244, False: 338] ------------------ 447| 732| for (auto& e : results) { ------------------ | Branch (447:20): [True: 732, False: 244] ------------------ 448| 732| e.outputhash.clear(); 449| 732| } 450| 244| } 451| | 452| 582| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 582| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 582] ------------------ 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| 582| } else { 474| 582| ret.implementations_agree = true; 475| 582| } 476| 582| return ret; 477| 582| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.74k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.74k| if constexpr (std::is_same_v) { 403| 1.74k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.74k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 1.16k| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_9clENSt3__14spanIKcLm18446744073709551615EEE: 554| 229| +[](std::span chardata) { \ 555| 229| const auto c = \ 556| 229| Conversion{ \ 559| 229| &I::lenfunc, &I::conversionfunc, \ 560| 229| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 229|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 229| ::std::void_t(); \ | | 1180| 229| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 229| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 229| constexpr auto _size = _name.size(); \ | | 1183| 229| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 229| return _nameof; }() ------------------ 561| 229| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 229|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 229| ::std::void_t(); \ | | 1180| 229| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 229| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 229| constexpr auto _size = _name.size(); \ | | 1183| 229| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 229| return _nameof; }() ------------------ 562| 229| c.fuzz(chardata); \ 563| 229| } \ _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: 268, False: 353] | Branch (208:28): [True: 4, False: 264] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 4| return; 211| 4| } 212| | 213| | // step 3 - run the conversion 214| 617| const auto [written, outputs_agree] = 215| 617| do_conversion(from, output_length, inputisvalid); 216| 617| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 617] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 617| return; 221| 617| } 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| 1.32k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.32k| return r.error == simdutf::SUCCESS; 277| 1.32k| }); _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| 617| const bool inputisvalid) const { 383| 617| conversion_result ret{}; 384| | 385| 617| const auto implementations = get_supported_implementations(); 386| | 387| 617| std::vector> results; 388| 617| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 617| std::vector> outputbuffers; 393| 617| outputbuffers.reserve(implementations.size()); 394| 2.46k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.85k, False: 617] ------------------ 395| 1.85k| auto impl = implementations[i]; 396| 1.85k| const ToType canary1{42}; 397| 1.85k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.85k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.85k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.85k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.85k| if constexpr (std::is_same_v) { 403| 1.85k| return r != 0; 404| 1.85k| } else { 405| 1.85k| return r.error == simdutf::error_code::SUCCESS; 406| 1.85k| } 407| 1.85k| }(implret1); 408| 1.85k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.85k| 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| 1.85k| const ToType canary2{25}; 414| 1.85k| const auto outputbuffer_first_run = outputbuffer; 415| 1.85k| std::ranges::fill(outputbuffer, canary2); 416| 1.85k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.85k| src.size(), outputbuffer.data()); 418| | 419| 1.85k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.85k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.85k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 1.05k, False: 792] | Branch (423:29): [True: 1.05k, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 1.05k| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 1.05k| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 1.05k] ------------------ 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| 1.05k| } 440| 1.85k| } 441| 1.85k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 1.05k, False: 801] ------------------ 442| 1.85k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 617| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 264, False: 353] ------------------ 447| 792| for (auto& e : results) { ------------------ | Branch (447:20): [True: 792, False: 264] ------------------ 448| 792| e.outputhash.clear(); 449| 792| } 450| 264| } 451| | 452| 617| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 617| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 617] ------------------ 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| 617| } else { 474| 617| ret.implementations_agree = true; 475| 617| } 476| 617| return ret; 477| 617| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.85k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.85k| if constexpr (std::is_same_v) { 403| 1.85k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.85k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 1.23k| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_10clENSt3__14spanIKcLm18446744073709551615EEE: 554| 47| +[](std::span chardata) { \ 555| 47| const auto c = \ 556| 47| Conversion{ \ 559| 47| &I::lenfunc, &I::conversionfunc, \ 560| 47| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 47|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 47| ::std::void_t(); \ | | 1180| 47| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 47| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 47| constexpr auto _size = _name.size(); \ | | 1183| 47| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 47| return _nameof; }() ------------------ 561| 47| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 47|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 47| ::std::void_t(); \ | | 1180| 47| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 47| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 47| constexpr auto _size = _name.size(); \ | | 1183| 47| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 47| return _nameof; }() ------------------ 562| 47| c.fuzz(chardata); \ 563| 47| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 47| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 47| const FromSpan from{reinterpret_cast(chardata.data()), 178| 47| chardata.size() / sizeof(FromType)}; 179| | 180| 47| static const bool do_print_testcase = 181| 47| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 47| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 47] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 47| do { 189| | // step 0 - is the input valid? 190| 47| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 47| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 47] | 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| 47| From == UtfEncodings::UTF8) { 198| 47| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 47] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 47| } 201| | 202| | // step 2 - what is the required size of the output? 203| 47| const auto [output_length, length_agree] = 204| 47| calculate_length(from, inputisvalid); 205| 47| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 47] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 47| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 9, False: 38] | Branch (208:28): [True: 0, False: 9] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 47| const auto [written, outputs_agree] = 215| 47| do_conversion(from, output_length, inputisvalid); 216| 47| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 47] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 47| return; 221| 47| } 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| 47| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 230| 47| validation_result verify_valid_input(FromSpan src) const { 231| 47| validation_result ret{}; 232| | 233| 47| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 47| const auto implementations = get_supported_implementations(); 235| 47| std::vector results; 236| 47| results.reserve(implementations.size()); 237| | 238| 141| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 141, False: 47] ------------------ 239| 141| results.push_back( 240| 141| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 141| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 141| const bool validation2 = 245| 141| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 141| src.data(), src.size()); 247| 141| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 141] ------------------ 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| 141| } 258| | 259| 47| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 47| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 47] ------------------ 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| 47| } else { 273| 47| ret.implementations_agree = true; 274| 47| } 275| 47| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 47| return r.error == simdutf::SUCCESS; 277| 47| }); 278| 47| return ret; 279| 47| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 94| 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| 123| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 123| return r.error == simdutf::SUCCESS; 277| 123| }); _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEE: 289| 47| bool count_the_input(FromSpan src) const { 290| 47| const auto implementations = get_supported_implementations(); 291| 47| std::vector results; 292| 47| results.reserve(implementations.size()); 293| | 294| 141| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 141, False: 47] ------------------ 295| 141| std::size_t ret; 296| 141| if constexpr (From == UtfEncodings::UTF16BE) { 297| 141| 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| 141| results.push_back(ret); 304| 141| } 305| 47| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 47| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 47] ------------------ 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| 47| return true; 321| 47| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 94| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 47| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 47| length_result ret{}; 344| | 345| 47| const auto implementations = get_supported_implementations(); 346| 47| std::vector results; 347| 47| results.reserve(implementations.size()); 348| | 349| 141| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 141, False: 47] ------------------ 350| 141| const auto len = invoke_lengthcalc(impl, src); 351| 141| results.push_back(len); 352| 141| ret.length.push_back(len); 353| 141| } 354| | 355| 47| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 47| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 47] ------------------ 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| 47| } else { 375| 47| ret.implementations_agree = true; 376| 47| } 377| 47| return ret; 378| 47| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 141| FromSpan src) const { 339| 141| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 141| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 94| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 47| const bool inputisvalid) const { 383| 47| conversion_result ret{}; 384| | 385| 47| const auto implementations = get_supported_implementations(); 386| | 387| 47| std::vector> results; 388| 47| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 47| std::vector> outputbuffers; 393| 47| outputbuffers.reserve(implementations.size()); 394| 188| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 141, False: 47] ------------------ 395| 141| auto impl = implementations[i]; 396| 141| const ToType canary1{42}; 397| 141| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 141| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 141| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 141| const auto success = [](const ConversionResult& r) -> bool { 402| 141| if constexpr (std::is_same_v) { 403| 141| return r != 0; 404| 141| } else { 405| 141| return r.error == simdutf::error_code::SUCCESS; 406| 141| } 407| 141| }(implret1); 408| 141| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 141| 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| 141| const ToType canary2{25}; 414| 141| const auto outputbuffer_first_run = outputbuffer; 415| 141| std::ranges::fill(outputbuffer, canary2); 416| 141| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 141| src.size(), outputbuffer.data()); 418| | 419| 141| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 141] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 141| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 114, False: 27] | Branch (423:29): [True: 60, False: 54] ------------------ 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| 141| } 441| 141| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 60, False: 81] ------------------ 442| 141| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 47| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 9, False: 38] ------------------ 447| 27| for (auto& e : results) { ------------------ | Branch (447:20): [True: 27, False: 9] ------------------ 448| 27| e.outputhash.clear(); 449| 27| } 450| 9| } 451| | 452| 47| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 47| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 47] ------------------ 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| 47| } else { 474| 47| ret.implementations_agree = true; 475| 47| } 476| 47| return ret; 477| 47| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 141| const auto success = [](const ConversionResult& r) -> bool { 402| 141| if constexpr (std::is_same_v) { 403| 141| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 141| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 94| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_11clENSt3__14spanIKcLm18446744073709551615EEE: 554| 110| +[](std::span chardata) { \ 555| 110| const auto c = \ 556| 110| Conversion{ \ 559| 110| &I::lenfunc, &I::conversionfunc, \ 560| 110| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 110|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 110| ::std::void_t(); \ | | 1180| 110| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 110| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 110| constexpr auto _size = _name.size(); \ | | 1183| 110| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 110| return _nameof; }() ------------------ 561| 110| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 110|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 110| ::std::void_t(); \ | | 1180| 110| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 110| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 110| constexpr auto _size = _name.size(); \ | | 1183| 110| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 110| return _nameof; }() ------------------ 562| 110| c.fuzz(chardata); \ 563| 110| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_12clENSt3__14spanIKcLm18446744073709551615EEE: 554| 176| +[](std::span chardata) { \ 555| 176| const auto c = \ 556| 176| Conversion{ \ 559| 176| &I::lenfunc, &I::conversionfunc, \ 560| 176| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 176|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 176| ::std::void_t(); \ | | 1180| 176| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 176| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 176| constexpr auto _size = _name.size(); \ | | 1183| 176| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 176| return _nameof; }() ------------------ 561| 176| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 176|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 176| ::std::void_t(); \ | | 1180| 176| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 176| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 176| constexpr auto _size = _name.size(); \ | | 1183| 176| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 176| return _nameof; }() ------------------ 562| 176| c.fuzz(chardata); \ 563| 176| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_13clENSt3__14spanIKcLm18446744073709551615EEE: 554| 67| +[](std::span chardata) { \ 555| 67| const auto c = \ 556| 67| Conversion{ \ 559| 67| &I::lenfunc, &I::conversionfunc, \ 560| 67| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 67|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 67| ::std::void_t(); \ | | 1180| 67| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 67| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 67| constexpr auto _size = _name.size(); \ | | 1183| 67| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 67| return _nameof; }() ------------------ 561| 67| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 67|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 67| ::std::void_t(); \ | | 1180| 67| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 67| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 67| constexpr auto _size = _name.size(); \ | | 1183| 67| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 67| return _nameof; }() ------------------ 562| 67| c.fuzz(chardata); \ 563| 67| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 67| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 67| const FromSpan from{reinterpret_cast(chardata.data()), 178| 67| chardata.size() / sizeof(FromType)}; 179| | 180| 67| static const bool do_print_testcase = 181| 67| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 67| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 67] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 67| do { 189| | // step 0 - is the input valid? 190| 67| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 67| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 67] | 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| 67| From == UtfEncodings::UTF8) { 198| 67| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 67] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 67| } 201| | 202| | // step 2 - what is the required size of the output? 203| 67| const auto [output_length, length_agree] = 204| 67| calculate_length(from, inputisvalid); 205| 67| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 67] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 67| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 19, False: 48] | Branch (208:28): [True: 0, False: 19] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 67| const auto [written, outputs_agree] = 215| 67| do_conversion(from, output_length, inputisvalid); 216| 67| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 67] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 67| return; 221| 67| } 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| 67| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 230| 67| validation_result verify_valid_input(FromSpan src) const { 231| 67| validation_result ret{}; 232| | 233| 67| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 67| const auto implementations = get_supported_implementations(); 235| 67| std::vector results; 236| 67| results.reserve(implementations.size()); 237| | 238| 201| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 201, False: 67] ------------------ 239| 201| results.push_back( 240| 201| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 201| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 201| const bool validation2 = 245| 201| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 201| src.data(), src.size()); 247| 201| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 201] ------------------ 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| 201| } 258| | 259| 67| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 67| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 67] ------------------ 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| 67| } else { 273| 67| ret.implementations_agree = true; 274| 67| } 275| 67| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 67| return r.error == simdutf::SUCCESS; 277| 67| }); 278| 67| return ret; 279| 67| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 134| 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| 163| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 163| return r.error == simdutf::SUCCESS; 277| 163| }); _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEE: 289| 67| bool count_the_input(FromSpan src) const { 290| 67| const auto implementations = get_supported_implementations(); 291| 67| std::vector results; 292| 67| results.reserve(implementations.size()); 293| | 294| 201| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 201, False: 67] ------------------ 295| 201| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 201| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 201| 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| 201| results.push_back(ret); 304| 201| } 305| 67| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 67| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 67] ------------------ 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| 67| return true; 321| 67| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 134| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 67| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 67| length_result ret{}; 344| | 345| 67| const auto implementations = get_supported_implementations(); 346| 67| std::vector results; 347| 67| results.reserve(implementations.size()); 348| | 349| 201| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 201, False: 67] ------------------ 350| 201| const auto len = invoke_lengthcalc(impl, src); 351| 201| results.push_back(len); 352| 201| ret.length.push_back(len); 353| 201| } 354| | 355| 67| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 67| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 67] ------------------ 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| 67| } else { 375| 67| ret.implementations_agree = true; 376| 67| } 377| 67| return ret; 378| 67| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 201| FromSpan src) const { 339| 201| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 201| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 134| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 67| const bool inputisvalid) const { 383| 67| conversion_result ret{}; 384| | 385| 67| const auto implementations = get_supported_implementations(); 386| | 387| 67| std::vector> results; 388| 67| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 67| std::vector> outputbuffers; 393| 67| outputbuffers.reserve(implementations.size()); 394| 268| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 201, False: 67] ------------------ 395| 201| auto impl = implementations[i]; 396| 201| const ToType canary1{42}; 397| 201| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 201| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 201| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 201| const auto success = [](const ConversionResult& r) -> bool { 402| 201| if constexpr (std::is_same_v) { 403| 201| return r != 0; 404| 201| } else { 405| 201| return r.error == simdutf::error_code::SUCCESS; 406| 201| } 407| 201| }(implret1); 408| 201| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 201| 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| 201| const ToType canary2{25}; 414| 201| const auto outputbuffer_first_run = outputbuffer; 415| 201| std::ranges::fill(outputbuffer, canary2); 416| 201| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 201| src.size(), outputbuffer.data()); 418| | 419| 201| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 201] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 201| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 144, False: 57] | Branch (423:29): [True: 69, False: 75] ------------------ 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| 201| } 441| 201| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 69, False: 132] ------------------ 442| 201| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 67| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 19, False: 48] ------------------ 447| 57| for (auto& e : results) { ------------------ | Branch (447:20): [True: 57, False: 19] ------------------ 448| 57| e.outputhash.clear(); 449| 57| } 450| 19| } 451| | 452| 67| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 67| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 67] ------------------ 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| 67| } else { 474| 67| ret.implementations_agree = true; 475| 67| } 476| 67| return ret; 477| 67| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 201| const auto success = [](const ConversionResult& r) -> bool { 402| 201| if constexpr (std::is_same_v) { 403| 201| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 201| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 134| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_14clENSt3__14spanIKcLm18446744073709551615EEE: 554| 153| +[](std::span chardata) { \ 555| 153| const auto c = \ 556| 153| Conversion{ \ 559| 153| &I::lenfunc, &I::conversionfunc, \ 560| 153| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 153|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 153| ::std::void_t(); \ | | 1180| 153| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 153| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 153| constexpr auto _size = _name.size(); \ | | 1183| 153| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 153| return _nameof; }() ------------------ 561| 153| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 153|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 153| ::std::void_t(); \ | | 1180| 153| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 153| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 153| constexpr auto _size = _name.size(); \ | | 1183| 153| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 153| return _nameof; }() ------------------ 562| 153| c.fuzz(chardata); \ 563| 153| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_15clENSt3__14spanIKcLm18446744073709551615EEE: 554| 208| +[](std::span chardata) { \ 555| 208| const auto c = \ 556| 208| Conversion{ \ 559| 208| &I::lenfunc, &I::conversionfunc, \ 560| 208| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 208|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 208| ::std::void_t(); \ | | 1180| 208| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 208| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 208| constexpr auto _size = _name.size(); \ | | 1183| 208| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 208| return _nameof; }() ------------------ 561| 208| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 208|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 208| ::std::void_t(); \ | | 1180| 208| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 208| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 208| constexpr auto _size = _name.size(); \ | | 1183| 208| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 208| return _nameof; }() ------------------ 562| 208| c.fuzz(chardata); \ 563| 208| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_16clENSt3__14spanIKcLm18446744073709551615EEE: 554| 86| +[](std::span chardata) { \ 555| 86| const auto c = \ 556| 86| Conversion{ \ 559| 86| &I::lenfunc, &I::conversionfunc, \ 560| 86| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 86|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 86| ::std::void_t(); \ | | 1180| 86| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 86| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 86| constexpr auto _size = _name.size(); \ | | 1183| 86| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 86| return _nameof; }() ------------------ 561| 86| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 86|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 86| ::std::void_t(); \ | | 1180| 86| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 86| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 86| constexpr auto _size = _name.size(); \ | | 1183| 86| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 86| return _nameof; }() ------------------ 562| 86| c.fuzz(chardata); \ 563| 86| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 86| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 86| const FromSpan from{reinterpret_cast(chardata.data()), 178| 86| chardata.size() / sizeof(FromType)}; 179| | 180| 86| static const bool do_print_testcase = 181| 86| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 86| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 86] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 86| do { 189| | // step 0 - is the input valid? 190| 86| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 86| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 86] | 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| 86| const auto [output_length, length_agree] = 204| 86| calculate_length(from, inputisvalid); 205| 86| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 86] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 86| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 59, False: 27] | Branch (208:28): [True: 0, False: 59] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 86| const auto [written, outputs_agree] = 215| 86| do_conversion(from, output_length, inputisvalid); 216| 86| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 86] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 86| return; 221| 86| } 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| 86| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 230| 86| validation_result verify_valid_input(FromSpan src) const { 231| 86| validation_result ret{}; 232| | 233| 86| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 86| const auto implementations = get_supported_implementations(); 235| 86| std::vector results; 236| 86| results.reserve(implementations.size()); 237| | 238| 258| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 258, False: 86] ------------------ 239| 258| results.push_back( 240| 258| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 258| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 258| const bool validation2 = 245| 258| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 258| src.data(), src.size()); 247| 258| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 258] ------------------ 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| 258| } 258| | 259| 86| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 86| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 86] ------------------ 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| 86| } else { 273| 86| ret.implementations_agree = true; 274| 86| } 275| 86| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 86| return r.error == simdutf::SUCCESS; 277| 86| }); 278| 86| return ret; 279| 86| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 172| 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| 140| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 140| return r.error == simdutf::SUCCESS; 277| 140| }); _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 86| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 86| length_result ret{}; 344| | 345| 86| const auto implementations = get_supported_implementations(); 346| 86| std::vector results; 347| 86| results.reserve(implementations.size()); 348| | 349| 258| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 258, False: 86] ------------------ 350| 258| const auto len = invoke_lengthcalc(impl, src); 351| 258| results.push_back(len); 352| 258| ret.length.push_back(len); 353| 258| } 354| | 355| 86| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 86| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 86] ------------------ 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| 86| } else { 375| 86| ret.implementations_agree = true; 376| 86| } 377| 86| return ret; 378| 86| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 258| FromSpan src) const { 339| 258| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 258| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 172| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 86| const bool inputisvalid) const { 383| 86| conversion_result ret{}; 384| | 385| 86| const auto implementations = get_supported_implementations(); 386| | 387| 86| std::vector> results; 388| 86| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 86| std::vector> outputbuffers; 393| 86| outputbuffers.reserve(implementations.size()); 394| 344| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 258, False: 86] ------------------ 395| 258| auto impl = implementations[i]; 396| 258| const ToType canary1{42}; 397| 258| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 258| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 258| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 258| const auto success = [](const ConversionResult& r) -> bool { 402| 258| if constexpr (std::is_same_v) { 403| 258| return r != 0; 404| 258| } else { 405| 258| return r.error == simdutf::error_code::SUCCESS; 406| 258| } 407| 258| }(implret1); 408| 258| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 258| 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| 258| const ToType canary2{25}; 414| 258| const auto outputbuffer_first_run = outputbuffer; 415| 258| std::ranges::fill(outputbuffer, canary2); 416| 258| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 258| src.size(), outputbuffer.data()); 418| | 419| 258| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 258] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 258| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 81, False: 177] | Branch (423:29): [True: 63, False: 18] ------------------ 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| 258| } 441| 258| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 63, False: 195] ------------------ 442| 258| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 86| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 59, False: 27] ------------------ 447| 177| for (auto& e : results) { ------------------ | Branch (447:20): [True: 177, False: 59] ------------------ 448| 177| e.outputhash.clear(); 449| 177| } 450| 59| } 451| | 452| 86| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 86| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 86] ------------------ 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| 86| } else { 474| 86| ret.implementations_agree = true; 475| 86| } 476| 86| return ret; 477| 86| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 258| const auto success = [](const ConversionResult& r) -> bool { 402| 258| if constexpr (std::is_same_v) { 403| 258| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 258| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 172| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_17clENSt3__14spanIKcLm18446744073709551615EEE: 554| 267| +[](std::span chardata) { \ 555| 267| const auto c = \ 556| 267| Conversion{ \ 559| 267| &I::lenfunc, &I::conversionfunc, \ 560| 267| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 267|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 267| ::std::void_t(); \ | | 1180| 267| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 267| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 267| constexpr auto _size = _name.size(); \ | | 1183| 267| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 267| return _nameof; }() ------------------ 561| 267| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 267|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 267| ::std::void_t(); \ | | 1180| 267| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 267| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 267| constexpr auto _size = _name.size(); \ | | 1183| 267| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 267| return _nameof; }() ------------------ 562| 267| c.fuzz(chardata); \ 563| 267| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_18clENSt3__14spanIKcLm18446744073709551615EEE: 554| 263| +[](std::span chardata) { \ 555| 263| const auto c = \ 556| 263| Conversion{ \ 559| 263| &I::lenfunc, &I::conversionfunc, \ 560| 263| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 263|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 263| ::std::void_t(); \ | | 1180| 263| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 263| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 263| constexpr auto _size = _name.size(); \ | | 1183| 263| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 263| return _nameof; }() ------------------ 561| 263| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 263|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 263| ::std::void_t(); \ | | 1180| 263| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 263| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 263| constexpr auto _size = _name.size(); \ | | 1183| 263| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 263| return _nameof; }() ------------------ 562| 263| c.fuzz(chardata); \ 563| 263| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_19clENSt3__14spanIKcLm18446744073709551615EEE: 554| 355| +[](std::span chardata) { \ 555| 355| const auto c = \ 556| 355| Conversion{ \ 559| 355| &I::lenfunc, &I::conversionfunc, \ 560| 355| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 355|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 355| ::std::void_t(); \ | | 1180| 355| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 355| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 355| constexpr auto _size = _name.size(); \ | | 1183| 355| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 355| return _nameof; }() ------------------ 561| 355| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 355|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 355| ::std::void_t(); \ | | 1180| 355| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 355| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 355| constexpr auto _size = _name.size(); \ | | 1183| 355| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 355| return _nameof; }() ------------------ 562| 355| c.fuzz(chardata); \ 563| 355| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_20clENSt3__14spanIKcLm18446744073709551615EEE: 554| 295| +[](std::span chardata) { \ 555| 295| const auto c = \ 556| 295| Conversion{ \ 559| 295| &I::lenfunc, &I::conversionfunc, \ 560| 295| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 295|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 295| ::std::void_t(); \ | | 1180| 295| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 295| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 295| constexpr auto _size = _name.size(); \ | | 1183| 295| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 295| return _nameof; }() ------------------ 561| 295| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 295|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 295| ::std::void_t(); \ | | 1180| 295| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 295| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 295| constexpr auto _size = _name.size(); \ | | 1183| 295| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 295| return _nameof; }() ------------------ 562| 295| c.fuzz(chardata); \ 563| 295| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 295| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 295| const FromSpan from{reinterpret_cast(chardata.data()), 178| 295| chardata.size() / sizeof(FromType)}; 179| | 180| 295| static const bool do_print_testcase = 181| 295| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 295| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 295] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 295| do { 189| | // step 0 - is the input valid? 190| 295| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 295| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 295] | 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| 295| From == UtfEncodings::UTF8) { 198| 295| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 295] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 295| } 201| | 202| | // step 2 - what is the required size of the output? 203| 295| const auto [output_length, length_agree] = 204| 295| calculate_length(from, inputisvalid); 205| 295| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 295] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 295| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 196, False: 99] | Branch (208:28): [True: 0, False: 196] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 295| const auto [written, outputs_agree] = 215| 295| do_conversion(from, output_length, inputisvalid); 216| 295| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 295] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 295| return; 221| 295| } 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| 295| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 295| validation_result verify_valid_input(FromSpan src) const { 231| 295| validation_result ret{}; 232| | 233| 295| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 295| const auto implementations = get_supported_implementations(); 235| 295| std::vector results; 236| 295| results.reserve(implementations.size()); 237| | 238| 885| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 885, False: 295] ------------------ 239| 885| results.push_back( 240| 885| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 885| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 885| const bool validation2 = 245| 885| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 885| src.data(), src.size()); 247| 885| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 885] ------------------ 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| 885| } 258| | 259| 295| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 295| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 295] ------------------ 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| 295| } else { 273| 295| ret.implementations_agree = true; 274| 295| } 275| 295| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 295| return r.error == simdutf::SUCCESS; 277| 295| }); 278| 295| return ret; 279| 295| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 590| 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| 493| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 493| return r.error == simdutf::SUCCESS; 277| 493| }); _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 295| bool count_the_input(FromSpan src) const { 290| 295| const auto implementations = get_supported_implementations(); 291| 295| std::vector results; 292| 295| results.reserve(implementations.size()); 293| | 294| 885| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 885, False: 295] ------------------ 295| 885| 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| 885| } else if constexpr (From == UtfEncodings::UTF8) { 301| 885| ret = impl->count_utf8(src.data(), src.size()); 302| 885| } 303| 885| results.push_back(ret); 304| 885| } 305| 295| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 295| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 295] ------------------ 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| 295| return true; 321| 295| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 590| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 295| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 295| length_result ret{}; 344| | 345| 295| const auto implementations = get_supported_implementations(); 346| 295| std::vector results; 347| 295| results.reserve(implementations.size()); 348| | 349| 885| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 885, False: 295] ------------------ 350| 885| const auto len = invoke_lengthcalc(impl, src); 351| 885| results.push_back(len); 352| 885| ret.length.push_back(len); 353| 885| } 354| | 355| 295| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 295| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 295] ------------------ 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| 295| } else { 375| 295| ret.implementations_agree = true; 376| 295| } 377| 295| return ret; 378| 295| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 885| FromSpan src) const { 331| 885| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 885| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 590| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 295| const bool inputisvalid) const { 383| 295| conversion_result ret{}; 384| | 385| 295| const auto implementations = get_supported_implementations(); 386| | 387| 295| std::vector> results; 388| 295| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 295| std::vector> outputbuffers; 393| 295| outputbuffers.reserve(implementations.size()); 394| 1.18k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 885, False: 295] ------------------ 395| 885| auto impl = implementations[i]; 396| 885| const ToType canary1{42}; 397| 885| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 885| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 885| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 885| const auto success = [](const ConversionResult& r) -> bool { 402| 885| if constexpr (std::is_same_v) { 403| 885| return r != 0; 404| 885| } else { 405| 885| return r.error == simdutf::error_code::SUCCESS; 406| 885| } 407| 885| }(implret1); 408| 885| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 885| 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| 885| const ToType canary2{25}; 414| 885| const auto outputbuffer_first_run = outputbuffer; 415| 885| std::ranges::fill(outputbuffer, canary2); 416| 885| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 885| src.size(), outputbuffer.data()); 418| | 419| 885| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 885] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 885| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 297, False: 588] | Branch (423:29): [True: 267, False: 30] ------------------ 424| | // only care about the output if the input is valid 425| 267| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 267| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 267] ------------------ 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| 267| } 440| 885| } 441| 885| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 267, False: 618] ------------------ 442| 885| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 295| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 196, False: 99] ------------------ 447| 588| for (auto& e : results) { ------------------ | Branch (447:20): [True: 588, False: 196] ------------------ 448| 588| e.outputhash.clear(); 449| 588| } 450| 196| } 451| | 452| 295| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 295| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 295] ------------------ 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| 295| } else { 474| 295| ret.implementations_agree = true; 475| 295| } 476| 295| return ret; 477| 295| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 885| const auto success = [](const ConversionResult& r) -> bool { 402| 885| if constexpr (std::is_same_v) { 403| 885| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 885| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 590| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_21clENSt3__14spanIKcLm18446744073709551615EEE: 554| 378| +[](std::span chardata) { \ 555| 378| const auto c = \ 556| 378| Conversion{ \ 559| 378| &I::lenfunc, &I::conversionfunc, \ 560| 378| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 378|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 378| ::std::void_t(); \ | | 1180| 378| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 378| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 378| constexpr auto _size = _name.size(); \ | | 1183| 378| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 378| return _nameof; }() ------------------ 561| 378| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 378|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 378| ::std::void_t(); \ | | 1180| 378| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 378| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 378| constexpr auto _size = _name.size(); \ | | 1183| 378| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 378| return _nameof; }() ------------------ 562| 378| c.fuzz(chardata); \ 563| 378| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_22clENSt3__14spanIKcLm18446744073709551615EEE: 554| 360| +[](std::span chardata) { \ 555| 360| const auto c = \ 556| 360| Conversion{ \ 559| 360| &I::lenfunc, &I::conversionfunc, \ 560| 360| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 360|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 360| ::std::void_t(); \ | | 1180| 360| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 360| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 360| constexpr auto _size = _name.size(); \ | | 1183| 360| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 360| return _nameof; }() ------------------ 561| 360| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 360|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 360| ::std::void_t(); \ | | 1180| 360| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 360| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 360| constexpr auto _size = _name.size(); \ | | 1183| 360| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 360| return _nameof; }() ------------------ 562| 360| c.fuzz(chardata); \ 563| 360| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_23clENSt3__14spanIKcLm18446744073709551615EEE: 554| 392| +[](std::span chardata) { \ 555| 392| const auto c = \ 556| 392| Conversion{ \ 559| 392| &I::lenfunc, &I::conversionfunc, \ 560| 392| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 392|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 392| ::std::void_t(); \ | | 1180| 392| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 392| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 392| constexpr auto _size = _name.size(); \ | | 1183| 392| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 392| return _nameof; }() ------------------ 561| 392| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 392|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 392| ::std::void_t(); \ | | 1180| 392| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 392| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 392| constexpr auto _size = _name.size(); \ | | 1183| 392| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 392| return _nameof; }() ------------------ 562| 392| c.fuzz(chardata); \ 563| 392| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_24clENSt3__14spanIKcLm18446744073709551615EEE: 554| 125| +[](std::span chardata) { \ 555| 125| const auto c = \ 556| 125| Conversion{ \ 559| 125| &I::lenfunc, &I::conversionfunc, \ 560| 125| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 125|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 125| ::std::void_t(); \ | | 1180| 125| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 125| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 125| constexpr auto _size = _name.size(); \ | | 1183| 125| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 125| return _nameof; }() ------------------ 561| 125| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 125|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 125| ::std::void_t(); \ | | 1180| 125| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 125| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 125| constexpr auto _size = _name.size(); \ | | 1183| 125| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 125| return _nameof; }() ------------------ 562| 125| c.fuzz(chardata); \ 563| 125| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 125| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 125| const FromSpan from{reinterpret_cast(chardata.data()), 178| 125| chardata.size() / sizeof(FromType)}; 179| | 180| 125| static const bool do_print_testcase = 181| 125| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 125| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 125] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 125| do { 189| | // step 0 - is the input valid? 190| 125| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 125| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 125] | 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| 125| From == UtfEncodings::UTF8) { 198| 125| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 125] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 125| } 201| | 202| | // step 2 - what is the required size of the output? 203| 125| const auto [output_length, length_agree] = 204| 125| calculate_length(from, inputisvalid); 205| 125| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 125] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 125| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 20, False: 105] | Branch (208:28): [True: 0, False: 20] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 125| const auto [written, outputs_agree] = 215| 125| do_conversion(from, output_length, inputisvalid); 216| 125| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 125] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 125| return; 221| 125| } 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| 125| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEE: 230| 125| validation_result verify_valid_input(FromSpan src) const { 231| 125| validation_result ret{}; 232| | 233| 125| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 125| const auto implementations = get_supported_implementations(); 235| 125| std::vector results; 236| 125| results.reserve(implementations.size()); 237| | 238| 375| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 375, False: 125] ------------------ 239| 375| results.push_back( 240| 375| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 375| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 375| const bool validation2 = 245| 375| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 375| src.data(), src.size()); 247| 375| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 375] ------------------ 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| 375| } 258| | 259| 125| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 125| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 125] ------------------ 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| 125| } else { 273| 125| ret.implementations_agree = true; 274| 125| } 275| 125| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 125| return r.error == simdutf::SUCCESS; 277| 125| }); 278| 125| return ret; 279| 125| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS5_S5_EEDaSJ_SM_: 259| 250| 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| 335| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 335| return r.error == simdutf::SUCCESS; 277| 335| }); _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEE: 289| 125| bool count_the_input(FromSpan src) const { 290| 125| const auto implementations = get_supported_implementations(); 291| 125| std::vector results; 292| 125| results.reserve(implementations.size()); 293| | 294| 375| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 375, False: 125] ------------------ 295| 375| std::size_t ret; 296| 375| if constexpr (From == UtfEncodings::UTF16BE) { 297| 375| 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| 375| results.push_back(ret); 304| 375| } 305| 125| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 125| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 125] ------------------ 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| 125| return true; 321| 125| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 250| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEb: 342| 125| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 125| length_result ret{}; 344| | 345| 125| const auto implementations = get_supported_implementations(); 346| 125| std::vector results; 347| 125| results.reserve(implementations.size()); 348| | 349| 375| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 375, False: 125] ------------------ 350| 375| const auto len = invoke_lengthcalc(impl, src); 351| 375| results.push_back(len); 352| 375| ret.length.push_back(len); 353| 375| } 354| | 355| 125| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 125| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 125] ------------------ 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| 125| } else { 375| 125| ret.implementations_agree = true; 376| 125| } 377| 125| return ret; 378| 125| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSF_NSt3__14spanIS6_Lm18446744073709551615EEE: 338| 375| FromSpan src) const { 339| 375| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 375| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 250| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 125| const bool inputisvalid) const { 383| 125| conversion_result ret{}; 384| | 385| 125| const auto implementations = get_supported_implementations(); 386| | 387| 125| std::vector> results; 388| 125| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 125| std::vector> outputbuffers; 393| 125| outputbuffers.reserve(implementations.size()); 394| 500| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 375, False: 125] ------------------ 395| 375| auto impl = implementations[i]; 396| 375| const ToType canary1{42}; 397| 375| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 375| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 375| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 375| const auto success = [](const ConversionResult& r) -> bool { 402| 375| if constexpr (std::is_same_v) { 403| 375| return r != 0; 404| 375| } else { 405| 375| return r.error == simdutf::error_code::SUCCESS; 406| 375| } 407| 375| }(implret1); 408| 375| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 375| 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| 375| const ToType canary2{25}; 414| 375| const auto outputbuffer_first_run = outputbuffer; 415| 375| std::ranges::fill(outputbuffer, canary2); 416| 375| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 375| src.size(), outputbuffer.data()); 418| | 419| 375| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 375] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 375| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 315, False: 60] | Branch (423:29): [True: 129, False: 186] ------------------ 424| | // only care about the output if the input is valid 425| 129| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 129| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 129] ------------------ 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| 129| } 440| 375| } 441| 375| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 129, False: 246] ------------------ 442| 375| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 125| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 20, False: 105] ------------------ 447| 60| for (auto& e : results) { ------------------ | Branch (447:20): [True: 60, False: 20] ------------------ 448| 60| e.outputhash.clear(); 449| 60| } 450| 20| } 451| | 452| 125| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 125| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 125] ------------------ 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| 125| } else { 474| 125| ret.implementations_agree = true; 475| 125| } 476| 125| return ret; 477| 125| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS5_E_clESM_: 401| 375| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 375| } else { 405| 375| return r.error == simdutf::error_code::SUCCESS; 406| 375| } 407| 375| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS5_ESU_EEDaSN_SQ_: 452| 250| 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: 90, False: 83] | 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| 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| 339| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 339| return r.error == simdutf::SUCCESS; 277| 339| }); _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: 249, False: 270] | Branch (423:29): [True: 249, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 249| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 249| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 249] ------------------ 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| 249| } 440| 519| } 441| 519| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 249, False: 270] ------------------ 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: 90, False: 83] ------------------ 447| 270| for (auto& e : results) { ------------------ | Branch (447:20): [True: 270, False: 90] ------------------ 448| 270| e.outputhash.clear(); 449| 270| } 450| 90| } 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| 364| +[](std::span chardata) { \ 555| 364| const auto c = \ 556| 364| Conversion{ \ 559| 364| &I::lenfunc, &I::conversionfunc, \ 560| 364| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 364|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 364| ::std::void_t(); \ | | 1180| 364| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 364| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 364| constexpr auto _size = _name.size(); \ | | 1183| 364| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 364| return _nameof; }() ------------------ 561| 364| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 364|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 364| ::std::void_t(); \ | | 1180| 364| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 364| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 364| constexpr auto _size = _name.size(); \ | | 1183| 364| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 364| return _nameof; }() ------------------ 562| 364| c.fuzz(chardata); \ 563| 364| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 364| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 364| const FromSpan from{reinterpret_cast(chardata.data()), 178| 364| chardata.size() / sizeof(FromType)}; 179| | 180| 364| static const bool do_print_testcase = 181| 364| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 364| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 364] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 364| do { 189| | // step 0 - is the input valid? 190| 364| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 364| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 364] | 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| 364| From == UtfEncodings::UTF8) { 198| 364| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 364] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 364| } 201| | 202| | // step 2 - what is the required size of the output? 203| 364| const auto [output_length, length_agree] = 204| 364| calculate_length(from, inputisvalid); 205| 364| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 364] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 364| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 153, False: 211] | Branch (208:28): [True: 0, False: 153] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 364| const auto [written, outputs_agree] = 215| 364| do_conversion(from, output_length, inputisvalid); 216| 364| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 364] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 364| return; 221| 364| } 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| 364| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 364| validation_result verify_valid_input(FromSpan src) const { 231| 364| validation_result ret{}; 232| | 233| 364| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 364| const auto implementations = get_supported_implementations(); 235| 364| std::vector results; 236| 364| results.reserve(implementations.size()); 237| | 238| 1.09k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.09k, False: 364] ------------------ 239| 1.09k| results.push_back( 240| 1.09k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.09k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.09k| const bool validation2 = 245| 1.09k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.09k| src.data(), src.size()); 247| 1.09k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.09k] ------------------ 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| 1.09k| } 258| | 259| 364| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 364| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 364] ------------------ 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| 364| } else { 273| 364| ret.implementations_agree = true; 274| 364| } 275| 364| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 364| return r.error == simdutf::SUCCESS; 277| 364| }); 278| 364| return ret; 279| 364| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 728| 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| 786| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 786| return r.error == simdutf::SUCCESS; 277| 786| }); _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 364| bool count_the_input(FromSpan src) const { 290| 364| const auto implementations = get_supported_implementations(); 291| 364| std::vector results; 292| 364| results.reserve(implementations.size()); 293| | 294| 1.09k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.09k, False: 364] ------------------ 295| 1.09k| std::size_t ret; 296| 1.09k| if constexpr (From == UtfEncodings::UTF16BE) { 297| 1.09k| 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| 1.09k| results.push_back(ret); 304| 1.09k| } 305| 364| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 364| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 364] ------------------ 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| 364| return true; 321| 364| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 728| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 364| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 364| length_result ret{}; 344| | 345| 364| const auto implementations = get_supported_implementations(); 346| 364| std::vector results; 347| 364| results.reserve(implementations.size()); 348| | 349| 1.09k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.09k, False: 364] ------------------ 350| 1.09k| const auto len = invoke_lengthcalc(impl, src); 351| 1.09k| results.push_back(len); 352| 1.09k| ret.length.push_back(len); 353| 1.09k| } 354| | 355| 364| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 364| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 364] ------------------ 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| 364| } else { 375| 364| ret.implementations_agree = true; 376| 364| } 377| 364| return ret; 378| 364| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.09k| FromSpan src) const { 331| 1.09k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.09k| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 728| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 364| const bool inputisvalid) const { 383| 364| conversion_result ret{}; 384| | 385| 364| const auto implementations = get_supported_implementations(); 386| | 387| 364| std::vector> results; 388| 364| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 364| std::vector> outputbuffers; 393| 364| outputbuffers.reserve(implementations.size()); 394| 1.45k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.09k, False: 364] ------------------ 395| 1.09k| auto impl = implementations[i]; 396| 1.09k| const ToType canary1{42}; 397| 1.09k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.09k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.09k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.09k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.09k| if constexpr (std::is_same_v) { 403| 1.09k| return r != 0; 404| 1.09k| } else { 405| 1.09k| return r.error == simdutf::error_code::SUCCESS; 406| 1.09k| } 407| 1.09k| }(implret1); 408| 1.09k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.09k| 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| 1.09k| const ToType canary2{25}; 414| 1.09k| const auto outputbuffer_first_run = outputbuffer; 415| 1.09k| std::ranges::fill(outputbuffer, canary2); 416| 1.09k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.09k| src.size(), outputbuffer.data()); 418| | 419| 1.09k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.09k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.09k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 633, False: 459] | Branch (423:29): [True: 633, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 633| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 633| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 633] ------------------ 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| 633| } 440| 1.09k| } 441| 1.09k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 633, False: 459] ------------------ 442| 1.09k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 364| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 153, False: 211] ------------------ 447| 459| for (auto& e : results) { ------------------ | Branch (447:20): [True: 459, False: 153] ------------------ 448| 459| e.outputhash.clear(); 449| 459| } 450| 153| } 451| | 452| 364| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 364| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 364] ------------------ 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| 364| } else { 474| 364| ret.implementations_agree = true; 475| 364| } 476| 364| return ret; 477| 364| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.09k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.09k| } else { 405| 1.09k| return r.error == simdutf::error_code::SUCCESS; 406| 1.09k| } 407| 1.09k| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 728| 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: 39, False: 100] | Branch (208:28): [True: 0, False: 39] ------------------ 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| 339| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 339| return r.error == simdutf::SUCCESS; 277| 339| }); _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: 300, False: 117] | Branch (423:29): [True: 150, False: 150] ------------------ 424| | // only care about the output if the input is valid 425| 150| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 150| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 150] ------------------ 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| 150| } 440| 417| } 441| 417| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 150, False: 267] ------------------ 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: 39, False: 100] ------------------ 447| 117| for (auto& e : results) { ------------------ | Branch (447:20): [True: 117, False: 39] ------------------ 448| 117| e.outputhash.clear(); 449| 117| } 450| 39| } 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| 209| +[](std::span chardata) { \ 555| 209| const auto c = \ 556| 209| Conversion{ \ 559| 209| &I::lenfunc, &I::conversionfunc, \ 560| 209| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 209|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 209| ::std::void_t(); \ | | 1180| 209| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 209| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 209| constexpr auto _size = _name.size(); \ | | 1183| 209| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 209| return _nameof; }() ------------------ 561| 209| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 209|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 209| ::std::void_t(); \ | | 1180| 209| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 209| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 209| constexpr auto _size = _name.size(); \ | | 1183| 209| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 209| return _nameof; }() ------------------ 562| 209| c.fuzz(chardata); \ 563| 209| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 209| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 209| const FromSpan from{reinterpret_cast(chardata.data()), 178| 209| chardata.size() / sizeof(FromType)}; 179| | 180| 209| static const bool do_print_testcase = 181| 209| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 209| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 209] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 209| do { 189| | // step 0 - is the input valid? 190| 209| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 209| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 209] | 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| 209| From == UtfEncodings::UTF8) { 198| 209| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 209] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 209| } 201| | 202| | // step 2 - what is the required size of the output? 203| 209| const auto [output_length, length_agree] = 204| 209| calculate_length(from, inputisvalid); 205| 209| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 209] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 209| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 118, False: 91] | Branch (208:28): [True: 0, False: 118] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 209| const auto [written, outputs_agree] = 215| 209| do_conversion(from, output_length, inputisvalid); 216| 209| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 209] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 209| return; 221| 209| } 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| 209| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 209| validation_result verify_valid_input(FromSpan src) const { 231| 209| validation_result ret{}; 232| | 233| 209| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 209| const auto implementations = get_supported_implementations(); 235| 209| std::vector results; 236| 209| results.reserve(implementations.size()); 237| | 238| 627| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 627, False: 209] ------------------ 239| 627| results.push_back( 240| 627| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 627| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 627| const bool validation2 = 245| 627| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 627| src.data(), src.size()); 247| 627| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 627] ------------------ 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| 627| } 258| | 259| 209| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 209| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 209] ------------------ 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| 209| } else { 273| 209| ret.implementations_agree = true; 274| 209| } 275| 209| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 209| return r.error == simdutf::SUCCESS; 277| 209| }); 278| 209| return ret; 279| 209| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 418| 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| 391| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 391| return r.error == simdutf::SUCCESS; 277| 391| }); _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 209| bool count_the_input(FromSpan src) const { 290| 209| const auto implementations = get_supported_implementations(); 291| 209| std::vector results; 292| 209| results.reserve(implementations.size()); 293| | 294| 627| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 627, False: 209] ------------------ 295| 627| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 627| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 627| 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| 627| results.push_back(ret); 304| 627| } 305| 209| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 209| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 209] ------------------ 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| 209| return true; 321| 209| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 418| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 209| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 209| length_result ret{}; 344| | 345| 209| const auto implementations = get_supported_implementations(); 346| 209| std::vector results; 347| 209| results.reserve(implementations.size()); 348| | 349| 627| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 627, False: 209] ------------------ 350| 627| const auto len = invoke_lengthcalc(impl, src); 351| 627| results.push_back(len); 352| 627| ret.length.push_back(len); 353| 627| } 354| | 355| 209| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 209| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 209] ------------------ 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| 209| } else { 375| 209| ret.implementations_agree = true; 376| 209| } 377| 209| return ret; 378| 209| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 627| FromSpan src) const { 331| 627| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 627| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 418| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 209| const bool inputisvalid) const { 383| 209| conversion_result ret{}; 384| | 385| 209| const auto implementations = get_supported_implementations(); 386| | 387| 209| std::vector> results; 388| 209| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 209| std::vector> outputbuffers; 393| 209| outputbuffers.reserve(implementations.size()); 394| 836| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 627, False: 209] ------------------ 395| 627| auto impl = implementations[i]; 396| 627| const ToType canary1{42}; 397| 627| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 627| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 627| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 627| const auto success = [](const ConversionResult& r) -> bool { 402| 627| if constexpr (std::is_same_v) { 403| 627| return r != 0; 404| 627| } else { 405| 627| return r.error == simdutf::error_code::SUCCESS; 406| 627| } 407| 627| }(implret1); 408| 627| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 627| 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| 627| const ToType canary2{25}; 414| 627| const auto outputbuffer_first_run = outputbuffer; 415| 627| std::ranges::fill(outputbuffer, canary2); 416| 627| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 627| src.size(), outputbuffer.data()); 418| | 419| 627| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 627] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 627| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 273, False: 354] | 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| 627| } 441| 627| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 273, False: 354] ------------------ 442| 627| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 209| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 118, False: 91] ------------------ 447| 354| for (auto& e : results) { ------------------ | Branch (447:20): [True: 354, False: 118] ------------------ 448| 354| e.outputhash.clear(); 449| 354| } 450| 118| } 451| | 452| 209| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 209| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 209] ------------------ 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| 209| } else { 474| 209| ret.implementations_agree = true; 475| 209| } 476| 209| return ret; 477| 209| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 627| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 627| } else { 405| 627| return r.error == simdutf::error_code::SUCCESS; 406| 627| } 407| 627| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 418| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_29clENSt3__14spanIKcLm18446744073709551615EEE: 554| 332| +[](std::span chardata) { \ 555| 332| const auto c = \ 556| 332| Conversion{ \ 559| 332| &I::lenfunc, &I::conversionfunc, \ 560| 332| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 332|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 332| ::std::void_t(); \ | | 1180| 332| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 332| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 332| constexpr auto _size = _name.size(); \ | | 1183| 332| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 332| return _nameof; }() ------------------ 561| 332| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 332|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 332| ::std::void_t(); \ | | 1180| 332| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 332| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 332| constexpr auto _size = _name.size(); \ | | 1183| 332| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 332| return _nameof; }() ------------------ 562| 332| c.fuzz(chardata); \ 563| 332| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 332| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 332| const FromSpan from{reinterpret_cast(chardata.data()), 178| 332| chardata.size() / sizeof(FromType)}; 179| | 180| 332| static const bool do_print_testcase = 181| 332| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 332| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 332] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 332| do { 189| | // step 0 - is the input valid? 190| 332| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 332| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 332] | 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| 332| From == UtfEncodings::UTF8) { 198| 332| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 332] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 332| } 201| | 202| | // step 2 - what is the required size of the output? 203| 332| const auto [output_length, length_agree] = 204| 332| calculate_length(from, inputisvalid); 205| 332| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 332] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 332| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 135, False: 197] | Branch (208:28): [True: 0, False: 135] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 332| const auto [written, outputs_agree] = 215| 332| do_conversion(from, output_length, inputisvalid); 216| 332| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 332] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 332| return; 221| 332| } 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| 332| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 332| validation_result verify_valid_input(FromSpan src) const { 231| 332| validation_result ret{}; 232| | 233| 332| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 332| const auto implementations = get_supported_implementations(); 235| 332| std::vector results; 236| 332| results.reserve(implementations.size()); 237| | 238| 996| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 996, False: 332] ------------------ 239| 996| results.push_back( 240| 996| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 996| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 996| const bool validation2 = 245| 996| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 996| src.data(), src.size()); 247| 996| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 996] ------------------ 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| 996| } 258| | 259| 332| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 332| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 332] ------------------ 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| 332| } else { 273| 332| ret.implementations_agree = true; 274| 332| } 275| 332| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 332| return r.error == simdutf::SUCCESS; 277| 332| }); 278| 332| return ret; 279| 332| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 664| 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| 726| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 726| return r.error == simdutf::SUCCESS; 277| 726| }); _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 332| bool count_the_input(FromSpan src) const { 290| 332| const auto implementations = get_supported_implementations(); 291| 332| std::vector results; 292| 332| results.reserve(implementations.size()); 293| | 294| 996| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 996, False: 332] ------------------ 295| 996| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 996| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 996| 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| 996| results.push_back(ret); 304| 996| } 305| 332| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 332| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 332] ------------------ 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| 332| return true; 321| 332| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 664| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 332| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 332| length_result ret{}; 344| | 345| 332| const auto implementations = get_supported_implementations(); 346| 332| std::vector results; 347| 332| results.reserve(implementations.size()); 348| | 349| 996| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 996, False: 332] ------------------ 350| 996| const auto len = invoke_lengthcalc(impl, src); 351| 996| results.push_back(len); 352| 996| ret.length.push_back(len); 353| 996| } 354| | 355| 332| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 332| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 332] ------------------ 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| 332| } else { 375| 332| ret.implementations_agree = true; 376| 332| } 377| 332| return ret; 378| 332| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 996| FromSpan src) const { 331| 996| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 996| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 664| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 332| const bool inputisvalid) const { 383| 332| conversion_result ret{}; 384| | 385| 332| const auto implementations = get_supported_implementations(); 386| | 387| 332| std::vector> results; 388| 332| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 332| std::vector> outputbuffers; 393| 332| outputbuffers.reserve(implementations.size()); 394| 1.32k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 996, False: 332] ------------------ 395| 996| auto impl = implementations[i]; 396| 996| const ToType canary1{42}; 397| 996| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 996| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 996| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 996| const auto success = [](const ConversionResult& r) -> bool { 402| 996| if constexpr (std::is_same_v) { 403| 996| return r != 0; 404| 996| } else { 405| 996| return r.error == simdutf::error_code::SUCCESS; 406| 996| } 407| 996| }(implret1); 408| 996| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 996| 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| 996| const ToType canary2{25}; 414| 996| const auto outputbuffer_first_run = outputbuffer; 415| 996| std::ranges::fill(outputbuffer, canary2); 416| 996| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 996| src.size(), outputbuffer.data()); 418| | 419| 996| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 996] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 996| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 591, False: 405] | Branch (423:29): [True: 591, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 591| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 591| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 591] ------------------ 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| 591| } 440| 996| } 441| 996| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 591, False: 405] ------------------ 442| 996| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 332| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 135, False: 197] ------------------ 447| 405| for (auto& e : results) { ------------------ | Branch (447:20): [True: 405, False: 135] ------------------ 448| 405| e.outputhash.clear(); 449| 405| } 450| 135| } 451| | 452| 332| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 332| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 332] ------------------ 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| 332| } else { 474| 332| ret.implementations_agree = true; 475| 332| } 476| 332| return ret; 477| 332| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 996| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 996| } else { 405| 996| return r.error == simdutf::error_code::SUCCESS; 406| 996| } 407| 996| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 664| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_30clENSt3__14spanIKcLm18446744073709551615EEE: 554| 204| +[](std::span chardata) { \ 555| 204| const auto c = \ 556| 204| Conversion{ \ 559| 204| &I::lenfunc, &I::conversionfunc, \ 560| 204| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 204|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 204| ::std::void_t(); \ | | 1180| 204| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 204| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 204| constexpr auto _size = _name.size(); \ | | 1183| 204| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 204| return _nameof; }() ------------------ 561| 204| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 204|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 204| ::std::void_t(); \ | | 1180| 204| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 204| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 204| constexpr auto _size = _name.size(); \ | | 1183| 204| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 204| return _nameof; }() ------------------ 562| 204| c.fuzz(chardata); \ 563| 204| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 204| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 204| const FromSpan from{reinterpret_cast(chardata.data()), 178| 204| chardata.size() / sizeof(FromType)}; 179| | 180| 204| static const bool do_print_testcase = 181| 204| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 204| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 204] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 204| do { 189| | // step 0 - is the input valid? 190| 204| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 204| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 204] | 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| 204| const auto [output_length, length_agree] = 204| 204| calculate_length(from, inputisvalid); 205| 204| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 204] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 204| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 173, False: 31] | Branch (208:28): [True: 0, False: 173] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 204| const auto [written, outputs_agree] = 215| 204| do_conversion(from, output_length, inputisvalid); 216| 204| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 204] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 204| return; 221| 204| } 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| 204| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEE: 230| 204| validation_result verify_valid_input(FromSpan src) const { 231| 204| validation_result ret{}; 232| | 233| 204| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 204| const auto implementations = get_supported_implementations(); 235| 204| std::vector results; 236| 204| results.reserve(implementations.size()); 237| | 238| 612| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 612, False: 204] ------------------ 239| 612| results.push_back( 240| 612| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 612| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 612| const bool validation2 = 245| 612| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 612| src.data(), src.size()); 247| 612| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 612] ------------------ 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| 612| } 258| | 259| 204| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 204| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 204] ------------------ 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| 204| } else { 273| 204| ret.implementations_agree = true; 274| 204| } 275| 204| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 204| return r.error == simdutf::SUCCESS; 277| 204| }); 278| 204| return ret; 279| 204| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS5_S5_EEDaSJ_SM_: 259| 408| 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| 266| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 266| return r.error == simdutf::SUCCESS; 277| 266| }); _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEb: 342| 204| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 204| length_result ret{}; 344| | 345| 204| const auto implementations = get_supported_implementations(); 346| 204| std::vector results; 347| 204| results.reserve(implementations.size()); 348| | 349| 612| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 612, False: 204] ------------------ 350| 612| const auto len = invoke_lengthcalc(impl, src); 351| 612| results.push_back(len); 352| 612| ret.length.push_back(len); 353| 612| } 354| | 355| 204| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 204| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 204] ------------------ 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| 204| } else { 375| 204| ret.implementations_agree = true; 376| 204| } 377| 204| return ret; 378| 204| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSF_NSt3__14spanIS6_Lm18446744073709551615EEE: 338| 612| FromSpan src) const { 339| 612| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 612| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 408| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 204| const bool inputisvalid) const { 383| 204| conversion_result ret{}; 384| | 385| 204| const auto implementations = get_supported_implementations(); 386| | 387| 204| std::vector> results; 388| 204| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 204| std::vector> outputbuffers; 393| 204| outputbuffers.reserve(implementations.size()); 394| 816| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 612, False: 204] ------------------ 395| 612| auto impl = implementations[i]; 396| 612| const ToType canary1{42}; 397| 612| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 612| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 612| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 612| const auto success = [](const ConversionResult& r) -> bool { 402| 612| if constexpr (std::is_same_v) { 403| 612| return r != 0; 404| 612| } else { 405| 612| return r.error == simdutf::error_code::SUCCESS; 406| 612| } 407| 612| }(implret1); 408| 612| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 612| 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| 612| const ToType canary2{25}; 414| 612| const auto outputbuffer_first_run = outputbuffer; 415| 612| std::ranges::fill(outputbuffer, canary2); 416| 612| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 612| src.size(), outputbuffer.data()); 418| | 419| 612| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 612] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 612| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 93, False: 519] | Branch (423:29): [True: 69, False: 24] ------------------ 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| 612| } 441| 612| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 69, False: 543] ------------------ 442| 612| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 204| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 173, False: 31] ------------------ 447| 519| for (auto& e : results) { ------------------ | Branch (447:20): [True: 519, False: 173] ------------------ 448| 519| e.outputhash.clear(); 449| 519| } 450| 173| } 451| | 452| 204| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 204| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 204] ------------------ 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| 204| } else { 474| 204| ret.implementations_agree = true; 475| 204| } 476| 204| return ret; 477| 204| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS5_E_clESM_: 401| 612| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 612| } else { 405| 612| return r.error == simdutf::error_code::SUCCESS; 406| 612| } 407| 612| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS5_ESU_EEDaSN_SQ_: 452| 408| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_31clENSt3__14spanIKcLm18446744073709551615EEE: 554| 315| +[](std::span chardata) { \ 555| 315| const auto c = \ 556| 315| Conversion{ \ 559| 315| &I::lenfunc, &I::conversionfunc, \ 560| 315| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 315|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 315| ::std::void_t(); \ | | 1180| 315| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 315| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 315| constexpr auto _size = _name.size(); \ | | 1183| 315| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 315| return _nameof; }() ------------------ 561| 315| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 315|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 315| ::std::void_t(); \ | | 1180| 315| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 315| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 315| constexpr auto _size = _name.size(); \ | | 1183| 315| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 315| return _nameof; }() ------------------ 562| 315| c.fuzz(chardata); \ 563| 315| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 315| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 315| const FromSpan from{reinterpret_cast(chardata.data()), 178| 315| chardata.size() / sizeof(FromType)}; 179| | 180| 315| static const bool do_print_testcase = 181| 315| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 315| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 315] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 315| do { 189| | // step 0 - is the input valid? 190| 315| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 315| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 315] | 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| 315| const auto [output_length, length_agree] = 204| 315| calculate_length(from, inputisvalid); 205| 315| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 315] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 315| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 229, False: 86] | Branch (208:28): [True: 0, False: 229] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 315| const auto [written, outputs_agree] = 215| 315| do_conversion(from, output_length, inputisvalid); 216| 315| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 315] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 315| return; 221| 315| } 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| 315| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 315| validation_result verify_valid_input(FromSpan src) const { 231| 315| validation_result ret{}; 232| | 233| 315| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 315| const auto implementations = get_supported_implementations(); 235| 315| std::vector results; 236| 315| results.reserve(implementations.size()); 237| | 238| 945| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 945, False: 315] ------------------ 239| 945| results.push_back( 240| 945| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 945| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 945| const bool validation2 = 245| 945| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 945| src.data(), src.size()); 247| 945| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 945] ------------------ 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| 945| } 258| | 259| 315| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 315| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 315] ------------------ 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| 315| } else { 273| 315| ret.implementations_agree = true; 274| 315| } 275| 315| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 315| return r.error == simdutf::SUCCESS; 277| 315| }); 278| 315| return ret; 279| 315| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 630| 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| 487| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 487| return r.error == simdutf::SUCCESS; 277| 487| }); _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 315| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 315| length_result ret{}; 344| | 345| 315| const auto implementations = get_supported_implementations(); 346| 315| std::vector results; 347| 315| results.reserve(implementations.size()); 348| | 349| 945| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 945, False: 315] ------------------ 350| 945| const auto len = invoke_lengthcalc(impl, src); 351| 945| results.push_back(len); 352| 945| ret.length.push_back(len); 353| 945| } 354| | 355| 315| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 315| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 315] ------------------ 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| 315| } else { 375| 315| ret.implementations_agree = true; 376| 315| } 377| 315| return ret; 378| 315| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 945| FromSpan src) const { 331| 945| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 945| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 630| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 315| const bool inputisvalid) const { 383| 315| conversion_result ret{}; 384| | 385| 315| const auto implementations = get_supported_implementations(); 386| | 387| 315| std::vector> results; 388| 315| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 315| std::vector> outputbuffers; 393| 315| outputbuffers.reserve(implementations.size()); 394| 1.26k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 945, False: 315] ------------------ 395| 945| auto impl = implementations[i]; 396| 945| const ToType canary1{42}; 397| 945| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 945| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 945| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 945| const auto success = [](const ConversionResult& r) -> bool { 402| 945| if constexpr (std::is_same_v) { 403| 945| return r != 0; 404| 945| } else { 405| 945| return r.error == simdutf::error_code::SUCCESS; 406| 945| } 407| 945| }(implret1); 408| 945| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 945| 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| 945| const ToType canary2{25}; 414| 945| const auto outputbuffer_first_run = outputbuffer; 415| 945| std::ranges::fill(outputbuffer, canary2); 416| 945| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 945| src.size(), outputbuffer.data()); 418| | 419| 945| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 945] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 945| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 258, False: 687] | Branch (423:29): [True: 258, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 258| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 258| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 258] ------------------ 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| 258| } 440| 945| } 441| 945| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 258, False: 687] ------------------ 442| 945| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 315| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 229, False: 86] ------------------ 447| 687| for (auto& e : results) { ------------------ | Branch (447:20): [True: 687, False: 229] ------------------ 448| 687| e.outputhash.clear(); 449| 687| } 450| 229| } 451| | 452| 315| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 315| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 315] ------------------ 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| 315| } else { 474| 315| ret.implementations_agree = true; 475| 315| } 476| 315| return ret; 477| 315| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 945| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 945| } else { 405| 945| return r.error == simdutf::error_code::SUCCESS; 406| 945| } 407| 945| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 630| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_32clENSt3__14spanIKcLm18446744073709551615EEE: 554| 342| +[](std::span chardata) { \ 555| 342| const auto c = \ 556| 342| Conversion{ \ 559| 342| &I::lenfunc, &I::conversionfunc, \ 560| 342| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 342|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 342| ::std::void_t(); \ | | 1180| 342| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 342| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 342| constexpr auto _size = _name.size(); \ | | 1183| 342| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 342| return _nameof; }() ------------------ 561| 342| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 342|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 342| ::std::void_t(); \ | | 1180| 342| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 342| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 342| constexpr auto _size = _name.size(); \ | | 1183| 342| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 342| return _nameof; }() ------------------ 562| 342| c.fuzz(chardata); \ 563| 342| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 342| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 342| const FromSpan from{reinterpret_cast(chardata.data()), 178| 342| chardata.size() / sizeof(FromType)}; 179| | 180| 342| static const bool do_print_testcase = 181| 342| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 342| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 342] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 342| do { 189| | // step 0 - is the input valid? 190| 342| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 342| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 342] | 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| 342| const auto [output_length, length_agree] = 204| 342| calculate_length(from, inputisvalid); 205| 342| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 342] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 342| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 252, False: 90] | 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| 342| const auto [written, outputs_agree] = 215| 342| do_conversion(from, output_length, inputisvalid); 216| 342| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 342] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 342| return; 221| 342| } 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| 342| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 342| validation_result verify_valid_input(FromSpan src) const { 231| 342| validation_result ret{}; 232| | 233| 342| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 342| const auto implementations = get_supported_implementations(); 235| 342| std::vector results; 236| 342| results.reserve(implementations.size()); 237| | 238| 1.02k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.02k, False: 342] ------------------ 239| 1.02k| results.push_back( 240| 1.02k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.02k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.02k| const bool validation2 = 245| 1.02k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.02k| src.data(), src.size()); 247| 1.02k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.02k] ------------------ 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| 1.02k| } 258| | 259| 342| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 342| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 342] ------------------ 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| 342| } else { 273| 342| ret.implementations_agree = true; 274| 342| } 275| 342| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 342| return r.error == simdutf::SUCCESS; 277| 342| }); 278| 342| return ret; 279| 342| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 684| 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| 522| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 522| return r.error == simdutf::SUCCESS; 277| 522| }); _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 342| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 342| length_result ret{}; 344| | 345| 342| const auto implementations = get_supported_implementations(); 346| 342| std::vector results; 347| 342| results.reserve(implementations.size()); 348| | 349| 1.02k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.02k, False: 342] ------------------ 350| 1.02k| const auto len = invoke_lengthcalc(impl, src); 351| 1.02k| results.push_back(len); 352| 1.02k| ret.length.push_back(len); 353| 1.02k| } 354| | 355| 342| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 342| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 342] ------------------ 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| 342| } else { 375| 342| ret.implementations_agree = true; 376| 342| } 377| 342| return ret; 378| 342| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.02k| FromSpan src) const { 331| 1.02k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.02k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 684| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 342| const bool inputisvalid) const { 383| 342| conversion_result ret{}; 384| | 385| 342| const auto implementations = get_supported_implementations(); 386| | 387| 342| std::vector> results; 388| 342| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 342| std::vector> outputbuffers; 393| 342| outputbuffers.reserve(implementations.size()); 394| 1.36k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.02k, False: 342] ------------------ 395| 1.02k| auto impl = implementations[i]; 396| 1.02k| const ToType canary1{42}; 397| 1.02k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.02k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.02k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.02k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.02k| if constexpr (std::is_same_v) { 403| 1.02k| return r != 0; 404| 1.02k| } else { 405| 1.02k| return r.error == simdutf::error_code::SUCCESS; 406| 1.02k| } 407| 1.02k| }(implret1); 408| 1.02k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.02k| 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| 1.02k| const ToType canary2{25}; 414| 1.02k| const auto outputbuffer_first_run = outputbuffer; 415| 1.02k| std::ranges::fill(outputbuffer, canary2); 416| 1.02k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.02k| src.size(), outputbuffer.data()); 418| | 419| 1.02k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.02k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.02k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 270, False: 756] | Branch (423:29): [True: 270, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 270| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 270| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 270] ------------------ 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| 270| } 440| 1.02k| } 441| 1.02k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 270, False: 756] ------------------ 442| 1.02k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 342| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 252, False: 90] ------------------ 447| 756| for (auto& e : results) { ------------------ | Branch (447:20): [True: 756, False: 252] ------------------ 448| 756| e.outputhash.clear(); 449| 756| } 450| 252| } 451| | 452| 342| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 342| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 342] ------------------ 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| 342| } else { 474| 342| ret.implementations_agree = true; 475| 342| } 476| 342| return ret; 477| 342| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.02k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.02k| } else { 405| 1.02k| return r.error == simdutf::error_code::SUCCESS; 406| 1.02k| } 407| 1.02k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 684| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_33clENSt3__14spanIKcLm18446744073709551615EEE: 554| 467| +[](std::span chardata) { \ 555| 467| const auto c = \ 556| 467| Conversion{ \ 559| 467| &I::lenfunc, &I::conversionfunc, \ 560| 467| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 467|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 467| ::std::void_t(); \ | | 1180| 467| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 467| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 467| constexpr auto _size = _name.size(); \ | | 1183| 467| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 467| return _nameof; }() ------------------ 561| 467| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 467|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 467| ::std::void_t(); \ | | 1180| 467| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 467| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 467| constexpr auto _size = _name.size(); \ | | 1183| 467| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 467| return _nameof; }() ------------------ 562| 467| c.fuzz(chardata); \ 563| 467| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 467| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 467| const FromSpan from{reinterpret_cast(chardata.data()), 178| 467| chardata.size() / sizeof(FromType)}; 179| | 180| 467| static const bool do_print_testcase = 181| 467| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 467| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 467] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 467| do { 189| | // step 0 - is the input valid? 190| 467| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 467| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 467] | 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| 467| const auto [output_length, length_agree] = 204| 467| calculate_length(from, inputisvalid); 205| 467| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 467] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 467| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 350, False: 117] | Branch (208:28): [True: 0, False: 350] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 467| const auto [written, outputs_agree] = 215| 467| do_conversion(from, output_length, inputisvalid); 216| 467| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 467] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 467| return; 221| 467| } 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| 467| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 467| validation_result verify_valid_input(FromSpan src) const { 231| 467| validation_result ret{}; 232| | 233| 467| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 467| const auto implementations = get_supported_implementations(); 235| 467| std::vector results; 236| 467| results.reserve(implementations.size()); 237| | 238| 1.40k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.40k, False: 467] ------------------ 239| 1.40k| results.push_back( 240| 1.40k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.40k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.40k| const bool validation2 = 245| 1.40k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.40k| src.data(), src.size()); 247| 1.40k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.40k] ------------------ 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| 1.40k| } 258| | 259| 467| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 467| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 467] ------------------ 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| 467| } else { 273| 467| ret.implementations_agree = true; 274| 467| } 275| 467| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 467| return r.error == simdutf::SUCCESS; 277| 467| }); 278| 467| return ret; 279| 467| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 934| 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| 701| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 701| return r.error == simdutf::SUCCESS; 277| 701| }); _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 467| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 467| length_result ret{}; 344| | 345| 467| const auto implementations = get_supported_implementations(); 346| 467| std::vector results; 347| 467| results.reserve(implementations.size()); 348| | 349| 1.40k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.40k, False: 467] ------------------ 350| 1.40k| const auto len = invoke_lengthcalc(impl, src); 351| 1.40k| results.push_back(len); 352| 1.40k| ret.length.push_back(len); 353| 1.40k| } 354| | 355| 467| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 467| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 467] ------------------ 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| 467| } else { 375| 467| ret.implementations_agree = true; 376| 467| } 377| 467| return ret; 378| 467| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.40k| FromSpan src) const { 331| 1.40k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.40k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 934| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 467| const bool inputisvalid) const { 383| 467| conversion_result ret{}; 384| | 385| 467| const auto implementations = get_supported_implementations(); 386| | 387| 467| std::vector> results; 388| 467| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 467| std::vector> outputbuffers; 393| 467| outputbuffers.reserve(implementations.size()); 394| 1.86k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.40k, False: 467] ------------------ 395| 1.40k| auto impl = implementations[i]; 396| 1.40k| const ToType canary1{42}; 397| 1.40k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.40k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.40k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.40k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.40k| if constexpr (std::is_same_v) { 403| 1.40k| return r != 0; 404| 1.40k| } else { 405| 1.40k| return r.error == simdutf::error_code::SUCCESS; 406| 1.40k| } 407| 1.40k| }(implret1); 408| 1.40k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.40k| 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| 1.40k| const ToType canary2{25}; 414| 1.40k| const auto outputbuffer_first_run = outputbuffer; 415| 1.40k| std::ranges::fill(outputbuffer, canary2); 416| 1.40k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.40k| src.size(), outputbuffer.data()); 418| | 419| 1.40k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.40k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.40k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 351, False: 1.05k] | Branch (423:29): [True: 351, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 351| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 351| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 351] ------------------ 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| 351| } 440| 1.40k| } 441| 1.40k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 351, False: 1.05k] ------------------ 442| 1.40k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 467| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 350, False: 117] ------------------ 447| 1.05k| for (auto& e : results) { ------------------ | Branch (447:20): [True: 1.05k, False: 350] ------------------ 448| 1.05k| e.outputhash.clear(); 449| 1.05k| } 450| 350| } 451| | 452| 467| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 467| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 467] ------------------ 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| 467| } else { 474| 467| ret.implementations_agree = true; 475| 467| } 476| 467| return ret; 477| 467| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.40k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.40k| } else { 405| 1.40k| return r.error == simdutf::error_code::SUCCESS; 406| 1.40k| } 407| 1.40k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 934| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_34clENSt3__14spanIKcLm18446744073709551615EEE: 554| 251| +[](std::span chardata) { \ 555| 251| const auto c = \ 556| 251| Conversion{ \ 559| 251| &I::lenfunc, &I::conversionfunc, \ 560| 251| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 251|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 251| ::std::void_t(); \ | | 1180| 251| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 251| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 251| constexpr auto _size = _name.size(); \ | | 1183| 251| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 251| return _nameof; }() ------------------ 561| 251| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 251|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 251| ::std::void_t(); \ | | 1180| 251| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 251| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 251| constexpr auto _size = _name.size(); \ | | 1183| 251| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 251| return _nameof; }() ------------------ 562| 251| c.fuzz(chardata); \ 563| 251| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 251| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 251| const FromSpan from{reinterpret_cast(chardata.data()), 178| 251| chardata.size() / sizeof(FromType)}; 179| | 180| 251| static const bool do_print_testcase = 181| 251| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 251| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 251] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 251| do { 189| | // step 0 - is the input valid? 190| 251| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 251| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 251] | 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| 251| From == UtfEncodings::UTF8) { 198| 251| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 251] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 251| } 201| | 202| | // step 2 - what is the required size of the output? 203| 251| const auto [output_length, length_agree] = 204| 251| calculate_length(from, inputisvalid); 205| 251| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 251] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 251| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 134, False: 117] | Branch (208:28): [True: 0, False: 134] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 251| const auto [written, outputs_agree] = 215| 251| do_conversion(from, output_length, inputisvalid); 216| 251| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 251] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 251| return; 221| 251| } 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| 251| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 251| validation_result verify_valid_input(FromSpan src) const { 231| 251| validation_result ret{}; 232| | 233| 251| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 251| const auto implementations = get_supported_implementations(); 235| 251| std::vector results; 236| 251| results.reserve(implementations.size()); 237| | 238| 753| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 753, False: 251] ------------------ 239| 753| results.push_back( 240| 753| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 753| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 753| const bool validation2 = 245| 753| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 753| src.data(), src.size()); 247| 753| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 753] ------------------ 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| 753| } 258| | 259| 251| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 251| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 251] ------------------ 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| 251| } else { 273| 251| ret.implementations_agree = true; 274| 251| } 275| 251| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 251| return r.error == simdutf::SUCCESS; 277| 251| }); 278| 251| return ret; 279| 251| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 502| 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| 485| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 485| return r.error == simdutf::SUCCESS; 277| 485| }); _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 251| bool count_the_input(FromSpan src) const { 290| 251| const auto implementations = get_supported_implementations(); 291| 251| std::vector results; 292| 251| results.reserve(implementations.size()); 293| | 294| 753| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 753, False: 251] ------------------ 295| 753| 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| 753| } else if constexpr (From == UtfEncodings::UTF8) { 301| 753| ret = impl->count_utf8(src.data(), src.size()); 302| 753| } 303| 753| results.push_back(ret); 304| 753| } 305| 251| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 251| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 251] ------------------ 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| 251| return true; 321| 251| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 502| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 251| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 251| length_result ret{}; 344| | 345| 251| const auto implementations = get_supported_implementations(); 346| 251| std::vector results; 347| 251| results.reserve(implementations.size()); 348| | 349| 753| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 753, False: 251] ------------------ 350| 753| const auto len = invoke_lengthcalc(impl, src); 351| 753| results.push_back(len); 352| 753| ret.length.push_back(len); 353| 753| } 354| | 355| 251| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 251| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 251] ------------------ 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| 251| } else { 375| 251| ret.implementations_agree = true; 376| 251| } 377| 251| return ret; 378| 251| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 753| FromSpan src) const { 331| 753| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 753| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 502| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 251| const bool inputisvalid) const { 383| 251| conversion_result ret{}; 384| | 385| 251| const auto implementations = get_supported_implementations(); 386| | 387| 251| std::vector> results; 388| 251| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 251| std::vector> outputbuffers; 393| 251| outputbuffers.reserve(implementations.size()); 394| 1.00k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 753, False: 251] ------------------ 395| 753| auto impl = implementations[i]; 396| 753| const ToType canary1{42}; 397| 753| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 753| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 753| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 753| const auto success = [](const ConversionResult& r) -> bool { 402| 753| if constexpr (std::is_same_v) { 403| 753| return r != 0; 404| 753| } else { 405| 753| return r.error == simdutf::error_code::SUCCESS; 406| 753| } 407| 753| }(implret1); 408| 753| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 753| 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| 753| const ToType canary2{25}; 414| 753| const auto outputbuffer_first_run = outputbuffer; 415| 753| std::ranges::fill(outputbuffer, canary2); 416| 753| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 753| src.size(), outputbuffer.data()); 418| | 419| 753| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 753] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 753| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 351, False: 402] | Branch (423:29): [True: 336, False: 15] ------------------ 424| | // only care about the output if the input is valid 425| 336| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 336| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 336] ------------------ 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| 336| } 440| 753| } 441| 753| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 336, False: 417] ------------------ 442| 753| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 251| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 134, False: 117] ------------------ 447| 402| for (auto& e : results) { ------------------ | Branch (447:20): [True: 402, False: 134] ------------------ 448| 402| e.outputhash.clear(); 449| 402| } 450| 134| } 451| | 452| 251| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 251| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 251] ------------------ 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| 251| } else { 474| 251| ret.implementations_agree = true; 475| 251| } 476| 251| return ret; 477| 251| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 753| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 753| } else { 405| 753| return r.error == simdutf::error_code::SUCCESS; 406| 753| } 407| 753| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 502| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_35clENSt3__14spanIKcLm18446744073709551615EEE: 554| 377| +[](std::span chardata) { \ 555| 377| const auto c = \ 556| 377| Conversion{ \ 559| 377| &I::lenfunc, &I::conversionfunc, \ 560| 377| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 377|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 377| ::std::void_t(); \ | | 1180| 377| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 377| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 377| constexpr auto _size = _name.size(); \ | | 1183| 377| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 377| return _nameof; }() ------------------ 561| 377| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 377|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 377| ::std::void_t(); \ | | 1180| 377| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 377| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 377| constexpr auto _size = _name.size(); \ | | 1183| 377| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 377| return _nameof; }() ------------------ 562| 377| c.fuzz(chardata); \ 563| 377| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 377| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 377| const FromSpan from{reinterpret_cast(chardata.data()), 178| 377| chardata.size() / sizeof(FromType)}; 179| | 180| 377| static const bool do_print_testcase = 181| 377| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 377| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 377] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 377| do { 189| | // step 0 - is the input valid? 190| 377| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 377| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 377] | 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| 377| From == UtfEncodings::UTF8) { 198| 377| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 377] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 377| } 201| | 202| | // step 2 - what is the required size of the output? 203| 377| const auto [output_length, length_agree] = 204| 377| calculate_length(from, inputisvalid); 205| 377| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 377] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 377| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 214, False: 163] | Branch (208:28): [True: 0, False: 214] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 377| const auto [written, outputs_agree] = 215| 377| do_conversion(from, output_length, inputisvalid); 216| 377| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 377] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 377| return; 221| 377| } 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| 377| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 377| validation_result verify_valid_input(FromSpan src) const { 231| 377| validation_result ret{}; 232| | 233| 377| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 377| const auto implementations = get_supported_implementations(); 235| 377| std::vector results; 236| 377| results.reserve(implementations.size()); 237| | 238| 1.13k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.13k, False: 377] ------------------ 239| 1.13k| results.push_back( 240| 1.13k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.13k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.13k| const bool validation2 = 245| 1.13k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.13k| src.data(), src.size()); 247| 1.13k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.13k] ------------------ 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| 1.13k| } 258| | 259| 377| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 377| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 377] ------------------ 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| 377| } else { 273| 377| ret.implementations_agree = true; 274| 377| } 275| 377| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 377| return r.error == simdutf::SUCCESS; 277| 377| }); 278| 377| return ret; 279| 377| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 754| 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| 703| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 703| return r.error == simdutf::SUCCESS; 277| 703| }); _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 377| bool count_the_input(FromSpan src) const { 290| 377| const auto implementations = get_supported_implementations(); 291| 377| std::vector results; 292| 377| results.reserve(implementations.size()); 293| | 294| 1.13k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.13k, False: 377] ------------------ 295| 1.13k| 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| 1.13k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.13k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.13k| } 303| 1.13k| results.push_back(ret); 304| 1.13k| } 305| 377| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 377| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 377] ------------------ 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| 377| return true; 321| 377| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 754| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 377| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 377| length_result ret{}; 344| | 345| 377| const auto implementations = get_supported_implementations(); 346| 377| std::vector results; 347| 377| results.reserve(implementations.size()); 348| | 349| 1.13k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.13k, False: 377] ------------------ 350| 1.13k| const auto len = invoke_lengthcalc(impl, src); 351| 1.13k| results.push_back(len); 352| 1.13k| ret.length.push_back(len); 353| 1.13k| } 354| | 355| 377| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 377| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 377] ------------------ 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| 377| } else { 375| 377| ret.implementations_agree = true; 376| 377| } 377| 377| return ret; 378| 377| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.13k| FromSpan src) const { 331| 1.13k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.13k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 754| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 377| const bool inputisvalid) const { 383| 377| conversion_result ret{}; 384| | 385| 377| const auto implementations = get_supported_implementations(); 386| | 387| 377| std::vector> results; 388| 377| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 377| std::vector> outputbuffers; 393| 377| outputbuffers.reserve(implementations.size()); 394| 1.50k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.13k, False: 377] ------------------ 395| 1.13k| auto impl = implementations[i]; 396| 1.13k| const ToType canary1{42}; 397| 1.13k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.13k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.13k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.13k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.13k| if constexpr (std::is_same_v) { 403| 1.13k| return r != 0; 404| 1.13k| } else { 405| 1.13k| return r.error == simdutf::error_code::SUCCESS; 406| 1.13k| } 407| 1.13k| }(implret1); 408| 1.13k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.13k| 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| 1.13k| const ToType canary2{25}; 414| 1.13k| const auto outputbuffer_first_run = outputbuffer; 415| 1.13k| std::ranges::fill(outputbuffer, canary2); 416| 1.13k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.13k| src.size(), outputbuffer.data()); 418| | 419| 1.13k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.13k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.13k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 489, False: 642] | Branch (423:29): [True: 489, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 489| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 489| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 489] ------------------ 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| 489| } 440| 1.13k| } 441| 1.13k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 489, False: 642] ------------------ 442| 1.13k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 377| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 214, False: 163] ------------------ 447| 642| for (auto& e : results) { ------------------ | Branch (447:20): [True: 642, False: 214] ------------------ 448| 642| e.outputhash.clear(); 449| 642| } 450| 214| } 451| | 452| 377| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 377| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 377] ------------------ 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| 377| } else { 474| 377| ret.implementations_agree = true; 475| 377| } 476| 377| return ret; 477| 377| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.13k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.13k| } else { 405| 1.13k| return r.error == simdutf::error_code::SUCCESS; 406| 1.13k| } 407| 1.13k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 754| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_36clENSt3__14spanIKcLm18446744073709551615EEE: 554| 306| +[](std::span chardata) { \ 555| 306| const auto c = \ 556| 306| Conversion{ \ 559| 306| &I::lenfunc, &I::conversionfunc, \ 560| 306| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 306|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 306| ::std::void_t(); \ | | 1180| 306| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 306| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 306| constexpr auto _size = _name.size(); \ | | 1183| 306| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 306| return _nameof; }() ------------------ 561| 306| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 306|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 306| ::std::void_t(); \ | | 1180| 306| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 306| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 306| constexpr auto _size = _name.size(); \ | | 1183| 306| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 306| return _nameof; }() ------------------ 562| 306| c.fuzz(chardata); \ 563| 306| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 306| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 306| const FromSpan from{reinterpret_cast(chardata.data()), 178| 306| chardata.size() / sizeof(FromType)}; 179| | 180| 306| static const bool do_print_testcase = 181| 306| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 306| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 306] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 306| do { 189| | // step 0 - is the input valid? 190| 306| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 306| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 306] | 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| 306| From == UtfEncodings::UTF8) { 198| 306| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 306] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 306| } 201| | 202| | // step 2 - what is the required size of the output? 203| 306| const auto [output_length, length_agree] = 204| 306| calculate_length(from, inputisvalid); 205| 306| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 306] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 306| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 177, False: 129] | Branch (208:28): [True: 0, False: 177] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 306| const auto [written, outputs_agree] = 215| 306| do_conversion(from, output_length, inputisvalid); 216| 306| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 306] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 306| return; 221| 306| } 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| 306| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 306| validation_result verify_valid_input(FromSpan src) const { 231| 306| validation_result ret{}; 232| | 233| 306| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 306| const auto implementations = get_supported_implementations(); 235| 306| std::vector results; 236| 306| results.reserve(implementations.size()); 237| | 238| 918| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 918, False: 306] ------------------ 239| 918| results.push_back( 240| 918| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 918| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 918| const bool validation2 = 245| 918| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 918| src.data(), src.size()); 247| 918| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 918] ------------------ 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| 918| } 258| | 259| 306| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 306| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 306] ------------------ 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| 306| } else { 273| 306| ret.implementations_agree = true; 274| 306| } 275| 306| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 306| return r.error == simdutf::SUCCESS; 277| 306| }); 278| 306| return ret; 279| 306| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 612| 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| 564| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 564| return r.error == simdutf::SUCCESS; 277| 564| }); _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 306| bool count_the_input(FromSpan src) const { 290| 306| const auto implementations = get_supported_implementations(); 291| 306| std::vector results; 292| 306| results.reserve(implementations.size()); 293| | 294| 918| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 918, False: 306] ------------------ 295| 918| 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| 918| } else if constexpr (From == UtfEncodings::UTF8) { 301| 918| ret = impl->count_utf8(src.data(), src.size()); 302| 918| } 303| 918| results.push_back(ret); 304| 918| } 305| 306| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 306| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 306] ------------------ 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| 306| return true; 321| 306| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 612| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 306| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 306| length_result ret{}; 344| | 345| 306| const auto implementations = get_supported_implementations(); 346| 306| std::vector results; 347| 306| results.reserve(implementations.size()); 348| | 349| 918| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 918, False: 306] ------------------ 350| 918| const auto len = invoke_lengthcalc(impl, src); 351| 918| results.push_back(len); 352| 918| ret.length.push_back(len); 353| 918| } 354| | 355| 306| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 306| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 306] ------------------ 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| 306| } else { 375| 306| ret.implementations_agree = true; 376| 306| } 377| 306| return ret; 378| 306| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 918| FromSpan src) const { 331| 918| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 918| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 612| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 306| const bool inputisvalid) const { 383| 306| conversion_result ret{}; 384| | 385| 306| const auto implementations = get_supported_implementations(); 386| | 387| 306| std::vector> results; 388| 306| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 306| std::vector> outputbuffers; 393| 306| outputbuffers.reserve(implementations.size()); 394| 1.22k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 918, False: 306] ------------------ 395| 918| auto impl = implementations[i]; 396| 918| const ToType canary1{42}; 397| 918| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 918| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 918| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 918| const auto success = [](const ConversionResult& r) -> bool { 402| 918| if constexpr (std::is_same_v) { 403| 918| return r != 0; 404| 918| } else { 405| 918| return r.error == simdutf::error_code::SUCCESS; 406| 918| } 407| 918| }(implret1); 408| 918| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 918| 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| 918| const ToType canary2{25}; 414| 918| const auto outputbuffer_first_run = outputbuffer; 415| 918| std::ranges::fill(outputbuffer, canary2); 416| 918| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 918| src.size(), outputbuffer.data()); 418| | 419| 918| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 918] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 918| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 387, False: 531] | Branch (423:29): [True: 387, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 387| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 387| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 387] ------------------ 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| 387| } 440| 918| } 441| 918| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 387, False: 531] ------------------ 442| 918| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 306| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 177, False: 129] ------------------ 447| 531| for (auto& e : results) { ------------------ | Branch (447:20): [True: 531, False: 177] ------------------ 448| 531| e.outputhash.clear(); 449| 531| } 450| 177| } 451| | 452| 306| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 306| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 306] ------------------ 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| 306| } else { 474| 306| ret.implementations_agree = true; 475| 306| } 476| 306| return ret; 477| 306| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 918| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 918| } else { 405| 918| return r.error == simdutf::error_code::SUCCESS; 406| 918| } 407| 918| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 612| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_37clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 397| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 397| const FromSpan from{reinterpret_cast(chardata.data()), 178| 397| chardata.size() / sizeof(FromType)}; 179| | 180| 397| static const bool do_print_testcase = 181| 397| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 397| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 397] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 397| do { 189| | // step 0 - is the input valid? 190| 397| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 397| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 397] | 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| 397| From == UtfEncodings::UTF8) { 198| 397| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 397] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 397| } 201| | 202| | // step 2 - what is the required size of the output? 203| 397| const auto [output_length, length_agree] = 204| 397| calculate_length(from, inputisvalid); 205| 397| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 397] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 397| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 217, False: 180] | 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| 397| const auto [written, outputs_agree] = 215| 397| do_conversion(from, output_length, inputisvalid); 216| 397| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 397] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 397| return; 221| 397| } 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| 397| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 397| validation_result verify_valid_input(FromSpan src) const { 231| 397| validation_result ret{}; 232| | 233| 397| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 397| const auto implementations = get_supported_implementations(); 235| 397| std::vector results; 236| 397| results.reserve(implementations.size()); 237| | 238| 1.19k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.19k, False: 397] ------------------ 239| 1.19k| results.push_back( 240| 1.19k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.19k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.19k| const bool validation2 = 245| 1.19k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.19k| src.data(), src.size()); 247| 1.19k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.19k] ------------------ 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| 1.19k| } 258| | 259| 397| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 397| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 397] ------------------ 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| 397| } else { 273| 397| ret.implementations_agree = true; 274| 397| } 275| 397| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 397| return r.error == simdutf::SUCCESS; 277| 397| }); 278| 397| return ret; 279| 397| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 794| 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| 757| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 757| return r.error == simdutf::SUCCESS; 277| 757| }); _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 397| bool count_the_input(FromSpan src) const { 290| 397| const auto implementations = get_supported_implementations(); 291| 397| std::vector results; 292| 397| results.reserve(implementations.size()); 293| | 294| 1.19k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.19k, False: 397] ------------------ 295| 1.19k| 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| 1.19k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.19k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.19k| } 303| 1.19k| results.push_back(ret); 304| 1.19k| } 305| 397| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 397| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 397] ------------------ 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| 397| return true; 321| 397| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 794| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 397| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 397| length_result ret{}; 344| | 345| 397| const auto implementations = get_supported_implementations(); 346| 397| std::vector results; 347| 397| results.reserve(implementations.size()); 348| | 349| 1.19k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.19k, False: 397] ------------------ 350| 1.19k| const auto len = invoke_lengthcalc(impl, src); 351| 1.19k| results.push_back(len); 352| 1.19k| ret.length.push_back(len); 353| 1.19k| } 354| | 355| 397| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 397| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 397] ------------------ 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| 397| } else { 375| 397| ret.implementations_agree = true; 376| 397| } 377| 397| return ret; 378| 397| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.19k| FromSpan src) const { 331| 1.19k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.19k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 794| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 397| const bool inputisvalid) const { 383| 397| conversion_result ret{}; 384| | 385| 397| const auto implementations = get_supported_implementations(); 386| | 387| 397| std::vector> results; 388| 397| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 397| std::vector> outputbuffers; 393| 397| outputbuffers.reserve(implementations.size()); 394| 1.58k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.19k, False: 397] ------------------ 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: 540, False: 651] | Branch (423:29): [True: 540, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 540| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 540| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 540] ------------------ 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| 540| } 440| 1.19k| } 441| 1.19k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 540, False: 651] ------------------ 442| 1.19k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 397| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 217, False: 180] ------------------ 447| 651| for (auto& e : results) { ------------------ | Branch (447:20): [True: 651, False: 217] ------------------ 448| 651| e.outputhash.clear(); 449| 651| } 450| 217| } 451| | 452| 397| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 397| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 397] ------------------ 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| 397| } else { 474| 397| ret.implementations_agree = true; 475| 397| } 476| 397| return ret; 477| 397| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.19k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.19k| } else { 405| 1.19k| return r.error == simdutf::error_code::SUCCESS; 406| 1.19k| } 407| 1.19k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 794| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_38clENSt3__14spanIKcLm18446744073709551615EEE: 554| 55| +[](std::span chardata) { \ 555| 55| const auto c = \ 556| 55| Conversion{ \ 559| 55| &I::lenfunc, &I::conversionfunc, \ 560| 55| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 55|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 55| ::std::void_t(); \ | | 1180| 55| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 55| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 55| constexpr auto _size = _name.size(); \ | | 1183| 55| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 55| return _nameof; }() ------------------ 561| 55| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 55|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 55| ::std::void_t(); \ | | 1180| 55| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 55| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 55| constexpr auto _size = _name.size(); \ | | 1183| 55| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 55| return _nameof; }() ------------------ 562| 55| c.fuzz(chardata); \ 563| 55| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE4fuzzENSt3__14spanIS5_Lm18446744073709551615EEE: 175| 55| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 55| const FromSpan from{reinterpret_cast(chardata.data()), 178| 55| chardata.size() / sizeof(FromType)}; 179| | 180| 55| static const bool do_print_testcase = 181| 55| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 55| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 55] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 55| do { 189| | // step 0 - is the input valid? 190| 55| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 55| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 55] | 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| 55| const auto [output_length, length_agree] = 204| 55| calculate_length(from, inputisvalid); 205| 55| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 55] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 55| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 55] | 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| 55| const auto [written, outputs_agree] = 215| 55| do_conversion(from, output_length, inputisvalid); 216| 55| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 55] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 55| return; 221| 55| } 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| 55| } _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 283| 55| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 55| return validation_result{.valid = true, .implementations_agree = true}; 287| 55| } _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 55| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 55| length_result ret{}; 344| | 345| 55| const auto implementations = get_supported_implementations(); 346| 55| std::vector results; 347| 55| results.reserve(implementations.size()); 348| | 349| 165| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 165, False: 55] ------------------ 350| 165| const auto len = invoke_lengthcalc(impl, src); 351| 165| results.push_back(len); 352| 165| ret.length.push_back(len); 353| 165| } 354| | 355| 55| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 55| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 55] ------------------ 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| 55| } else { 375| 55| ret.implementations_agree = true; 376| 55| } 377| 55| return ret; 378| 55| } _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 165| FromSpan src) const { 339| 165| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 165| } _ZZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 110| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 55| const bool inputisvalid) const { 383| 55| conversion_result ret{}; 384| | 385| 55| const auto implementations = get_supported_implementations(); 386| | 387| 55| std::vector> results; 388| 55| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 55| std::vector> outputbuffers; 393| 55| outputbuffers.reserve(implementations.size()); 394| 220| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 165, False: 55] ------------------ 395| 165| auto impl = implementations[i]; 396| 165| const ToType canary1{42}; 397| 165| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 165| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 165| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 165| const auto success = [](const ConversionResult& r) -> bool { 402| 165| if constexpr (std::is_same_v) { 403| 165| return r != 0; 404| 165| } else { 405| 165| return r.error == simdutf::error_code::SUCCESS; 406| 165| } 407| 165| }(implret1); 408| 165| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 165| 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| 165| const ToType canary2{25}; 414| 165| const auto outputbuffer_first_run = outputbuffer; 415| 165| std::ranges::fill(outputbuffer, canary2); 416| 165| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 165| src.size(), outputbuffer.data()); 418| | 419| 165| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 165] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 165| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 165, False: 0] | Branch (423:29): [True: 159, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 159| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 159| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 159] ------------------ 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| 159| } 440| 165| } 441| 165| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 159, False: 6] ------------------ 442| 165| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 55| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 55] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 55| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 55| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 55] ------------------ 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| 55| } else { 474| 55| ret.implementations_agree = true; 475| 55| } 476| 55| return ret; 477| 55| } _ZZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 165| const auto success = [](const ConversionResult& r) -> bool { 402| 165| if constexpr (std::is_same_v) { 403| 165| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 165| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 110| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_39clENSt3__14spanIKcLm18446744073709551615EEE: 554| 44| +[](std::span chardata) { \ 555| 44| const auto c = \ 556| 44| Conversion{ \ 559| 44| &I::lenfunc, &I::conversionfunc, \ 560| 44| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 44|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 44| ::std::void_t(); \ | | 1180| 44| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 44| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 44| constexpr auto _size = _name.size(); \ | | 1183| 44| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 44| return _nameof; }() ------------------ 561| 44| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 44|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 44| ::std::void_t(); \ | | 1180| 44| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 44| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 44| constexpr auto _size = _name.size(); \ | | 1183| 44| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 44| return _nameof; }() ------------------ 562| 44| c.fuzz(chardata); \ 563| 44| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE4fuzzENSt3__14spanIS5_Lm18446744073709551615EEE: 175| 44| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 44| const FromSpan from{reinterpret_cast(chardata.data()), 178| 44| chardata.size() / sizeof(FromType)}; 179| | 180| 44| static const bool do_print_testcase = 181| 44| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 44| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 44] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 44| do { 189| | // step 0 - is the input valid? 190| 44| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 44| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 44] | 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| 44| const auto [output_length, length_agree] = 204| 44| calculate_length(from, inputisvalid); 205| 44| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 44] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 44| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 44] | 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| 44| const auto [written, outputs_agree] = 215| 44| do_conversion(from, output_length, inputisvalid); 216| 44| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 44] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 44| return; 221| 44| } 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| 44| } _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 283| 44| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 44| return validation_result{.valid = true, .implementations_agree = true}; 287| 44| } _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 44| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 44| length_result ret{}; 344| | 345| 44| const auto implementations = get_supported_implementations(); 346| 44| std::vector results; 347| 44| results.reserve(implementations.size()); 348| | 349| 132| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 132, False: 44] ------------------ 350| 132| const auto len = invoke_lengthcalc(impl, src); 351| 132| results.push_back(len); 352| 132| ret.length.push_back(len); 353| 132| } 354| | 355| 44| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 44| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 44] ------------------ 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| 44| } else { 375| 44| ret.implementations_agree = true; 376| 44| } 377| 44| return ret; 378| 44| } _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 132| FromSpan src) const { 339| 132| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 132| } _ZZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 88| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 44| const bool inputisvalid) const { 383| 44| conversion_result ret{}; 384| | 385| 44| const auto implementations = get_supported_implementations(); 386| | 387| 44| std::vector> results; 388| 44| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 44| std::vector> outputbuffers; 393| 44| outputbuffers.reserve(implementations.size()); 394| 176| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 132, False: 44] ------------------ 395| 132| auto impl = implementations[i]; 396| 132| const ToType canary1{42}; 397| 132| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 132| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 132| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 132| const auto success = [](const ConversionResult& r) -> bool { 402| 132| if constexpr (std::is_same_v) { 403| 132| return r != 0; 404| 132| } else { 405| 132| return r.error == simdutf::error_code::SUCCESS; 406| 132| } 407| 132| }(implret1); 408| 132| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 132| 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| 132| const ToType canary2{25}; 414| 132| const auto outputbuffer_first_run = outputbuffer; 415| 132| std::ranges::fill(outputbuffer, canary2); 416| 132| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 132| src.size(), outputbuffer.data()); 418| | 419| 132| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 132] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 132| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 132, False: 0] | Branch (423:29): [True: 126, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 126| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 126| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 126] ------------------ 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| 126| } 440| 132| } 441| 132| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 126, False: 6] ------------------ 442| 132| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 44| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 44] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 44| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 44| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 44] ------------------ 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| 44| } else { 474| 44| ret.implementations_agree = true; 475| 44| } 476| 44| return ret; 477| 44| } _ZZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 132| const auto success = [](const ConversionResult& r) -> bool { 402| 132| if constexpr (std::is_same_v) { 403| 132| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 132| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 88| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_40clENSt3__14spanIKcLm18446744073709551615EEE: 554| 43| +[](std::span chardata) { \ 555| 43| const auto c = \ 556| 43| Conversion{ \ 559| 43| &I::lenfunc, &I::conversionfunc, \ 560| 43| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 43|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 43| ::std::void_t(); \ | | 1180| 43| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 43| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 43| constexpr auto _size = _name.size(); \ | | 1183| 43| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 43| return _nameof; }() ------------------ 561| 43| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 43|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 43| ::std::void_t(); \ | | 1180| 43| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 43| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 43| constexpr auto _size = _name.size(); \ | | 1183| 43| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 43| return _nameof; }() ------------------ 562| 43| c.fuzz(chardata); \ 563| 43| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE4fuzzENSt3__14spanIS5_Lm18446744073709551615EEE: 175| 43| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 43| const FromSpan from{reinterpret_cast(chardata.data()), 178| 43| chardata.size() / sizeof(FromType)}; 179| | 180| 43| static const bool do_print_testcase = 181| 43| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 43| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 43] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 43| do { 189| | // step 0 - is the input valid? 190| 43| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 43| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 43] | 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| 43| const auto [output_length, length_agree] = 204| 43| calculate_length(from, inputisvalid); 205| 43| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 43] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 43| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 43] | 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| 43| const auto [written, outputs_agree] = 215| 43| do_conversion(from, output_length, inputisvalid); 216| 43| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 43] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 43| return; 221| 43| } 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| 43| } _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 283| 43| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 43| return validation_result{.valid = true, .implementations_agree = true}; 287| 43| } _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 43| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 43| length_result ret{}; 344| | 345| 43| const auto implementations = get_supported_implementations(); 346| 43| std::vector results; 347| 43| results.reserve(implementations.size()); 348| | 349| 129| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 129, False: 43] ------------------ 350| 129| const auto len = invoke_lengthcalc(impl, src); 351| 129| results.push_back(len); 352| 129| ret.length.push_back(len); 353| 129| } 354| | 355| 43| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 43| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 43] ------------------ 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| 43| } else { 375| 43| ret.implementations_agree = true; 376| 43| } 377| 43| return ret; 378| 43| } _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 129| FromSpan src) const { 339| 129| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 129| } _ZZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 86| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 43| const bool inputisvalid) const { 383| 43| conversion_result ret{}; 384| | 385| 43| const auto implementations = get_supported_implementations(); 386| | 387| 43| std::vector> results; 388| 43| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 43| std::vector> outputbuffers; 393| 43| outputbuffers.reserve(implementations.size()); 394| 172| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 129, False: 43] ------------------ 395| 129| auto impl = implementations[i]; 396| 129| const ToType canary1{42}; 397| 129| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 129| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 129| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 129| const auto success = [](const ConversionResult& r) -> bool { 402| 129| if constexpr (std::is_same_v) { 403| 129| return r != 0; 404| 129| } else { 405| 129| return r.error == simdutf::error_code::SUCCESS; 406| 129| } 407| 129| }(implret1); 408| 129| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 129| 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| 129| const ToType canary2{25}; 414| 129| const auto outputbuffer_first_run = outputbuffer; 415| 129| std::ranges::fill(outputbuffer, canary2); 416| 129| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 129| src.size(), outputbuffer.data()); 418| | 419| 129| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 129] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 129| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 129, False: 0] | Branch (423:29): [True: 123, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 123| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 123| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 123] ------------------ 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| 123| } 440| 129| } 441| 129| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 123, False: 6] ------------------ 442| 129| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 43| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 43] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 43| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 43| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 43] ------------------ 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| 43| } else { 474| 43| ret.implementations_agree = true; 475| 43| } 476| 43| return ret; 477| 43| } _ZZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 129| const auto success = [](const ConversionResult& r) -> bool { 402| 129| if constexpr (std::is_same_v) { 403| 129| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 129| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 86| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_41clENSt3__14spanIKcLm18446744073709551615EEE: 554| 303| +[](std::span chardata) { \ 555| 303| const auto c = \ 556| 303| Conversion{ \ 559| 303| &I::lenfunc, &I::conversionfunc, \ 560| 303| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 303|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 303| ::std::void_t(); \ | | 1180| 303| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 303| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 303| constexpr auto _size = _name.size(); \ | | 1183| 303| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 303| return _nameof; }() ------------------ 561| 303| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 303|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 303| ::std::void_t(); \ | | 1180| 303| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 303| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 303| constexpr auto _size = _name.size(); \ | | 1183| 303| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 303| return _nameof; }() ------------------ 562| 303| c.fuzz(chardata); \ 563| 303| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 303| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 303| const FromSpan from{reinterpret_cast(chardata.data()), 178| 303| chardata.size() / sizeof(FromType)}; 179| | 180| 303| static const bool do_print_testcase = 181| 303| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 303| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 303] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 303| do { 189| | // step 0 - is the input valid? 190| 303| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 303| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 303] | 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| 303| const auto [output_length, length_agree] = 204| 303| calculate_length(from, inputisvalid); 205| 303| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 303] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 303| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 303] | 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| 303| const auto [written, outputs_agree] = 215| 303| do_conversion(from, output_length, inputisvalid); 216| 303| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 303] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 303| return; 221| 303| } 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| 303| } _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 283| 303| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 303| return validation_result{.valid = true, .implementations_agree = true}; 287| 303| } _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 303| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 303| length_result ret{}; 344| | 345| 303| const auto implementations = get_supported_implementations(); 346| 303| std::vector results; 347| 303| results.reserve(implementations.size()); 348| | 349| 909| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 909, False: 303] ------------------ 350| 909| const auto len = invoke_lengthcalc(impl, src); 351| 909| results.push_back(len); 352| 909| ret.length.push_back(len); 353| 909| } 354| | 355| 303| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 303| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 303] ------------------ 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| 303| } else { 375| 303| ret.implementations_agree = true; 376| 303| } 377| 303| return ret; 378| 303| } _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 909| FromSpan src) const { 331| 909| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 909| } _ZZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 606| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 303| const bool inputisvalid) const { 383| 303| conversion_result ret{}; 384| | 385| 303| const auto implementations = get_supported_implementations(); 386| | 387| 303| std::vector> results; 388| 303| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 303| std::vector> outputbuffers; 393| 303| outputbuffers.reserve(implementations.size()); 394| 1.21k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 909, False: 303] ------------------ 395| 909| auto impl = implementations[i]; 396| 909| const ToType canary1{42}; 397| 909| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 909| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 909| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 909| const auto success = [](const ConversionResult& r) -> bool { 402| 909| if constexpr (std::is_same_v) { 403| 909| return r != 0; 404| 909| } else { 405| 909| return r.error == simdutf::error_code::SUCCESS; 406| 909| } 407| 909| }(implret1); 408| 909| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 909| 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| 909| const ToType canary2{25}; 414| 909| const auto outputbuffer_first_run = outputbuffer; 415| 909| std::ranges::fill(outputbuffer, canary2); 416| 909| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 909| src.size(), outputbuffer.data()); 418| | 419| 909| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 909] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 909| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 909, False: 0] | Branch (423:29): [True: 903, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 903| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 903| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 903] ------------------ 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| 903| } 440| 909| } 441| 909| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 903, False: 6] ------------------ 442| 909| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 303| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 303] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 303| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 303| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 303] ------------------ 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| 303| } else { 474| 303| ret.implementations_agree = true; 475| 303| } 476| 303| return ret; 477| 303| } _ZZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 909| const auto success = [](const ConversionResult& r) -> bool { 402| 909| if constexpr (std::is_same_v) { 403| 909| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 909| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 606| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZN10FNV1A_hash14fnv1ahash_implENSt3__14spanIKcLm18446744073709551615EEE: 75| 17.7k| static constexpr std::uint64_t fnv1ahash_impl(std::span bytes) { 76| 17.7k| auto hash = offset; 77| | 78| 601M| for (auto byte : bytes) { ------------------ | Branch (78:20): [True: 601M, False: 17.7k] ------------------ 79| 601M| hash ^= static_cast(byte); 80| 601M| hash *= prime; 81| 601M| } 82| | 83| 17.7k| return hash; 84| 17.7k| } _ZN10FNV1A_hash14fnv1ahash_implIKDsLm18446744073709551615EQaaaa11is_hashableIT_Entsr3stdE9is_same_vIS2_cEntsr3stdE9is_same_vIS2_hEEEmNSt3__14spanIS2_XT0_EEE: 89| 14.6k| static constexpr std::uint64_t fnv1ahash_impl(std::span data) { 90| 14.6k| return fnv1ahash_impl({reinterpret_cast(data.data()), 91| 14.6k| data.size_bytes()}); 92| 14.6k| } _ZN10FNV1A_hash14fnv1ahash_implENSt3__14spanIKhLm18446744073709551615EEE: 65| 23.2k| fnv1ahash_impl(std::span bytes) { 66| 23.2k| auto hash = offset; 67| | 68| 2.23G| for (std::uint64_t byte : bytes) { ------------------ | Branch (68:29): [True: 2.23G, False: 23.2k] ------------------ 69| 2.23G| hash ^= byte; 70| 2.23G| hash *= prime; 71| 2.23G| } 72| | 73| 23.2k| return hash; 74| 23.2k| } _Z29get_supported_implementationsv: 26| 33.3k|get_supported_implementations() { 27| 33.3k| static const auto impl = []() -> auto { 28| 33.3k| std::vector ret; 29| 33.3k| for (auto e : simdutf::get_available_implementations()) { 30| 33.3k| std::cerr << "implementation " << e->name() << " is available? " 31| 33.3k| << e->supported_by_runtime_system() << '\n'; 32| 33.3k| if (e->supported_by_runtime_system()) { 33| 33.3k| ret.push_back(e); 34| 33.3k| } 35| 33.3k| } 36| 33.3k| return ret; 37| 33.3k| }(); 38| 33.3k| return {impl.data(), impl.size()}; 39| 33.3k|} _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| 29.5k|inline bool operator!=(const simdutf::result& a, const simdutf::result& b) { 43| 29.5k| return a.count != b.count || a.error != b.error; ------------------ | Branch (43:10): [True: 0, False: 29.5k] | Branch (43:32): [True: 0, False: 29.5k] ------------------ 44| 29.5k|} _ZN10FNV1A_hash6as_strIJNSt3__16vectorIDiNS1_9allocatorIDiEEEEEEENS1_12basic_stringIcNS1_11char_traitsIcEENS3_IcEEEEDpRKT_: 100| 8.54k| template static std::string as_str(const Data&... data) { 101| 8.54k| static_assert(sizeof...(Data) > 0, "must hash with at least one argument"); 102| 8.54k| 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| 8.54k| } else { 108| 8.54k| h = fnv1ahash_impl(data...); 109| 8.54k| } 110| 8.54k| constexpr std::size_t expected_chars = 16; 111| 8.54k| std::string ret(expected_chars, '0'); 112| 8.54k| auto c = std::to_chars(ret.data(), ret.data() + ret.size(), h, 16); 113| 8.54k| assert(c.ec == std::errc{}); ------------------ | Branch (113:5): [True: 8.54k, False: 0] ------------------ 114| 8.54k| auto nwritten = c.ptr - ret.data(); 115| 8.54k| assert(nwritten <= expected_chars); ------------------ | Branch (115:5): [True: 8.54k, False: 0] ------------------ 116| 8.54k| std::rotate(ret.data(), c.ptr, ret.data() + expected_chars); 117| 8.54k| return ret; 118| 8.54k| } _ZN10FNV1A_hash14fnv1ahash_implIDiQ11is_hashableIT_EEEmRKNSt3__16vectorIS1_NS2_9allocatorIS1_EEEE: 96| 8.54k| static constexpr std::uint64_t fnv1ahash_impl(const std::vector& data) { 97| 8.54k| return fnv1ahash_impl(std::span(data)); 98| 8.54k| } _ZN10FNV1A_hash14fnv1ahash_implIKDiLm18446744073709551615EQaaaa11is_hashableIT_Entsr3stdE9is_same_vIS2_cEntsr3stdE9is_same_vIS2_hEEEmNSt3__14spanIS2_XT0_EEE: 89| 8.54k| static constexpr std::uint64_t fnv1ahash_impl(std::span data) { 90| 8.54k| return fnv1ahash_impl({reinterpret_cast(data.data()), 91| 8.54k| data.size_bytes()}); 92| 8.54k| } _ZN10FNV1A_hash6as_strIJNSt3__16vectorIcNS1_9allocatorIcEEEEEEENS1_12basic_stringIcNS1_11char_traitsIcEES4_EEDpRKT_: 100| 17.7k| template static std::string as_str(const Data&... data) { 101| 17.7k| static_assert(sizeof...(Data) > 0, "must hash with at least one argument"); 102| 17.7k| 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| 17.7k| } else { 108| 17.7k| h = fnv1ahash_impl(data...); 109| 17.7k| } 110| 17.7k| constexpr std::size_t expected_chars = 16; 111| 17.7k| std::string ret(expected_chars, '0'); 112| 17.7k| auto c = std::to_chars(ret.data(), ret.data() + ret.size(), h, 16); 113| 17.7k| assert(c.ec == std::errc{}); ------------------ | Branch (113:5): [True: 17.7k, False: 0] ------------------ 114| 17.7k| auto nwritten = c.ptr - ret.data(); 115| 17.7k| assert(nwritten <= expected_chars); ------------------ | Branch (115:5): [True: 17.7k, False: 0] ------------------ 116| 17.7k| std::rotate(ret.data(), c.ptr, ret.data() + expected_chars); 117| 17.7k| return ret; 118| 17.7k| } _ZN10FNV1A_hash14fnv1ahash_implIcQ11is_hashableIT_EEEmRKNSt3__16vectorIS1_NS2_9allocatorIS1_EEEE: 96| 17.7k| static constexpr std::uint64_t fnv1ahash_impl(const std::vector& data) { 97| 17.7k| return fnv1ahash_impl(std::span(data)); 98| 17.7k| } _ZN10FNV1A_hash6as_strIJNSt3__16vectorIDsNS1_9allocatorIDsEEEEEEENS1_12basic_stringIcNS1_11char_traitsIcEENS3_IcEEEEDpRKT_: 100| 14.6k| template static std::string as_str(const Data&... data) { 101| 14.6k| static_assert(sizeof...(Data) > 0, "must hash with at least one argument"); 102| 14.6k| 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| 14.6k| } else { 108| 14.6k| h = fnv1ahash_impl(data...); 109| 14.6k| } 110| 14.6k| constexpr std::size_t expected_chars = 16; 111| 14.6k| std::string ret(expected_chars, '0'); 112| 14.6k| auto c = std::to_chars(ret.data(), ret.data() + ret.size(), h, 16); 113| 14.6k| assert(c.ec == std::errc{}); ------------------ | Branch (113:5): [True: 14.6k, False: 0] ------------------ 114| 14.6k| auto nwritten = c.ptr - ret.data(); 115| 14.6k| assert(nwritten <= expected_chars); ------------------ | Branch (115:5): [True: 14.6k, False: 0] ------------------ 116| 14.6k| std::rotate(ret.data(), c.ptr, ret.data() + expected_chars); 117| 14.6k| return ret; 118| 14.6k| } _ZN10FNV1A_hash14fnv1ahash_implIDsQ11is_hashableIT_EEEmRKNSt3__16vectorIS1_NS2_9allocatorIS1_EEEE: 96| 14.6k| static constexpr std::uint64_t fnv1ahash_impl(const std::vector& data) { 97| 14.6k| return fnv1ahash_impl(std::span(data)); 98| 14.6k| } _ZeqRKN7simdutf6resultES2_: 45| 8.00k|inline bool operator==(const simdutf::result& a, const simdutf::result& b) { 46| 8.00k| return a.count == b.count && a.error == b.error; ------------------ | Branch (46:10): [True: 8.00k, False: 0] | Branch (46:32): [True: 8.00k, False: 0] ------------------ 47| 8.00k|} _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$_0clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 99|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 99| ::std::void_t(); \ 1180| 99| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 99| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 99| constexpr auto _size = _name.size(); \ 1183| 99| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 99| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_0clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 99|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 99| ::std::void_t(); \ 1180| 99| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 99| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 99| constexpr auto _size = _name.size(); \ 1183| 99| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 99| return _nameof; }() _ZNK6nameof7cstringILt30EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 354| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt30EE4dataEv: 233| 354| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt30EE4sizeEv: 235| 354| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_1clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 216|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 216| ::std::void_t(); \ 1180| 216| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 216| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 216| constexpr auto _size = _name.size(); \ 1183| 216| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 216| return _nameof; }() _ZNK6nameof7cstringILt24EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 3.13k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt24EE4dataEv: 233| 3.13k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt24EE4sizeEv: 235| 3.13k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_1clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 216|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 216| ::std::void_t(); \ 1180| 216| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 216| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 216| constexpr auto _size = _name.size(); \ 1183| 216| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 216| return _nameof; }() _ZNK6nameof7cstringILt29EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 939| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt29EE4dataEv: 233| 939| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt29EE4sizeEv: 235| 939| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_2clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 58|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 58| ::std::void_t(); \ 1180| 58| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 58| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 58| constexpr auto _size = _name.size(); \ 1183| 58| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 58| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_2clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 58|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 58| ::std::void_t(); \ 1180| 58| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 58| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 58| constexpr auto _size = _name.size(); \ 1183| 58| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 58| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_3clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 236|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 236| ::std::void_t(); \ 1180| 236| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 236| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 236| constexpr auto _size = _name.size(); \ 1183| 236| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 236| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_3clENSt3__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$_4clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 66|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 66| ::std::void_t(); \ 1180| 66| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 66| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 66| constexpr auto _size = _name.size(); \ 1183| 66| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 66| return _nameof; }() _ZNK6nameof7cstringILt23EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 3.49k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt23EE4dataEv: 233| 3.49k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt23EE4sizeEv: 235| 3.49k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_4clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 66|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 66| ::std::void_t(); \ 1180| 66| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 66| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 66| constexpr auto _size = _name.size(); \ 1183| 66| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 66| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_5clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 131|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 131| ::std::void_t(); \ 1180| 131| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 131| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 131| constexpr auto _size = _name.size(); \ 1183| 131| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 131| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_5clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 131|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 131| ::std::void_t(); \ 1180| 131| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 131| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 131| constexpr auto _size = _name.size(); \ 1183| 131| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 131| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_6clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 115|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 115| ::std::void_t(); \ 1180| 115| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 115| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 115| constexpr auto _size = _name.size(); \ 1183| 115| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 115| return _nameof; }() _ZNK6nameof7cstringILt22EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 4.46k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt22EE4dataEv: 233| 4.46k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt22EE4sizeEv: 235| 4.46k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_6clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 115|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 115| ::std::void_t(); \ 1180| 115| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 115| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 115| constexpr auto _size = _name.size(); \ 1183| 115| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 115| return _nameof; }() _ZNK6nameof7cstringILt27EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 344| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt27EE4dataEv: 233| 344| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt27EE4sizeEv: 235| 344| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_7clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 248|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 248| ::std::void_t(); \ 1180| 248| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 248| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 248| constexpr auto _size = _name.size(); \ 1183| 248| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 248| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_7clENSt3__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; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_8clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 239|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 239| ::std::void_t(); \ 1180| 239| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 239| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 239| constexpr auto _size = _name.size(); \ 1183| 239| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 239| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_8clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 239|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 239| ::std::void_t(); \ 1180| 239| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 239| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 239| constexpr auto _size = _name.size(); \ 1183| 239| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 239| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_9clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 229|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 229| ::std::void_t(); \ 1180| 229| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 229| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 229| constexpr auto _size = _name.size(); \ 1183| 229| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 229| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_9clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 229|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 229| ::std::void_t(); \ 1180| 229| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 229| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 229| constexpr auto _size = _name.size(); \ 1183| 229| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 229| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_10clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 47|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 47| ::std::void_t(); \ 1180| 47| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 47| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 47| constexpr auto _size = _name.size(); \ 1183| 47| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 47| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_10clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 47|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 47| ::std::void_t(); \ 1180| 47| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 47| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 47| constexpr auto _size = _name.size(); \ 1183| 47| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 47| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_11clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 110|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 110| ::std::void_t(); \ 1180| 110| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 110| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 110| constexpr auto _size = _name.size(); \ 1183| 110| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 110| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_11clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 110|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 110| ::std::void_t(); \ 1180| 110| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 110| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 110| constexpr auto _size = _name.size(); \ 1183| 110| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 110| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_12clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 176|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 176| ::std::void_t(); \ 1180| 176| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 176| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 176| constexpr auto _size = _name.size(); \ 1183| 176| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 176| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_12clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 176|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 176| ::std::void_t(); \ 1180| 176| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 176| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 176| constexpr auto _size = _name.size(); \ 1183| 176| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 176| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_13clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 67|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 67| ::std::void_t(); \ 1180| 67| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 67| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 67| constexpr auto _size = _name.size(); \ 1183| 67| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 67| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_13clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 67|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 67| ::std::void_t(); \ 1180| 67| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 67| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 67| constexpr auto _size = _name.size(); \ 1183| 67| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 67| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_14clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 153|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 153| ::std::void_t(); \ 1180| 153| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 153| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 153| constexpr auto _size = _name.size(); \ 1183| 153| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 153| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_14clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 153|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 153| ::std::void_t(); \ 1180| 153| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 153| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 153| constexpr auto _size = _name.size(); \ 1183| 153| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 153| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_15clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 208|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 208| ::std::void_t(); \ 1180| 208| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 208| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 208| constexpr auto _size = _name.size(); \ 1183| 208| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 208| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_15clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 208|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 208| ::std::void_t(); \ 1180| 208| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 208| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 208| constexpr auto _size = _name.size(); \ 1183| 208| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 208| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_16clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 86|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 86| ::std::void_t(); \ 1180| 86| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 86| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 86| constexpr auto _size = _name.size(); \ 1183| 86| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 86| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_16clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 86|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 86| ::std::void_t(); \ 1180| 86| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 86| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 86| constexpr auto _size = _name.size(); \ 1183| 86| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 86| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_17clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 267|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 267| ::std::void_t(); \ 1180| 267| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 267| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 267| constexpr auto _size = _name.size(); \ 1183| 267| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 267| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_17clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 267|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 267| ::std::void_t(); \ 1180| 267| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 267| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 267| constexpr auto _size = _name.size(); \ 1183| 267| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 267| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_18clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 263|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 263| ::std::void_t(); \ 1180| 263| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 263| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 263| constexpr auto _size = _name.size(); \ 1183| 263| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 263| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_18clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 263|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 263| ::std::void_t(); \ 1180| 263| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 263| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 263| constexpr auto _size = _name.size(); \ 1183| 263| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 263| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_19clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 355|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 355| ::std::void_t(); \ 1180| 355| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 355| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 355| constexpr auto _size = _name.size(); \ 1183| 355| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 355| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_19clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 355|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 355| ::std::void_t(); \ 1180| 355| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 355| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 355| constexpr auto _size = _name.size(); \ 1183| 355| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 355| return _nameof; }() _ZNK6nameof7cstringILt21EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 747| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt21EE4dataEv: 233| 747| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt21EE4sizeEv: 235| 747| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_20clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 295|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 295| ::std::void_t(); \ 1180| 295| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 295| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 295| constexpr auto _size = _name.size(); \ 1183| 295| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 295| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_20clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 295|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 295| ::std::void_t(); \ 1180| 295| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 295| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 295| constexpr auto _size = _name.size(); \ 1183| 295| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 295| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_21clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 378|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 378| ::std::void_t(); \ 1180| 378| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 378| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 378| constexpr auto _size = _name.size(); \ 1183| 378| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 378| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_21clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 378|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 378| ::std::void_t(); \ 1180| 378| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 378| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 378| constexpr auto _size = _name.size(); \ 1183| 378| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 378| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_22clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 360|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 360| ::std::void_t(); \ 1180| 360| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 360| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 360| constexpr auto _size = _name.size(); \ 1183| 360| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 360| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_22clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 360|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 360| ::std::void_t(); \ 1180| 360| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 360| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 360| constexpr auto _size = _name.size(); \ 1183| 360| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 360| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_23clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 392|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 392| ::std::void_t(); \ 1180| 392| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 392| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 392| constexpr auto _size = _name.size(); \ 1183| 392| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 392| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_23clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 392|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 392| ::std::void_t(); \ 1180| 392| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 392| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 392| constexpr auto _size = _name.size(); \ 1183| 392| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 392| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_24clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 125|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 125| ::std::void_t(); \ 1180| 125| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 125| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 125| constexpr auto _size = _name.size(); \ 1183| 125| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 125| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_24clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 125|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 125| ::std::void_t(); \ 1180| 125| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 125| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 125| constexpr auto _size = _name.size(); \ 1183| 125| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 125| return _nameof; }() _ZNK6nameof7cstringILt37EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 264| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt37EE4dataEv: 233| 264| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt37EE4sizeEv: 235| 264| [[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| 1.03k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt36EE4dataEv: 233| 1.03k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt36EE4sizeEv: 235| 1.03k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_26clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 364|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 364| ::std::void_t(); \ 1180| 364| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 364| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 364| constexpr auto _size = _name.size(); \ 1183| 364| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 364| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_26clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 364|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 364| ::std::void_t(); \ 1180| 364| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 364| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 364| constexpr auto _size = _name.size(); \ 1183| 364| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 364| return _nameof; }() _ZNK6nameof7cstringILt35EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 1.58k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt35EE4dataEv: 233| 1.58k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt35EE4sizeEv: 235| 1.58k| [[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$_28clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 209|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 209| ::std::void_t(); \ 1180| 209| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 209| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 209| constexpr auto _size = _name.size(); \ 1183| 209| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 209| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_28clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 209|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 209| ::std::void_t(); \ 1180| 209| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 209| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 209| constexpr auto _size = _name.size(); \ 1183| 209| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 209| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_29clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 332|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 332| ::std::void_t(); \ 1180| 332| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 332| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 332| constexpr auto _size = _name.size(); \ 1183| 332| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 332| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_29clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 332|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 332| ::std::void_t(); \ 1180| 332| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 332| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 332| constexpr auto _size = _name.size(); \ 1183| 332| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 332| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_30clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 204|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 204| ::std::void_t(); \ 1180| 204| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 204| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 204| constexpr auto _size = _name.size(); \ 1183| 204| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 204| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_30clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 204|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 204| ::std::void_t(); \ 1180| 204| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 204| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 204| constexpr auto _size = _name.size(); \ 1183| 204| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 204| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_31clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 315|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 315| ::std::void_t(); \ 1180| 315| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 315| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 315| constexpr auto _size = _name.size(); \ 1183| 315| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 315| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_31clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 315|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 315| ::std::void_t(); \ 1180| 315| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 315| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 315| constexpr auto _size = _name.size(); \ 1183| 315| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 315| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_32clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 342|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 342| ::std::void_t(); \ 1180| 342| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 342| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 342| constexpr auto _size = _name.size(); \ 1183| 342| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 342| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_32clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 342|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 342| ::std::void_t(); \ 1180| 342| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 342| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 342| constexpr auto _size = _name.size(); \ 1183| 342| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 342| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_33clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 467|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 467| ::std::void_t(); \ 1180| 467| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 467| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 467| constexpr auto _size = _name.size(); \ 1183| 467| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 467| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_33clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 467|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 467| ::std::void_t(); \ 1180| 467| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 467| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 467| constexpr auto _size = _name.size(); \ 1183| 467| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 467| return _nameof; }() _ZNK6nameof7cstringILt33EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 864| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt33EE4dataEv: 233| 864| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt33EE4sizeEv: 235| 864| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_34clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 251|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 251| ::std::void_t(); \ 1180| 251| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 251| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 251| constexpr auto _size = _name.size(); \ 1183| 251| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 251| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_34clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 251|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 251| ::std::void_t(); \ 1180| 251| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 251| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 251| constexpr auto _size = _name.size(); \ 1183| 251| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 251| return _nameof; }() _ZNK6nameof7cstringILt34EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 251| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt34EE4dataEv: 233| 251| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt34EE4sizeEv: 235| 251| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_35clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 377|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 377| ::std::void_t(); \ 1180| 377| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 377| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 377| constexpr auto _size = _name.size(); \ 1183| 377| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 377| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_35clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 377|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 377| ::std::void_t(); \ 1180| 377| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 377| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 377| constexpr auto _size = _name.size(); \ 1183| 377| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 377| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_36clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 306|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 306| ::std::void_t(); \ 1180| 306| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 306| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 306| constexpr auto _size = _name.size(); \ 1183| 306| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 306| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_36clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 306|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 306| ::std::void_t(); \ 1180| 306| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 306| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 306| constexpr auto _size = _name.size(); \ 1183| 306| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 306| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_37clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 397|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 397| ::std::void_t(); \ 1180| 397| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 397| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 397| constexpr auto _size = _name.size(); \ 1183| 397| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 397| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_37clENSt3__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$_38clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 55|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 55| ::std::void_t(); \ 1180| 55| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 55| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 55| constexpr auto _size = _name.size(); \ 1183| 55| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 55| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_38clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 55|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 55| ::std::void_t(); \ 1180| 55| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 55| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 55| constexpr auto _size = _name.size(); \ 1183| 55| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 55| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_39clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 44|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 44| ::std::void_t(); \ 1180| 44| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 44| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 44| constexpr auto _size = _name.size(); \ 1183| 44| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 44| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_39clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 44|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 44| ::std::void_t(); \ 1180| 44| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 44| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 44| constexpr auto _size = _name.size(); \ 1183| 44| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 44| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_40clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 43|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 43| ::std::void_t(); \ 1180| 43| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 43| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 43| constexpr auto _size = _name.size(); \ 1183| 43| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 43| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_40clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 43|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 43| ::std::void_t(); \ 1180| 43| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 43| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 43| constexpr auto _size = _name.size(); \ 1183| 43| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 43| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_41clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 303|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 303| ::std::void_t(); \ 1180| 303| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 303| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 303| constexpr auto _size = _name.size(); \ 1183| 303| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 303| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_41clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 303|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 303| ::std::void_t(); \ 1180| 303| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 303| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 303| constexpr auto _size = _name.size(); \ 1183| 303| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 303| return _nameof; }() _ZN7simdutf6resultC2ENS_10error_codeEm: 87| 77.3k| : error{err}, count{pos} {} _ZNK7simdutf6result6is_errEv: 93| 5.35k| simdutf_really_inline simdutf_constexpr23 bool is_err() const noexcept { 94| 5.35k| return error != error_code::SUCCESS; 95| 5.35k| } _ZN7simdutf11full_resultC2ENS_10error_codeEmm: 111| 3.47k| : error{err}, input_count{pos_in}, output_count{pos_out} {} _ZNK7simdutf11full_resultcvNS_6resultEEv: 117| 3.47k| simdutf_really_inline simdutf_constexpr23 operator result() const noexcept { 118| 3.47k| if (error == error_code::SUCCESS) { ------------------ | Branch (118:9): [True: 2.43k, False: 1.04k] ------------------ 119| 2.43k| return result{error, output_count}; 120| 2.43k| } else { 121| 1.04k| return result{error, input_count}; 122| 1.04k| } 123| 3.47k| } _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| 261| utf16_length_from_latin1(size_t length) const noexcept { 6348| 261| return length; 6349| 261| } _ZNK7simdutf14implementation24latin1_length_from_utf32Em: 6535| 870| latin1_length_from_utf32(size_t length) const noexcept { 6536| 870| return length; 6537| 870| } _ZNK7simdutf14implementation24latin1_length_from_utf16Em: 6572| 1.13k| latin1_length_from_utf16(size_t length) const noexcept { 6573| 1.13k| return length; 6574| 1.13k| } _ZNK7simdutf14implementation24utf32_length_from_latin1Em: 6605| 165| utf32_length_from_latin1(size_t length) const noexcept { 6606| 165| return length; 6607| 165| } _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| 606| size_t len) { 11| 606| const uint8_t *c = reinterpret_cast(buf); 12| 606| size_t answer = 0; 13| 4.72k| for (size_t i = 0; i < len; i++) { ------------------ | Branch (13:22): [True: 4.11k, False: 606] ------------------ 14| 4.11k| if ((c[i] >> 7)) { ------------------ | Branch (14:9): [True: 1.16k, False: 2.95k] ------------------ 15| 1.16k| answer++; 16| 1.16k| } 17| 4.11k| } 18| 606| return answer + len; 19| 606|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115latin1_to_utf167convertILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 14| 222| char16_t *utf16_output) { 15| 222| size_t pos = 0; 16| 222| char16_t *start{utf16_output}; 17| | 18| 1.70M| while (pos < len) { ------------------ | Branch (18:10): [True: 1.70M, False: 222] ------------------ 19| 1.70M| uint16_t word = 20| 1.70M| uint8_t(data[pos]); // extend Latin-1 char to 16-bit Unicode code point 21| 1.70M| *utf16_output++ = 22| 1.70M| char16_t(match_system(big_endian) ? word : u16_swap_bytes(word)); ------------------ | Branch (22:18): [True: 1.70M, Folded] ------------------ 23| 1.70M| pos++; 24| 1.70M| } 25| | 26| 222| return utf16_output - start; 27| 222|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115latin1_to_utf167convertILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 14| 224| char16_t *utf16_output) { 15| 224| size_t pos = 0; 16| 224| char16_t *start{utf16_output}; 17| | 18| 1.93M| while (pos < len) { ------------------ | Branch (18:10): [True: 1.93M, False: 224] ------------------ 19| 1.93M| uint16_t word = 20| 1.93M| uint8_t(data[pos]); // extend Latin-1 char to 16-bit Unicode code point 21| 1.93M| *utf16_output++ = 22| 1.93M| char16_t(match_system(big_endian) ? word : u16_swap_bytes(word)); ------------------ | Branch (22:18): [Folded, False: 1.93M] ------------------ 23| 1.93M| pos++; 24| 1.93M| } 25| | 26| 224| return utf16_output - start; 27| 224|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115latin1_to_utf327convertIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mPDi: 14| 272| char32_t *utf32_output) { 15| 272| char32_t *start{utf32_output}; 16| 8.29M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (16:22): [True: 8.29M, False: 272] ------------------ 17| 8.29M| *utf32_output++ = uint8_t(data[i]); 18| 8.29M| } 19| 272| return utf32_output - start; 20| 272|} 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| 6.10M| while (pos < len) { ------------------ | Branch (22:10): [True: 6.09M, False: 1.71k] ------------------ 23| |#if SIMDUTF_CPLUSPLUS23 24| | if !consteval 25| |#endif 26| 6.09M| { 27| | // try to convert the next block of 16 ASCII bytes 28| 6.09M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (28:11): [True: 6.08M, False: 15.6k] ------------------ 29| | // they are ascii 30| 6.08M| uint64_t v1; 31| 6.08M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 32| 6.08M| uint64_t v2; 33| 6.08M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 34| 6.08M| uint64_t v{v1 | 35| 6.08M| v2}; // We are only interested in these bits: 1000 1000 1000 36| | // 1000, so it makes sense to concatenate everything 37| 6.08M| if ((v & 0x8080808080808080) == ------------------ | Branch (37:13): [True: 2.34M, False: 3.74M] ------------------ 38| 6.08M| 0) { // if NONE of these are set, e.g. all of them are zero, then 39| | // everything is ASCII 40| 2.34M| size_t final_pos = pos + 16; 41| 39.8M| while (pos < final_pos) { ------------------ | Branch (41:18): [True: 37.4M, False: 2.34M] ------------------ 42| 37.4M| utf8_output[utf8_pos++] = char(data[pos]); 43| 37.4M| pos++; 44| 37.4M| } 45| 2.34M| continue; 46| 2.34M| } 47| 6.08M| } // if (pos + 16 <= len) 48| 6.09M| } // !consteval scope 49| | 50| 3.75M| unsigned char byte = data[pos]; 51| 3.75M| if ((byte & 0x80) == 0) { // if ASCII ------------------ | Branch (51:9): [True: 2.33M, False: 1.41M] ------------------ 52| | // will generate one UTF-8 bytes 53| 2.33M| utf8_output[utf8_pos++] = char(byte); 54| 2.33M| pos++; 55| 2.33M| } else { 56| | // will generate two UTF-8 bytes 57| 1.41M| utf8_output[utf8_pos++] = char((byte >> 6) | 0b11000000); 58| 1.41M| utf8_output[utf8_pos++] = char((byte & 0b111111) | 0b10000000); 59| 1.41M| pos++; 60| 1.41M| } 61| 3.75M| } // 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| 303|utf8_length_from_latin1(InputPtr input, size_t length) noexcept { 153| 303| size_t answer = length; 154| 303| size_t i = 0; 155| | 156| |#if SIMDUTF_CPLUSPLUS23 157| | if !consteval 158| |#endif 159| 303| { 160| 303| auto pop = [](uint64_t v) { 161| 303| return (size_t)(((v >> 7) & UINT64_C(0x0101010101010101)) * 162| 303| UINT64_C(0x0101010101010101) >> 163| 303| 56); 164| 303| }; 165| 644k| for (; i + 32 <= length; i += 32) { ------------------ | Branch (165:12): [True: 643k, False: 303] ------------------ 166| 643k| uint64_t v; 167| 643k| memcpy(&v, input + i, 8); 168| 643k| answer += pop(v); 169| 643k| memcpy(&v, input + i + 8, sizeof(v)); 170| 643k| answer += pop(v); 171| 643k| memcpy(&v, input + i + 16, sizeof(v)); 172| 643k| answer += pop(v); 173| 643k| memcpy(&v, input + i + 24, sizeof(v)); 174| 643k| answer += pop(v); 175| 643k| } 176| 583| for (; i + 8 <= length; i += 8) { ------------------ | Branch (176:12): [True: 280, False: 303] ------------------ 177| 280| uint64_t v; 178| 280| memcpy(&v, input + i, sizeof(v)); 179| 280| answer += pop(v); 180| 280| } 181| 303| } // !consteval scope 182| 882| for (; i + 1 <= length; i += 1) { ------------------ | Branch (182:10): [True: 579, False: 303] ------------------ 183| 579| answer += static_cast(input[i]) >> 7; 184| 579| } 185| 303| return answer; 186| 303|} simdutf.cpp:_ZZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf823utf8_length_from_latin1IPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mENKUlmE_clEm: 160| 2.57M| auto pop = [](uint64_t v) { 161| 2.57M| return (size_t)(((v >> 7) & UINT64_C(0x0101010101010101)) * 162| | UINT64_C(0x0101010101010101) >> 163| 2.57M| 56); 164| 2.57M| }; _ZN7simdutf6scalar14u16_swap_bytesEt: 8| 205M|u16_swap_bytes(const uint16_t word) { 9| 205M| return uint16_t((word >> 8) | (word << 8)); 10| 205M|} _ZN7simdutf6scalar5utf1614swap_if_neededILNS_10endiannessE1EEEtt: 27| 92.4M|template constexpr uint16_t swap_if_needed(uint16_t c) { 28| 92.4M| return !match_system(big_endian) ? scalar::u16_swap_bytes(c) : c; ------------------ | Branch (28:10): [True: 92.4M, Folded] ------------------ 29| 92.4M|} _ZN7simdutf6scalar5utf1614swap_if_neededILNS_10endiannessE0EEEtt: 27| 95.9M|template constexpr uint16_t swap_if_needed(uint16_t c) { 28| 95.9M| return !match_system(big_endian) ? scalar::u16_swap_bytes(c) : c; ------------------ | Branch (28:10): [Folded, False: 95.9M] ------------------ 29| 95.9M|} _ZN7simdutf6scalar5utf168validateILNS_10endiannessE0EEEbPKDsm: 22| 3.88k|validate(const char16_t *data, size_t len) noexcept { 23| 3.88k| uint64_t pos = 0; 24| 21.4M| while (pos < len) { ------------------ | Branch (24:10): [True: 21.4M, False: 2.61k] ------------------ 25| 21.4M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 26| 21.4M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (26:9): [True: 20.4k, False: 21.4M] ------------------ 27| 20.4k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 219, False: 20.2k] ------------------ 28| 219| return false; 29| 219| } 30| 20.2k| char16_t diff = char16_t(word - 0xD800); 31| 20.2k| if (diff > 0x3FF) { ------------------ | Branch (31:11): [True: 450, False: 19.8k] ------------------ 32| 450| return false; 33| 450| } 34| 19.8k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (34:28): [Folded, False: 19.8k] ------------------ 35| 19.8k| ? u16_swap_bytes(data[pos + 1]) 36| 19.8k| : data[pos + 1]; 37| 19.8k| char16_t diff2 = char16_t(next_word - 0xDC00); 38| 19.8k| if (diff2 > 0x3FF) { ------------------ | Branch (38:11): [True: 603, False: 19.2k] ------------------ 39| 603| return false; 40| 603| } 41| 19.2k| pos += 2; 42| 21.4M| } else { 43| 21.4M| pos++; 44| 21.4M| } 45| 21.4M| } 46| 2.61k| return true; 47| 3.88k|} _ZN7simdutf6scalar5utf168validateILNS_10endiannessE1EEEbPKDsm: 22| 3.68k|validate(const char16_t *data, size_t len) noexcept { 23| 3.68k| uint64_t pos = 0; 24| 21.3M| while (pos < len) { ------------------ | Branch (24:10): [True: 21.3M, False: 2.55k] ------------------ 25| 21.3M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 26| 21.3M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (26:9): [True: 24.9k, False: 21.3M] ------------------ 27| 24.9k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 222, False: 24.7k] ------------------ 28| 222| return false; 29| 222| } 30| 24.7k| char16_t diff = char16_t(word - 0xD800); 31| 24.7k| if (diff > 0x3FF) { ------------------ | Branch (31:11): [True: 419, False: 24.2k] ------------------ 32| 419| return false; 33| 419| } 34| 24.2k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (34:28): [True: 24.2k, Folded] ------------------ 35| 24.2k| ? u16_swap_bytes(data[pos + 1]) 36| 24.2k| : data[pos + 1]; 37| 24.2k| char16_t diff2 = char16_t(next_word - 0xDC00); 38| 24.2k| if (diff2 > 0x3FF) { ------------------ | Branch (38:11): [True: 493, False: 23.7k] ------------------ 39| 493| return false; 40| 493| } 41| 23.7k| pos += 2; 42| 21.3M| } else { 43| 21.3M| pos++; 44| 21.3M| } 45| 21.3M| } 46| 2.55k| return true; 47| 3.68k|} _ZN7simdutf6scalar5utf1620validate_with_errorsILNS_10endiannessE0EEENS_6resultEPKDsm: 51| 4.17k|validate_with_errors(const char16_t *data, size_t len) noexcept { 52| 4.17k| size_t pos = 0; 53| 21.4M| while (pos < len) { ------------------ | Branch (53:10): [True: 21.4M, False: 2.61k] ------------------ 54| 21.4M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 55| 21.4M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (55:9): [True: 20.8k, False: 21.4M] ------------------ 56| 20.8k| if (pos + 1 >= len) { ------------------ | Branch (56:11): [True: 219, False: 20.6k] ------------------ 57| 219| return result(error_code::SURROGATE, pos); 58| 219| } 59| 20.6k| char16_t diff = char16_t(word - 0xD800); 60| 20.6k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 585, False: 20.0k] ------------------ 61| 585| return result(error_code::SURROGATE, pos); 62| 585| } 63| 20.0k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (63:28): [Folded, False: 20.0k] ------------------ 64| 20.0k| ? u16_swap_bytes(data[pos + 1]) 65| 20.0k| : data[pos + 1]; 66| 20.0k| char16_t diff2 = uint16_t(next_word - 0xDC00); 67| 20.0k| if (diff2 > 0x3FF) { ------------------ | Branch (67:11): [True: 753, False: 19.2k] ------------------ 68| 753| return result(error_code::SURROGATE, pos); 69| 753| } 70| 19.2k| pos += 2; 71| 21.4M| } else { 72| 21.4M| pos++; 73| 21.4M| } 74| 21.4M| } 75| 2.61k| return result(error_code::SUCCESS, pos); 76| 4.17k|} _ZN7simdutf6scalar5utf1620validate_with_errorsILNS_10endiannessE1EEENS_6resultEPKDsm: 51| 3.89k|validate_with_errors(const char16_t *data, size_t len) noexcept { 52| 3.89k| size_t pos = 0; 53| 21.3M| while (pos < len) { ------------------ | Branch (53:10): [True: 21.3M, False: 2.55k] ------------------ 54| 21.3M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 55| 21.3M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (55:9): [True: 25.1k, False: 21.3M] ------------------ 56| 25.1k| if (pos + 1 >= len) { ------------------ | Branch (56:11): [True: 222, False: 24.9k] ------------------ 57| 222| return result(error_code::SURROGATE, pos); 58| 222| } 59| 24.9k| char16_t diff = char16_t(word - 0xD800); 60| 24.9k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 531, False: 24.4k] ------------------ 61| 531| return result(error_code::SURROGATE, pos); 62| 531| } 63| 24.4k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (63:28): [True: 24.4k, Folded] ------------------ 64| 24.4k| ? u16_swap_bytes(data[pos + 1]) 65| 24.4k| : data[pos + 1]; 66| 24.4k| char16_t diff2 = uint16_t(next_word - 0xDC00); 67| 24.4k| if (diff2 > 0x3FF) { ------------------ | Branch (67:11): [True: 594, False: 23.8k] ------------------ 68| 594| return result(error_code::SURROGATE, pos); 69| 594| } 70| 23.8k| pos += 2; 71| 21.3M| } else { 72| 21.3M| pos++; 73| 21.3M| } 74| 21.3M| } 75| 2.55k| return result(error_code::SUCCESS, pos); 76| 3.89k|} _ZN7simdutf6scalar5utf1617count_code_pointsILNS_10endiannessE0EEEmPKDsm: 79| 5.04k|simdutf_constexpr23 size_t count_code_points(const char16_t *p, size_t len) { 80| | // We are not BOM aware. 81| 5.04k| size_t counter{0}; 82| 29.7M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (82:22): [True: 29.7M, False: 5.04k] ------------------ 83| 29.7M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 84| 29.7M| counter += ((word & 0xFC00) != 0xDC00); 85| 29.7M| } 86| 5.04k| return counter; 87| 5.04k|} _ZN7simdutf6scalar5utf1617count_code_pointsILNS_10endiannessE1EEEmPKDsm: 79| 4.69k|simdutf_constexpr23 size_t count_code_points(const char16_t *p, size_t len) { 80| | // We are not BOM aware. 81| 4.69k| size_t counter{0}; 82| 27.1M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (82:22): [True: 27.1M, False: 4.69k] ------------------ 83| 27.1M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 84| 27.1M| counter += ((word & 0xFC00) != 0xDC00); 85| 27.1M| } 86| 4.69k| return counter; 87| 4.69k|} _ZN7simdutf6scalar5utf1622utf8_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 91| 2.32k| size_t len) { 92| | // We are not BOM aware. 93| 2.32k| size_t counter{0}; 94| 17.2M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (94:22): [True: 17.2M, False: 2.32k] ------------------ 95| 17.2M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 96| 17.2M| counter++; // ASCII 97| 17.2M| counter += static_cast( 98| 17.2M| word > 99| 17.2M| 0x7F); // non-ASCII is at least 2 bytes, surrogates are 2*2 == 4 bytes 100| 17.2M| counter += static_cast((word > 0x7FF && word <= 0xD7FF) || ------------------ | Branch (100:37): [True: 3.15M, False: 14.0M] | Branch (100:53): [True: 2.85M, False: 297k] ------------------ 101| 14.3M| (word >= 0xE000)); // three-byte ------------------ | Branch (101:36): [True: 235k, False: 14.1M] ------------------ 102| 17.2M| } 103| 2.32k| return counter; 104| 2.32k|} _ZN7simdutf6scalar5utf1622utf8_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 91| 2.26k| size_t len) { 92| | // We are not BOM aware. 93| 2.26k| size_t counter{0}; 94| 16.5M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (94:22): [True: 16.5M, False: 2.26k] ------------------ 95| 16.5M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 96| 16.5M| counter++; // ASCII 97| 16.5M| counter += static_cast( 98| 16.5M| word > 99| 16.5M| 0x7F); // non-ASCII is at least 2 bytes, surrogates are 2*2 == 4 bytes 100| 16.5M| counter += static_cast((word > 0x7FF && word <= 0xD7FF) || ------------------ | Branch (100:37): [True: 3.10M, False: 13.4M] | Branch (100:53): [True: 2.86M, False: 241k] ------------------ 101| 13.6M| (word >= 0xE000)); // three-byte ------------------ | Branch (101:36): [True: 201k, False: 13.4M] ------------------ 102| 16.5M| } 103| 2.26k| return counter; 104| 2.26k|} _ZN7simdutf6scalar5utf1623utf32_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 108| 420| size_t len) { 109| | // We are not BOM aware. 110| 420| size_t counter{0}; 111| 5.29M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (111:22): [True: 5.29M, False: 420] ------------------ 112| 5.29M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 113| 5.29M| counter += ((word & 0xFC00) != 0xDC00); 114| 5.29M| } 115| 420| return counter; 116| 420|} _ZN7simdutf6scalar5utf1623utf32_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 108| 382| size_t len) { 109| | // We are not BOM aware. 110| 382| size_t counter{0}; 111| 4.91M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (111:22): [True: 4.91M, False: 382] ------------------ 112| 4.91M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 113| 4.91M| counter += ((word & 0xFC00) != 0xDC00); 114| 4.91M| } 115| 382| return counter; 116| 382|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin17convertILNS_10endiannessE0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 266| OutputPtr latin_output) { 18| 266| if (len == 0) { ------------------ | Branch (18:7): [True: 6, False: 260] ------------------ 19| 6| return 0; 20| 6| } 21| 260| size_t pos = 0; 22| 260| const auto latin_output_start = latin_output; 23| 260| uint16_t word = 0; 24| 260| uint16_t too_large = 0; 25| | 26| 4.20M| while (pos < len) { ------------------ | Branch (26:10): [True: 4.20M, False: 260] ------------------ 27| 4.20M| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (27:12): [Folded, False: 4.20M] ------------------ 28| 4.20M| too_large |= word; 29| 4.20M| *latin_output++ = char(word & 0xFF); 30| 4.20M| pos++; 31| 4.20M| } 32| 260| if ((too_large & 0xFF00) != 0) { ------------------ | Branch (32:7): [True: 150, False: 110] ------------------ 33| 150| return 0; 34| 150| } 35| | 36| 110| return latin_output - latin_output_start; 37| 260|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin17convertILNS_10endiannessE1EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 202| OutputPtr latin_output) { 18| 202| if (len == 0) { ------------------ | Branch (18:7): [True: 6, False: 196] ------------------ 19| 6| return 0; 20| 6| } 21| 196| size_t pos = 0; 22| 196| const auto latin_output_start = latin_output; 23| 196| uint16_t word = 0; 24| 196| uint16_t too_large = 0; 25| | 26| 617k| while (pos < len) { ------------------ | Branch (26:10): [True: 617k, False: 196] ------------------ 27| 617k| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (27:12): [True: 617k, Folded] ------------------ 28| 617k| too_large |= word; 29| 617k| *latin_output++ = char(word & 0xFF); 30| 617k| pos++; 31| 617k| } 32| 196| if ((too_large & 0xFF00) != 0) { ------------------ | Branch (32:7): [True: 104, False: 92] ------------------ 33| 104| return 0; 34| 104| } 35| | 36| 92| return latin_output - latin_output_start; 37| 196|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin119convert_with_errorsILNS_10endiannessE0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEENS_6resultES8_mS9_: 45| 476| OutputPtr latin_output) { 46| 476| if (len == 0) { ------------------ | Branch (46:7): [True: 4, False: 472] ------------------ 47| 4| return result(error_code::SUCCESS, 0); 48| 4| } 49| 472| size_t pos = 0; 50| 472| auto start = latin_output; 51| 472| uint16_t word; 52| | 53| 413k| while (pos < len) { ------------------ | Branch (53:10): [True: 413k, False: 202] ------------------ 54| |#if SIMDUTF_CPLUSPLUS23 55| | if !consteval 56| |#endif 57| 413k| { 58| 413k| if (pos + 16 <= len) { // if it is safe to read 32 more bytes, check that ------------------ | Branch (58:11): [True: 411k, False: 1.68k] ------------------ 59| | // they are Latin1 60| 411k| uint64_t v1, v2, v3, v4; 61| 411k| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 62| 411k| ::memcpy(&v2, data + pos + 4, sizeof(uint64_t)); 63| 411k| ::memcpy(&v3, data + pos + 8, sizeof(uint64_t)); 64| 411k| ::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| 411k| if (((v1 | v2 | v3 | v4) & 0xFF00FF00FF00FF00) == 0) { ------------------ | Branch (79:13): [True: 411k, False: 346] ------------------ 80| 411k| size_t final_pos = pos + 16; 81| 6.99M| while (pos < final_pos) { ------------------ | Branch (81:18): [True: 6.58M, False: 411k] ------------------ 82| 6.58M| *latin_output++ = !match_system(big_endian) ------------------ | Branch (82:31): [Folded, False: 6.58M] ------------------ 83| 6.58M| ? char(u16_swap_bytes(data[pos])) 84| 6.58M| : char(data[pos]); 85| 6.58M| pos++; 86| 6.58M| } 87| 411k| continue; 88| 411k| } 89| 411k| } 90| 413k| } 91| | 92| 2.02k| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (92:12): [Folded, False: 2.02k] ------------------ 93| 2.02k| if ((word & 0xFF00) == 0) { ------------------ | Branch (93:9): [True: 1.75k, False: 270] ------------------ 94| 1.75k| *latin_output++ = char(word & 0xFF); 95| 1.75k| pos++; 96| 1.75k| } else { 97| 270| return result(error_code::TOO_LARGE, pos); 98| 270| } 99| 2.02k| } 100| 202| return result(error_code::SUCCESS, latin_output - start); 101| 472|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin119convert_with_errorsILNS_10endiannessE1EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEENS_6resultES8_mS9_: 45| 432| OutputPtr latin_output) { 46| 432| if (len == 0) { ------------------ | Branch (46:7): [True: 4, False: 428] ------------------ 47| 4| return result(error_code::SUCCESS, 0); 48| 4| } 49| 428| size_t pos = 0; 50| 428| auto start = latin_output; 51| 428| uint16_t word; 52| | 53| 465k| while (pos < len) { ------------------ | Branch (53:10): [True: 465k, False: 182] ------------------ 54| |#if SIMDUTF_CPLUSPLUS23 55| | if !consteval 56| |#endif 57| 465k| { 58| 465k| if (pos + 16 <= len) { // if it is safe to read 32 more bytes, check that ------------------ | Branch (58:11): [True: 463k, False: 1.37k] ------------------ 59| | // they are Latin1 60| 463k| uint64_t v1, v2, v3, v4; 61| 463k| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 62| 463k| ::memcpy(&v2, data + pos + 4, sizeof(uint64_t)); 63| 463k| ::memcpy(&v3, data + pos + 8, sizeof(uint64_t)); 64| 463k| ::memcpy(&v4, data + pos + 12, sizeof(uint64_t)); 65| | 66| 463k| if constexpr (!match_system(big_endian)) { 67| 463k| v1 = (v1 >> 8) | (v1 << (64 - 8)); 68| 463k| } 69| 463k| if constexpr (!match_system(big_endian)) { 70| 463k| v2 = (v2 >> 8) | (v2 << (64 - 8)); 71| 463k| } 72| 463k| if constexpr (!match_system(big_endian)) { 73| 463k| v3 = (v3 >> 8) | (v3 << (64 - 8)); 74| 463k| } 75| 463k| if constexpr (!match_system(big_endian)) { 76| 463k| v4 = (v4 >> 8) | (v4 << (64 - 8)); 77| 463k| } 78| | 79| 463k| if (((v1 | v2 | v3 | v4) & 0xFF00FF00FF00FF00) == 0) { ------------------ | Branch (79:13): [True: 463k, False: 390] ------------------ 80| 463k| size_t final_pos = pos + 16; 81| 7.87M| while (pos < final_pos) { ------------------ | Branch (81:18): [True: 7.41M, False: 463k] ------------------ 82| 7.41M| *latin_output++ = !match_system(big_endian) ------------------ | Branch (82:31): [True: 7.41M, Folded] ------------------ 83| 7.41M| ? char(u16_swap_bytes(data[pos])) 84| 7.41M| : char(data[pos]); 85| 7.41M| pos++; 86| 7.41M| } 87| 463k| continue; 88| 463k| } 89| 463k| } 90| 465k| } 91| | 92| 1.76k| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (92:12): [True: 1.76k, Folded] ------------------ 93| 1.76k| if ((word & 0xFF00) == 0) { ------------------ | Branch (93:9): [True: 1.51k, False: 246] ------------------ 94| 1.51k| *latin_output++ = char(word & 0xFF); 95| 1.51k| pos++; 96| 1.51k| } else { 97| 246| return result(error_code::TOO_LARGE, pos); 98| 246| } 99| 1.76k| } 100| 182| return result(error_code::SUCCESS, latin_output - start); 101| 428|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf327convertILNS_10endiannessE0EEEmPKDsmPDi: 11| 774| char32_t *utf32_output) { 12| 774| size_t pos = 0; 13| 774| char32_t *start{utf32_output}; 14| 1.23M| while (pos < len) { ------------------ | Branch (14:10): [True: 1.23M, False: 476] ------------------ 15| 1.23M| uint16_t word = 16| 1.23M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [Folded, False: 1.23M] ------------------ 17| 1.23M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 1.22M, False: 2.94k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 1.22M| *utf32_output++ = char32_t(word); 20| 1.22M| pos++; 21| 1.22M| } else { 22| | // must be a surrogate pair 23| 2.94k| uint16_t diff = uint16_t(word - 0xD800); 24| 2.94k| if (diff > 0x3FF) { ------------------ | Branch (24:11): [True: 112, False: 2.82k] ------------------ 25| 112| return 0; 26| 112| } 27| 2.82k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 78, False: 2.75k] ------------------ 28| 78| return 0; 29| 78| } // minimal bound checking 30| 2.75k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (30:28): [Folded, False: 2.75k] ------------------ 31| 2.75k| ? u16_swap_bytes(data[pos + 1]) 32| 2.75k| : data[pos + 1]; 33| 2.75k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 34| 2.75k| if (diff2 > 0x3FF) { ------------------ | Branch (34:11): [True: 108, False: 2.64k] ------------------ 35| 108| return 0; 36| 108| } 37| 2.64k| uint32_t value = (diff << 10) + diff2 + 0x10000; 38| 2.64k| *utf32_output++ = char32_t(value); 39| 2.64k| pos += 2; 40| 2.64k| } 41| 1.23M| } 42| 476| return utf32_output - start; 43| 774|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf327convertILNS_10endiannessE1EEEmPKDsmPDi: 11| 762| char32_t *utf32_output) { 12| 762| size_t pos = 0; 13| 762| char32_t *start{utf32_output}; 14| 1.20M| while (pos < len) { ------------------ | Branch (14:10): [True: 1.20M, False: 474] ------------------ 15| 1.20M| uint16_t word = 16| 1.20M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [True: 1.20M, Folded] ------------------ 17| 1.20M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 1.20M, False: 1.57k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 1.20M| *utf32_output++ = char32_t(word); 20| 1.20M| pos++; 21| 1.20M| } else { 22| | // must be a surrogate pair 23| 1.57k| uint16_t diff = uint16_t(word - 0xD800); 24| 1.57k| if (diff > 0x3FF) { ------------------ | Branch (24:11): [True: 84, False: 1.49k] ------------------ 25| 84| return 0; 26| 84| } 27| 1.49k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 108, False: 1.38k] ------------------ 28| 108| return 0; 29| 108| } // minimal bound checking 30| 1.38k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (30:28): [True: 1.38k, Folded] ------------------ 31| 1.38k| ? u16_swap_bytes(data[pos + 1]) 32| 1.38k| : data[pos + 1]; 33| 1.38k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 34| 1.38k| if (diff2 > 0x3FF) { ------------------ | Branch (34:11): [True: 96, False: 1.28k] ------------------ 35| 96| return 0; 36| 96| } 37| 1.28k| uint32_t value = (diff << 10) + diff2 + 0x10000; 38| 1.28k| *utf32_output++ = char32_t(value); 39| 1.28k| pos += 2; 40| 1.28k| } 41| 1.20M| } 42| 474| return utf32_output - start; 43| 762|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3219convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKDsmPDi: 47| 858| char32_t *utf32_output) { 48| 858| size_t pos = 0; 49| 858| char32_t *start{utf32_output}; 50| 4.29M| while (pos < len) { ------------------ | Branch (50:10): [True: 4.29M, False: 464] ------------------ 51| 4.29M| uint16_t word = 52| 4.29M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (52:9): [Folded, False: 4.29M] ------------------ 53| 4.29M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (53:9): [True: 4.27M, False: 15.1k] ------------------ 54| | // No surrogate pair, extend 16-bit word to 32-bit word 55| 4.27M| *utf32_output++ = char32_t(word); 56| 4.27M| pos++; 57| 4.27M| } else { 58| | // must be a surrogate pair 59| 15.1k| uint16_t diff = uint16_t(word - 0xD800); 60| 15.1k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 122, False: 15.0k] ------------------ 61| 122| return result(error_code::SURROGATE, pos); 62| 122| } 63| 15.0k| if (pos + 1 >= len) { ------------------ | Branch (63:11): [True: 96, False: 14.9k] ------------------ 64| 96| return result(error_code::SURROGATE, pos); 65| 96| } // minimal bound checking 66| 14.9k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (66:28): [Folded, False: 14.9k] ------------------ 67| 14.9k| ? u16_swap_bytes(data[pos + 1]) 68| 14.9k| : data[pos + 1]; 69| 14.9k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 70| 14.9k| if (diff2 > 0x3FF) { ------------------ | Branch (70:11): [True: 176, False: 14.7k] ------------------ 71| 176| return result(error_code::SURROGATE, pos); 72| 176| } 73| 14.7k| uint32_t value = (diff << 10) + diff2 + 0x10000; 74| 14.7k| *utf32_output++ = char32_t(value); 75| 14.7k| pos += 2; 76| 14.7k| } 77| 4.29M| } 78| 464| return result(error_code::SUCCESS, utf32_output - start); 79| 858|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3219convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKDsmPDi: 47| 780| char32_t *utf32_output) { 48| 780| size_t pos = 0; 49| 780| char32_t *start{utf32_output}; 50| 2.52M| while (pos < len) { ------------------ | Branch (50:10): [True: 2.52M, False: 424] ------------------ 51| 2.52M| uint16_t word = 52| 2.52M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (52:9): [True: 2.52M, Folded] ------------------ 53| 2.52M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (53:9): [True: 2.51M, False: 10.8k] ------------------ 54| | // No surrogate pair, extend 16-bit word to 32-bit word 55| 2.51M| *utf32_output++ = char32_t(word); 56| 2.51M| pos++; 57| 2.51M| } else { 58| | // must be a surrogate pair 59| 10.8k| uint16_t diff = uint16_t(word - 0xD800); 60| 10.8k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 106, False: 10.7k] ------------------ 61| 106| return result(error_code::SURROGATE, pos); 62| 106| } 63| 10.7k| if (pos + 1 >= len) { ------------------ | Branch (63:11): [True: 96, False: 10.6k] ------------------ 64| 96| return result(error_code::SURROGATE, pos); 65| 96| } // minimal bound checking 66| 10.6k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (66:28): [True: 10.6k, Folded] ------------------ 67| 10.6k| ? u16_swap_bytes(data[pos + 1]) 68| 10.6k| : data[pos + 1]; 69| 10.6k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 70| 10.6k| if (diff2 > 0x3FF) { ------------------ | Branch (70:11): [True: 154, False: 10.4k] ------------------ 71| 154| return result(error_code::SURROGATE, pos); 72| 154| } 73| 10.4k| uint32_t value = (diff << 10) + diff2 + 0x10000; 74| 10.4k| *utf32_output++ = char32_t(value); 75| 10.4k| pos += 2; 76| 10.4k| } 77| 2.52M| } 78| 424| return result(error_code::SUCCESS, utf32_output - start); 79| 780|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3213convert_validILNS_10endiannessE0EEEmPKDsmPDi: 11| 106| char32_t *utf32_output) { 12| 106| size_t pos = 0; 13| 106| char32_t *start{utf32_output}; 14| 1.10M| while (pos < len) { ------------------ | Branch (14:10): [True: 1.10M, False: 106] ------------------ 15| 1.10M| uint16_t word = 16| 1.10M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [Folded, False: 1.10M] ------------------ 17| 1.10M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 1.10M, False: 2.86k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 1.10M| *utf32_output++ = char32_t(word); 20| 1.10M| pos++; 21| 1.10M| } else { 22| | // must be a surrogate pair 23| 2.86k| uint16_t diff = uint16_t(word - 0xD800); 24| 2.86k| if (pos + 1 >= len) { ------------------ | Branch (24:11): [True: 0, False: 2.86k] ------------------ 25| 0| return 0; 26| 0| } // minimal bound checking 27| 2.86k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (27:28): [Folded, False: 2.86k] ------------------ 28| 2.86k| ? u16_swap_bytes(data[pos + 1]) 29| 2.86k| : data[pos + 1]; 30| 2.86k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 31| 2.86k| uint32_t value = (diff << 10) + diff2 + 0x10000; 32| 2.86k| *utf32_output++ = char32_t(value); 33| 2.86k| pos += 2; 34| 2.86k| } 35| 1.10M| } 36| 106| return utf32_output - start; 37| 106|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3213convert_validILNS_10endiannessE1EEEmPKDsmPDi: 11| 172| char32_t *utf32_output) { 12| 172| size_t pos = 0; 13| 172| char32_t *start{utf32_output}; 14| 3.78M| while (pos < len) { ------------------ | Branch (14:10): [True: 3.78M, False: 172] ------------------ 15| 3.78M| uint16_t word = 16| 3.78M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [True: 3.78M, Folded] ------------------ 17| 3.78M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 3.78M, False: 1.17k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 3.78M| *utf32_output++ = char32_t(word); 20| 3.78M| pos++; 21| 3.78M| } else { 22| | // must be a surrogate pair 23| 1.17k| uint16_t diff = uint16_t(word - 0xD800); 24| 1.17k| if (pos + 1 >= len) { ------------------ | Branch (24:11): [True: 0, False: 1.17k] ------------------ 25| 0| return 0; 26| 0| } // minimal bound checking 27| 1.17k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (27:28): [True: 1.17k, Folded] ------------------ 28| 1.17k| ? u16_swap_bytes(data[pos + 1]) 29| 1.17k| : data[pos + 1]; 30| 1.17k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 31| 1.17k| uint32_t value = (diff << 10) + diff2 + 0x10000; 32| 1.17k| *utf32_output++ = char32_t(value); 33| 1.17k| pos += 2; 34| 1.17k| } 35| 3.78M| } 36| 172| return utf32_output - start; 37| 172|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf87convertILNS_10endiannessE0EPKDsPcQsr7simdutf6detailE18indexes_into_utf16IT0_EEEmS8_mT1_: 17| 1.86k| OutputPtr utf8_output) { 18| 1.86k| size_t pos = 0; 19| 1.86k| const auto start = utf8_output; 20| 4.80M| while (pos < len) { ------------------ | Branch (20:10): [True: 4.80M, False: 1.50k] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 4.80M| { 25| | // try to convert the next block of 8 bytes 26| 4.80M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 4.80M, False: 3.41k] ------------------ 27| | // they are ascii 28| 4.80M| uint64_t v; 29| 4.80M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| | if constexpr (!match_system(big_endian)) { 31| | v = (v >> 8) | (v << (64 - 8)); 32| | } 33| 4.80M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.27M, False: 3.52M] ------------------ 34| 1.27M| size_t final_pos = pos + 4; 35| 6.39M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 5.11M, False: 1.27M] ------------------ 36| 5.11M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [Folded, False: 5.11M] ------------------ 37| 5.11M| ? char(u16_swap_bytes(data[pos])) 38| 5.11M| : char(data[pos]); 39| 5.11M| pos++; 40| 5.11M| } 41| 1.27M| continue; 42| 1.27M| } 43| 4.80M| } 44| 4.80M| } 45| 3.52M| uint16_t word = 46| 3.52M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (46:9): [Folded, False: 3.52M] ------------------ 47| 3.52M| if ((word & 0xFF80) == 0) { ------------------ | Branch (47:9): [True: 122k, False: 3.40M] ------------------ 48| | // will generate one UTF-8 bytes 49| 122k| *utf8_output++ = char(word); 50| 122k| pos++; 51| 3.40M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (51:16): [True: 3.20M, False: 204k] ------------------ 52| | // will generate two UTF-8 bytes 53| | // we have 0b110XXXXX 0b10XXXXXX 54| 3.20M| *utf8_output++ = char((word >> 6) | 0b11000000); 55| 3.20M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 56| 3.20M| pos++; 57| 3.20M| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (57:16): [True: 199k, False: 5.12k] ------------------ 58| | // will generate three UTF-8 bytes 59| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 60| 199k| *utf8_output++ = char((word >> 12) | 0b11100000); 61| 199k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 62| 199k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 63| 199k| pos++; 64| 199k| } else { 65| | // must be a surrogate pair 66| 5.12k| if (pos + 1 >= len) { ------------------ | Branch (66:11): [True: 54, False: 5.06k] ------------------ 67| 54| return 0; 68| 54| } 69| 5.06k| uint16_t diff = uint16_t(word - 0xD800); 70| 5.06k| if (diff > 0x3FF) { ------------------ | Branch (70:11): [True: 150, False: 4.91k] ------------------ 71| 150| return 0; 72| 150| } 73| 4.91k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (73:28): [Folded, False: 4.91k] ------------------ 74| 4.91k| ? u16_swap_bytes(data[pos + 1]) 75| 4.91k| : data[pos + 1]; 76| 4.91k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 77| 4.91k| if (diff2 > 0x3FF) { ------------------ | Branch (77:11): [True: 154, False: 4.76k] ------------------ 78| 154| return 0; 79| 154| } 80| 4.76k| uint32_t value = (diff << 10) + diff2 + 0x10000; 81| | // will generate four UTF-8 bytes 82| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 83| 4.76k| *utf8_output++ = char((value >> 18) | 0b11110000); 84| 4.76k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 85| 4.76k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 86| 4.76k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 87| 4.76k| pos += 2; 88| 4.76k| } 89| 3.52M| } 90| 1.50k| return utf8_output - start; 91| 1.86k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf87convertILNS_10endiannessE1EPKDsPcQsr7simdutf6detailE18indexes_into_utf16IT0_EEEmS8_mT1_: 17| 1.64k| OutputPtr utf8_output) { 18| 1.64k| size_t pos = 0; 19| 1.64k| const auto start = utf8_output; 20| 3.57M| while (pos < len) { ------------------ | Branch (20:10): [True: 3.57M, False: 1.37k] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 3.57M| { 25| | // try to convert the next block of 8 bytes 26| 3.57M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 3.56M, False: 2.91k] ------------------ 27| | // they are ascii 28| 3.56M| uint64_t v; 29| 3.56M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| 3.56M| if constexpr (!match_system(big_endian)) { 31| 3.56M| v = (v >> 8) | (v << (64 - 8)); 32| 3.56M| } 33| 3.56M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.37M, False: 2.19M] ------------------ 34| 1.37M| size_t final_pos = pos + 4; 35| 6.85M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 5.48M, False: 1.37M] ------------------ 36| 5.48M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [True: 5.48M, Folded] ------------------ 37| 5.48M| ? char(u16_swap_bytes(data[pos])) 38| 5.48M| : char(data[pos]); 39| 5.48M| pos++; 40| 5.48M| } 41| 1.37M| continue; 42| 1.37M| } 43| 3.56M| } 44| 3.57M| } 45| 2.20M| uint16_t word = 46| 2.20M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (46:9): [True: 2.20M, Folded] ------------------ 47| 2.20M| if ((word & 0xFF80) == 0) { ------------------ | Branch (47:9): [True: 116k, False: 2.08M] ------------------ 48| | // will generate one UTF-8 bytes 49| 116k| *utf8_output++ = char(word); 50| 116k| pos++; 51| 2.08M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (51:16): [True: 36.6k, False: 2.04M] ------------------ 52| | // will generate two UTF-8 bytes 53| | // we have 0b110XXXXX 0b10XXXXXX 54| 36.6k| *utf8_output++ = char((word >> 6) | 0b11000000); 55| 36.6k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 56| 36.6k| pos++; 57| 2.04M| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (57:16): [True: 2.04M, False: 3.98k] ------------------ 58| | // will generate three UTF-8 bytes 59| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 60| 2.04M| *utf8_output++ = char((word >> 12) | 0b11100000); 61| 2.04M| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 62| 2.04M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 63| 2.04M| pos++; 64| 2.04M| } else { 65| | // must be a surrogate pair 66| 3.98k| if (pos + 1 >= len) { ------------------ | Branch (66:11): [True: 30, False: 3.95k] ------------------ 67| 30| return 0; 68| 30| } 69| 3.95k| uint16_t diff = uint16_t(word - 0xD800); 70| 3.95k| if (diff > 0x3FF) { ------------------ | Branch (70:11): [True: 126, False: 3.83k] ------------------ 71| 126| return 0; 72| 126| } 73| 3.83k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (73:28): [True: 3.83k, Folded] ------------------ 74| 3.83k| ? u16_swap_bytes(data[pos + 1]) 75| 3.83k| : data[pos + 1]; 76| 3.83k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 77| 3.83k| if (diff2 > 0x3FF) { ------------------ | Branch (77:11): [True: 114, False: 3.71k] ------------------ 78| 114| return 0; 79| 114| } 80| 3.71k| uint32_t value = (diff << 10) + diff2 + 0x10000; 81| | // will generate four UTF-8 bytes 82| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 83| 3.71k| *utf8_output++ = char((value >> 18) | 0b11110000); 84| 3.71k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 85| 3.71k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 86| 3.71k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 87| 3.71k| pos += 2; 88| 3.71k| } 89| 2.20M| } 90| 1.37k| return utf8_output - start; 91| 1.64k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf819convert_with_errorsILNS_10endiannessE0ELb0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT1_Esr7simdutf6detailE26index_assignable_from_charIT2_EEENS_11full_resultES8_mS9_m: 101| 1.67k| size_t utf8_len = 0) { 102| 1.67k| if (check_output && utf8_len == 0) { ------------------ | Branch (102:7): [Folded, False: 1.67k] | Branch (102:23): [True: 0, False: 0] ------------------ 103| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, 0, 0); 104| 0| } 105| | 106| 1.67k| size_t pos = 0; 107| 1.67k| auto start = utf8_output; 108| 1.67k| auto end = utf8_output + utf8_len; 109| | 110| 5.47M| while (pos < len) { ------------------ | Branch (110:10): [True: 5.47M, False: 1.17k] ------------------ 111| |#if SIMDUTF_CPLUSPLUS23 112| | if !consteval 113| |#endif 114| 5.47M| { 115| | // try to convert the next block of 8 bytes 116| 5.47M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (116:11): [True: 5.47M, False: 2.69k] ------------------ 117| | // they are ascii 118| 5.47M| uint64_t v; 119| 5.47M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 120| | if constexpr (!match_system(big_endian)) 121| | v = (v >> 8) | (v << (64 - 8)); 122| 5.47M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (122:13): [True: 1.11M, False: 4.36M] ------------------ 123| 1.11M| size_t final_pos = pos + 4; 124| 5.57M| while (pos < final_pos) { ------------------ | Branch (124:18): [True: 4.46M, False: 1.11M] ------------------ 125| 4.46M| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (125:17): [Folded, False: 4.46M] | 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| 4.46M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (129:30): [Folded, False: 4.46M] ------------------ 130| 4.46M| ? char(u16_swap_bytes(data[pos])) 131| 4.46M| : char(data[pos]); 132| 4.46M| pos++; 133| 4.46M| } 134| 1.11M| continue; 135| 1.11M| } 136| 5.47M| } 137| 5.47M| } 138| | 139| 4.36M| uint16_t word = 140| 4.36M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (140:9): [Folded, False: 4.36M] ------------------ 141| 4.36M| if ((word & 0xFF80) == 0) { ------------------ | Branch (141:9): [True: 209k, False: 4.15M] ------------------ 142| | // will generate one UTF-8 bytes 143| 209k| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (143:11): [Folded, False: 209k] | 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| 209k| *utf8_output++ = char(word); 148| 209k| pos++; 149| 4.15M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (149:16): [True: 1.38M, False: 2.77M] ------------------ 150| | // will generate two UTF-8 bytes 151| | // we have 0b110XXXXX 0b10XXXXXX 152| 1.38M| if (check_output && size_t(end - utf8_output) < 2) { ------------------ | Branch (152:11): [Folded, False: 1.38M] | 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| 1.38M| *utf8_output++ = char((word >> 6) | 0b11000000); 157| 1.38M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 158| 1.38M| pos++; 159| | 160| 2.77M| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (160:16): [True: 2.76M, False: 3.21k] ------------------ 161| | // will generate three UTF-8 bytes 162| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 163| 2.76M| if (check_output && size_t(end - utf8_output) < 3) { ------------------ | Branch (163:11): [Folded, False: 2.76M] | 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| 2.76M| *utf8_output++ = char((word >> 12) | 0b11100000); 168| 2.76M| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 169| 2.76M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 170| 2.76M| pos++; 171| 2.76M| } else { 172| | 173| 3.21k| if (check_output && size_t(end - utf8_output) < 4) { ------------------ | Branch (173:11): [Folded, False: 3.21k] | 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| 3.21k| if (pos + 1 >= len) { ------------------ | Branch (178:11): [True: 60, False: 3.15k] ------------------ 179| 60| return full_result(error_code::SURROGATE, pos, utf8_output - start); 180| 60| } 181| 3.15k| uint16_t diff = uint16_t(word - 0xD800); 182| 3.15k| if (diff > 0x3FF) { ------------------ | Branch (182:11): [True: 162, False: 2.98k] ------------------ 183| 162| return full_result(error_code::SURROGATE, pos, utf8_output - start); 184| 162| } 185| 2.98k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (185:28): [Folded, False: 2.98k] ------------------ 186| 2.98k| ? u16_swap_bytes(data[pos + 1]) 187| 2.98k| : data[pos + 1]; 188| 2.98k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 189| 2.98k| if (diff2 > 0x3FF) { ------------------ | Branch (189:11): [True: 274, False: 2.71k] ------------------ 190| 274| return full_result(error_code::SURROGATE, pos, utf8_output - start); 191| 274| } 192| 2.71k| uint32_t value = (diff << 10) + diff2 + 0x10000; 193| | // will generate four UTF-8 bytes 194| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 195| 2.71k| *utf8_output++ = char((value >> 18) | 0b11110000); 196| 2.71k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 197| 2.71k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 198| 2.71k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 199| 2.71k| pos += 2; 200| 2.71k| } 201| 4.36M| } 202| 1.17k| return full_result(error_code::SUCCESS, pos, utf8_output - start); 203| 1.67k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf819convert_with_errorsILNS_10endiannessE1ELb0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT1_Esr7simdutf6detailE26index_assignable_from_charIT2_EEENS_11full_resultES8_mS9_m: 101| 1.80k| size_t utf8_len = 0) { 102| 1.80k| if (check_output && utf8_len == 0) { ------------------ | Branch (102:7): [Folded, False: 1.80k] | Branch (102:23): [True: 0, False: 0] ------------------ 103| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, 0, 0); 104| 0| } 105| | 106| 1.80k| size_t pos = 0; 107| 1.80k| auto start = utf8_output; 108| 1.80k| auto end = utf8_output + utf8_len; 109| | 110| 4.39M| while (pos < len) { ------------------ | Branch (110:10): [True: 4.39M, False: 1.25k] ------------------ 111| |#if SIMDUTF_CPLUSPLUS23 112| | if !consteval 113| |#endif 114| 4.39M| { 115| | // try to convert the next block of 8 bytes 116| 4.39M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (116:11): [True: 4.39M, False: 2.79k] ------------------ 117| | // they are ascii 118| 4.39M| uint64_t v; 119| 4.39M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 120| | if constexpr (!match_system(big_endian)) 121| 4.39M| v = (v >> 8) | (v << (64 - 8)); 122| 4.39M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (122:13): [True: 2.21M, False: 2.18M] ------------------ 123| 2.21M| size_t final_pos = pos + 4; 124| 11.0M| while (pos < final_pos) { ------------------ | Branch (124:18): [True: 8.84M, False: 2.21M] ------------------ 125| 8.84M| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (125:17): [Folded, False: 8.84M] | 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| 8.84M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (129:30): [True: 8.84M, Folded] ------------------ 130| 8.84M| ? char(u16_swap_bytes(data[pos])) 131| 8.84M| : char(data[pos]); 132| 8.84M| pos++; 133| 8.84M| } 134| 2.21M| continue; 135| 2.21M| } 136| 4.39M| } 137| 4.39M| } 138| | 139| 2.18M| uint16_t word = 140| 2.18M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (140:9): [True: 2.18M, Folded] ------------------ 141| 2.18M| if ((word & 0xFF80) == 0) { ------------------ | Branch (141:9): [True: 749k, False: 1.43M] ------------------ 142| | // will generate one UTF-8 bytes 143| 749k| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (143:11): [Folded, False: 749k] | 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| 749k| *utf8_output++ = char(word); 148| 749k| pos++; 149| 1.43M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (149:16): [True: 317k, False: 1.11M] ------------------ 150| | // will generate two UTF-8 bytes 151| | // we have 0b110XXXXX 0b10XXXXXX 152| 317k| if (check_output && size_t(end - utf8_output) < 2) { ------------------ | Branch (152:11): [Folded, False: 317k] | 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| 317k| *utf8_output++ = char((word >> 6) | 0b11000000); 157| 317k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 158| 317k| pos++; 159| | 160| 1.11M| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (160:16): [True: 1.09M, False: 26.5k] ------------------ 161| | // will generate three UTF-8 bytes 162| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 163| 1.09M| if (check_output && size_t(end - utf8_output) < 3) { ------------------ | Branch (163:11): [Folded, False: 1.09M] | 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| 1.09M| *utf8_output++ = char((word >> 12) | 0b11100000); 168| 1.09M| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 169| 1.09M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 170| 1.09M| pos++; 171| 1.09M| } else { 172| | 173| 26.5k| if (check_output && size_t(end - utf8_output) < 4) { ------------------ | Branch (173:11): [Folded, False: 26.5k] | 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| 26.5k| if (pos + 1 >= len) { ------------------ | Branch (178:11): [True: 72, False: 26.4k] ------------------ 179| 72| return full_result(error_code::SURROGATE, pos, utf8_output - start); 180| 72| } 181| 26.4k| uint16_t diff = uint16_t(word - 0xD800); 182| 26.4k| if (diff > 0x3FF) { ------------------ | Branch (182:11): [True: 172, False: 26.2k] ------------------ 183| 172| return full_result(error_code::SURROGATE, pos, utf8_output - start); 184| 172| } 185| 26.2k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (185:28): [True: 26.2k, Folded] ------------------ 186| 26.2k| ? u16_swap_bytes(data[pos + 1]) 187| 26.2k| : data[pos + 1]; 188| 26.2k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 189| 26.2k| if (diff2 > 0x3FF) { ------------------ | Branch (189:11): [True: 300, False: 25.9k] ------------------ 190| 300| return full_result(error_code::SURROGATE, pos, utf8_output - start); 191| 300| } 192| 25.9k| uint32_t value = (diff << 10) + diff2 + 0x10000; 193| | // will generate four UTF-8 bytes 194| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 195| 25.9k| *utf8_output++ = char((value >> 18) | 0b11110000); 196| 25.9k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 197| 25.9k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 198| 25.9k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 199| 25.9k| pos += 2; 200| 25.9k| } 201| 2.18M| } 202| 1.25k| return full_result(error_code::SUCCESS, pos, utf8_output - start); 203| 1.80k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf813convert_validILNS_10endiannessE0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 440| OutputPtr utf8_output) { 18| 440| size_t pos = 0; 19| 440| auto start = utf8_output; 20| 2.66M| while (pos < len) { ------------------ | Branch (20:10): [True: 2.66M, False: 440] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 2.66M| { 25| | // try to convert the next block of 4 ASCII characters 26| 2.66M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 2.65M, False: 938] ------------------ 27| | // they are ascii 28| 2.65M| uint64_t v; 29| 2.65M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| | if constexpr (!match_system(big_endian)) { 31| | v = (v >> 8) | (v << (64 - 8)); 32| | } 33| 2.65M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.63M, False: 1.02M] ------------------ 34| 1.63M| size_t final_pos = pos + 4; 35| 8.17M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 6.54M, False: 1.63M] ------------------ 36| 6.54M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [Folded, False: 6.54M] ------------------ 37| 6.54M| ? char(u16_swap_bytes(data[pos])) 38| 6.54M| : char(data[pos]); 39| 6.54M| pos++; 40| 6.54M| } 41| 1.63M| continue; 42| 1.63M| } 43| 2.65M| } 44| 2.66M| } 45| | 46| 1.02M| uint16_t word = 47| 1.02M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (47:9): [Folded, False: 1.02M] ------------------ 48| 1.02M| if ((word & 0xFF80) == 0) { ------------------ | Branch (48:9): [True: 429k, False: 594k] ------------------ 49| | // will generate one UTF-8 bytes 50| 429k| *utf8_output++ = char(word); 51| 429k| pos++; 52| 594k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (52:16): [True: 224k, False: 370k] ------------------ 53| | // will generate two UTF-8 bytes 54| | // we have 0b110XXXXX 0b10XXXXXX 55| 224k| *utf8_output++ = char((word >> 6) | 0b11000000); 56| 224k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 57| 224k| pos++; 58| 370k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (58:16): [True: 363k, False: 6.83k] ------------------ 59| | // will generate three UTF-8 bytes 60| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 61| 363k| *utf8_output++ = char((word >> 12) | 0b11100000); 62| 363k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 63| 363k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 64| 363k| pos++; 65| 363k| } else { 66| | // must be a surrogate pair 67| 6.83k| uint16_t diff = uint16_t(word - 0xD800); 68| 6.83k| if (pos + 1 >= len) { ------------------ | Branch (68:11): [True: 0, False: 6.83k] ------------------ 69| 0| return 0; 70| 0| } // minimal bound checking 71| 6.83k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (71:28): [Folded, False: 6.83k] ------------------ 72| 6.83k| ? u16_swap_bytes(data[pos + 1]) 73| 6.83k| : data[pos + 1]; 74| 6.83k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 75| 6.83k| uint32_t value = (diff << 10) + diff2 + 0x10000; 76| | // will generate four UTF-8 bytes 77| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 78| 6.83k| *utf8_output++ = char((value >> 18) | 0b11110000); 79| 6.83k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 80| 6.83k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 81| 6.83k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 82| 6.83k| pos += 2; 83| 6.83k| } 84| 1.02M| } 85| 440| return utf8_output - start; 86| 440|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf813convert_validILNS_10endiannessE1EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 398| OutputPtr utf8_output) { 18| 398| size_t pos = 0; 19| 398| auto start = utf8_output; 20| 2.40M| while (pos < len) { ------------------ | Branch (20:10): [True: 2.40M, False: 398] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 2.40M| { 25| | // try to convert the next block of 4 ASCII characters 26| 2.40M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 2.40M, False: 774] ------------------ 27| | // they are ascii 28| 2.40M| uint64_t v; 29| 2.40M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| 2.40M| if constexpr (!match_system(big_endian)) { 31| 2.40M| v = (v >> 8) | (v << (64 - 8)); 32| 2.40M| } 33| 2.40M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.55M, False: 851k] ------------------ 34| 1.55M| size_t final_pos = pos + 4; 35| 7.75M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 6.20M, False: 1.55M] ------------------ 36| 6.20M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [True: 6.20M, Folded] ------------------ 37| 6.20M| ? char(u16_swap_bytes(data[pos])) 38| 6.20M| : char(data[pos]); 39| 6.20M| pos++; 40| 6.20M| } 41| 1.55M| continue; 42| 1.55M| } 43| 2.40M| } 44| 2.40M| } 45| | 46| 852k| uint16_t word = 47| 852k| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (47:9): [True: 852k, Folded] ------------------ 48| 852k| if ((word & 0xFF80) == 0) { ------------------ | Branch (48:9): [True: 241k, False: 610k] ------------------ 49| | // will generate one UTF-8 bytes 50| 241k| *utf8_output++ = char(word); 51| 241k| pos++; 52| 610k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (52:16): [True: 72.5k, False: 537k] ------------------ 53| | // will generate two UTF-8 bytes 54| | // we have 0b110XXXXX 0b10XXXXXX 55| 72.5k| *utf8_output++ = char((word >> 6) | 0b11000000); 56| 72.5k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 57| 72.5k| pos++; 58| 537k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (58:16): [True: 535k, False: 1.98k] ------------------ 59| | // will generate three UTF-8 bytes 60| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 61| 535k| *utf8_output++ = char((word >> 12) | 0b11100000); 62| 535k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 63| 535k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 64| 535k| pos++; 65| 535k| } else { 66| | // must be a surrogate pair 67| 1.98k| uint16_t diff = uint16_t(word - 0xD800); 68| 1.98k| if (pos + 1 >= len) { ------------------ | Branch (68:11): [True: 0, False: 1.98k] ------------------ 69| 0| return 0; 70| 0| } // minimal bound checking 71| 1.98k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (71:28): [True: 1.98k, Folded] ------------------ 72| 1.98k| ? u16_swap_bytes(data[pos + 1]) 73| 1.98k| : data[pos + 1]; 74| 1.98k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 75| 1.98k| uint32_t value = (diff << 10) + diff2 + 0x10000; 76| | // will generate four UTF-8 bytes 77| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 78| 1.98k| *utf8_output++ = char((value >> 18) | 0b11110000); 79| 1.98k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 80| 1.98k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 81| 1.98k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 82| 1.98k| pos += 2; 83| 1.98k| } 84| 852k| } 85| 398| return utf8_output - start; 86| 398|} _ZN7simdutf6scalar5utf328validateEPKDim: 25| 5.68k| size_t len) noexcept { 26| 5.68k| return validate(reinterpret_cast(buf), len); 27| 5.68k|} _ZN7simdutf6scalar5utf3220validate_with_errorsEPKDim: 49| 7.78k|validate_with_errors(const char32_t *buf, size_t len) noexcept { 50| 7.78k| return validate_with_errors(reinterpret_cast(buf), len); 51| 7.78k|} _ZN7simdutf6scalar5utf3222utf8_length_from_utf32EPKDim: 54| 2.81k| size_t len) { 55| | // We are not BOM aware. 56| 2.81k| size_t counter{0}; 57| 2.41M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (57:22): [True: 2.41M, False: 2.81k] ------------------ 58| | // credit: @ttsugriy for the vectorizable approach 59| 2.41M| counter++; // ASCII 60| 2.41M| counter += static_cast(p[i] > 0x7F); // two-byte 61| 2.41M| counter += static_cast(p[i] > 0x7FF); // three-byte 62| 2.41M| counter += static_cast(p[i] > 0xFFFF); // four-bytes 63| 2.41M| } 64| 2.81k| return counter; 65| 2.81k|} _ZN7simdutf6scalar5utf3223utf16_length_from_utf32EPKDim: 68| 4.15k|utf16_length_from_utf32(const char32_t *p, size_t len) { 69| | // We are not BOM aware. 70| 4.15k| size_t counter{0}; 71| 6.11M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (71:22): [True: 6.11M, False: 4.15k] ------------------ 72| 6.11M| counter++; // non-surrogate word 73| 6.11M| counter += static_cast(p[i] > 0xFFFF); // surrogate pair 74| 6.11M| } 75| 4.15k| return counter; 76| 4.15k|} _ZN7simdutf6scalar5utf328validateIPKjQsr7simdutf6detailE19indexes_into_uint32IT_EEEbS5_m: 13| 5.68k| size_t len) noexcept { 14| 5.68k| uint64_t pos = 0; 15| 1.74M| for (; pos < len; pos++) { ------------------ | Branch (15:10): [True: 1.74M, False: 2.51k] ------------------ 16| 1.74M| uint32_t word = data[pos]; 17| 1.74M| if (word > 0x10FFFF || (word >= 0xD800 && word <= 0xDFFF)) { ------------------ | Branch (17:9): [True: 2.67k, False: 1.74M] | Branch (17:29): [True: 40.1k, False: 1.70M] | Branch (17:47): [True: 495, False: 39.6k] ------------------ 18| 3.17k| return false; 19| 3.17k| } 20| 1.74M| } 21| 2.51k| return true; 22| 5.68k|} _ZN7simdutf6scalar5utf3220validate_with_errorsIPKjQsr7simdutf6detailE19indexes_into_uint32IT_EEENS_6resultES5_m: 34| 7.78k|validate_with_errors(InputPtr data, size_t len) noexcept { 35| 7.78k| size_t pos = 0; 36| 1.75M| for (; pos < len; pos++) { ------------------ | Branch (36:10): [True: 1.75M, False: 2.51k] ------------------ 37| 1.75M| uint32_t word = data[pos]; 38| 1.75M| if (word > 0x10FFFF) { ------------------ | Branch (38:9): [True: 4.37k, False: 1.74M] ------------------ 39| 4.37k| return result(error_code::TOO_LARGE, pos); 40| 4.37k| } 41| 1.74M| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (41:9): [True: 41.5k, False: 1.70M] | Branch (41:27): [True: 894, False: 40.6k] ------------------ 42| 894| return result(error_code::SURROGATE, pos); 43| 894| } 44| 1.74M| } 45| 2.51k| return result(error_code::SUCCESS, pos); 46| 7.78k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf32_to_latin17convertEPKDimPc: 12| 426| char *latin1_output) { 13| 426| char *start = latin1_output; 14| 426| uint32_t utf32_char; 15| 426| size_t pos = 0; 16| 426| uint32_t too_large = 0; 17| | 18| 19.6k| while (pos < len) { ------------------ | Branch (18:10): [True: 19.1k, False: 426] ------------------ 19| 19.1k| utf32_char = (uint32_t)data[pos]; 20| 19.1k| too_large |= utf32_char; 21| 19.1k| *latin1_output++ = (char)(utf32_char & 0xFF); 22| 19.1k| pos++; 23| 19.1k| } 24| 426| if ((too_large & 0xFFFFFF00) != 0) { ------------------ | Branch (24:7): [True: 332, False: 94] ------------------ 25| 332| return 0; 26| 332| } 27| 94| return latin1_output - start; 28| 426|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf32_to_latin119convert_with_errorsEPKDimPc: 32| 1.18k| char *latin1_output) { 33| 1.18k| char *start{latin1_output}; 34| 1.18k| size_t pos = 0; 35| 236k| while (pos < len) { ------------------ | Branch (35:10): [True: 236k, False: 94] ------------------ 36| |#if SIMDUTF_CPLUSPLUS23 37| | if !consteval 38| |#endif 39| 236k| { 40| 236k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (40:11): [True: 236k, False: 312] ------------------ 41| | // they are Latin1 42| 236k| uint64_t v; 43| 236k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 44| 236k| if ((v & 0xFFFFFF00FFFFFF00) == 0) { ------------------ | Branch (44:13): [True: 235k, False: 996] ------------------ 45| 235k| *latin1_output++ = char(data[pos]); 46| 235k| *latin1_output++ = char(data[pos + 1]); 47| 235k| pos += 2; 48| 235k| continue; 49| 235k| } 50| 236k| } 51| 236k| } 52| | 53| 1.30k| uint32_t utf32_char = data[pos]; 54| 1.30k| if ((utf32_char & 0xFFFFFF00) == ------------------ | Branch (54:9): [True: 222, False: 1.08k] ------------------ 55| 1.30k| 0) { // Check if the character can be represented in Latin-1 56| 222| *latin1_output++ = (char)(utf32_char & 0xFF); 57| 222| pos++; 58| 1.08k| } else { 59| 1.08k| return result(error_code::TOO_LARGE, pos); 60| 1.08k| }; 61| 222| } 62| 94| return result(error_code::SUCCESS, latin1_output - start); 63| 1.18k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf167convertILNS_10endiannessE0EEEmPKDimPDs: 11| 1.82k| char16_t *utf16_output) { 12| 1.82k| size_t pos = 0; 13| 1.82k| char16_t *start{utf16_output}; 14| 538k| while (pos < len) { ------------------ | Branch (14:10): [True: 537k, False: 938] ------------------ 15| 537k| uint32_t word = data[pos]; 16| 537k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 534k, False: 3.35k] ------------------ 17| 534k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (17:11): [True: 1.75k, False: 532k] | Branch (17:29): [True: 192, False: 1.55k] ------------------ 18| 192| return 0; 19| 192| } 20| | // will not generate a surrogate pair 21| 534k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (21:25): [Folded, False: 534k] ------------------ 22| 534k| ? char16_t(u16_swap_bytes(uint16_t(word))) 23| 534k| : char16_t(word); 24| 534k| } else { 25| | // will generate a surrogate pair 26| 3.35k| if (word > 0x10FFFF) { ------------------ | Branch (26:11): [True: 690, False: 2.66k] ------------------ 27| 690| return 0; 28| 690| } 29| 2.66k| word -= 0x10000; 30| 2.66k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 31| 2.66k| 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| 2.66k| *utf16_output++ = char16_t(high_surrogate); 37| 2.66k| *utf16_output++ = char16_t(low_surrogate); 38| 2.66k| } 39| 536k| pos++; 40| 536k| } 41| 938| return utf16_output - start; 42| 1.82k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf167convertILNS_10endiannessE1EEEmPKDimPDs: 11| 1.53k| char16_t *utf16_output) { 12| 1.53k| size_t pos = 0; 13| 1.53k| char16_t *start{utf16_output}; 14| 176k| while (pos < len) { ------------------ | Branch (14:10): [True: 176k, False: 714] ------------------ 15| 176k| uint32_t word = data[pos]; 16| 176k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 173k, False: 2.83k] ------------------ 17| 173k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (17:11): [True: 1.18k, False: 171k] | Branch (17:29): [True: 166, False: 1.01k] ------------------ 18| 166| return 0; 19| 166| } 20| | // will not generate a surrogate pair 21| 173k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (21:25): [True: 173k, Folded] ------------------ 22| 173k| ? char16_t(u16_swap_bytes(uint16_t(word))) 23| 173k| : char16_t(word); 24| 173k| } else { 25| | // will generate a surrogate pair 26| 2.83k| if (word > 0x10FFFF) { ------------------ | Branch (26:11): [True: 656, False: 2.18k] ------------------ 27| 656| return 0; 28| 656| } 29| 2.18k| word -= 0x10000; 30| 2.18k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 31| 2.18k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 32| 2.18k| if constexpr (!match_system(big_endian)) { 33| 2.18k| high_surrogate = u16_swap_bytes(high_surrogate); 34| 2.18k| low_surrogate = u16_swap_bytes(low_surrogate); 35| 2.18k| } 36| 2.18k| *utf16_output++ = char16_t(high_surrogate); 37| 2.18k| *utf16_output++ = char16_t(low_surrogate); 38| 2.18k| } 39| 175k| pos++; 40| 175k| } 41| 714| return utf16_output - start; 42| 1.53k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1619convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKDimPDs: 46| 2.02k| char16_t *utf16_output) { 47| 2.02k| size_t pos = 0; 48| 2.02k| char16_t *start{utf16_output}; 49| 445k| while (pos < len) { ------------------ | Branch (49:10): [True: 445k, False: 512] ------------------ 50| 445k| uint32_t word = data[pos]; 51| 445k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (51:9): [True: 436k, False: 8.99k] ------------------ 52| 436k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (52:11): [True: 1.61k, False: 434k] | Branch (52:29): [True: 300, False: 1.31k] ------------------ 53| 300| return result(error_code::SURROGATE, pos); 54| 300| } 55| | // will not generate a surrogate pair 56| 436k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (56:25): [Folded, False: 436k] ------------------ 57| 436k| ? char16_t(u16_swap_bytes(uint16_t(word))) 58| 436k| : char16_t(word); 59| 436k| } else { 60| | // will generate a surrogate pair 61| 8.99k| if (word > 0x10FFFF) { ------------------ | Branch (61:11): [True: 1.21k, False: 7.77k] ------------------ 62| 1.21k| return result(error_code::TOO_LARGE, pos); 63| 1.21k| } 64| 7.77k| word -= 0x10000; 65| 7.77k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 66| 7.77k| 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| 7.77k| *utf16_output++ = char16_t(high_surrogate); 72| 7.77k| *utf16_output++ = char16_t(low_surrogate); 73| 7.77k| } 74| 443k| pos++; 75| 443k| } 76| 512| return result(error_code::SUCCESS, utf16_output - start); 77| 2.02k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1619convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKDimPDs: 46| 1.86k| char16_t *utf16_output) { 47| 1.86k| size_t pos = 0; 48| 1.86k| char16_t *start{utf16_output}; 49| 28.3k| while (pos < len) { ------------------ | Branch (49:10): [True: 27.8k, False: 490] ------------------ 50| 27.8k| uint32_t word = data[pos]; 51| 27.8k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (51:9): [True: 23.6k, False: 4.17k] ------------------ 52| 23.6k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (52:11): [True: 1.53k, False: 22.1k] | Branch (52:29): [True: 192, False: 1.34k] ------------------ 53| 192| return result(error_code::SURROGATE, pos); 54| 192| } 55| | // will not generate a surrogate pair 56| 23.4k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (56:25): [True: 23.4k, Folded] ------------------ 57| 23.4k| ? char16_t(u16_swap_bytes(uint16_t(word))) 58| 23.4k| : char16_t(word); 59| 23.4k| } else { 60| | // will generate a surrogate pair 61| 4.17k| if (word > 0x10FFFF) { ------------------ | Branch (61:11): [True: 1.18k, False: 2.99k] ------------------ 62| 1.18k| return result(error_code::TOO_LARGE, pos); 63| 1.18k| } 64| 2.99k| word -= 0x10000; 65| 2.99k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 66| 2.99k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 67| 2.99k| if constexpr (!match_system(big_endian)) { 68| 2.99k| high_surrogate = u16_swap_bytes(high_surrogate); 69| 2.99k| low_surrogate = u16_swap_bytes(low_surrogate); 70| 2.99k| } 71| 2.99k| *utf16_output++ = char16_t(high_surrogate); 72| 2.99k| *utf16_output++ = char16_t(low_surrogate); 73| 2.99k| } 74| 26.4k| pos++; 75| 26.4k| } 76| 490| return result(error_code::SUCCESS, utf16_output - start); 77| 1.86k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1613convert_validILNS_10endiannessE0EEEmPKDimPDs: 11| 248| char16_t *utf16_output) { 12| 248| size_t pos = 0; 13| 248| char16_t *start{utf16_output}; 14| 532k| while (pos < len) { ------------------ | Branch (14:10): [True: 532k, False: 248] ------------------ 15| 532k| uint32_t word = data[pos]; 16| 532k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 520k, False: 11.7k] ------------------ 17| | // will not generate a surrogate pair 18| 520k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (18:25): [Folded, False: 520k] ------------------ 19| 520k| ? char16_t(u16_swap_bytes(uint16_t(word))) 20| 520k| : char16_t(word); 21| 520k| pos++; 22| 520k| } else { 23| | // will generate a surrogate pair 24| 11.7k| word -= 0x10000; 25| 11.7k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 26| 11.7k| 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| 11.7k| *utf16_output++ = char16_t(high_surrogate); 32| 11.7k| *utf16_output++ = char16_t(low_surrogate); 33| 11.7k| pos++; 34| 11.7k| } 35| 532k| } 36| 248| return utf16_output - start; 37| 248|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1613convert_validILNS_10endiannessE1EEEmPKDimPDs: 11| 128| char16_t *utf16_output) { 12| 128| size_t pos = 0; 13| 128| char16_t *start{utf16_output}; 14| 30.3k| while (pos < len) { ------------------ | Branch (14:10): [True: 30.2k, False: 128] ------------------ 15| 30.2k| uint32_t word = data[pos]; 16| 30.2k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 11.6k, False: 18.6k] ------------------ 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| 18.6k| } else { 23| | // will generate a surrogate pair 24| 18.6k| word -= 0x10000; 25| 18.6k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 26| 18.6k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 27| 18.6k| if constexpr (!match_system(big_endian)) { 28| 18.6k| high_surrogate = u16_swap_bytes(high_surrogate); 29| 18.6k| low_surrogate = u16_swap_bytes(low_surrogate); 30| 18.6k| } 31| 18.6k| *utf16_output++ = char16_t(high_surrogate); 32| 18.6k| *utf16_output++ = char16_t(low_surrogate); 33| 18.6k| pos++; 34| 18.6k| } 35| 30.2k| } 36| 128| return utf16_output - start; 37| 128|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf32_to_utf87convertIPKDiPcQaasr7simdutf6detailE18indexes_into_utf32IT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEEmS7_mS8_: 17| 1.89k| OutputPtr utf8_output) { 18| 1.89k| size_t pos = 0; 19| 1.89k| auto start = utf8_output; 20| 242k| while (pos < len) { ------------------ | Branch (20:10): [True: 242k, False: 736] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 242k| { // try to convert the next block of 2 ASCII characters 25| 242k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (25:11): [True: 241k, False: 1.02k] ------------------ 26| | // they are ascii 27| 241k| uint64_t v; 28| 241k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 29| 241k| if ((v & 0xFFFFFF80FFFFFF80) == 0) { ------------------ | Branch (29:13): [True: 218k, False: 22.1k] ------------------ 30| 218k| *utf8_output++ = char(data[pos]); 31| 218k| *utf8_output++ = char(data[pos + 1]); 32| 218k| pos += 2; 33| 218k| continue; 34| 218k| } 35| 241k| } 36| 242k| } 37| | 38| 23.1k| uint32_t word = data[pos]; 39| 23.1k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (39:9): [True: 3.76k, False: 19.4k] ------------------ 40| | // will generate one UTF-8 bytes 41| 3.76k| *utf8_output++ = char(word); 42| 3.76k| pos++; 43| 19.4k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (43:16): [True: 3.88k, False: 15.5k] ------------------ 44| | // will generate two UTF-8 bytes 45| | // we have 0b110XXXXX 0b10XXXXXX 46| 3.88k| *utf8_output++ = char((word >> 6) | 0b11000000); 47| 3.88k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 48| 3.88k| pos++; 49| 15.5k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (49:16): [True: 9.83k, False: 5.70k] ------------------ 50| | // will generate three UTF-8 bytes 51| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 52| 9.83k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (52:11): [True: 7.03k, False: 2.80k] | Branch (52:29): [True: 228, False: 6.80k] ------------------ 53| 228| return 0; 54| 228| } 55| 9.60k| *utf8_output++ = char((word >> 12) | 0b11100000); 56| 9.60k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 57| 9.60k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 58| 9.60k| pos++; 59| 9.60k| } else { 60| | // will generate four UTF-8 bytes 61| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 62| 5.70k| if (word > 0x10FFFF) { ------------------ | Branch (62:11): [True: 930, False: 4.77k] ------------------ 63| 930| return 0; 64| 930| } 65| 4.77k| *utf8_output++ = char((word >> 18) | 0b11110000); 66| 4.77k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 67| 4.77k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 68| 4.77k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 69| 4.77k| pos++; 70| 4.77k| } 71| 23.1k| } 72| 736| return utf8_output - start; 73| 1.89k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf32_to_utf819convert_with_errorsIPKDiPcQaasr7simdutf6detailE18indexes_into_utf32IT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEENS_6resultES7_mS8_: 81| 2.79k| OutputPtr utf8_output) { 82| 2.79k| size_t pos = 0; 83| 2.79k| auto start = utf8_output; 84| 424k| while (pos < len) { ------------------ | Branch (84:10): [True: 424k, False: 694] ------------------ 85| |#if SIMDUTF_CPLUSPLUS23 86| | if !consteval 87| |#endif 88| 424k| { // try to convert the next block of 2 ASCII characters 89| 424k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (89:11): [True: 423k, False: 1.13k] ------------------ 90| | // they are ascii 91| 423k| uint64_t v; 92| 423k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 93| 423k| if ((v & 0xFFFFFF80FFFFFF80) == 0) { ------------------ | Branch (93:13): [True: 368k, False: 54.2k] ------------------ 94| 368k| *utf8_output++ = char(data[pos]); 95| 368k| *utf8_output++ = char(data[pos + 1]); 96| 368k| pos += 2; 97| 368k| continue; 98| 368k| } 99| 423k| } 100| 424k| } 101| | 102| 55.4k| uint32_t word = data[pos]; 103| 55.4k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (103:9): [True: 7.62k, False: 47.7k] ------------------ 104| | // will generate one UTF-8 bytes 105| 7.62k| *utf8_output++ = char(word); 106| 7.62k| pos++; 107| 47.7k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (107:16): [True: 22.8k, False: 24.9k] ------------------ 108| | // will generate two UTF-8 bytes 109| | // we have 0b110XXXXX 0b10XXXXXX 110| 22.8k| *utf8_output++ = char((word >> 6) | 0b11000000); 111| 22.8k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 112| 22.8k| pos++; 113| 24.9k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (113:16): [True: 16.5k, False: 8.45k] ------------------ 114| | // will generate three UTF-8 bytes 115| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 116| 16.5k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (116:11): [True: 11.8k, False: 4.64k] | Branch (116:29): [True: 342, False: 11.5k] ------------------ 117| 342| return result(error_code::SURROGATE, pos); 118| 342| } 119| 16.1k| *utf8_output++ = char((word >> 12) | 0b11100000); 120| 16.1k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 121| 16.1k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 122| 16.1k| pos++; 123| 16.1k| } else { 124| | // will generate four UTF-8 bytes 125| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 126| 8.45k| if (word > 0x10FFFF) { ------------------ | Branch (126:11): [True: 1.75k, False: 6.69k] ------------------ 127| 1.75k| return result(error_code::TOO_LARGE, pos); 128| 1.75k| } 129| 6.69k| *utf8_output++ = char((word >> 18) | 0b11110000); 130| 6.69k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 131| 6.69k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 132| 6.69k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 133| 6.69k| pos++; 134| 6.69k| } 135| 55.4k| } 136| 694| return result(error_code::SUCCESS, utf8_output - start); 137| 2.79k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf32_to_utf813convert_validIPKDiPcQaasr7simdutf6detailE18indexes_into_utf32IT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEEmS7_mS8_: 17| 206| OutputPtr utf8_output) { 18| 206| size_t pos = 0; 19| 206| auto start = utf8_output; 20| 18.6k| while (pos < len) { ------------------ | Branch (20:10): [True: 18.4k, False: 206] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 18.4k| { // try to convert the next block of 2 ASCII characters 25| 18.4k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (25:11): [True: 18.2k, False: 186] ------------------ 26| | // they are ascii 27| 18.2k| uint64_t v; 28| 18.2k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 29| 18.2k| if ((v & 0xFFFFFF80FFFFFF80) == 0) { ------------------ | Branch (29:13): [True: 1.59k, False: 16.6k] ------------------ 30| 1.59k| *utf8_output++ = char(data[pos]); 31| 1.59k| *utf8_output++ = char(data[pos + 1]); 32| 1.59k| pos += 2; 33| 1.59k| continue; 34| 1.59k| } 35| 18.2k| } 36| 18.4k| } 37| | 38| 16.8k| uint32_t word = data[pos]; 39| 16.8k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (39:9): [True: 608, False: 16.2k] ------------------ 40| | // will generate one UTF-8 bytes 41| 608| *utf8_output++ = char(word); 42| 608| pos++; 43| 16.2k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (43:16): [True: 13.0k, False: 3.19k] ------------------ 44| | // will generate two UTF-8 bytes 45| | // we have 0b110XXXXX 0b10XXXXXX 46| 13.0k| *utf8_output++ = char((word >> 6) | 0b11000000); 47| 13.0k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 48| 13.0k| pos++; 49| 13.0k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (49:16): [True: 1.63k, False: 1.56k] ------------------ 50| | // will generate three UTF-8 bytes 51| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 52| 1.63k| *utf8_output++ = char((word >> 12) | 0b11100000); 53| 1.63k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 54| 1.63k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 55| 1.63k| pos++; 56| 1.63k| } else { 57| | // will generate four UTF-8 bytes 58| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 59| 1.56k| *utf8_output++ = char((word >> 18) | 0b11110000); 60| 1.56k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 61| 1.56k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 62| 1.56k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 63| 1.56k| pos++; 64| 1.56k| } 65| 16.8k| } 66| 206| return utf8_output - start; 67| 206|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf88validateEPKcm: 113| 3.47k| size_t len) noexcept { 114| 3.47k| return validate(reinterpret_cast(buf), len); 115| 3.47k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf88validateIPKhEEbT_m: 14| 3.47k| size_t len) noexcept { 15| 3.47k| static_assert( 16| 3.47k| std::is_same::type, uint8_t>::value, 17| 3.47k| "dereferencing the data pointer must result in a uint8_t"); 18| 3.47k| uint64_t pos = 0; 19| 3.47k| uint32_t code_point = 0; 20| 8.61M| while (pos < len) { ------------------ | Branch (20:10): [True: 8.61M, False: 631] ------------------ 21| 8.61M| uint64_t next_pos; 22| |#if SIMDUTF_CPLUSPLUS23 23| | if !consteval 24| |#endif 25| 8.61M| { // check if the next 16 bytes are ascii. 26| 8.61M| next_pos = pos + 16; 27| 8.61M| if (next_pos <= len) { // if it is safe to read 16 more bytes, check ------------------ | Branch (27:11): [True: 8.61M, False: 3.38k] ------------------ 28| | // that they are ascii 29| 8.61M| uint64_t v1{}; 30| 8.61M| std::memcpy(&v1, data + pos, sizeof(uint64_t)); 31| 8.61M| uint64_t v2{}; 32| 8.61M| std::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 33| 8.61M| uint64_t v{v1 | v2}; 34| 8.61M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (34:13): [True: 7.41M, False: 1.20M] ------------------ 35| 7.41M| pos = next_pos; 36| 7.41M| continue; 37| 7.41M| } 38| 8.61M| } 39| 8.61M| } 40| | 41| 1.20M| unsigned char byte = data[pos]; 42| | 43| 5.71M| while (byte < 0b10000000) { ------------------ | Branch (43:12): [True: 4.51M, False: 1.20M] ------------------ 44| 4.51M| if (++pos == len) { ------------------ | Branch (44:11): [True: 1.08k, False: 4.51M] ------------------ 45| 1.08k| return true; 46| 1.08k| } 47| 4.51M| byte = data[pos]; 48| 4.51M| } 49| | 50| 1.20M| if ((byte & 0b11100000) == 0b11000000) { ------------------ | Branch (50:9): [True: 829k, False: 372k] ------------------ 51| 829k| next_pos = pos + 2; 52| 829k| if (next_pos > len) { ------------------ | Branch (52:11): [True: 69, False: 829k] ------------------ 53| 69| return false; 54| 69| } 55| 829k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (55:11): [True: 244, False: 829k] ------------------ 56| 244| return false; 57| 244| } 58| | // range check 59| 829k| code_point = (byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 60| 829k| if (code_point < 0x80) { ------------------ | Branch (60:11): [True: 38, False: 829k] ------------------ 61| 38| return false; 62| 38| } 63| 829k| } else if ((byte & 0b11110000) == 0b11100000) { ------------------ | Branch (63:16): [True: 283k, False: 89.2k] ------------------ 64| 283k| next_pos = pos + 3; 65| 283k| if (next_pos > len) { ------------------ | Branch (65:11): [True: 52, False: 283k] ------------------ 66| 52| return false; 67| 52| } 68| 283k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (68:11): [True: 119, False: 283k] ------------------ 69| 119| return false; 70| 119| } 71| 283k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (71:11): [True: 52, False: 283k] ------------------ 72| 52| return false; 73| 52| } 74| | // range check 75| 283k| code_point = (byte & 0b00001111) << 12 | 76| 283k| (data[pos + 1] & 0b00111111) << 6 | 77| 283k| (data[pos + 2] & 0b00111111); 78| 283k| if ((code_point < 0x800) || ------------------ | Branch (78:11): [True: 43, False: 283k] ------------------ 79| 283k| (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (79:12): [True: 8.29k, False: 275k] | Branch (79:35): [True: 18, False: 8.27k] ------------------ 80| 61| return false; 81| 61| } 82| 283k| } else if ((byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (82:16): [True: 88.3k, False: 826] ------------------ 83| 88.3k| next_pos = pos + 4; 84| 88.3k| if (next_pos > len) { ------------------ | Branch (84:11): [True: 41, False: 88.3k] ------------------ 85| 41| return false; 86| 41| } 87| 88.3k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (87:11): [True: 97, False: 88.2k] ------------------ 88| 97| return false; 89| 97| } 90| 88.2k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (90:11): [True: 43, False: 88.2k] ------------------ 91| 43| return false; 92| 43| } 93| 88.2k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (93:11): [True: 41, False: 88.1k] ------------------ 94| 41| return false; 95| 41| } 96| | // range check 97| 88.1k| code_point = 98| 88.1k| (byte & 0b00000111) << 18 | (data[pos + 1] & 0b00111111) << 12 | 99| 88.1k| (data[pos + 2] & 0b00111111) << 6 | (data[pos + 3] & 0b00111111); 100| 88.1k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (100:11): [True: 38, False: 88.1k] | Branch (100:35): [True: 35, False: 88.1k] ------------------ 101| 73| return false; 102| 73| } 103| 88.1k| } else { 104| | // we may have a continuation 105| 826| return false; 106| 826| } 107| 1.20M| pos = next_pos; 108| 1.20M| } 109| 631| return true; 110| 3.47k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf820validate_with_errorsEPKcm: 221| 6.71k|validate_with_errors(const char *buf, size_t len) noexcept { 222| 6.71k| return validate_with_errors(reinterpret_cast(buf), len); 223| 6.71k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf820validate_with_errorsIPKhEENS_6resultET_m: 119| 6.71k|validate_with_errors(BytePtr data, size_t len) noexcept { 120| 6.71k| static_assert( 121| 6.71k| std::is_same::type, uint8_t>::value, 122| 6.71k| "dereferencing the data pointer must result in a uint8_t"); 123| 6.71k| size_t pos = 0; 124| 6.71k| uint32_t code_point = 0; 125| 8.62M| while (pos < len) { ------------------ | Branch (125:10): [True: 8.62M, False: 631] ------------------ 126| | // check of the next 16 bytes are ascii. 127| 8.62M| size_t next_pos = pos + 16; 128| 8.62M| if (next_pos <= ------------------ | Branch (128:9): [True: 8.61M, False: 5.17k] ------------------ 129| 8.62M| len) { // if it is safe to read 16 more bytes, check that they are ascii 130| 8.61M| uint64_t v1; 131| 8.61M| std::memcpy(&v1, data + pos, sizeof(uint64_t)); 132| 8.61M| uint64_t v2; 133| 8.61M| std::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 134| 8.61M| uint64_t v{v1 | v2}; 135| 8.61M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (135:11): [True: 7.41M, False: 1.20M] ------------------ 136| 7.41M| pos = next_pos; 137| 7.41M| continue; 138| 7.41M| } 139| 8.61M| } 140| 1.21M| unsigned char byte = data[pos]; 141| | 142| 5.73M| while (byte < 0b10000000) { ------------------ | Branch (142:12): [True: 4.53M, False: 1.20M] ------------------ 143| 4.53M| if (++pos == len) { ------------------ | Branch (143:11): [True: 1.08k, False: 4.52M] ------------------ 144| 1.08k| return result(error_code::SUCCESS, len); 145| 1.08k| } 146| 4.52M| byte = data[pos]; 147| 4.52M| } 148| | 149| 1.20M| if ((byte & 0b11100000) == 0b11000000) { ------------------ | Branch (149:9): [True: 832k, False: 376k] ------------------ 150| 832k| next_pos = pos + 2; 151| 832k| if (next_pos > len) { ------------------ | Branch (151:11): [True: 207, False: 832k] ------------------ 152| 207| return result(error_code::TOO_SHORT, pos); 153| 207| } 154| 832k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (154:11): [True: 732, False: 831k] ------------------ 155| 732| return result(error_code::TOO_SHORT, pos); 156| 732| } 157| | // range check 158| 831k| code_point = (byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 159| 831k| if (code_point < 0x80) { ------------------ | Branch (159:11): [True: 114, False: 831k] ------------------ 160| 114| return result(error_code::OVERLONG, pos); 161| 114| } 162| 831k| } else if ((byte & 0b11110000) == 0b11100000) { ------------------ | Branch (162:16): [True: 285k, False: 91.6k] ------------------ 163| 285k| next_pos = pos + 3; 164| 285k| if (next_pos > len) { ------------------ | Branch (164:11): [True: 156, False: 284k] ------------------ 165| 156| return result(error_code::TOO_SHORT, pos); 166| 156| } 167| 284k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (167:11): [True: 357, False: 284k] ------------------ 168| 357| return result(error_code::TOO_SHORT, pos); 169| 357| } 170| 284k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (170:11): [True: 156, False: 284k] ------------------ 171| 156| return result(error_code::TOO_SHORT, pos); 172| 156| } 173| | // range check 174| 284k| code_point = (byte & 0b00001111) << 12 | 175| 284k| (data[pos + 1] & 0b00111111) << 6 | 176| 284k| (data[pos + 2] & 0b00111111); 177| 284k| if (code_point < 0x800) { ------------------ | Branch (177:11): [True: 129, False: 284k] ------------------ 178| 129| return result(error_code::OVERLONG, pos); 179| 129| } 180| 284k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (180:11): [True: 8.91k, False: 275k] | Branch (180:34): [True: 54, False: 8.86k] ------------------ 181| 54| return result(error_code::SURROGATE, pos); 182| 54| } 183| 284k| } else if ((byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (183:16): [True: 89.4k, False: 2.21k] ------------------ 184| 89.4k| next_pos = pos + 4; 185| 89.4k| if (next_pos > len) { ------------------ | Branch (185:11): [True: 123, False: 89.2k] ------------------ 186| 123| return result(error_code::TOO_SHORT, pos); 187| 123| } 188| 89.2k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (188:11): [True: 291, False: 89.0k] ------------------ 189| 291| return result(error_code::TOO_SHORT, pos); 190| 291| } 191| 89.0k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (191:11): [True: 129, False: 88.8k] ------------------ 192| 129| return result(error_code::TOO_SHORT, pos); 193| 129| } 194| 88.8k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (194:11): [True: 123, False: 88.7k] ------------------ 195| 123| return result(error_code::TOO_SHORT, pos); 196| 123| } 197| | // range check 198| 88.7k| code_point = 199| 88.7k| (byte & 0b00000111) << 18 | (data[pos + 1] & 0b00111111) << 12 | 200| 88.7k| (data[pos + 2] & 0b00111111) << 6 | (data[pos + 3] & 0b00111111); 201| 88.7k| if (code_point <= 0xffff) { ------------------ | Branch (201:11): [True: 114, False: 88.6k] ------------------ 202| 114| return result(error_code::OVERLONG, pos); 203| 114| } 204| 88.6k| if (0x10ffff < code_point) { ------------------ | Branch (204:11): [True: 105, False: 88.5k] ------------------ 205| 105| return result(error_code::TOO_LARGE, pos); 206| 105| } 207| 88.6k| } else { 208| | // we either have too many continuation bytes or an invalid leading byte 209| 2.21k| if ((byte & 0b11000000) == 0b10000000) { ------------------ | Branch (209:11): [True: 1.15k, False: 1.05k] ------------------ 210| 1.15k| return result(error_code::TOO_LONG, pos); 211| 1.15k| } else { 212| 1.05k| return result(error_code::HEADER_BITS, pos); 213| 1.05k| } 214| 2.21k| } 215| 1.20M| pos = next_pos; 216| 1.20M| } 217| 631| return result(error_code::SUCCESS, len); 218| 6.71k|} 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: 266, False: 3.24k] ------------------ 235| 266| return result(error_code::TOO_LONG, 0); 236| 266| } 237| 3.24k| size_t extra_len{0}; 238| | // A leading byte cannot be further than 4 bytes away 239| 3.43k| for (int i = 0; i < 5; i++) { ------------------ | Branch (239:19): [True: 3.43k, False: 0] ------------------ 240| 3.43k| unsigned char byte = *buf; 241| 3.43k| 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| 3.43k| } 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| 15.1k|simdutf_constexpr23 size_t count_code_points(InputPtr data, size_t len) { 259| 15.1k| size_t counter{0}; 260| 256M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (260:22): [True: 256M, False: 15.1k] ------------------ 261| | // -65 is 0b10111111, anything larger in two-complement's should start a new 262| | // code point. 263| 256M| if (int8_t(data[i]) > -65) { ------------------ | Branch (263:9): [True: 253M, False: 3.21M] ------------------ 264| 253M| counter++; 265| 253M| } 266| 256M| } 267| 15.1k| return counter; 268| 15.1k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf822utf16_length_from_utf8IPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_m: 274| 5.72k|simdutf_constexpr23 size_t utf16_length_from_utf8(InputPtr data, size_t len) { 275| 5.72k| size_t counter{0}; 276| 90.5M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (276:22): [True: 90.5M, False: 5.72k] ------------------ 277| 90.5M| if (int8_t(data[i]) > -65) { ------------------ | Branch (277:9): [True: 89.0M, False: 1.51M] ------------------ 278| 89.0M| counter++; 279| 89.0M| } 280| 90.5M| if (uint8_t(data[i]) >= 240) { ------------------ | Branch (280:9): [True: 141k, False: 90.4M] ------------------ 281| 141k| counter++; 282| 141k| } 283| 90.5M| } 284| 5.72k| return counter; 285| 5.72k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf8_to_latin130rewind_and_convert_with_errorsEmPKcmPc: 176| 996| char *latin1_output) { 177| 996| 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| 996| 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| 996| bool found_leading_bytes{false}; 185| | // important: it is i <= how_far_back and not 'i < how_far_back'. 186| 1.06k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (186:22): [True: 1.00k, False: 60] ------------------ 187| 1.00k| unsigned char byte = buf[-static_cast(i)]; 188| 1.00k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 189| 1.00k| if (found_leading_bytes) { ------------------ | Branch (189:9): [True: 936, False: 64] ------------------ 190| 936| if (i > 0 && byte < 128) { ------------------ | Branch (190:11): [True: 4, False: 932] | 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| 932| buf -= i; 196| 932| extra_len = i; 197| 932| break; 198| 936| } 199| 1.00k| } 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| 992| if (!found_leading_bytes) { ------------------ | Branch (208:7): [True: 60, False: 932] ------------------ 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| 60| return result(error_code::TOO_LONG, 0 - how_far_back); 214| 60| } 215| 932| result res = convert_with_errors(buf, len + extra_len, latin1_output); 216| 932| if (res.error) { ------------------ | Branch (216:7): [True: 492, False: 440] ------------------ 217| 492| res.count -= extra_len; 218| 492| } 219| 932| return res; 220| 992|} 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| 1.45M| while (pos < len) { ------------------ | Branch (96:10): [True: 1.45M, False: 664] ------------------ 97| |#if SIMDUTF_CPLUSPLUS23 98| | if !consteval 99| |#endif 100| 1.45M| { 101| | // try to convert the next block of 16 ASCII bytes 102| 1.45M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (102:11): [True: 1.44M, False: 7.40k] ------------------ 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: 1.20M, False: 238k] ------------------ 111| 1.44M| 0) { // if NONE of these are set, e.g. all of them are zero, then 112| | // everything is ASCII 113| 1.20M| size_t final_pos = pos + 16; 114| 20.5M| while (pos < final_pos) { ------------------ | Branch (114:18): [True: 19.3M, False: 1.20M] ------------------ 115| 19.3M| *latin_output++ = char(data[pos]); 116| 19.3M| pos++; 117| 19.3M| } 118| 1.20M| continue; 119| 1.20M| } 120| 1.44M| } 121| 1.45M| } 122| | // suppose it is not an all ASCII byte sequence 123| 245k| uint8_t leading_byte = data[pos]; // leading byte 124| 245k| if (leading_byte < 0b10000000) { ------------------ | Branch (124:9): [True: 213k, False: 32.4k] ------------------ 125| | // converting one ASCII byte !!! 126| 213k| *latin_output++ = char(leading_byte); 127| 213k| pos++; 128| 213k| } else if ((leading_byte & 0b11100000) == ------------------ | Branch (128:16): [True: 31.9k, False: 470] ------------------ 129| 32.4k| 0b11000000) { // the first three bits indicate: 130| | // We have a two-byte UTF-8 131| 31.9k| if (pos + 1 >= len) { ------------------ | Branch (131:11): [True: 66, False: 31.8k] ------------------ 132| 66| return result(error_code::TOO_SHORT, pos); 133| 66| } // minimal bound checking 134| 31.8k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (134:11): [True: 180, False: 31.7k] ------------------ 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| 31.7k| uint32_t code_point = 140| 31.7k| (leading_byte & 0b00011111) << 6 | 141| 31.7k| (data[pos + 1] & 142| 31.7k| 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| 31.7k| if (code_point < 0x80) { ------------------ | Branch (147:11): [True: 12, False: 31.6k] ------------------ 148| 12| return result(error_code::OVERLONG, pos); 149| 12| } 150| 31.6k| if (0xFF < code_point) { ------------------ | Branch (150:11): [True: 42, False: 31.6k] ------------------ 151| 42| return result(error_code::TOO_LARGE, pos); 152| 42| } // We only care about the range 129-255 which is Non-ASCII latin1 153| | // characters 154| 31.6k| *latin_output++ = char(code_point); 155| 31.6k| pos += 2; 156| 31.6k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (156:16): [True: 60, False: 410] ------------------ 157| | // We have a three-byte UTF-8 158| 60| return result(error_code::TOO_LARGE, pos); 159| 410| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (159:16): [True: 54, False: 356] ------------------ 160| | // we have a 4-byte UTF-8 word. 161| 54| return result(error_code::TOO_LARGE, pos); 162| 356| } else { 163| | // we either have too many continuation bytes or an invalid leading byte 164| 356| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (164:11): [True: 158, False: 198] ------------------ 165| 158| return result(error_code::TOO_LONG, pos); 166| 158| } 167| | 168| 198| return result(error_code::HEADER_BITS, pos); 169| 356| } 170| 245k| } 171| 664| 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| 1.25k| OutputPtr latin_output) { 18| 1.25k| size_t pos = 0; 19| 1.25k| auto start = latin_output; 20| | 21| 1.50M| while (pos < len) { ------------------ | Branch (21:10): [True: 1.50M, False: 538] ------------------ 22| |#if SIMDUTF_CPLUSPLUS23 23| | if !consteval 24| |#endif 25| 1.50M| { 26| | // try to convert the next block of 16 ASCII bytes 27| 1.50M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (27:11): [True: 1.50M, False: 4.11k] ------------------ 28| | // they are ascii 29| 1.50M| uint64_t v1; 30| 1.50M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 31| 1.50M| uint64_t v2; 32| 1.50M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 33| 1.50M| uint64_t v{v1 | v2}; // We are only interested in these bits: 1000 1000 34| | // 1000 1000 .... etc 35| 1.50M| if ((v & 0x8080808080808080) == ------------------ | Branch (35:13): [True: 1.35M, False: 147k] ------------------ 36| 1.50M| 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| 23.0M| 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| 1.50M| } 46| 1.50M| } 47| | 48| | // suppose it is not an all ASCII byte sequence 49| 151k| uint8_t leading_byte = data[pos]; // leading byte 50| 151k| if (leading_byte < 0b10000000) { ------------------ | Branch (50:9): [True: 133k, False: 18.1k] ------------------ 51| | // converting one ASCII byte !!! 52| 133k| *latin_output++ = char(leading_byte); 53| 133k| pos++; 54| 133k| } else if ((leading_byte & 0b11100000) == ------------------ | Branch (54:16): [True: 17.6k, False: 452] ------------------ 55| 18.1k| 0b11000000) { // the first three bits indicate: 56| | // We have a two-byte UTF-8 57| 17.6k| if (pos + 1 >= len) { ------------------ | Branch (57:11): [True: 42, False: 17.6k] ------------------ 58| 42| return 0; 59| 42| } // minimal bound checking 60| 17.6k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (60:11): [True: 150, False: 17.4k] ------------------ 61| 150| return 0; 62| 150| } // 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| 17.4k| uint32_t code_point = 66| 17.4k| (leading_byte & 0b00011111) << 6 | 67| 17.4k| (data[pos + 1] & 68| 17.4k| 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| 17.4k| if (code_point < 0x80 || 0xFF < code_point) { ------------------ | Branch (73:11): [True: 18, False: 17.4k] | Branch (73:32): [True: 54, False: 17.3k] ------------------ 74| 72| 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| 72| } 78| 17.3k| *latin_output++ = char(code_point); 79| 17.3k| pos += 2; 80| 17.3k| } else { 81| 452| return 0; 82| 452| } 83| 151k| } 84| 538| return latin_output - start; 85| 1.25k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf167convertILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.56k| char16_t *utf16_output) { 17| 1.56k| size_t pos = 0; 18| 1.56k| char16_t *start{utf16_output}; 19| 1.63M| while (pos < len) { ------------------ | Branch (19:10): [True: 1.63M, False: 692] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| | // try to convert the next block of 16 ASCII bytes 24| 1.63M| { 25| 1.63M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (25:11): [True: 1.63M, False: 4.83k] ------------------ 26| | // they are ascii 27| 1.63M| uint64_t v1; 28| 1.63M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 29| 1.63M| uint64_t v2; 30| 1.63M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 31| 1.63M| uint64_t v{v1 | v2}; 32| 1.63M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (32:13): [True: 949k, False: 680k] ------------------ 33| 949k| size_t final_pos = pos + 16; 34| 16.1M| while (pos < final_pos) { ------------------ | Branch (34:18): [True: 15.1M, False: 949k] ------------------ 35| 15.1M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (35:31): [Folded, False: 15.1M] ------------------ 36| 15.1M| ? char16_t(u16_swap_bytes(data[pos])) 37| 15.1M| : char16_t(data[pos]); 38| 15.1M| pos++; 39| 15.1M| } 40| 949k| continue; 41| 949k| } 42| 1.63M| } 43| 1.63M| } 44| | 45| 685k| uint8_t leading_byte = data[pos]; // leading byte 46| 685k| if (leading_byte < 0b10000000) { ------------------ | Branch (46:9): [True: 534k, False: 150k] ------------------ 47| | // converting one ASCII byte !!! 48| 534k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (48:25): [Folded, False: 534k] ------------------ 49| 534k| ? char16_t(u16_swap_bytes(leading_byte)) 50| 534k| : char16_t(leading_byte); 51| 534k| pos++; 52| 534k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (52:16): [True: 104k, False: 46.4k] ------------------ 53| | // We have a two-byte UTF-8, it should become 54| | // a single UTF-16 word. 55| 104k| if (pos + 1 >= len) { ------------------ | Branch (55:11): [True: 36, False: 104k] ------------------ 56| 36| return 0; 57| 36| } // minimal bound checking 58| 104k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (58:11): [True: 90, False: 104k] ------------------ 59| 90| return 0; 60| 90| } 61| | // range check 62| 104k| uint32_t code_point = 63| 104k| (leading_byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 64| 104k| if (code_point < 0x80) { ------------------ | Branch (64:11): [True: 20, False: 104k] ------------------ 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| 104k| *utf16_output++ = char16_t(code_point); 71| 104k| pos += 2; 72| 104k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (72:16): [True: 34.8k, False: 11.6k] ------------------ 73| | // We have a three-byte UTF-8, it should become 74| | // a single UTF-16 word. 75| 34.8k| if (pos + 2 >= len) { ------------------ | Branch (75:11): [True: 42, False: 34.7k] ------------------ 76| 42| return 0; 77| 42| } // minimal bound checking 78| | 79| 34.7k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (79:11): [True: 50, False: 34.7k] ------------------ 80| 50| return 0; 81| 50| } 82| 34.7k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (82:11): [True: 44, False: 34.6k] ------------------ 83| 44| return 0; 84| 44| } 85| | // range check 86| 34.6k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 87| 34.6k| (data[pos + 1] & 0b00111111) << 6 | 88| 34.6k| (data[pos + 2] & 0b00111111); 89| 34.6k| if (code_point < 0x800 || (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (89:11): [True: 44, False: 34.6k] | Branch (89:34): [True: 966, False: 33.6k] | Branch (89:57): [True: 14, False: 952] ------------------ 90| 58| return 0; 91| 58| } 92| | if constexpr (!match_system(big_endian)) { 93| | code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 94| | } 95| 34.6k| *utf16_output++ = char16_t(code_point); 96| 34.6k| pos += 3; 97| 34.6k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (97:16): [True: 11.2k, False: 326] ------------------ 98| | // we have a 4-byte UTF-8 word. 99| 11.2k| if (pos + 3 >= len) { ------------------ | Branch (99:11): [True: 42, False: 11.2k] ------------------ 100| 42| return 0; 101| 42| } // minimal bound checking 102| 11.2k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (102:11): [True: 62, False: 11.1k] ------------------ 103| 62| return 0; 104| 62| } 105| 11.1k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (105:11): [True: 20, False: 11.1k] ------------------ 106| 20| return 0; 107| 20| } 108| 11.1k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (108:11): [True: 26, False: 11.1k] ------------------ 109| 26| return 0; 110| 26| } 111| | 112| | // range check 113| 11.1k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 114| 11.1k| (data[pos + 1] & 0b00111111) << 12 | 115| 11.1k| (data[pos + 2] & 0b00111111) << 6 | 116| 11.1k| (data[pos + 3] & 0b00111111); 117| 11.1k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (117:11): [True: 10, False: 11.1k] | Branch (117:35): [True: 42, False: 11.0k] ------------------ 118| 52| return 0; 119| 52| } 120| 11.0k| code_point -= 0x10000; 121| 11.0k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 122| 11.0k| 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| 11.0k| *utf16_output++ = char16_t(high_surrogate); 128| 11.0k| *utf16_output++ = char16_t(low_surrogate); 129| 11.0k| pos += 4; 130| 11.0k| } else { 131| 326| return 0; 132| 326| } 133| 685k| } 134| 692| return utf16_output - start; 135| 1.56k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf167convertILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.54k| char16_t *utf16_output) { 17| 1.54k| size_t pos = 0; 18| 1.54k| char16_t *start{utf16_output}; 19| 1.33M| while (pos < len) { ------------------ | Branch (19:10): [True: 1.33M, False: 590] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| | // try to convert the next block of 16 ASCII bytes 24| 1.33M| { 25| 1.33M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (25:11): [True: 1.33M, False: 3.69k] ------------------ 26| | // they are ascii 27| 1.33M| uint64_t v1; 28| 1.33M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 29| 1.33M| uint64_t v2; 30| 1.33M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 31| 1.33M| uint64_t v{v1 | v2}; 32| 1.33M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (32:13): [True: 840k, False: 491k] ------------------ 33| 840k| size_t final_pos = pos + 16; 34| 14.2M| while (pos < final_pos) { ------------------ | Branch (34:18): [True: 13.4M, False: 840k] ------------------ 35| 13.4M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (35:31): [True: 13.4M, Folded] ------------------ 36| 13.4M| ? char16_t(u16_swap_bytes(data[pos])) 37| 13.4M| : char16_t(data[pos]); 38| 13.4M| pos++; 39| 13.4M| } 40| 840k| continue; 41| 840k| } 42| 1.33M| } 43| 1.33M| } 44| | 45| 495k| uint8_t leading_byte = data[pos]; // leading byte 46| 495k| if (leading_byte < 0b10000000) { ------------------ | Branch (46:9): [True: 397k, False: 98.0k] ------------------ 47| | // converting one ASCII byte !!! 48| 397k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (48:25): [True: 397k, Folded] ------------------ 49| 397k| ? char16_t(u16_swap_bytes(leading_byte)) 50| 397k| : char16_t(leading_byte); 51| 397k| pos++; 52| 397k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (52:16): [True: 62.0k, False: 35.9k] ------------------ 53| | // We have a two-byte UTF-8, it should become 54| | // a single UTF-16 word. 55| 62.0k| if (pos + 1 >= len) { ------------------ | Branch (55:11): [True: 30, False: 62.0k] ------------------ 56| 30| return 0; 57| 30| } // minimal bound checking 58| 62.0k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (58:11): [True: 84, False: 61.9k] ------------------ 59| 84| return 0; 60| 84| } 61| | // range check 62| 61.9k| uint32_t code_point = 63| 61.9k| (leading_byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 64| 61.9k| if (code_point < 0x80) { ------------------ | Branch (64:11): [True: 22, False: 61.9k] ------------------ 65| 22| return 0; 66| 22| } 67| 61.9k| if constexpr (!match_system(big_endian)) { 68| 61.9k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 69| 61.9k| } 70| 61.9k| *utf16_output++ = char16_t(code_point); 71| 61.9k| pos += 2; 72| 61.9k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (72:16): [True: 27.1k, False: 8.82k] ------------------ 73| | // We have a three-byte UTF-8, it should become 74| | // a single UTF-16 word. 75| 27.1k| if (pos + 2 >= len) { ------------------ | Branch (75:11): [True: 18, False: 27.0k] ------------------ 76| 18| return 0; 77| 18| } // minimal bound checking 78| | 79| 27.0k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (79:11): [True: 64, False: 27.0k] ------------------ 80| 64| return 0; 81| 64| } 82| 27.0k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (82:11): [True: 46, False: 26.9k] ------------------ 83| 46| return 0; 84| 46| } 85| | // range check 86| 26.9k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 87| 26.9k| (data[pos + 1] & 0b00111111) << 6 | 88| 26.9k| (data[pos + 2] & 0b00111111); 89| 26.9k| if (code_point < 0x800 || (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (89:11): [True: 54, False: 26.9k] | Branch (89:34): [True: 948, False: 25.9k] | Branch (89:57): [True: 12, False: 936] ------------------ 90| 66| return 0; 91| 66| } 92| 26.9k| if constexpr (!match_system(big_endian)) { 93| 26.9k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 94| 26.9k| } 95| 26.9k| *utf16_output++ = char16_t(code_point); 96| 26.9k| pos += 3; 97| 26.9k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (97:16): [True: 8.41k, False: 410] ------------------ 98| | // we have a 4-byte UTF-8 word. 99| 8.41k| if (pos + 3 >= len) { ------------------ | Branch (99:11): [True: 30, False: 8.38k] ------------------ 100| 30| return 0; 101| 30| } // minimal bound checking 102| 8.38k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (102:11): [True: 62, False: 8.32k] ------------------ 103| 62| return 0; 104| 62| } 105| 8.32k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (105:11): [True: 22, False: 8.30k] ------------------ 106| 22| return 0; 107| 22| } 108| 8.30k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (108:11): [True: 42, False: 8.26k] ------------------ 109| 42| return 0; 110| 42| } 111| | 112| | // range check 113| 8.26k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 114| 8.26k| (data[pos + 1] & 0b00111111) << 12 | 115| 8.26k| (data[pos + 2] & 0b00111111) << 6 | 116| 8.26k| (data[pos + 3] & 0b00111111); 117| 8.26k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (117:11): [True: 36, False: 8.22k] | Branch (117:35): [True: 26, False: 8.20k] ------------------ 118| 62| return 0; 119| 62| } 120| 8.20k| code_point -= 0x10000; 121| 8.20k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 122| 8.20k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 123| 8.20k| if constexpr (!match_system(big_endian)) { 124| 8.20k| high_surrogate = u16_swap_bytes(high_surrogate); 125| 8.20k| low_surrogate = u16_swap_bytes(low_surrogate); 126| 8.20k| } 127| 8.20k| *utf16_output++ = char16_t(high_surrogate); 128| 8.20k| *utf16_output++ = char16_t(low_surrogate); 129| 8.20k| pos += 4; 130| 8.20k| } else { 131| 410| return 0; 132| 410| } 133| 495k| } 134| 590| return utf16_output - start; 135| 1.54k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1619convert_with_errorsILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEENS_6resultES7_mPDs: 142| 1.76k| char16_t *utf16_output) { 143| 1.76k| size_t pos = 0; 144| 1.76k| char16_t *start{utf16_output}; 145| 2.97M| while (pos < len) { ------------------ | Branch (145:10): [True: 2.97M, False: 766] ------------------ 146| |#if SIMDUTF_CPLUSPLUS23 147| | if !consteval 148| |#endif 149| 2.97M| { 150| | // try to convert the next block of 16 ASCII bytes 151| 2.97M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (151:11): [True: 2.97M, False: 6.11k] ------------------ 152| | // they are ascii 153| 2.97M| uint64_t v1; 154| 2.97M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 155| 2.97M| uint64_t v2; 156| 2.97M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 157| 2.97M| uint64_t v{v1 | v2}; 158| 2.97M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (158:13): [True: 1.37M, False: 1.59M] ------------------ 159| 1.37M| size_t final_pos = pos + 16; 160| 23.4M| while (pos < final_pos) { ------------------ | Branch (160:18): [True: 22.0M, False: 1.37M] ------------------ 161| 22.0M| const char16_t byte = uint8_t(data[pos]); 162| 22.0M| *utf16_output++ = 163| 22.0M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (163:17): [Folded, False: 22.0M] ------------------ 164| 22.0M| pos++; 165| 22.0M| } 166| 1.37M| continue; 167| 1.37M| } 168| 2.97M| } 169| 2.97M| } 170| | 171| 1.59M| auto leading_byte = uint8_t(data[pos]); // leading byte 172| 1.59M| if (leading_byte < 0b10000000) { ------------------ | Branch (172:9): [True: 1.32M, False: 276k] ------------------ 173| | // converting one ASCII byte !!! 174| 1.32M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (174:25): [Folded, False: 1.32M] ------------------ 175| 1.32M| ? char16_t(u16_swap_bytes(leading_byte)) 176| 1.32M| : char16_t(leading_byte); 177| 1.32M| pos++; 178| 1.32M| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (178:16): [True: 179k, False: 96.8k] ------------------ 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: 48, False: 179k] ------------------ 182| 48| return result(error_code::TOO_SHORT, pos); 183| 48| } // minimal bound checking 184| 179k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (184:11): [True: 90, False: 179k] ------------------ 185| 90| return result(error_code::TOO_SHORT, pos); 186| 90| } 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: 72.7k, False: 24.0k] ------------------ 199| | // We have a three-byte UTF-8, it should become 200| | // a single UTF-16 word. 201| 72.7k| if (pos + 2 >= len) { ------------------ | Branch (201:11): [True: 42, False: 72.7k] ------------------ 202| 42| return result(error_code::TOO_SHORT, pos); 203| 42| } // minimal bound checking 204| | 205| 72.7k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (205:11): [True: 66, False: 72.6k] ------------------ 206| 66| return result(error_code::TOO_SHORT, pos); 207| 66| } 208| 72.6k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (208:11): [True: 18, False: 72.6k] ------------------ 209| 18| return result(error_code::TOO_SHORT, pos); 210| 18| } 211| | // range check 212| 72.6k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 213| 72.6k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 214| 72.6k| (uint8_t(data[pos + 2]) & 0b00111111); 215| 72.6k| if (code_point < 0x800) { ------------------ | Branch (215:11): [True: 36, False: 72.6k] ------------------ 216| 36| return result(error_code::OVERLONG, pos); 217| 36| } 218| 72.6k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (218:11): [True: 1.53k, False: 71.0k] | Branch (218:34): [True: 18, False: 1.52k] ------------------ 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| 72.5k| *utf16_output++ = char16_t(code_point); 225| 72.5k| pos += 3; 226| 72.5k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (226:16): [True: 23.6k, False: 440] ------------------ 227| | // we have a 4-byte UTF-8 word. 228| 23.6k| if (pos + 3 >= len) { ------------------ | Branch (228:11): [True: 42, False: 23.5k] ------------------ 229| 42| return result(error_code::TOO_SHORT, pos); 230| 42| } // minimal bound checking 231| 23.5k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (231:11): [True: 60, False: 23.5k] ------------------ 232| 60| return result(error_code::TOO_SHORT, pos); 233| 60| } 234| 23.5k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (234:11): [True: 18, False: 23.5k] ------------------ 235| 18| return result(error_code::TOO_SHORT, pos); 236| 18| } 237| 23.5k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (237:11): [True: 18, False: 23.4k] ------------------ 238| 18| return result(error_code::TOO_SHORT, pos); 239| 18| } 240| | 241| | // range check 242| 23.4k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 243| 23.4k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 244| 23.4k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 245| 23.4k| (uint8_t(data[pos + 3]) & 0b00111111); 246| 23.4k| if (code_point <= 0xffff) { ------------------ | Branch (246:11): [True: 42, False: 23.4k] ------------------ 247| 42| return result(error_code::OVERLONG, pos); 248| 42| } 249| 23.4k| if (0x10ffff < code_point) { ------------------ | Branch (249:11): [True: 36, False: 23.4k] ------------------ 250| 36| return result(error_code::TOO_LARGE, pos); 251| 36| } 252| 23.4k| code_point -= 0x10000; 253| 23.4k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 254| 23.4k| 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| 23.4k| *utf16_output++ = char16_t(high_surrogate); 260| 23.4k| *utf16_output++ = char16_t(low_surrogate); 261| 23.4k| pos += 4; 262| 23.4k| } else { 263| | // we either have too many continuation bytes or an invalid leading byte 264| 440| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (264:11): [True: 182, False: 258] ------------------ 265| 182| return result(error_code::TOO_LONG, pos); 266| 258| } else { 267| 258| return result(error_code::HEADER_BITS, pos); 268| 258| } 269| 440| } 270| 1.59M| } 271| 766| return result(error_code::SUCCESS, utf16_output - start); 272| 1.76k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1619convert_with_errorsILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEENS_6resultES7_mPDs: 142| 2.15k| char16_t *utf16_output) { 143| 2.15k| size_t pos = 0; 144| 2.15k| char16_t *start{utf16_output}; 145| 3.05M| while (pos < len) { ------------------ | Branch (145:10): [True: 3.05M, False: 970] ------------------ 146| |#if SIMDUTF_CPLUSPLUS23 147| | if !consteval 148| |#endif 149| 3.05M| { 150| | // try to convert the next block of 16 ASCII bytes 151| 3.05M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (151:11): [True: 3.04M, False: 9.04k] ------------------ 152| | // they are ascii 153| 3.04M| uint64_t v1; 154| 3.04M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 155| 3.04M| uint64_t v2; 156| 3.04M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 157| 3.04M| uint64_t v{v1 | v2}; 158| 3.04M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (158:13): [True: 1.54M, False: 1.49M] ------------------ 159| 1.54M| size_t final_pos = pos + 16; 160| 26.2M| while (pos < final_pos) { ------------------ | Branch (160:18): [True: 24.6M, False: 1.54M] ------------------ 161| 24.6M| const char16_t byte = uint8_t(data[pos]); 162| 24.6M| *utf16_output++ = 163| 24.6M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (163:17): [True: 24.6M, Folded] ------------------ 164| 24.6M| pos++; 165| 24.6M| } 166| 1.54M| continue; 167| 1.54M| } 168| 3.04M| } 169| 3.05M| } 170| | 171| 1.50M| auto leading_byte = uint8_t(data[pos]); // leading byte 172| 1.50M| if (leading_byte < 0b10000000) { ------------------ | Branch (172:9): [True: 1.15M, False: 356k] ------------------ 173| | // converting one ASCII byte !!! 174| 1.15M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (174:25): [True: 1.15M, Folded] ------------------ 175| 1.15M| ? char16_t(u16_swap_bytes(leading_byte)) 176| 1.15M| : char16_t(leading_byte); 177| 1.15M| pos++; 178| 1.15M| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (178:16): [True: 251k, False: 105k] ------------------ 179| | // We have a two-byte UTF-8, it should become 180| | // a single UTF-16 word. 181| 251k| if (pos + 1 >= len) { ------------------ | Branch (181:11): [True: 36, False: 250k] ------------------ 182| 36| return result(error_code::TOO_SHORT, pos); 183| 36| } // minimal bound checking 184| 250k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (184:11): [True: 162, False: 250k] ------------------ 185| 162| return result(error_code::TOO_SHORT, pos); 186| 162| } 187| | // range check 188| 250k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 189| 250k| (uint8_t(data[pos + 1]) & 0b00111111); 190| 250k| if (code_point < 0x80) { ------------------ | Branch (190:11): [True: 18, False: 250k] ------------------ 191| 18| return result(error_code::OVERLONG, pos); 192| 18| } 193| 250k| if constexpr (!match_system(big_endian)) { 194| 250k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 195| 250k| } 196| 250k| *utf16_output++ = char16_t(code_point); 197| 250k| pos += 2; 198| 250k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (198:16): [True: 74.9k, False: 30.2k] ------------------ 199| | // We have a three-byte UTF-8, it should become 200| | // a single UTF-16 word. 201| 74.9k| if (pos + 2 >= len) { ------------------ | Branch (201:11): [True: 36, False: 74.8k] ------------------ 202| 36| return result(error_code::TOO_SHORT, pos); 203| 36| } // minimal bound checking 204| | 205| 74.8k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (205:11): [True: 120, False: 74.7k] ------------------ 206| 120| return result(error_code::TOO_SHORT, pos); 207| 120| } 208| 74.7k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (208:11): [True: 60, False: 74.7k] ------------------ 209| 60| return result(error_code::TOO_SHORT, pos); 210| 60| } 211| | // range check 212| 74.7k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 213| 74.7k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 214| 74.7k| (uint8_t(data[pos + 2]) & 0b00111111); 215| 74.7k| if (code_point < 0x800) { ------------------ | Branch (215:11): [True: 42, False: 74.6k] ------------------ 216| 42| return result(error_code::OVERLONG, pos); 217| 42| } 218| 74.6k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (218:11): [True: 2.89k, False: 71.7k] | Branch (218:34): [True: 6, False: 2.88k] ------------------ 219| 6| return result(error_code::SURROGATE, pos); 220| 6| } 221| 74.6k| if constexpr (!match_system(big_endian)) { 222| 74.6k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 223| 74.6k| } 224| 74.6k| *utf16_output++ = char16_t(code_point); 225| 74.6k| pos += 3; 226| 74.6k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (226:16): [True: 29.7k, False: 484] ------------------ 227| | // we have a 4-byte UTF-8 word. 228| 29.7k| if (pos + 3 >= len) { ------------------ | Branch (228:11): [True: 48, False: 29.7k] ------------------ 229| 48| return result(error_code::TOO_SHORT, pos); 230| 48| } // minimal bound checking 231| 29.7k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (231:11): [True: 42, False: 29.7k] ------------------ 232| 42| return result(error_code::TOO_SHORT, pos); 233| 42| } 234| 29.7k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (234:11): [True: 54, False: 29.6k] ------------------ 235| 54| return result(error_code::TOO_SHORT, pos); 236| 54| } 237| 29.6k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (237:11): [True: 54, False: 29.5k] ------------------ 238| 54| return result(error_code::TOO_SHORT, pos); 239| 54| } 240| | 241| | // range check 242| 29.5k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 243| 29.5k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 244| 29.5k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 245| 29.5k| (uint8_t(data[pos + 3]) & 0b00111111); 246| 29.5k| if (code_point <= 0xffff) { ------------------ | Branch (246:11): [True: 6, False: 29.5k] ------------------ 247| 6| return result(error_code::OVERLONG, pos); 248| 6| } 249| 29.5k| if (0x10ffff < code_point) { ------------------ | Branch (249:11): [True: 12, False: 29.5k] ------------------ 250| 12| return result(error_code::TOO_LARGE, pos); 251| 12| } 252| 29.5k| code_point -= 0x10000; 253| 29.5k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 254| 29.5k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 255| 29.5k| if constexpr (!match_system(big_endian)) { 256| 29.5k| high_surrogate = u16_swap_bytes(high_surrogate); 257| 29.5k| low_surrogate = u16_swap_bytes(low_surrogate); 258| 29.5k| } 259| 29.5k| *utf16_output++ = char16_t(high_surrogate); 260| 29.5k| *utf16_output++ = char16_t(low_surrogate); 261| 29.5k| pos += 4; 262| 29.5k| } else { 263| | // we either have too many continuation bytes or an invalid leading byte 264| 484| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (264:11): [True: 232, False: 252] ------------------ 265| 232| return result(error_code::TOO_LONG, pos); 266| 252| } else { 267| 252| return result(error_code::HEADER_BITS, pos); 268| 252| } 269| 484| } 270| 1.50M| } 271| 970| return result(error_code::SUCCESS, utf16_output - start); 272| 2.15k|} 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| 1.28k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (302:22): [True: 1.22k, False: 52] ------------------ 303| 1.22k| unsigned char byte = buf[-static_cast(i)]; 304| 1.22k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 305| 1.22k| if (found_leading_bytes) { ------------------ | Branch (305:9): [True: 1.16k, False: 64] ------------------ 306| 1.16k| if (i > 0 && byte < 128) { ------------------ | Branch (306:11): [True: 12, False: 1.15k] | Branch (306:20): [True: 12, False: 0] ------------------ 307| | // If we had to go back and the leading byte is ascii 308| | // then we can stop right away. 309| 12| return result(error_code::TOO_LONG, 0 - i + 1); 310| 12| } 311| 1.15k| buf -= i; 312| 1.15k| extra_len = i; 313| 1.15k| break; 314| 1.16k| } 315| 1.22k| } 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: 52, False: 1.15k] ------------------ 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| 52| return result(error_code::TOO_LONG, 0 - how_far_back); 330| 52| } 331| 1.15k| result res = convert_with_errors(buf, len + extra_len, utf16_output); 332| 1.15k| if (res.error) { ------------------ | Branch (332:7): [True: 644, False: 508] ------------------ 333| 644| res.count -= extra_len; 334| 644| } 335| 1.15k| return res; 336| 1.20k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1630rewind_and_convert_with_errorsILNS_10endiannessE1EEENS_6resultEmPKcmPDs: 292| 1.50k| char16_t *utf16_output) { 293| 1.50k| 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| 1.50k| 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| 1.50k| bool found_leading_bytes{false}; 301| | // important: it is i <= how_far_back and not 'i < how_far_back'. 302| 1.60k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (302:22): [True: 1.52k, False: 76] ------------------ 303| 1.52k| unsigned char byte = buf[-static_cast(i)]; 304| 1.52k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 305| 1.52k| if (found_leading_bytes) { ------------------ | Branch (305:9): [True: 1.42k, False: 104] ------------------ 306| 1.42k| if (i > 0 && byte < 128) { ------------------ | Branch (306:11): [True: 28, False: 1.39k] | Branch (306:20): [True: 28, False: 0] ------------------ 307| | // If we had to go back and the leading byte is ascii 308| | // then we can stop right away. 309| 28| return result(error_code::TOO_LONG, 0 - i + 1); 310| 28| } 311| 1.39k| buf -= i; 312| 1.39k| extra_len = i; 313| 1.39k| break; 314| 1.42k| } 315| 1.52k| } 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| 1.47k| if (!found_leading_bytes) { ------------------ | Branch (324:7): [True: 76, False: 1.39k] ------------------ 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| 76| return result(error_code::TOO_LONG, 0 - how_far_back); 330| 76| } 331| 1.39k| result res = convert_with_errors(buf, len + extra_len, utf16_output); 332| 1.39k| if (res.error) { ------------------ | Branch (332:7): [True: 752, False: 644] ------------------ 333| 752| res.count -= extra_len; 334| 752| } 335| 1.39k| return res; 336| 1.47k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.33k| char16_t *utf16_output) { 17| 1.33k| size_t pos = 0; 18| 1.33k| char16_t *start{utf16_output}; 19| 4.31M| while (pos < len) { ------------------ | Branch (19:10): [True: 4.30M, False: 1.33k] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 4.30M| { // try to convert the next block of 8 ASCII bytes 24| 4.30M| if (pos + 8 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (24:11): [True: 4.30M, False: 4.44k] ------------------ 25| | // they are ascii 26| 4.30M| uint64_t v; 27| 4.30M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 28| 4.30M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (28:13): [True: 2.82M, False: 1.47M] ------------------ 29| 2.82M| size_t final_pos = pos + 8; 30| 25.4M| while (pos < final_pos) { ------------------ | Branch (30:18): [True: 22.6M, False: 2.82M] ------------------ 31| 22.6M| const char16_t byte = uint8_t(data[pos]); 32| 22.6M| *utf16_output++ = 33| 22.6M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (33:17): [Folded, False: 22.6M] ------------------ 34| 22.6M| pos++; 35| 22.6M| } 36| 2.82M| continue; 37| 2.82M| } 38| 4.30M| } 39| 4.30M| } 40| | 41| 1.48M| auto leading_byte = uint8_t(data[pos]); // leading byte 42| 1.48M| if (leading_byte < 0b10000000) { ------------------ | Branch (42:9): [True: 1.00M, False: 481k] ------------------ 43| | // converting one ASCII byte !!! 44| 1.00M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (44:25): [Folded, False: 1.00M] ------------------ 45| 1.00M| ? char16_t(u16_swap_bytes(leading_byte)) 46| 1.00M| : char16_t(leading_byte); 47| 1.00M| pos++; 48| 1.00M| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (48:16): [True: 277k, False: 203k] ------------------ 49| | // We have a two-byte UTF-8, it should become 50| | // a single UTF-16 word. 51| 277k| if (pos + 1 >= len) { ------------------ | Branch (51:11): [True: 0, False: 277k] ------------------ 52| 0| break; 53| 0| } // minimal bound checking 54| 277k| uint16_t code_point = uint16_t(((leading_byte & 0b00011111) << 6) | 55| 277k| (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| 277k| *utf16_output++ = char16_t(code_point); 60| 277k| pos += 2; 61| 277k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (61:16): [True: 185k, False: 18.1k] ------------------ 62| | // We have a three-byte UTF-8, it should become 63| | // a single UTF-16 word. 64| 185k| if (pos + 2 >= len) { ------------------ | Branch (64:11): [True: 0, False: 185k] ------------------ 65| 0| break; 66| 0| } // minimal bound checking 67| 185k| uint16_t code_point = 68| 185k| uint16_t(((leading_byte & 0b00001111) << 12) | 69| 185k| ((uint8_t(data[pos + 1]) & 0b00111111) << 6) | 70| 185k| (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| 185k| *utf16_output++ = char16_t(code_point); 75| 185k| pos += 3; 76| 185k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (76:16): [True: 18.1k, False: 0] ------------------ 77| | // we have a 4-byte UTF-8 word. 78| 18.1k| if (pos + 3 >= len) { ------------------ | Branch (78:11): [True: 0, False: 18.1k] ------------------ 79| 0| break; 80| 0| } // minimal bound checking 81| 18.1k| uint32_t code_point = ((leading_byte & 0b00000111) << 18) | 82| 18.1k| ((uint8_t(data[pos + 1]) & 0b00111111) << 12) | 83| 18.1k| ((uint8_t(data[pos + 2]) & 0b00111111) << 6) | 84| 18.1k| (uint8_t(data[pos + 3]) & 0b00111111); 85| 18.1k| code_point -= 0x10000; 86| 18.1k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 87| 18.1k| 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| 18.1k| *utf16_output++ = char16_t(high_surrogate); 93| 18.1k| *utf16_output++ = char16_t(low_surrogate); 94| 18.1k| pos += 4; 95| 18.1k| } else { 96| | // we may have a continuation but we do not do error checking 97| 0| return 0; 98| 0| } 99| 1.48M| } 100| 1.33k| return utf16_output - start; 101| 1.33k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.42k| char16_t *utf16_output) { 17| 1.42k| size_t pos = 0; 18| 1.42k| char16_t *start{utf16_output}; 19| 4.23M| while (pos < len) { ------------------ | Branch (19:10): [True: 4.23M, False: 1.42k] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 4.23M| { // try to convert the next block of 8 ASCII bytes 24| 4.23M| if (pos + 8 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (24:11): [True: 4.23M, False: 4.23k] ------------------ 25| | // they are ascii 26| 4.23M| uint64_t v; 27| 4.23M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 28| 4.23M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (28:13): [True: 3.63M, False: 598k] ------------------ 29| 3.63M| size_t final_pos = pos + 8; 30| 32.7M| while (pos < final_pos) { ------------------ | Branch (30:18): [True: 29.0M, False: 3.63M] ------------------ 31| 29.0M| const char16_t byte = uint8_t(data[pos]); 32| 29.0M| *utf16_output++ = 33| 29.0M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (33:17): [True: 29.0M, Folded] ------------------ 34| 29.0M| pos++; 35| 29.0M| } 36| 3.63M| continue; 37| 3.63M| } 38| 4.23M| } 39| 4.23M| } 40| | 41| 603k| auto leading_byte = uint8_t(data[pos]); // leading byte 42| 603k| if (leading_byte < 0b10000000) { ------------------ | Branch (42:9): [True: 332k, False: 270k] ------------------ 43| | // converting one ASCII byte !!! 44| 332k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (44:25): [True: 332k, Folded] ------------------ 45| 332k| ? char16_t(u16_swap_bytes(leading_byte)) 46| 332k| : char16_t(leading_byte); 47| 332k| pos++; 48| 332k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (48:16): [True: 215k, False: 55.1k] ------------------ 49| | // We have a two-byte UTF-8, it should become 50| | // a single UTF-16 word. 51| 215k| if (pos + 1 >= len) { ------------------ | Branch (51:11): [True: 0, False: 215k] ------------------ 52| 0| break; 53| 0| } // minimal bound checking 54| 215k| uint16_t code_point = uint16_t(((leading_byte & 0b00011111) << 6) | 55| 215k| (uint8_t(data[pos + 1]) & 0b00111111)); 56| 215k| if constexpr (!match_system(big_endian)) { 57| 215k| code_point = u16_swap_bytes(uint16_t(code_point)); 58| 215k| } 59| 215k| *utf16_output++ = char16_t(code_point); 60| 215k| pos += 2; 61| 215k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (61:16): [True: 31.6k, False: 23.4k] ------------------ 62| | // We have a three-byte UTF-8, it should become 63| | // a single UTF-16 word. 64| 31.6k| if (pos + 2 >= len) { ------------------ | Branch (64:11): [True: 0, False: 31.6k] ------------------ 65| 0| break; 66| 0| } // minimal bound checking 67| 31.6k| uint16_t code_point = 68| 31.6k| uint16_t(((leading_byte & 0b00001111) << 12) | 69| 31.6k| ((uint8_t(data[pos + 1]) & 0b00111111) << 6) | 70| 31.6k| (uint8_t(data[pos + 2]) & 0b00111111)); 71| 31.6k| if constexpr (!match_system(big_endian)) { 72| 31.6k| code_point = u16_swap_bytes(uint16_t(code_point)); 73| 31.6k| } 74| 31.6k| *utf16_output++ = char16_t(code_point); 75| 31.6k| pos += 3; 76| 31.6k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (76:16): [True: 23.4k, False: 0] ------------------ 77| | // we have a 4-byte UTF-8 word. 78| 23.4k| if (pos + 3 >= len) { ------------------ | Branch (78:11): [True: 0, False: 23.4k] ------------------ 79| 0| break; 80| 0| } // minimal bound checking 81| 23.4k| uint32_t code_point = ((leading_byte & 0b00000111) << 18) | 82| 23.4k| ((uint8_t(data[pos + 1]) & 0b00111111) << 12) | 83| 23.4k| ((uint8_t(data[pos + 2]) & 0b00111111) << 6) | 84| 23.4k| (uint8_t(data[pos + 3]) & 0b00111111); 85| 23.4k| code_point -= 0x10000; 86| 23.4k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 87| 23.4k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 88| 23.4k| if constexpr (!match_system(big_endian)) { 89| 23.4k| high_surrogate = u16_swap_bytes(high_surrogate); 90| 23.4k| low_surrogate = u16_swap_bytes(low_surrogate); 91| 23.4k| } 92| 23.4k| *utf16_output++ = char16_t(high_surrogate); 93| 23.4k| *utf16_output++ = char16_t(low_surrogate); 94| 23.4k| pos += 4; 95| 23.4k| } else { 96| | // we may have a continuation but we do not do error checking 97| 0| return 0; 98| 0| } 99| 603k| } 100| 1.42k| return utf16_output - start; 101| 1.42k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf3230rewind_and_convert_with_errorsEmPKcmPDi: 249| 1.58k| char32_t *utf32_output) { 250| 1.58k| size_t extra_len{0}; 251| | // We potentially need to go back in time and find a leading byte. 252| 1.58k| size_t how_far_back = 3; // 3 bytes in the past + current position 253| 1.58k| if (how_far_back > prior_bytes) { ------------------ | Branch (253:7): [True: 836, False: 744] ------------------ 254| 836| how_far_back = prior_bytes; 255| 836| } 256| 1.58k| bool found_leading_bytes{false}; 257| | // important: it is i <= how_far_back and not 'i < how_far_back'. 258| 1.67k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (258:22): [True: 1.59k, False: 80] ------------------ 259| 1.59k| unsigned char byte = buf[-static_cast(i)]; 260| 1.59k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 261| 1.59k| if (found_leading_bytes) { ------------------ | Branch (261:9): [True: 1.50k, False: 92] ------------------ 262| 1.50k| if (i > 0 && byte < 128) { ------------------ | Branch (262:11): [True: 12, False: 1.48k] | Branch (262:20): [True: 12, False: 0] ------------------ 263| | // If we had to go back and the leading byte is ascii 264| | // then we can stop right away. 265| 12| return result(error_code::TOO_LONG, 0 - i + 1); 266| 12| } 267| 1.48k| buf -= i; 268| 1.48k| extra_len = i; 269| 1.48k| break; 270| 1.50k| } 271| 1.59k| } 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| 1.56k| if (!found_leading_bytes) { ------------------ | Branch (280:7): [True: 80, False: 1.48k] ------------------ 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| 80| return result(error_code::TOO_LONG, 0 - how_far_back); 286| 80| } 287| | 288| 1.48k| result res = convert_with_errors(buf, len + extra_len, utf32_output); 289| 1.48k| if (res.error) { ------------------ | Branch (289:7): [True: 776, False: 712] ------------------ 290| 776| res.count -= extra_len; 291| 776| } 292| 1.48k| return res; 293| 1.56k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf3219convert_with_errorsIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEENS_6resultES6_mPDi: 121| 2.28k| char32_t *utf32_output) { 122| 2.28k| size_t pos = 0; 123| 2.28k| char32_t *start{utf32_output}; 124| 3.11M| while (pos < len) { ------------------ | Branch (124:10): [True: 3.11M, False: 1.07k] ------------------ 125| |#if SIMDUTF_CPLUSPLUS23 126| | if !consteval 127| |#endif 128| 3.11M| { 129| | // try to convert the next block of 16 ASCII bytes 130| 3.11M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (130:11): [True: 3.10M, False: 8.95k] ------------------ 131| | // they are ascii 132| 3.10M| uint64_t v1; 133| 3.10M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 134| 3.10M| uint64_t v2; 135| 3.10M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 136| 3.10M| uint64_t v{v1 | v2}; 137| 3.10M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (137:13): [True: 1.67M, False: 1.43M] ------------------ 138| 1.67M| size_t final_pos = pos + 16; 139| 28.4M| while (pos < final_pos) { ------------------ | Branch (139:18): [True: 26.7M, False: 1.67M] ------------------ 140| 26.7M| *utf32_output++ = uint8_t(data[pos]); 141| 26.7M| pos++; 142| 26.7M| } 143| 1.67M| continue; 144| 1.67M| } 145| 3.10M| } 146| 3.11M| } 147| 1.44M| auto leading_byte = uint8_t(data[pos]); // leading byte 148| 1.44M| if (leading_byte < 0b10000000) { ------------------ | Branch (148:9): [True: 1.10M, False: 332k] ------------------ 149| | // converting one ASCII byte !!! 150| 1.10M| *utf32_output++ = char32_t(leading_byte); 151| 1.10M| pos++; 152| 1.10M| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (152:16): [True: 235k, False: 97.0k] ------------------ 153| | // We have a two-byte UTF-8 154| 235k| if (pos + 1 >= len) { ------------------ | Branch (154:11): [True: 60, False: 235k] ------------------ 155| 60| return result(error_code::TOO_SHORT, pos); 156| 60| } // minimal bound checking 157| 235k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (157:11): [True: 180, False: 234k] ------------------ 158| 180| return result(error_code::TOO_SHORT, pos); 159| 180| } 160| | // range check 161| 234k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 162| 234k| (uint8_t(data[pos + 1]) & 0b00111111); 163| 234k| if (code_point < 0x80) { ------------------ | Branch (163:11): [True: 30, False: 234k] ------------------ 164| 30| return result(error_code::OVERLONG, pos); 165| 30| } 166| 234k| *utf32_output++ = char32_t(code_point); 167| 234k| pos += 2; 168| 234k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (168:16): [True: 72.2k, False: 24.7k] ------------------ 169| | // We have a three-byte UTF-8 170| 72.2k| if (pos + 2 >= len) { ------------------ | Branch (170:11): [True: 48, False: 72.2k] ------------------ 171| 48| return result(error_code::TOO_SHORT, pos); 172| 48| } // minimal bound checking 173| | 174| 72.2k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (174:11): [True: 114, False: 72.1k] ------------------ 175| 114| return result(error_code::TOO_SHORT, pos); 176| 114| } 177| 72.1k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (177:11): [True: 36, False: 72.0k] ------------------ 178| 36| return result(error_code::TOO_SHORT, pos); 179| 36| } 180| | // range check 181| 72.0k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 182| 72.0k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 183| 72.0k| (uint8_t(data[pos + 2]) & 0b00111111); 184| 72.0k| if (code_point < 0x800) { ------------------ | Branch (184:11): [True: 36, False: 72.0k] ------------------ 185| 36| return result(error_code::OVERLONG, pos); 186| 36| } 187| 72.0k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (187:11): [True: 2.69k, False: 69.3k] | Branch (187:34): [True: 6, False: 2.69k] ------------------ 188| 6| return result(error_code::SURROGATE, pos); 189| 6| } 190| 72.0k| *utf32_output++ = char32_t(code_point); 191| 72.0k| pos += 3; 192| 72.0k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (192:16): [True: 24.2k, False: 472] ------------------ 193| | // we have a 4-byte UTF-8 word. 194| 24.2k| if (pos + 3 >= len) { ------------------ | Branch (194:11): [True: 30, False: 24.2k] ------------------ 195| 30| return result(error_code::TOO_SHORT, pos); 196| 30| } // minimal bound checking 197| 24.2k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (197:11): [True: 84, False: 24.1k] ------------------ 198| 84| return result(error_code::TOO_SHORT, pos); 199| 84| } 200| 24.1k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (200:11): [True: 36, False: 24.1k] ------------------ 201| 36| return result(error_code::TOO_SHORT, pos); 202| 36| } 203| 24.1k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (203:11): [True: 18, False: 24.1k] ------------------ 204| 18| return result(error_code::TOO_SHORT, pos); 205| 18| } 206| | 207| | // range check 208| 24.1k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 209| 24.1k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 210| 24.1k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 211| 24.1k| (uint8_t(data[pos + 3]) & 0b00111111); 212| 24.1k| if (code_point <= 0xffff) { ------------------ | Branch (212:11): [True: 30, False: 24.0k] ------------------ 213| 30| return result(error_code::OVERLONG, pos); 214| 30| } 215| 24.0k| if (0x10ffff < code_point) { ------------------ | Branch (215:11): [True: 30, False: 24.0k] ------------------ 216| 30| return result(error_code::TOO_LARGE, pos); 217| 30| } 218| 24.0k| *utf32_output++ = char32_t(code_point); 219| 24.0k| pos += 4; 220| 24.0k| } else { 221| | // we either have too many continuation bytes or an invalid leading byte 222| 472| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (222:11): [True: 244, False: 228] ------------------ 223| 244| return result(error_code::TOO_LONG, pos); 224| 244| } else { 225| 228| return result(error_code::HEADER_BITS, pos); 226| 228| } 227| 472| } 228| 1.44M| } 229| 1.07k| return result(error_code::SUCCESS, utf32_output - start); 230| 2.28k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf327convertIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mPDi: 16| 1.73k| char32_t *utf32_output) { 17| 1.73k| size_t pos = 0; 18| 1.73k| char32_t *start{utf32_output}; 19| 1.95M| while (pos < len) { ------------------ | Branch (19:10): [True: 1.95M, False: 764] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 1.95M| { 24| | // try to convert the next block of 16 ASCII bytes 25| 1.95M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (25:11): [True: 1.94M, False: 5.83k] ------------------ 26| | // they are ascii 27| 1.94M| uint64_t v1; 28| 1.94M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 29| 1.94M| uint64_t v2; 30| 1.94M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 31| 1.94M| uint64_t v{v1 | v2}; 32| 1.94M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (32:13): [True: 1.16M, False: 782k] ------------------ 33| 1.16M| size_t final_pos = pos + 16; 34| 19.8M| while (pos < final_pos) { ------------------ | Branch (34:18): [True: 18.6M, False: 1.16M] ------------------ 35| 18.6M| *utf32_output++ = uint8_t(data[pos]); 36| 18.6M| pos++; 37| 18.6M| } 38| 1.16M| continue; 39| 1.16M| } 40| 1.94M| } 41| 1.95M| } 42| 788k| auto leading_byte = uint8_t(data[pos]); // leading byte 43| 788k| if (leading_byte < 0b10000000) { ------------------ | Branch (43:9): [True: 639k, False: 148k] ------------------ 44| | // converting one ASCII byte !!! 45| 639k| *utf32_output++ = char32_t(leading_byte); 46| 639k| pos++; 47| 639k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (47:16): [True: 91.7k, False: 56.5k] ------------------ 48| | // We have a two-byte UTF-8 49| 91.7k| if (pos + 1 >= len) { ------------------ | Branch (49:11): [True: 60, False: 91.7k] ------------------ 50| 60| return 0; 51| 60| } // minimal bound checking 52| 91.7k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (52:11): [True: 86, False: 91.6k] ------------------ 53| 86| return 0; 54| 86| } 55| | // range check 56| 91.6k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 57| 91.6k| (uint8_t(data[pos + 1]) & 0b00111111); 58| 91.6k| if (code_point < 0x80) { ------------------ | Branch (58:11): [True: 36, False: 91.6k] ------------------ 59| 36| return 0; 60| 36| } 61| 91.6k| *utf32_output++ = char32_t(code_point); 62| 91.6k| pos += 2; 63| 91.6k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (63:16): [True: 49.7k, False: 6.74k] ------------------ 64| | // We have a three-byte UTF-8 65| 49.7k| if (pos + 2 >= len) { ------------------ | Branch (65:11): [True: 54, False: 49.7k] ------------------ 66| 54| return 0; 67| 54| } // minimal bound checking 68| | 69| 49.7k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (69:11): [True: 68, False: 49.6k] ------------------ 70| 68| return 0; 71| 68| } 72| 49.6k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (72:11): [True: 42, False: 49.6k] ------------------ 73| 42| return 0; 74| 42| } 75| | // range check 76| 49.6k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 77| 49.6k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 78| 49.6k| (uint8_t(data[pos + 2]) & 0b00111111); 79| 49.6k| if (code_point < 0x800 || (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (79:11): [True: 26, False: 49.5k] | Branch (79:34): [True: 746, False: 48.8k] | Branch (79:57): [True: 16, False: 730] ------------------ 80| 42| return 0; 81| 42| } 82| 49.5k| *utf32_output++ = char32_t(code_point); 83| 49.5k| pos += 3; 84| 49.5k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (84:16): [True: 6.37k, False: 370] ------------------ 85| | // we have a 4-byte UTF-8 word. 86| 6.37k| if (pos + 3 >= len) { ------------------ | Branch (86:11): [True: 30, False: 6.34k] ------------------ 87| 30| return 0; 88| 30| } // minimal bound checking 89| 6.34k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (89:11): [True: 50, False: 6.29k] ------------------ 90| 50| return 0; 91| 50| } 92| 6.29k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (92:11): [True: 38, False: 6.25k] ------------------ 93| 38| return 0; 94| 38| } 95| 6.25k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (95:11): [True: 16, False: 6.23k] ------------------ 96| 16| return 0; 97| 16| } 98| | 99| | // range check 100| 6.23k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 101| 6.23k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 102| 6.23k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 103| 6.23k| (uint8_t(data[pos + 3]) & 0b00111111); 104| 6.23k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (104:11): [True: 56, False: 6.18k] | Branch (104:35): [True: 24, False: 6.15k] ------------------ 105| 80| return 0; 106| 80| } 107| 6.15k| *utf32_output++ = char32_t(code_point); 108| 6.15k| pos += 4; 109| 6.15k| } else { 110| 370| return 0; 111| 370| } 112| 788k| } 113| 764| return utf32_output - start; 114| 1.73k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf3213convert_validIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mPDi: 16| 1.35k| char32_t *utf32_output) { 17| 1.35k| size_t pos = 0; 18| 1.35k| char32_t *start{utf32_output}; 19| 3.73M| while (pos < len) { ------------------ | Branch (19:10): [True: 3.72M, False: 1.35k] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 3.72M| { 24| | // try to convert the next block of 8 ASCII bytes 25| 3.72M| if (pos + 8 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (25:11): [True: 3.72M, False: 3.98k] ------------------ 26| | // they are ascii 27| 3.72M| uint64_t v; 28| 3.72M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 29| 3.72M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (29:13): [True: 3.17M, False: 546k] ------------------ 30| 3.17M| size_t final_pos = pos + 8; 31| 28.6M| while (pos < final_pos) { ------------------ | Branch (31:18): [True: 25.4M, False: 3.17M] ------------------ 32| 25.4M| *utf32_output++ = uint8_t(data[pos]); 33| 25.4M| pos++; 34| 25.4M| } 35| 3.17M| continue; 36| 3.17M| } 37| 3.72M| } 38| 3.72M| } 39| 550k| auto leading_byte = uint8_t(data[pos]); // leading byte 40| 550k| if (leading_byte < 0b10000000) { ------------------ | Branch (40:9): [True: 296k, False: 253k] ------------------ 41| | // converting one ASCII byte !!! 42| 296k| *utf32_output++ = char32_t(leading_byte); 43| 296k| pos++; 44| 296k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (44:16): [True: 198k, False: 55.7k] ------------------ 45| | // We have a two-byte UTF-8 46| 198k| if (pos + 1 >= len) { ------------------ | Branch (46:11): [True: 0, False: 198k] ------------------ 47| 0| break; 48| 0| } // minimal bound checking 49| 198k| *utf32_output++ = char32_t(((leading_byte & 0b00011111) << 6) | 50| 198k| (uint8_t(data[pos + 1]) & 0b00111111)); 51| 198k| pos += 2; 52| 198k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (52:16): [True: 21.4k, False: 34.3k] ------------------ 53| | // We have a three-byte UTF-8 54| 21.4k| if (pos + 2 >= len) { ------------------ | Branch (54:11): [True: 0, False: 21.4k] ------------------ 55| 0| break; 56| 0| } // minimal bound checking 57| 21.4k| *utf32_output++ = char32_t(((leading_byte & 0b00001111) << 12) | 58| 21.4k| ((uint8_t(data[pos + 1]) & 0b00111111) << 6) | 59| 21.4k| (uint8_t(data[pos + 2]) & 0b00111111)); 60| 21.4k| pos += 3; 61| 34.3k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (61:16): [True: 34.3k, False: 0] ------------------ 62| | // we have a 4-byte UTF-8 word. 63| 34.3k| if (pos + 3 >= len) { ------------------ | Branch (63:11): [True: 0, False: 34.3k] ------------------ 64| 0| break; 65| 0| } // minimal bound checking 66| 34.3k| uint32_t code_word = ((leading_byte & 0b00000111) << 18) | 67| 34.3k| ((uint8_t(data[pos + 1]) & 0b00111111) << 12) | 68| 34.3k| ((uint8_t(data[pos + 2]) & 0b00111111) << 6) | 69| 34.3k| (uint8_t(data[pos + 3]) & 0b00111111); 70| 34.3k| *utf32_output++ = char32_t(code_word); 71| 34.3k| pos += 4; 72| 34.3k| } else { 73| | // we may have a continuation but we do not do error checking 74| 0| return 0; 75| 0| } 76| 550k| } 77| 1.35k| return utf32_output - start; 78| 1.35k|} _ZNK7simdutf8fallback14implementation13validate_utf8EPKcm: 37| 3.47k|implementation::validate_utf8(const char *buf, size_t len) const noexcept { 38| 3.47k| return scalar::utf8::validate(buf, len); 39| 3.47k|} _ZNK7simdutf8fallback14implementation25validate_utf8_with_errorsEPKcm: 44| 3.47k| const char *buf, size_t len) const noexcept { 45| 3.47k| return scalar::utf8::validate_with_errors(buf, len); 46| 3.47k|} _ZNK7simdutf8fallback14implementation16validate_utf16leEPKDsm: 77| 1.40k| size_t len) const noexcept { 78| 1.40k| return scalar::utf16::validate(buf, len); 79| 1.40k|} _ZNK7simdutf8fallback14implementation16validate_utf16beEPKDsm: 85| 1.31k| size_t len) const noexcept { 86| 1.31k| return scalar::utf16::validate(buf, len); 87| 1.31k|} _ZNK7simdutf8fallback14implementation28validate_utf16le_with_errorsEPKDsm: 90| 1.40k| const char16_t *buf, size_t len) const noexcept { 91| 1.40k| return scalar::utf16::validate_with_errors(buf, len); 92| 1.40k|} _ZNK7simdutf8fallback14implementation28validate_utf16be_with_errorsEPKDsm: 95| 1.31k| const char16_t *buf, size_t len) const noexcept { 96| 1.31k| return scalar::utf16::validate_with_errors(buf, len); 97| 1.31k|} _ZNK7simdutf8fallback14implementation14validate_utf32EPKDim: 114| 2.61k|implementation::validate_utf32(const char32_t *buf, size_t len) const noexcept { 115| 2.61k| return scalar::utf32::validate(buf, len); 116| 2.61k|} _ZNK7simdutf8fallback14implementation26validate_utf32_with_errorsEPKDim: 121| 2.61k| const char32_t *buf, size_t len) const noexcept { 122| 2.61k| return scalar::utf32::validate_with_errors(buf, len); 123| 2.61k|} _ZNK7simdutf8fallback14implementation22convert_latin1_to_utf8EPKcmPc: 128| 606| const char *buf, size_t len, char *utf8_output) const noexcept { 129| 606| return scalar::latin1_to_utf8::convert(buf, len, utf8_output); 130| 606|} _ZNK7simdutf8fallback14implementation25convert_latin1_to_utf16leEPKcmPDs: 135| 86| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 136| 86| return scalar::latin1_to_utf16::convert(buf, len, 137| 86| utf16_output); 138| 86|} _ZNK7simdutf8fallback14implementation25convert_latin1_to_utf16beEPKcmPDs: 141| 88| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 142| 88| return scalar::latin1_to_utf16::convert(buf, len, 143| 88| utf16_output); 144| 88|} _ZNK7simdutf8fallback14implementation23convert_latin1_to_utf32EPKcmPDi: 149| 110| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 150| 110| return scalar::latin1_to_utf32::convert(buf, len, utf32_output); 151| 110|} _ZNK7simdutf8fallback14implementation22convert_utf8_to_latin1EPKcmPc: 156| 590| const char *buf, size_t len, char *latin1_output) const noexcept { 157| 590| return scalar::utf8_to_latin1::convert(buf, len, latin1_output); 158| 590|} _ZNK7simdutf8fallback14implementation34convert_utf8_to_latin1_with_errorsEPKcmPc: 161| 502| const char *buf, size_t len, char *latin1_output) const noexcept { 162| 502| return scalar::utf8_to_latin1::convert_with_errors(buf, len, latin1_output); 163| 502|} _ZNK7simdutf8fallback14implementation23convert_utf8_to_utf16leEPKcmPDs: 173| 720| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 174| 720| return scalar::utf8_to_utf16::convert(buf, len, 175| 720| utf16_output); 176| 720|} _ZNK7simdutf8fallback14implementation23convert_utf8_to_utf16beEPKcmPDs: 179| 756| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 180| 756| return scalar::utf8_to_utf16::convert(buf, len, 181| 756| utf16_output); 182| 756|} _ZNK7simdutf8fallback14implementation35convert_utf8_to_utf16le_with_errorsEPKcmPDs: 185| 612| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 186| 612| return scalar::utf8_to_utf16::convert_with_errors( 187| 612| buf, len, utf16_output); 188| 612|} _ZNK7simdutf8fallback14implementation35convert_utf8_to_utf16be_with_errorsEPKcmPDs: 191| 754| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 192| 754| return scalar::utf8_to_utf16::convert_with_errors( 193| 754| buf, len, utf16_output); 194| 754|} _ZNK7simdutf8fallback14implementation29convert_valid_utf8_to_utf16leEPKcmPDs: 197| 444| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 198| 444| return scalar::utf8_to_utf16::convert_valid(buf, len, 199| 444| utf16_output); 200| 444|} _ZNK7simdutf8fallback14implementation29convert_valid_utf8_to_utf16beEPKcmPDs: 203| 476| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 204| 476| return scalar::utf8_to_utf16::convert_valid(buf, len, 205| 476| utf16_output); 206| 476|} _ZNK7simdutf8fallback14implementation21convert_utf8_to_utf32EPKcmPDi: 211| 784| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 212| 784| return scalar::utf8_to_utf32::convert(buf, len, utf32_output); 213| 784|} _ZNK7simdutf8fallback14implementation33convert_utf8_to_utf32_with_errorsEPKcmPDi: 216| 794| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 217| 794| return scalar::utf8_to_utf32::convert_with_errors(buf, len, utf32_output); 218| 794|} _ZNK7simdutf8fallback14implementation27convert_valid_utf8_to_utf32EPKcmPDi: 221| 450| const char *input, size_t size, char32_t *utf32_output) const noexcept { 222| 450| return scalar::utf8_to_utf32::convert_valid(input, size, utf32_output); 223| 450|} _ZNK7simdutf8fallback14implementation25convert_utf16le_to_latin1EPKDsmPc: 228| 134| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 229| 134| return scalar::utf16_to_latin1::convert(buf, len, 230| 134| latin1_output); 231| 134|} _ZNK7simdutf8fallback14implementation25convert_utf16be_to_latin1EPKDsmPc: 234| 94| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 235| 94| return scalar::utf16_to_latin1::convert(buf, len, 236| 94| latin1_output); 237| 94|} _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| 250| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 249| 250| return scalar::utf16_to_latin1::convert_with_errors( 250| 250| buf, len, latin1_output); 251| 250|} _ZNK7simdutf8fallback14implementation23convert_utf16le_to_utf8EPKDsmPc: 268| 416| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 269| 416| return scalar::utf16_to_utf8::convert(buf, len, 270| 416| utf8_output); 271| 416|} _ZNK7simdutf8fallback14implementation23convert_utf16be_to_utf8EPKDsmPc: 274| 352| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 275| 352| return scalar::utf16_to_utf8::convert(buf, len, utf8_output); 276| 352|} _ZNK7simdutf8fallback14implementation35convert_utf16le_to_utf8_with_errorsEPKDsmPc: 279| 664| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 280| 664| return scalar::utf16_to_utf8::convert_with_errors( 281| 664| buf, len, utf8_output); 282| 664|} _ZNK7simdutf8fallback14implementation35convert_utf16be_to_utf8_with_errorsEPKDsmPc: 285| 728| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 286| 728| return scalar::utf16_to_utf8::convert_with_errors( 287| 728| buf, len, utf8_output); 288| 728|} _ZNK7simdutf8fallback14implementation29convert_valid_utf16le_to_utf8EPKDsmPc: 291| 440| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 292| 440| return scalar::utf16_to_utf8::convert_valid(buf, len, 293| 440| utf8_output); 294| 440|} _ZNK7simdutf8fallback14implementation29convert_valid_utf16be_to_utf8EPKDsmPc: 297| 398| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 298| 398| return scalar::utf16_to_utf8::convert_valid(buf, len, 299| 398| utf8_output); 300| 398|} _ZNK7simdutf8fallback14implementation23convert_utf32_to_latin1EPKDimPc: 305| 172| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 306| 172| return scalar::utf32_to_latin1::convert(buf, len, latin1_output); 307| 172|} _ZNK7simdutf8fallback14implementation35convert_utf32_to_latin1_with_errorsEPKDimPc: 310| 408| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 311| 408| return scalar::utf32_to_latin1::convert_with_errors(buf, len, latin1_output); 312| 408|} _ZNK7simdutf8fallback14implementation21convert_utf32_to_utf8EPKDimPc: 322| 710| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 323| 710| return scalar::utf32_to_utf8::convert(buf, len, utf8_output); 324| 710|} _ZNK7simdutf8fallback14implementation33convert_utf32_to_utf8_with_errorsEPKDimPc: 327| 934| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 328| 934| return scalar::utf32_to_utf8::convert_with_errors(buf, len, utf8_output); 329| 934|} _ZNK7simdutf8fallback14implementation27convert_valid_utf32_to_utf8EPKDimPc: 332| 206| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 333| 206| return scalar::utf32_to_utf8::convert_valid(buf, len, utf8_output); 334| 206|} _ZNK7simdutf8fallback14implementation24convert_utf32_to_utf16leEPKDimPDs: 339| 526| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 340| 526| return scalar::utf32_to_utf16::convert(buf, len, 341| 526| utf16_output); 342| 526|} _ZNK7simdutf8fallback14implementation24convert_utf32_to_utf16beEPKDimPDs: 345| 534| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 346| 534| return scalar::utf32_to_utf16::convert(buf, len, 347| 534| utf16_output); 348| 534|} _ZNK7simdutf8fallback14implementation36convert_utf32_to_utf16le_with_errorsEPKDimPDs: 351| 684| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 352| 684| return scalar::utf32_to_utf16::convert_with_errors( 353| 684| buf, len, utf16_output); 354| 684|} _ZNK7simdutf8fallback14implementation36convert_utf32_to_utf16be_with_errorsEPKDimPDs: 357| 630| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 358| 630| return scalar::utf32_to_utf16::convert_with_errors( 359| 630| buf, len, utf16_output); 360| 630|} _ZNK7simdutf8fallback14implementation30convert_valid_utf32_to_utf16leEPKDimPDs: 363| 248| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 364| 248| return scalar::utf32_to_utf16::convert_valid( 365| 248| buf, len, utf16_output); 366| 248|} _ZNK7simdutf8fallback14implementation30convert_valid_utf32_to_utf16beEPKDimPDs: 369| 128| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 370| 128| return scalar::utf32_to_utf16::convert_valid(buf, len, 371| 128| utf16_output); 372| 128|} _ZNK7simdutf8fallback14implementation24convert_utf16le_to_utf32EPKDsmPDi: 375| 306| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 376| 306| return scalar::utf16_to_utf32::convert(buf, len, 377| 306| utf32_output); 378| 306|} _ZNK7simdutf8fallback14implementation24convert_utf16be_to_utf32EPKDsmPDi: 381| 220| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 382| 220| return scalar::utf16_to_utf32::convert(buf, len, 383| 220| utf32_output); 384| 220|} _ZNK7simdutf8fallback14implementation36convert_utf16le_to_utf32_with_errorsEPKDsmPDi: 387| 418| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 388| 418| return scalar::utf16_to_utf32::convert_with_errors( 389| 418| buf, len, utf32_output); 390| 418|} _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| 106| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 400| 106| return scalar::utf16_to_utf32::convert_valid( 401| 106| buf, len, utf32_output); 402| 106|} _ZNK7simdutf8fallback14implementation30convert_valid_utf16be_to_utf32EPKDsmPDi: 405| 172| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 406| 172| return scalar::utf16_to_utf32::convert_valid(buf, len, 407| 172| utf32_output); 408| 172|} _ZNK7simdutf8fallback14implementation13count_utf16leEPKDsm: 419| 1.40k| const char16_t *input, size_t length) const noexcept { 420| 1.40k| return scalar::utf16::count_code_points(input, length); 421| 1.40k|} _ZNK7simdutf8fallback14implementation13count_utf16beEPKDsm: 424| 1.31k| const char16_t *input, size_t length) const noexcept { 425| 1.31k| return scalar::utf16::count_code_points(input, length); 426| 1.31k|} _ZNK7simdutf8fallback14implementation10count_utf8EPKcm: 431| 3.47k|implementation::count_utf8(const char *input, size_t length) const noexcept { 432| 3.47k| return scalar::utf8::count_code_points(input, length); 433| 3.47k|} _ZNK7simdutf8fallback14implementation23latin1_length_from_utf8EPKcm: 438| 546| const char *buf, size_t len) const noexcept { 439| 546| return scalar::utf8::count_code_points(buf, len); 440| 546|} _ZNK7simdutf8fallback14implementation23utf8_length_from_latin1EPKcm: 445| 303| const char *input, size_t length) const noexcept { 446| 303| return scalar::latin1_to_utf8::utf8_length_from_latin1(input, length); 447| 303|} _ZNK7simdutf8fallback14implementation24utf8_length_from_utf16leEPKDsm: 452| 776| const char16_t *input, size_t length) const noexcept { 453| 776| return scalar::utf16::utf8_length_from_utf16(input, 454| 776| length); 455| 776|} _ZNK7simdutf8fallback14implementation24utf8_length_from_utf16beEPKDsm: 458| 756| const char16_t *input, size_t length) const noexcept { 459| 756| return scalar::utf16::utf8_length_from_utf16(input, length); 460| 756|} _ZNK7simdutf8fallback14implementation25utf32_length_from_utf16leEPKDsm: 465| 420| const char16_t *input, size_t length) const noexcept { 466| 420| return scalar::utf16::utf32_length_from_utf16(input, 467| 420| length); 468| 420|} _ZNK7simdutf8fallback14implementation25utf32_length_from_utf16beEPKDsm: 471| 382| const char16_t *input, size_t length) const noexcept { 472| 382| return scalar::utf16::utf32_length_from_utf16(input, length); 473| 382|} _ZNK7simdutf8fallback14implementation22utf16_length_from_utf8EPKcm: 478| 1.90k| const char *input, size_t length) const noexcept { 479| 1.90k| return scalar::utf8::utf16_length_from_utf8(input, length); 480| 1.90k|} _ZNK7simdutf8fallback14implementation22utf8_length_from_utf32EPKDim: 513| 937| const char32_t *input, size_t length) const noexcept { 514| 937| return scalar::utf32::utf8_length_from_utf32(input, length); 515| 937|} _ZNK7simdutf8fallback14implementation23utf16_length_from_utf32EPKDim: 520| 1.38k| const char32_t *input, size_t length) const noexcept { 521| 1.38k| return scalar::utf32::utf16_length_from_utf32(input, length); 522| 1.38k|} _ZNK7simdutf8fallback14implementation22utf32_length_from_utf8EPKcm: 527| 1.01k| const char *input, size_t length) const noexcept { 528| 1.01k| return scalar::utf8::count_code_points(input, length); 529| 1.01k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE13get_remainderEPh: 57| 6.09k|buf_block_reader::get_remainder(uint8_t *dst) const { 58| 6.09k| if (len == idx) { ------------------ | Branch (58:7): [True: 36, False: 6.05k] ------------------ 59| 36| return 0; 60| 36| } // memcpy(dst, null, 0) will trigger an error with some sanitizers 61| 6.05k| std::memset(dst, 0x20, 62| 6.05k| STEP_SIZE); // std::memset STEP_SIZE because it is more efficient 63| | // to write out 8 or 16 bytes at once. 64| 6.05k| std::memcpy(dst, buf + idx, len - idx); 65| 6.05k| return len - idx; 66| 6.09k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE13get_remainderEPh: 57| 6.09k|buf_block_reader::get_remainder(uint8_t *dst) const { 58| 6.09k| if (len == idx) { ------------------ | Branch (58:7): [True: 36, False: 6.05k] ------------------ 59| 36| return 0; 60| 36| } // memcpy(dst, null, 0) will trigger an error with some sanitizers 61| 6.05k| std::memset(dst, 0x20, 62| 6.05k| STEP_SIZE); // std::memset STEP_SIZE because it is more efficient 63| | // to write out 8 or 16 bytes at once. 64| 6.05k| std::memcpy(dst, buf + idx, len - idx); 65| 6.05k| return len - idx; 66| 6.09k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EEC2EPKhm: 36| 6.94k| : buf{_buf}, len{_len}, lenminusstep{len < STEP_SIZE ? 0 : len - STEP_SIZE}, ------------------ | Branch (36:42): [True: 2.64k, False: 4.29k] ------------------ 37| 6.94k| idx{0} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE14has_full_blockEv: 45| 4.68M|simdutf_really_inline bool buf_block_reader::has_full_block() const { 46| 4.68M| return idx < lenminusstep; 47| 4.68M|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE10full_blockEv: 51| 4.67M|buf_block_reader::full_block() const { 52| 4.67M| return &buf[idx]; 53| 4.67M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE7advanceEv: 69| 4.68M|simdutf_really_inline void buf_block_reader::advance() { 70| 4.68M| idx += STEP_SIZE; 71| 4.68M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EEC2EPKhm: 36| 6.94k| : buf{_buf}, len{_len}, lenminusstep{len < STEP_SIZE ? 0 : len - STEP_SIZE}, ------------------ | Branch (36:42): [True: 2.64k, False: 4.29k] ------------------ 37| 6.94k| idx{0} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE14has_full_blockEv: 45| 4.68M|simdutf_really_inline bool buf_block_reader::has_full_block() const { 46| 4.68M| return idx < lenminusstep; 47| 4.68M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE10full_blockEv: 51| 4.67M|buf_block_reader::full_block() const { 52| 4.67M| return &buf[idx]; 53| 4.67M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE7advanceEv: 69| 4.68M|simdutf_really_inline void buf_block_reader::advance() { 70| 4.68M| idx += STEP_SIZE; 71| 4.68M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE0EEEmPKDsm: 8| 1.82k| size_t size) { 9| 1.82k| size_t pos = 0; 10| 1.82k| size_t count = 0; 11| 1.09M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 1.09M, False: 1.82k] ------------------ 12| 1.09M| simd16x32 input(reinterpret_cast(in + pos)); 13| | if constexpr (!match_system(big_endian)) { 14| | input.swap_bytes(); 15| | } 16| 1.09M| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 1.09M| count += count_ones(not_pair) / 2; 18| 1.09M| } 19| 1.82k| return count + 20| 1.82k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.82k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE1EEEmPKDsm: 8| 1.69k| size_t size) { 9| 1.69k| size_t pos = 0; 10| 1.69k| size_t count = 0; 11| 1.00M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 999k, False: 1.69k] ------------------ 12| 999k| simd16x32 input(reinterpret_cast(in + pos)); 13| 999k| if constexpr (!match_system(big_endian)) { 14| 999k| input.swap_bytes(); 15| 999k| } 16| 999k| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 999k| count += count_ones(not_pair) / 2; 18| 999k| } 19| 1.69k| return count + 20| 1.69k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.69k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 51| 420| size_t size) { 52| 420| return count_code_points(in, size); 53| 420|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 51| 382| size_t size) { 52| 382| return count_code_points(in, size); 53| 382|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE0EEEmPKDsm: 8| 1.82k| size_t size) { 9| 1.82k| size_t pos = 0; 10| 1.82k| size_t count = 0; 11| 1.09M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 1.09M, False: 1.82k] ------------------ 12| 1.09M| simd16x32 input(reinterpret_cast(in + pos)); 13| | if constexpr (!match_system(big_endian)) { 14| | input.swap_bytes(); 15| | } 16| 1.09M| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 1.09M| count += count_ones(not_pair) / 2; 18| 1.09M| } 19| 1.82k| return count + 20| 1.82k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.82k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE1EEEmPKDsm: 8| 1.69k| size_t size) { 9| 1.69k| size_t pos = 0; 10| 1.69k| size_t count = 0; 11| 1.00M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 999k, False: 1.69k] ------------------ 12| 999k| simd16x32 input(reinterpret_cast(in + pos)); 13| 999k| if constexpr (!match_system(big_endian)) { 14| 999k| input.swap_bytes(); 15| 999k| } 16| 999k| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 999k| count += count_ones(not_pair) / 2; 18| 999k| } 19| 1.69k| return count + 20| 1.69k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.69k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 51| 420| size_t size) { 52| 420| return count_code_points(in, size); 53| 420|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 51| 382| size_t size) { 52| 382| return count_code_points(in, size); 53| 382|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE0EEEmPKDsm: 10| 776| size_t size) { 11| 776| size_t pos = 0; 12| | 13| 776| using vector_u16 = simd16; 14| 776| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 776| const auto one = vector_u16::splat(1); 17| | 18| 776| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 776| 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| 776| constexpr size_t max_iterations = 65535 / 2; 26| 776| size_t iteration = max_iterations; 27| | 28| 1.07M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 1.07M, False: 776] ------------------ 29| 1.07M| 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| 1.07M| 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| 1.07M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 1.07M| 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| 1.07M| v_count += c0; 68| 1.07M| v_count += c1; 69| 1.07M| v_count += vector_u16(is_surrogate); 70| | 71| 1.07M| iteration -= 1; 72| 1.07M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 18, False: 1.07M] ------------------ 73| 18| count += v_count.sum(); 74| 18| v_count = vector_u16::zero(); 75| 18| iteration = max_iterations; 76| 18| } 77| 1.07M| } 78| | 79| 776| if (iteration > 0) { ------------------ | Branch (79:7): [True: 776, False: 0] ------------------ 80| 776| count += v_count.sum(); 81| 776| } 82| | 83| 776| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 776| size - pos); 85| 776|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE1EEEmPKDsm: 10| 756| size_t size) { 11| 756| size_t pos = 0; 12| | 13| 756| using vector_u16 = simd16; 14| 756| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 756| const auto one = vector_u16::splat(1); 17| | 18| 756| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 756| 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| 756| constexpr size_t max_iterations = 65535 / 2; 26| 756| size_t iteration = max_iterations; 27| | 28| 1.03M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 1.03M, False: 756] ------------------ 29| 1.03M| auto input = vector_u16::load(reinterpret_cast(in + pos)); 30| 1.03M| if constexpr (!match_system(big_endian)) { 31| 1.03M| input = input.swap_bytes(); 32| 1.03M| } 33| | // 0xd800 .. 0xdbff - low surrogate 34| | // 0xdc00 .. 0xdfff - high surrogate 35| 1.03M| 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| 1.03M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 1.03M| 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| 1.03M| v_count += c0; 68| 1.03M| v_count += c1; 69| 1.03M| v_count += vector_u16(is_surrogate); 70| | 71| 1.03M| iteration -= 1; 72| 1.03M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 16, False: 1.03M] ------------------ 73| 16| count += v_count.sum(); 74| 16| v_count = vector_u16::zero(); 75| 16| iteration = max_iterations; 76| 16| } 77| 1.03M| } 78| | 79| 756| if (iteration > 0) { ------------------ | Branch (79:7): [True: 756, False: 0] ------------------ 80| 756| count += v_count.sum(); 81| 756| } 82| | 83| 756| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 756| size - pos); 85| 756|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE0EEEmPKDsm: 10| 776| size_t size) { 11| 776| size_t pos = 0; 12| | 13| 776| using vector_u16 = simd16; 14| 776| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 776| const auto one = vector_u16::splat(1); 17| | 18| 776| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 776| 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| 776| constexpr size_t max_iterations = 65535 / 2; 26| 776| size_t iteration = max_iterations; 27| | 28| 2.15M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 2.15M, False: 776] ------------------ 29| 2.15M| 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| 2.15M| 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| 2.15M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 2.15M| 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| 2.15M| v_count += c0; 68| 2.15M| v_count += c1; 69| 2.15M| v_count += vector_u16(is_surrogate); 70| | 71| 2.15M| iteration -= 1; 72| 2.15M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 48, False: 2.15M] ------------------ 73| 48| count += v_count.sum(); 74| 48| v_count = vector_u16::zero(); 75| 48| iteration = max_iterations; 76| 48| } 77| 2.15M| } 78| | 79| 776| if (iteration > 0) { ------------------ | Branch (79:7): [True: 776, False: 0] ------------------ 80| 776| count += v_count.sum(); 81| 776| } 82| | 83| 776| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 776| size - pos); 85| 776|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE1EEEmPKDsm: 10| 756| size_t size) { 11| 756| size_t pos = 0; 12| | 13| 756| using vector_u16 = simd16; 14| 756| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 756| const auto one = vector_u16::splat(1); 17| | 18| 756| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 756| 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| 756| constexpr size_t max_iterations = 65535 / 2; 26| 756| size_t iteration = max_iterations; 27| | 28| 2.06M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 2.06M, False: 756] ------------------ 29| 2.06M| auto input = vector_u16::load(reinterpret_cast(in + pos)); 30| 2.06M| if constexpr (!match_system(big_endian)) { 31| 2.06M| input = input.swap_bytes(); 32| 2.06M| } 33| | // 0xd800 .. 0xdbff - low surrogate 34| | // 0xdc00 .. 0xdfff - high surrogate 35| 2.06M| 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| 2.06M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 2.06M| 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| 2.06M| v_count += c0; 68| 2.06M| v_count += c1; 69| 2.06M| v_count += vector_u16(is_surrogate); 70| | 71| 2.06M| iteration -= 1; 72| 2.06M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 47, False: 2.06M] ------------------ 73| 47| count += v_count.sum(); 74| 47| v_count = vector_u16::zero(); 75| 47| iteration = max_iterations; 76| 47| } 77| 2.06M| } 78| | 79| 756| if (iteration > 0) { ------------------ | Branch (79:7): [True: 756, False: 0] ------------------ 80| 756| count += v_count.sum(); 81| 756| } 82| | 83| 756| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 756| size - pos); 85| 756|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf3222utf8_length_from_utf32EPKDim: 11| 937| size_t length) { 12| 937| using vector_u32 = simd32; 13| | 14| 937| 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| 937| const size_t max_increment = 3; 19| | 20| 937| 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| 937| const auto v_ffffff80 = vector_u32::splat(0xffffff80); 28| 937| const auto v_fffff800 = vector_u32::splat(0xfffff800); 29| 937| const auto v_ffff0000 = vector_u32::splat(0xffff0000); 30| 937| const auto one = vector_u32::splat(1); 31| 937|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 32| | 33| 937| size_t counter = 0; 34| | 35| | // 1. vectorized loop unrolled 4 times 36| 937| { 37| | // we use vector of uint32 counters, this is why this limit is used 38| 937| const size_t max_iterations = 39| 937| std::numeric_limits::max() / (max_increment * 4); 40| 937| size_t blocks = length / (N * 4); 41| 937| length -= blocks * (N * 4); 42| 1.14k| while (blocks != 0) { ------------------ | Branch (42:12): [True: 210, False: 937] ------------------ 43| 210| const size_t iterations = min(blocks, max_iterations); 44| 210| blocks -= iterations; 45| | 46| 210| simd32 acc = vector_u32::zero(); 47| 75.1k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (47:26): [True: 74.9k, False: 210] ------------------ 48| 74.9k| const auto in0 = vector_u32(input + 0 * N); 49| 74.9k| const auto in1 = vector_u32(input + 1 * N); 50| 74.9k| const auto in2 = vector_u32(input + 2 * N); 51| 74.9k| 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| 74.9k| acc += min(one, in0 & v_ffffff80); 70| 74.9k| acc += min(one, in1 & v_ffffff80); 71| 74.9k| acc += min(one, in2 & v_ffffff80); 72| 74.9k| acc += min(one, in3 & v_ffffff80); 73| | 74| 74.9k| acc += min(one, in0 & v_fffff800); 75| 74.9k| acc += min(one, in1 & v_fffff800); 76| 74.9k| acc += min(one, in2 & v_fffff800); 77| 74.9k| acc += min(one, in3 & v_fffff800); 78| | 79| 74.9k| acc += min(one, in0 & v_ffff0000); 80| 74.9k| acc += min(one, in1 & v_ffff0000); 81| 74.9k| acc += min(one, in2 & v_ffff0000); 82| 74.9k| acc += min(one, in3 & v_ffff0000); 83| 74.9k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 84| | 85| 74.9k| input += 4 * N; 86| 74.9k| } 87| | 88| 210| counter += acc.sum(); 89| 210| } 90| 937| } 91| | 92| | // 2. vectorized loop for tail 93| 937| { 94| 937| const size_t max_iterations = 95| 937| std::numeric_limits::max() / max_increment; 96| 937| size_t blocks = length / N; 97| 937| length -= blocks * N; 98| 1.43k| while (blocks != 0) { ------------------ | Branch (98:12): [True: 495, False: 937] ------------------ 99| 495| const size_t iterations = min(blocks, max_iterations); 100| 495| blocks -= iterations; 101| | 102| 495| auto acc = vector_u32::zero(); 103| 1.70k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (103:26): [True: 1.20k, False: 495] ------------------ 104| 1.20k| 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| 1.20k| acc += min(one, in & v_ffffff80); 112| 1.20k| acc += min(one, in & v_fffff800); 113| 1.20k| acc += min(one, in & v_ffff0000); 114| 1.20k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 115| | 116| 1.20k| input += N; 117| 1.20k| } 118| | 119| 495| counter += acc.sum(); 120| 495| } 121| 937| } 122| | 123| 937| const size_t consumed = input - start; 124| 937| if (consumed != 0) { ------------------ | Branch (124:7): [True: 537, False: 400] ------------------ 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| 537| counter += consumed; 128| 537| } 129| | 130| 937| return counter + scalar::utf32::utf8_length_from_utf32(input, length); 131| 937|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf323minImEET_S4_S4_: 8| 705|template T min(T a, T b) { return a <= b ? a : b; } ------------------ | Branch (8:48): [True: 705, False: 0] ------------------ simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf3222utf8_length_from_utf32EPKDim: 11| 937| size_t length) { 12| 937| using vector_u32 = simd32; 13| | 14| 937| 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| 937| const size_t max_increment = 3; 19| | 20| 937| 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| 937| const auto v_ffffff80 = vector_u32::splat(0xffffff80); 28| 937| const auto v_fffff800 = vector_u32::splat(0xfffff800); 29| 937| const auto v_ffff0000 = vector_u32::splat(0xffff0000); 30| 937| const auto one = vector_u32::splat(1); 31| 937|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 32| | 33| 937| size_t counter = 0; 34| | 35| | // 1. vectorized loop unrolled 4 times 36| 937| { 37| | // we use vector of uint32 counters, this is why this limit is used 38| 937| const size_t max_iterations = 39| 937| std::numeric_limits::max() / (max_increment * 4); 40| 937| size_t blocks = length / (N * 4); 41| 937| length -= blocks * (N * 4); 42| 1.43k| while (blocks != 0) { ------------------ | Branch (42:12): [True: 499, False: 937] ------------------ 43| 499| const size_t iterations = min(blocks, max_iterations); 44| 499| blocks -= iterations; 45| | 46| 499| simd32 acc = vector_u32::zero(); 47| 150k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (47:26): [True: 150k, False: 499] ------------------ 48| 150k| const auto in0 = vector_u32(input + 0 * N); 49| 150k| const auto in1 = vector_u32(input + 1 * N); 50| 150k| const auto in2 = vector_u32(input + 2 * N); 51| 150k| 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| 150k| acc += min(one, in0 & v_ffffff80); 70| 150k| acc += min(one, in1 & v_ffffff80); 71| 150k| acc += min(one, in2 & v_ffffff80); 72| 150k| acc += min(one, in3 & v_ffffff80); 73| | 74| 150k| acc += min(one, in0 & v_fffff800); 75| 150k| acc += min(one, in1 & v_fffff800); 76| 150k| acc += min(one, in2 & v_fffff800); 77| 150k| acc += min(one, in3 & v_fffff800); 78| | 79| 150k| acc += min(one, in0 & v_ffff0000); 80| 150k| acc += min(one, in1 & v_ffff0000); 81| 150k| acc += min(one, in2 & v_ffff0000); 82| 150k| acc += min(one, in3 & v_ffff0000); 83| 150k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 84| | 85| 150k| input += 4 * N; 86| 150k| } 87| | 88| 499| counter += acc.sum(); 89| 499| } 90| 937| } 91| | 92| | // 2. vectorized loop for tail 93| 937| { 94| 937| const size_t max_iterations = 95| 937| std::numeric_limits::max() / max_increment; 96| 937| size_t blocks = length / N; 97| 937| length -= blocks * N; 98| 1.46k| while (blocks != 0) { ------------------ | Branch (98:12): [True: 528, False: 937] ------------------ 99| 528| const size_t iterations = min(blocks, max_iterations); 100| 528| blocks -= iterations; 101| | 102| 528| auto acc = vector_u32::zero(); 103| 1.84k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (103:26): [True: 1.31k, False: 528] ------------------ 104| 1.31k| 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| 1.31k| acc += min(one, in & v_ffffff80); 112| 1.31k| acc += min(one, in & v_fffff800); 113| 1.31k| acc += min(one, in & v_ffff0000); 114| 1.31k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 115| | 116| 1.31k| input += N; 117| 1.31k| } 118| | 119| 528| counter += acc.sum(); 120| 528| } 121| 937| } 122| | 123| 937| const size_t consumed = input - start; 124| 937| if (consumed != 0) { ------------------ | Branch (124:7): [True: 575, False: 362] ------------------ 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| 575| counter += consumed; 128| 575| } 129| | 130| 937| return counter + scalar::utf32::utf8_length_from_utf32(input, length); 131| 937|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf323minImEET_S4_S4_: 8| 1.02k|template T min(T a, T b) { return a <= b ? a : b; } ------------------ | Branch (8:48): [True: 1.02k, False: 0] ------------------ simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14utf826count_code_points_bytemaskEPKcm: 21| 4.01k|count_code_points_bytemask(const char *in, size_t size) { 22| 4.01k| using vector_i8 = simd8; 23| 4.01k| using vector_u8 = simd8; 24| 4.01k| using vector_u64 = simd64; 25| | 26| 4.01k| constexpr size_t N = vector_i8::SIZE; 27| 4.01k| constexpr size_t max_iterations = 255 / 4; 28| | 29| 4.01k| size_t pos = 0; 30| 4.01k| size_t count = 0; 31| | 32| 4.01k| auto counters = vector_u64::zero(); 33| 4.01k| auto local = vector_u8::zero(); 34| 4.01k| size_t iterations = 0; 35| 1.61M| for (; pos + 4 * N <= size; pos += 4 * N) { ------------------ | Branch (35:10): [True: 1.61M, False: 4.01k] ------------------ 36| 1.61M| const auto input0 = 37| 1.61M| simd8::load(reinterpret_cast(in + pos + 0 * N)); 38| 1.61M| const auto input1 = 39| 1.61M| simd8::load(reinterpret_cast(in + pos + 1 * N)); 40| 1.61M| const auto input2 = 41| 1.61M| simd8::load(reinterpret_cast(in + pos + 2 * N)); 42| 1.61M| const auto input3 = 43| 1.61M| simd8::load(reinterpret_cast(in + pos + 3 * N)); 44| 1.61M| const auto mask0 = input0 > int8_t(-65); 45| 1.61M| const auto mask1 = input1 > int8_t(-65); 46| 1.61M| const auto mask2 = input2 > int8_t(-65); 47| 1.61M| const auto mask3 = input3 > int8_t(-65); 48| | 49| 1.61M| local -= vector_u8(mask0); 50| 1.61M| local -= vector_u8(mask1); 51| 1.61M| local -= vector_u8(mask2); 52| 1.61M| local -= vector_u8(mask3); 53| | 54| 1.61M| iterations += 1; 55| 1.61M| if (iterations == max_iterations) { ------------------ | Branch (55:9): [True: 25.3k, False: 1.58M] ------------------ 56| 25.3k| counters += sum_8bytes(local); 57| 25.3k| local = vector_u8::zero(); 58| 25.3k| iterations = 0; 59| 25.3k| } 60| 1.61M| } 61| | 62| 4.01k| if (iterations > 0) { ------------------ | Branch (62:7): [True: 1.47k, False: 2.54k] ------------------ 63| 1.47k| count += local.sum_bytes(); 64| 1.47k| } 65| | 66| 4.01k| count += counters.sum(); 67| | 68| 4.01k| return count + scalar::utf8::count_code_points(in + pos, size - pos); 69| 4.01k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14utf817count_code_pointsEPKcm: 8| 1.01k|simdutf_really_inline size_t count_code_points(const char *in, size_t size) { 9| 1.01k| size_t pos = 0; 10| 1.01k| size_t count = 0; 11| 774k| for (; pos + 64 <= size; pos += 64) { ------------------ | Branch (11:10): [True: 773k, False: 1.01k] ------------------ 12| 773k| simd8x64 input(reinterpret_cast(in + pos)); 13| 773k| uint64_t utf8_continuation_mask = input.gt(-65); 14| 773k| count += count_ones(utf8_continuation_mask); 15| 773k| } 16| 1.01k| return count + scalar::utf8::count_code_points(in + pos, size - pos); 17| 1.01k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14utf826count_code_points_bytemaskEPKcm: 21| 4.01k|count_code_points_bytemask(const char *in, size_t size) { 22| 4.01k| using vector_i8 = simd8; 23| 4.01k| using vector_u8 = simd8; 24| 4.01k| using vector_u64 = simd64; 25| | 26| 4.01k| constexpr size_t N = vector_i8::SIZE; 27| 4.01k| constexpr size_t max_iterations = 255 / 4; 28| | 29| 4.01k| size_t pos = 0; 30| 4.01k| size_t count = 0; 31| | 32| 4.01k| auto counters = vector_u64::zero(); 33| 4.01k| auto local = vector_u8::zero(); 34| 4.01k| size_t iterations = 0; 35| 3.23M| for (; pos + 4 * N <= size; pos += 4 * N) { ------------------ | Branch (35:10): [True: 3.22M, False: 4.01k] ------------------ 36| 3.22M| const auto input0 = 37| 3.22M| simd8::load(reinterpret_cast(in + pos + 0 * N)); 38| 3.22M| const auto input1 = 39| 3.22M| simd8::load(reinterpret_cast(in + pos + 1 * N)); 40| 3.22M| const auto input2 = 41| 3.22M| simd8::load(reinterpret_cast(in + pos + 2 * N)); 42| 3.22M| const auto input3 = 43| 3.22M| simd8::load(reinterpret_cast(in + pos + 3 * N)); 44| 3.22M| const auto mask0 = input0 > int8_t(-65); 45| 3.22M| const auto mask1 = input1 > int8_t(-65); 46| 3.22M| const auto mask2 = input2 > int8_t(-65); 47| 3.22M| const auto mask3 = input3 > int8_t(-65); 48| | 49| 3.22M| local -= vector_u8(mask0); 50| 3.22M| local -= vector_u8(mask1); 51| 3.22M| local -= vector_u8(mask2); 52| 3.22M| local -= vector_u8(mask3); 53| | 54| 3.22M| iterations += 1; 55| 3.22M| if (iterations == max_iterations) { ------------------ | Branch (55:9): [True: 50.9k, False: 3.17M] ------------------ 56| 50.9k| counters += sum_8bytes(local); 57| 50.9k| local = vector_u8::zero(); 58| 50.9k| iterations = 0; 59| 50.9k| } 60| 3.22M| } 61| | 62| 4.01k| if (iterations > 0) { ------------------ | Branch (62:7): [True: 2.44k, False: 1.57k] ------------------ 63| 2.44k| count += local.sum_bytes(); 64| 2.44k| } 65| | 66| 4.01k| count += counters.sum(); 67| | 68| 4.01k| return count + scalar::utf8::count_code_points(in + pos, size - pos); 69| 4.01k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14utf817count_code_pointsEPKcm: 8| 1.01k|simdutf_really_inline size_t count_code_points(const char *in, size_t size) { 9| 1.01k| size_t pos = 0; 10| 1.01k| size_t count = 0; 11| 774k| for (; pos + 64 <= size; pos += 64) { ------------------ | Branch (11:10): [True: 773k, False: 1.01k] ------------------ 12| 773k| simd8x64 input(reinterpret_cast(in + pos)); 13| 773k| uint64_t utf8_continuation_mask = input.gt(-65); 14| 773k| count += count_ones(utf8_continuation_mask); 15| 773k| } 16| 1.01k| return count + scalar::utf8::count_code_points(in + pos, size - pos); 17| 1.01k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14utf831utf16_length_from_utf8_bytemaskEPKcm: 9| 1.90k| size_t size) { 10| 1.90k| using vector_i8 = simd8; 11| 1.90k| using vector_u8 = simd8; 12| 1.90k| using vector_u64 = simd64; 13| | 14| 1.90k| constexpr size_t N = vector_i8::SIZE; 15| 1.90k| constexpr size_t max_iterations = 255 / 2; 16| | 17| 1.90k| auto counters = vector_u64::zero(); 18| 1.90k| auto local = vector_u8::zero(); 19| | 20| 1.90k| size_t iterations = 0; 21| 1.90k| size_t pos = 0; 22| 1.90k| size_t count = 0; 23| 2.83M| for (; pos + N <= size; pos += N) { ------------------ | Branch (23:10): [True: 2.82M, False: 1.90k] ------------------ 24| 2.82M| const auto input = 25| 2.82M| vector_i8::load(reinterpret_cast(in + pos)); 26| | 27| 2.82M| const auto continuation = input > int8_t(-65); 28| 2.82M| const auto utf_4bytes = vector_u8(input.value) >= uint8_t(240); 29| | 30| 2.82M| local -= vector_u8(continuation); 31| 2.82M| local -= vector_u8(utf_4bytes); 32| | 33| 2.82M| iterations += 1; 34| 2.82M| if (iterations == max_iterations) { ------------------ | Branch (34:9): [True: 22.1k, False: 2.80M] ------------------ 35| 22.1k| counters += sum_8bytes(local); 36| 22.1k| local = vector_u8::zero(); 37| 22.1k| iterations = 0; 38| 22.1k| } 39| 2.82M| } 40| | 41| 1.90k| if (iterations > 0) { ------------------ | Branch (41:7): [True: 1.23k, False: 677] ------------------ 42| 1.23k| count += local.sum_bytes(); 43| 1.23k| } 44| | 45| 1.90k| count += counters.sum(); 46| | 47| 1.90k| return count + scalar::utf8::utf16_length_from_utf8(in + pos, size - pos); 48| 1.90k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14utf831utf16_length_from_utf8_bytemaskEPKcm: 9| 1.90k| size_t size) { 10| 1.90k| using vector_i8 = simd8; 11| 1.90k| using vector_u8 = simd8; 12| 1.90k| using vector_u64 = simd64; 13| | 14| 1.90k| constexpr size_t N = vector_i8::SIZE; 15| 1.90k| constexpr size_t max_iterations = 255 / 2; 16| | 17| 1.90k| auto counters = vector_u64::zero(); 18| 1.90k| auto local = vector_u8::zero(); 19| | 20| 1.90k| size_t iterations = 0; 21| 1.90k| size_t pos = 0; 22| 1.90k| size_t count = 0; 23| 5.66M| for (; pos + N <= size; pos += N) { ------------------ | Branch (23:10): [True: 5.65M, False: 1.90k] ------------------ 24| 5.65M| const auto input = 25| 5.65M| vector_i8::load(reinterpret_cast(in + pos)); 26| | 27| 5.65M| const auto continuation = input > int8_t(-65); 28| 5.65M| const auto utf_4bytes = vector_u8(input.value) >= uint8_t(240); 29| | 30| 5.65M| local -= vector_u8(continuation); 31| 5.65M| local -= vector_u8(utf_4bytes); 32| | 33| 5.65M| iterations += 1; 34| 5.65M| if (iterations == max_iterations) { ------------------ | Branch (34:9): [True: 44.3k, False: 5.61M] ------------------ 35| 44.3k| counters += sum_8bytes(local); 36| 44.3k| local = vector_u8::zero(); 37| 44.3k| iterations = 0; 38| 44.3k| } 39| 5.65M| } 40| | 41| 1.90k| if (iterations > 0) { ------------------ | Branch (41:7): [True: 1.33k, False: 569] ------------------ 42| 1.33k| count += local.sum_bytes(); 43| 1.33k| } 44| | 45| 1.90k| count += counters.sum(); 46| | 47| 1.90k| return count + scalar::utf8::utf16_length_from_utf8(in + pos, size - pos); 48| 1.90k|} 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| 590| char *latin1_output) { 116| 590| size_t pos = 0; 117| 590| 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| 590| size_t leading_byte = 0; 125| 590| size_t margin = size; 126| 9.57k| for (; margin > 0 && leading_byte < 16; margin--) { ------------------ | Branch (126:12): [True: 9.42k, False: 150] | Branch (126:26): [True: 8.98k, False: 440] ------------------ 127| 8.98k| leading_byte += (int8_t(in[margin - 1]) > 128| 8.98k| -65); // twos complement of -65 is 1011 1111 ... 129| 8.98k| } 130| | // If the input is long enough, then we have that margin-1 is the eight last 131| | // leading byte. 132| 590| const size_t safety_margin = size - margin + 1; // to avoid overruns! 133| 613k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (133:12): [True: 613k, False: 562] ------------------ 134| 613k| simd8x64 input(reinterpret_cast(in + pos)); 135| 613k| if (input.is_ascii()) { ------------------ | Branch (135:11): [True: 564k, False: 48.6k] ------------------ 136| 564k| input.store((int8_t *)latin1_output); 137| 564k| latin1_output += 64; 138| 564k| pos += 64; 139| 564k| } else { 140| | // you might think that a for-loop would work, but under Visual Studio, 141| | // it is not good enough. 142| 48.6k| static_assert( 143| 48.6k| (simd8x64::NUM_CHUNKS == 2) || 144| 48.6k| (simd8x64::NUM_CHUNKS == 4), 145| 48.6k| "We support either two or four chunks per 64-byte block."); 146| 48.6k| auto zero = simd8{uint8_t(0)}; 147| 48.6k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 148| 48.6k| this->check_utf8_bytes(input.chunks[0], zero); 149| 48.6k| 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| 48.6k| uint64_t utf8_continuation_mask = 157| 48.6k| input.lt(-65 + 1); // -64 is 1100 0000 in twos complement. Note: in 158| | // this case, we also have ASCII to account for. 159| 48.6k| if (utf8_continuation_mask & 1) { ------------------ | Branch (159:13): [True: 28, False: 48.6k] ------------------ 160| 28| return 0; // error 161| 28| } 162| 48.6k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 163| 48.6k| 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| 48.6k| size_t max_starting_point = (pos + 64) - 12; 168| | // Next loop is going to run at least five times. 169| 481k| while (pos < max_starting_point) { ------------------ | Branch (169:16): [True: 432k, False: 48.6k] ------------------ 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| 432k| size_t consumed = convert_masked_utf8_to_latin1( 180| 432k| in + pos, utf8_end_of_code_point_mask, latin1_output); 181| 432k| pos += consumed; 182| 432k| utf8_end_of_code_point_mask >>= consumed; 183| 432k| } 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| 48.6k| } 189| 613k| } 190| 562| if (errors()) { ------------------ | Branch (190:9): [True: 226, False: 336] ------------------ 191| 226| return 0; 192| 226| } 193| 336| if (pos < size) { ------------------ | Branch (193:9): [True: 332, False: 4] ------------------ 194| 332| size_t howmany = 195| 332| scalar::utf8_to_latin1::convert(in + pos, size - pos, latin1_output); 196| 332| if (howmany == 0) { ------------------ | Branch (196:11): [True: 154, False: 178] ------------------ 197| 154| return 0; 198| 154| } 199| 178| latin1_output += howmany; 200| 178| } 201| 182| return latin1_output - start; 202| 336| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 106| 135k| 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| 135k| simd8 prev1 = input.prev<1>(prev_input); 111| 135k| this->error |= check_special_cases(input, prev1); 112| 135k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin119check_special_casesENS1_4simd5simd8IhEES5_: 8| 135k|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| 135k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 20| | // 11______ 11______ 21| 135k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 22| 135k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 23| 135k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 24| 135k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 25| 135k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 26| 135k| 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| 135k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 35| | // 11110101 1000____ 36| | // 1111011_ 1000____ 37| | // 11111___ 1000____ 38| 135k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 39| 135k| constexpr const uint8_t FORBIDDEN = 0xff; 40| | 41| 135k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 42| | // 0_______ ________ 43| 135k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 44| 135k| TOO_LONG, 45| | // 10______ ________ 46| 135k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 47| | // 1100____ ________ 48| 135k| TOO_SHORT | OVERLONG_2, 49| | // 1101____ ________ 50| 135k| FORBIDDEN, 51| | // 1110____ ________ 52| 135k| FORBIDDEN, 53| | // 1111____ ________ 54| 135k| FORBIDDEN); 55| 135k| constexpr const uint8_t CARRY = 56| 135k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 57| 135k| const simd8 byte_1_low = 58| 135k| (prev1 & 0x0F) 59| 135k| .lookup_16( 60| | // ____0000 ________ 61| 135k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 62| | // ____0001 ________ 63| 135k| CARRY | OVERLONG_2, 64| | // ____001_ ________ 65| 135k| CARRY, CARRY, 66| | 67| | // ____0100 ________ 68| 135k| FORBIDDEN, 69| | // ____0101 ________ 70| 135k| FORBIDDEN, 71| | // ____011_ ________ 72| 135k| FORBIDDEN, FORBIDDEN, 73| | 74| | // ____1___ ________ 75| 135k| FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, 76| | // ____1101 ________ 77| 135k| FORBIDDEN, FORBIDDEN, FORBIDDEN); 78| 135k| const simd8 byte_2_high = input.shr<4>().lookup_16( 79| | // ________ 0_______ 80| 135k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 81| 135k| TOO_SHORT, TOO_SHORT, 82| | 83| | // ________ 1000____ 84| 135k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 85| 135k| OVERLONG_4, 86| | // ________ 1001____ 87| 135k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 88| | // ________ 101_____ 89| 135k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 90| 135k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 91| | 92| | // ________ 11______ 93| 135k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 94| 135k| return (byte_1_high & byte_1_low & byte_2_high); 95| 135k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder6errorsEv: 308| 20.3k| simdutf_really_inline bool errors() const { 309| 20.3k| return this->error.any_bits_set_anywhere(); 310| 20.3k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder19convert_with_errorsEPKcmPc: 205| 502| char *latin1_output) { 206| 502| size_t pos = 0; 207| 502| 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| 502| size_t leading_byte = 0; 215| 502| size_t margin = size; 216| 4.38k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (216:12): [True: 4.27k, False: 110] | Branch (216:26): [True: 3.88k, False: 392] ------------------ 217| 3.88k| leading_byte += (int8_t(in[margin - 1]) > -65); 218| 3.88k| } 219| | // If the input is long enough, then we have that margin-1 is the eight last 220| | // leading byte. 221| 502| const size_t safety_margin = size - margin + 1; // to avoid overruns! 222| 309k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (222:12): [True: 308k, False: 422] ------------------ 223| 308k| simd8x64 input(reinterpret_cast(in + pos)); 224| 308k| if (input.is_ascii()) { ------------------ | Branch (224:11): [True: 289k, False: 19.3k] ------------------ 225| 289k| input.store((int8_t *)latin1_output); 226| 289k| latin1_output += 64; 227| 289k| pos += 64; 228| 289k| } else { 229| | // you might think that a for-loop would work, but under Visual Studio, 230| | // it is not good enough. 231| 19.3k| static_assert( 232| 19.3k| (simd8x64::NUM_CHUNKS == 2) || 233| 19.3k| (simd8x64::NUM_CHUNKS == 4), 234| 19.3k| "We support either two or four chunks per 64-byte block."); 235| 19.3k| auto zero = simd8{uint8_t(0)}; 236| 19.3k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 237| 19.3k| this->check_utf8_bytes(input.chunks[0], zero); 238| 19.3k| 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| 19.3k| if (errors()) { ------------------ | Branch (245:13): [True: 80, False: 19.2k] ------------------ 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| 80| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 250| 80| pos, in + pos, size - pos, latin1_output); 251| 80| res.count += pos; 252| 80| return res; 253| 80| } 254| 19.2k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 255| 19.2k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 256| 19.2k| 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| 19.2k| size_t max_starting_point = (pos + 64) - 12; 261| | // Next loop is going to run at least five times. 262| 192k| while (pos < max_starting_point) { ------------------ | Branch (262:16): [True: 172k, False: 19.2k] ------------------ 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| 172k| size_t consumed = convert_masked_utf8_to_latin1( 273| 172k| in + pos, utf8_end_of_code_point_mask, latin1_output); 274| 172k| pos += consumed; 275| 172k| utf8_end_of_code_point_mask >>= consumed; 276| 172k| } 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| 19.2k| } 282| 308k| } 283| 422| if (errors()) { ------------------ | Branch (283:9): [True: 0, False: 422] ------------------ 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| 422| if (pos < size) { ------------------ | Branch (292:9): [True: 418, 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| 418| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 297| 418| pos, in + pos, size - pos, latin1_output); 298| 418| if (res.error) { // In case of error, we want the error position ------------------ | Branch (298:11): [True: 198, False: 220] ------------------ 299| 198| res.count += pos; 300| 198| return res; 301| 220| } else { // In case of success, we want the number of word written 302| 220| latin1_output += res.count; 303| 220| } 304| 418| } 305| 224| return result(error_code::SUCCESS, latin1_output - start); 306| 422| } 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| 590| char *latin1_output) { 116| 590| size_t pos = 0; 117| 590| 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| 590| size_t leading_byte = 0; 125| 590| size_t margin = size; 126| 9.57k| for (; margin > 0 && leading_byte < 16; margin--) { ------------------ | Branch (126:12): [True: 9.42k, False: 150] | Branch (126:26): [True: 8.98k, False: 440] ------------------ 127| 8.98k| leading_byte += (int8_t(in[margin - 1]) > 128| 8.98k| -65); // twos complement of -65 is 1011 1111 ... 129| 8.98k| } 130| | // If the input is long enough, then we have that margin-1 is the eight last 131| | // leading byte. 132| 590| const size_t safety_margin = size - margin + 1; // to avoid overruns! 133| 613k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (133:12): [True: 613k, False: 562] ------------------ 134| 613k| simd8x64 input(reinterpret_cast(in + pos)); 135| 613k| if (input.is_ascii()) { ------------------ | Branch (135:11): [True: 564k, False: 48.6k] ------------------ 136| 564k| input.store((int8_t *)latin1_output); 137| 564k| latin1_output += 64; 138| 564k| pos += 64; 139| 564k| } else { 140| | // you might think that a for-loop would work, but under Visual Studio, 141| | // it is not good enough. 142| 48.6k| static_assert( 143| 48.6k| (simd8x64::NUM_CHUNKS == 2) || 144| 48.6k| (simd8x64::NUM_CHUNKS == 4), 145| 48.6k| "We support either two or four chunks per 64-byte block."); 146| 48.6k| 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| 48.6k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 151| 48.6k| this->check_utf8_bytes(input.chunks[0], zero); 152| 48.6k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 153| 48.6k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 154| 48.6k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 155| 48.6k| } 156| 48.6k| uint64_t utf8_continuation_mask = 157| 48.6k| input.lt(-65 + 1); // -64 is 1100 0000 in twos complement. Note: in 158| | // this case, we also have ASCII to account for. 159| 48.6k| if (utf8_continuation_mask & 1) { ------------------ | Branch (159:13): [True: 28, False: 48.6k] ------------------ 160| 28| return 0; // error 161| 28| } 162| 48.6k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 163| 48.6k| 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| 48.6k| size_t max_starting_point = (pos + 64) - 12; 168| | // Next loop is going to run at least five times. 169| 481k| while (pos < max_starting_point) { ------------------ | Branch (169:16): [True: 432k, False: 48.6k] ------------------ 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| 432k| size_t consumed = convert_masked_utf8_to_latin1( 180| 432k| in + pos, utf8_end_of_code_point_mask, latin1_output); 181| 432k| pos += consumed; 182| 432k| utf8_end_of_code_point_mask >>= consumed; 183| 432k| } 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| 48.6k| } 189| 613k| } 190| 562| if (errors()) { ------------------ | Branch (190:9): [True: 226, False: 336] ------------------ 191| 226| return 0; 192| 226| } 193| 336| if (pos < size) { ------------------ | Branch (193:9): [True: 332, False: 4] ------------------ 194| 332| size_t howmany = 195| 332| scalar::utf8_to_latin1::convert(in + pos, size - pos, latin1_output); 196| 332| if (howmany == 0) { ------------------ | Branch (196:11): [True: 154, False: 178] ------------------ 197| 154| return 0; 198| 154| } 199| 178| latin1_output += howmany; 200| 178| } 201| 182| return latin1_output - start; 202| 336| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 106| 271k| 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| 271k| simd8 prev1 = input.prev<1>(prev_input); 111| 271k| this->error |= check_special_cases(input, prev1); 112| 271k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin119check_special_casesENS1_4simd5simd8IhEES5_: 8| 271k|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| 271k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 20| | // 11______ 11______ 21| 271k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 22| 271k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 23| 271k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 24| 271k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 25| 271k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 26| 271k| 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| 271k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 35| | // 11110101 1000____ 36| | // 1111011_ 1000____ 37| | // 11111___ 1000____ 38| 271k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 39| 271k| constexpr const uint8_t FORBIDDEN = 0xff; 40| | 41| 271k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 42| | // 0_______ ________ 43| 271k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 44| 271k| TOO_LONG, 45| | // 10______ ________ 46| 271k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 47| | // 1100____ ________ 48| 271k| TOO_SHORT | OVERLONG_2, 49| | // 1101____ ________ 50| 271k| FORBIDDEN, 51| | // 1110____ ________ 52| 271k| FORBIDDEN, 53| | // 1111____ ________ 54| 271k| FORBIDDEN); 55| 271k| constexpr const uint8_t CARRY = 56| 271k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 57| 271k| const simd8 byte_1_low = 58| 271k| (prev1 & 0x0F) 59| 271k| .lookup_16( 60| | // ____0000 ________ 61| 271k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 62| | // ____0001 ________ 63| 271k| CARRY | OVERLONG_2, 64| | // ____001_ ________ 65| 271k| CARRY, CARRY, 66| | 67| | // ____0100 ________ 68| 271k| FORBIDDEN, 69| | // ____0101 ________ 70| 271k| FORBIDDEN, 71| | // ____011_ ________ 72| 271k| FORBIDDEN, FORBIDDEN, 73| | 74| | // ____1___ ________ 75| 271k| FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, 76| | // ____1101 ________ 77| 271k| FORBIDDEN, FORBIDDEN, FORBIDDEN); 78| 271k| const simd8 byte_2_high = input.shr<4>().lookup_16( 79| | // ________ 0_______ 80| 271k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 81| 271k| TOO_SHORT, TOO_SHORT, 82| | 83| | // ________ 1000____ 84| 271k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 85| 271k| OVERLONG_4, 86| | // ________ 1001____ 87| 271k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 88| | // ________ 101_____ 89| 271k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 90| 271k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 91| | 92| | // ________ 11______ 93| 271k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 94| 271k| return (byte_1_high & byte_1_low & byte_2_high); 95| 271k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder6errorsEv: 308| 20.3k| simdutf_really_inline bool errors() const { 309| 20.3k| return this->error.any_bits_set_anywhere(); 310| 20.3k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder19convert_with_errorsEPKcmPc: 205| 502| char *latin1_output) { 206| 502| size_t pos = 0; 207| 502| 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| 502| size_t leading_byte = 0; 215| 502| size_t margin = size; 216| 4.38k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (216:12): [True: 4.27k, False: 110] | Branch (216:26): [True: 3.88k, False: 392] ------------------ 217| 3.88k| leading_byte += (int8_t(in[margin - 1]) > -65); 218| 3.88k| } 219| | // If the input is long enough, then we have that margin-1 is the eight last 220| | // leading byte. 221| 502| const size_t safety_margin = size - margin + 1; // to avoid overruns! 222| 309k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (222:12): [True: 308k, False: 422] ------------------ 223| 308k| simd8x64 input(reinterpret_cast(in + pos)); 224| 308k| if (input.is_ascii()) { ------------------ | Branch (224:11): [True: 289k, False: 19.3k] ------------------ 225| 289k| input.store((int8_t *)latin1_output); 226| 289k| latin1_output += 64; 227| 289k| pos += 64; 228| 289k| } else { 229| | // you might think that a for-loop would work, but under Visual Studio, 230| | // it is not good enough. 231| 19.3k| static_assert( 232| 19.3k| (simd8x64::NUM_CHUNKS == 2) || 233| 19.3k| (simd8x64::NUM_CHUNKS == 4), 234| 19.3k| "We support either two or four chunks per 64-byte block."); 235| 19.3k| 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| 19.3k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 240| 19.3k| this->check_utf8_bytes(input.chunks[0], zero); 241| 19.3k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 242| 19.3k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 243| 19.3k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 244| 19.3k| } 245| 19.3k| if (errors()) { ------------------ | Branch (245:13): [True: 80, False: 19.2k] ------------------ 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| 80| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 250| 80| pos, in + pos, size - pos, latin1_output); 251| 80| res.count += pos; 252| 80| return res; 253| 80| } 254| 19.2k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 255| 19.2k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 256| 19.2k| 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| 19.2k| size_t max_starting_point = (pos + 64) - 12; 261| | // Next loop is going to run at least five times. 262| 192k| while (pos < max_starting_point) { ------------------ | Branch (262:16): [True: 172k, False: 19.2k] ------------------ 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| 172k| size_t consumed = convert_masked_utf8_to_latin1( 273| 172k| in + pos, utf8_end_of_code_point_mask, latin1_output); 274| 172k| pos += consumed; 275| 172k| utf8_end_of_code_point_mask >>= consumed; 276| 172k| } 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| 19.2k| } 282| 308k| } 283| 422| if (errors()) { ------------------ | Branch (283:9): [True: 0, False: 422] ------------------ 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| 422| if (pos < size) { ------------------ | Branch (292:9): [True: 418, 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| 418| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 297| 418| pos, in + pos, size - pos, latin1_output); 298| 418| if (res.error) { // In case of error, we want the error position ------------------ | Branch (298:11): [True: 198, False: 220] ------------------ 299| 198| res.count += pos; 300| 198| return res; 301| 220| } else { // In case of success, we want the number of word written 302| 220| latin1_output += res.count; 303| 220| } 304| 418| } 305| 224| return result(error_code::SUCCESS, latin1_output - start); 306| 422| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoderC2Ev: 114| 2.84k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoderC2Ev: 114| 2.84k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE0EEEmPKcmPDs: 130| 720| char16_t *utf16_output) { 131| 720| size_t pos = 0; 132| 720| 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| 720| size_t leading_byte = 0; 140| 720| size_t margin = size; 141| 6.58k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 6.35k, False: 232] | Branch (141:26): [True: 5.86k, False: 488] ------------------ 142| 5.86k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 5.86k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 720| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 488k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 487k, False: 704] ------------------ 148| 487k| simd8x64 input(reinterpret_cast(in + pos)); 149| 487k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 412k, False: 74.8k] ------------------ 150| 412k| input.store_ascii_as_utf16(utf16_output); 151| 412k| utf16_output += 64; 152| 412k| pos += 64; 153| 412k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 74.8k| static_assert( 157| 74.8k| (simd8x64::NUM_CHUNKS == 2) || 158| 74.8k| (simd8x64::NUM_CHUNKS == 4), 159| 74.8k| "We support either two or four chunks per 64-byte block."); 160| 74.8k| auto zero = simd8{uint8_t(0)}; 161| 74.8k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 162| 74.8k| this->check_utf8_bytes(input.chunks[0], zero); 163| 74.8k| 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| 74.8k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 74.8k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 16, False: 74.8k] ------------------ 172| 16| return 0; // error 173| 16| } 174| 74.8k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 74.8k| 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| 74.8k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 730k| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 655k, False: 74.8k] ------------------ 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| 655k| size_t consumed = convert_masked_utf8_to_utf16( 192| 655k| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 655k| pos += consumed; 194| 655k| utf8_end_of_code_point_mask >>= consumed; 195| 655k| } 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| 74.8k| } 201| 487k| } 202| 704| if (errors()) { ------------------ | Branch (202:9): [True: 282, False: 422] ------------------ 203| 282| return 0; 204| 282| } 205| 422| if (pos < size) { ------------------ | Branch (205:9): [True: 420, False: 2] ------------------ 206| 420| size_t howmany = scalar::utf8_to_utf16::convert( 207| 420| in + pos, size - pos, utf16_output); 208| 420| if (howmany == 0) { ------------------ | Branch (208:11): [True: 190, False: 230] ------------------ 209| 190| return 0; 210| 190| } 211| 230| utf16_output += howmany; 212| 230| } 213| 232| return utf16_output - start; 214| 422| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 920k| 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| 920k| simd8 prev1 = input.prev<1>(prev_input); 124| 920k| simd8 sc = check_special_cases(input, prev1); 125| 920k| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 920k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1619check_special_casesENS1_4simd5simd8IhEES5_: 8| 920k|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| 920k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 920k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 920k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 920k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 920k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 920k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 920k| 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| 920k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 920k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 920k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 920k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 920k| TOO_LONG, 40| | // 10______ ________ 41| 920k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 920k| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 920k| TOO_SHORT, 46| | // 1110____ ________ 47| 920k| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 920k| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 920k| constexpr const uint8_t CARRY = 51| 920k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 920k| const simd8 byte_1_low = 53| 920k| (prev1 & 0x0F) 54| 920k| .lookup_16( 55| | // ____0000 ________ 56| 920k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 920k| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 920k| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 920k| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 920k| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 920k| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 920k| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 920k| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 920k| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 920k| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 920k| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 920k| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 920k| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 920k| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 920k| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 920k| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 920k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 920k| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 920k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 920k| OVERLONG_4, 88| | // ________ 1001____ 89| 920k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 920k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 920k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 920k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 920k| return (byte_1_high & byte_1_low & byte_2_high); 97| 920k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1623check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 920k| const simd8 sc) { 102| 920k| simd8 prev2 = input.prev<2>(prev_input); 103| 920k| simd8 prev3 = input.prev<3>(prev_input); 104| 920k| simd8 must23 = 105| 920k| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 920k| simd8 must23_80 = must23 & uint8_t(0x80); 107| 920k| return must23_80 ^ sc; 108| 920k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder6errorsEv: 324| 190k| simdutf_really_inline bool errors() const { 325| 190k| return this->error.any_bits_set_anywhere(); 326| 190k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE1EEEmPKcmPDs: 130| 756| char16_t *utf16_output) { 131| 756| size_t pos = 0; 132| 756| 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| 756| size_t leading_byte = 0; 140| 756| 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: 6.45k, False: 504] ------------------ 142| 6.45k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 6.45k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 756| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 544k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 544k, False: 738] ------------------ 148| 544k| simd8x64 input(reinterpret_cast(in + pos)); 149| 544k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 346k, False: 197k] ------------------ 150| 346k| input.store_ascii_as_utf16(utf16_output); 151| 346k| utf16_output += 64; 152| 346k| pos += 64; 153| 346k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 197k| static_assert( 157| 197k| (simd8x64::NUM_CHUNKS == 2) || 158| 197k| (simd8x64::NUM_CHUNKS == 4), 159| 197k| "We support either two or four chunks per 64-byte block."); 160| 197k| auto zero = simd8{uint8_t(0)}; 161| 197k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 162| 197k| this->check_utf8_bytes(input.chunks[0], zero); 163| 197k| 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| 197k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 197k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 18, False: 197k] ------------------ 172| 18| return 0; // error 173| 18| } 174| 197k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 197k| 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| 197k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 1.95M| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 1.76M, False: 197k] ------------------ 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| 1.76M| size_t consumed = convert_masked_utf8_to_utf16( 192| 1.76M| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 1.76M| pos += consumed; 194| 1.76M| utf8_end_of_code_point_mask >>= consumed; 195| 1.76M| } 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| 197k| } 201| 544k| } 202| 738| if (errors()) { ------------------ | Branch (202:9): [True: 340, False: 398] ------------------ 203| 340| return 0; 204| 340| } 205| 398| if (pos < size) { ------------------ | Branch (205:9): [True: 396, False: 2] ------------------ 206| 396| size_t howmany = scalar::utf8_to_utf16::convert( 207| 396| in + pos, size - pos, utf16_output); 208| 396| if (howmany == 0) { ------------------ | Branch (208:11): [True: 200, False: 196] ------------------ 209| 200| return 0; 210| 200| } 211| 196| utf16_output += howmany; 212| 196| } 213| 198| return utf16_output - start; 214| 398| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKcmPDs: 218| 612| char16_t *utf16_output) { 219| 612| size_t pos = 0; 220| 612| 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| 612| size_t leading_byte = 0; 228| 612| size_t margin = size; 229| 5.26k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 5.02k, False: 244] | Branch (229:26): [True: 4.65k, False: 368] ------------------ 230| 4.65k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 4.65k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 612| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 389k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 388k, False: 518] ------------------ 236| 388k| simd8x64 input(reinterpret_cast(in + pos)); 237| 388k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 294k, False: 93.9k] ------------------ 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| 93.9k| static_assert( 245| 93.9k| (simd8x64::NUM_CHUNKS == 2) || 246| 93.9k| (simd8x64::NUM_CHUNKS == 4), 247| 93.9k| "We support either two or four chunks per 64-byte block."); 248| 93.9k| auto zero = simd8{uint8_t(0)}; 249| 93.9k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 250| 93.9k| this->check_utf8_bytes(input.chunks[0], zero); 251| 93.9k| 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| 93.9k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 93.9k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 94, False: 93.8k] | Branch (259:25): [True: 0, False: 93.8k] ------------------ 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| 94| result res = 264| 94| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 94| pos, in + pos, size - pos, utf16_output); 266| 94| res.count += pos; 267| 94| return res; 268| 94| } 269| 93.8k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 93.8k| 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| 93.8k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 926k| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 832k, False: 93.8k] ------------------ 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| 832k| size_t consumed = convert_masked_utf8_to_utf16( 287| 832k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 832k| pos += consumed; 289| 832k| utf8_end_of_code_point_mask >>= consumed; 290| 832k| } 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| 93.8k| } 296| 388k| } 297| 518| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 518] ------------------ 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| 518| if (pos < size) { ------------------ | Branch (307:9): [True: 514, 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| 514| result res = 312| 514| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 514| pos, in + pos, size - pos, utf16_output); 314| 514| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 260, False: 254] ------------------ 315| 260| res.count += pos; 316| 260| return res; 317| 260| } else { // In case of success, we want the number of word written 318| 254| utf16_output += res.count; 319| 254| } 320| 514| } 321| 258| return result(error_code::SUCCESS, utf16_output - start); 322| 518| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKcmPDs: 218| 754| char16_t *utf16_output) { 219| 754| size_t pos = 0; 220| 754| 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| 754| size_t leading_byte = 0; 228| 754| size_t margin = size; 229| 6.69k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 6.44k, False: 248] | Branch (229:26): [True: 5.94k, False: 506] ------------------ 230| 5.94k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 5.94k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 754| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 429k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 429k, False: 662] ------------------ 236| 429k| simd8x64 input(reinterpret_cast(in + pos)); 237| 429k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 335k, False: 94.0k] ------------------ 238| 335k| input.store_ascii_as_utf16(utf16_output); 239| 335k| utf16_output += 64; 240| 335k| pos += 64; 241| 335k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 94.0k| static_assert( 245| 94.0k| (simd8x64::NUM_CHUNKS == 2) || 246| 94.0k| (simd8x64::NUM_CHUNKS == 4), 247| 94.0k| "We support either two or four chunks per 64-byte block."); 248| 94.0k| auto zero = simd8{uint8_t(0)}; 249| 94.0k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 250| 94.0k| this->check_utf8_bytes(input.chunks[0], zero); 251| 94.0k| 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| 94.0k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 94.0k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 92, False: 93.9k] | Branch (259:25): [True: 0, False: 93.9k] ------------------ 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| 92| result res = 264| 92| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 92| pos, in + pos, size - pos, utf16_output); 266| 92| res.count += pos; 267| 92| return res; 268| 92| } 269| 93.9k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 93.9k| 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| 93.9k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 912k| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 818k, False: 93.9k] ------------------ 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| 818k| size_t consumed = convert_masked_utf8_to_utf16( 287| 818k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 818k| pos += consumed; 289| 818k| utf8_end_of_code_point_mask >>= consumed; 290| 818k| } 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| 93.9k| } 296| 429k| } 297| 662| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 662] ------------------ 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| 662| if (pos < size) { ------------------ | Branch (307:9): [True: 658, 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| 658| result res = 312| 658| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 658| pos, in + pos, size - pos, utf16_output); 314| 658| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 336, False: 322] ------------------ 315| 336| res.count += pos; 316| 336| return res; 317| 336| } else { // In case of success, we want the number of word written 318| 322| utf16_output += res.count; 319| 322| } 320| 658| } 321| 326| return result(error_code::SUCCESS, utf16_output - start); 322| 662| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE0EEEmPKcmPDs: 130| 720| char16_t *utf16_output) { 131| 720| size_t pos = 0; 132| 720| 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| 720| size_t leading_byte = 0; 140| 720| size_t margin = size; 141| 6.58k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 6.35k, False: 232] | Branch (141:26): [True: 5.86k, False: 488] ------------------ 142| 5.86k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 5.86k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 720| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 488k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 487k, False: 704] ------------------ 148| 487k| simd8x64 input(reinterpret_cast(in + pos)); 149| 487k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 412k, False: 74.8k] ------------------ 150| 412k| input.store_ascii_as_utf16(utf16_output); 151| 412k| utf16_output += 64; 152| 412k| pos += 64; 153| 412k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 74.8k| static_assert( 157| 74.8k| (simd8x64::NUM_CHUNKS == 2) || 158| 74.8k| (simd8x64::NUM_CHUNKS == 4), 159| 74.8k| "We support either two or four chunks per 64-byte block."); 160| 74.8k| 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| 74.8k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 165| 74.8k| this->check_utf8_bytes(input.chunks[0], zero); 166| 74.8k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 167| 74.8k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 168| 74.8k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 169| 74.8k| } 170| 74.8k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 74.8k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 16, False: 74.8k] ------------------ 172| 16| return 0; // error 173| 16| } 174| 74.8k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 74.8k| 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| 74.8k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 730k| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 655k, False: 74.8k] ------------------ 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| 655k| size_t consumed = convert_masked_utf8_to_utf16( 192| 655k| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 655k| pos += consumed; 194| 655k| utf8_end_of_code_point_mask >>= consumed; 195| 655k| } 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| 74.8k| } 201| 487k| } 202| 704| if (errors()) { ------------------ | Branch (202:9): [True: 282, False: 422] ------------------ 203| 282| return 0; 204| 282| } 205| 422| if (pos < size) { ------------------ | Branch (205:9): [True: 420, False: 2] ------------------ 206| 420| size_t howmany = scalar::utf8_to_utf16::convert( 207| 420| in + pos, size - pos, utf16_output); 208| 420| if (howmany == 0) { ------------------ | Branch (208:11): [True: 190, False: 230] ------------------ 209| 190| return 0; 210| 190| } 211| 230| utf16_output += howmany; 212| 230| } 213| 232| return utf16_output - start; 214| 422| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 1.84M| 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| 1.84M| simd8 prev1 = input.prev<1>(prev_input); 124| 1.84M| simd8 sc = check_special_cases(input, prev1); 125| 1.84M| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 1.84M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1619check_special_casesENS1_4simd5simd8IhEES5_: 8| 1.84M|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| 1.84M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 1.84M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 1.84M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 1.84M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 1.84M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 1.84M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 1.84M| 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| 1.84M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 1.84M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 1.84M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 1.84M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 1.84M| TOO_LONG, 40| | // 10______ ________ 41| 1.84M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 1.84M| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 1.84M| TOO_SHORT, 46| | // 1110____ ________ 47| 1.84M| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 1.84M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 1.84M| constexpr const uint8_t CARRY = 51| 1.84M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 1.84M| const simd8 byte_1_low = 53| 1.84M| (prev1 & 0x0F) 54| 1.84M| .lookup_16( 55| | // ____0000 ________ 56| 1.84M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 1.84M| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 1.84M| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 1.84M| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 1.84M| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 1.84M| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 1.84M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 1.84M| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 1.84M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 1.84M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 1.84M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 1.84M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 1.84M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 1.84M| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 1.84M| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 1.84M| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 1.84M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 1.84M| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 1.84M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 1.84M| OVERLONG_4, 88| | // ________ 1001____ 89| 1.84M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 1.84M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 1.84M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 1.84M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 1.84M| return (byte_1_high & byte_1_low & byte_2_high); 97| 1.84M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1623check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 1.84M| const simd8 sc) { 102| 1.84M| simd8 prev2 = input.prev<2>(prev_input); 103| 1.84M| simd8 prev3 = input.prev<3>(prev_input); 104| 1.84M| simd8 must23 = 105| 1.84M| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 1.84M| simd8 must23_80 = must23 & uint8_t(0x80); 107| 1.84M| return must23_80 ^ sc; 108| 1.84M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder6errorsEv: 324| 190k| simdutf_really_inline bool errors() const { 325| 190k| return this->error.any_bits_set_anywhere(); 326| 190k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE1EEEmPKcmPDs: 130| 756| char16_t *utf16_output) { 131| 756| size_t pos = 0; 132| 756| 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| 756| size_t leading_byte = 0; 140| 756| 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: 6.45k, False: 504] ------------------ 142| 6.45k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 6.45k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 756| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 544k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 544k, False: 738] ------------------ 148| 544k| simd8x64 input(reinterpret_cast(in + pos)); 149| 544k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 346k, False: 197k] ------------------ 150| 346k| input.store_ascii_as_utf16(utf16_output); 151| 346k| utf16_output += 64; 152| 346k| pos += 64; 153| 346k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 197k| static_assert( 157| 197k| (simd8x64::NUM_CHUNKS == 2) || 158| 197k| (simd8x64::NUM_CHUNKS == 4), 159| 197k| "We support either two or four chunks per 64-byte block."); 160| 197k| 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| 197k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 165| 197k| this->check_utf8_bytes(input.chunks[0], zero); 166| 197k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 167| 197k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 168| 197k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 169| 197k| } 170| 197k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 197k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 18, False: 197k] ------------------ 172| 18| return 0; // error 173| 18| } 174| 197k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 197k| 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| 197k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 1.95M| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 1.76M, False: 197k] ------------------ 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| 1.76M| size_t consumed = convert_masked_utf8_to_utf16( 192| 1.76M| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 1.76M| pos += consumed; 194| 1.76M| utf8_end_of_code_point_mask >>= consumed; 195| 1.76M| } 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| 197k| } 201| 544k| } 202| 738| if (errors()) { ------------------ | Branch (202:9): [True: 340, False: 398] ------------------ 203| 340| return 0; 204| 340| } 205| 398| if (pos < size) { ------------------ | Branch (205:9): [True: 396, False: 2] ------------------ 206| 396| size_t howmany = scalar::utf8_to_utf16::convert( 207| 396| in + pos, size - pos, utf16_output); 208| 396| if (howmany == 0) { ------------------ | Branch (208:11): [True: 200, False: 196] ------------------ 209| 200| return 0; 210| 200| } 211| 196| utf16_output += howmany; 212| 196| } 213| 198| return utf16_output - start; 214| 398| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKcmPDs: 218| 612| char16_t *utf16_output) { 219| 612| size_t pos = 0; 220| 612| 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| 612| size_t leading_byte = 0; 228| 612| size_t margin = size; 229| 5.26k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 5.02k, False: 244] | Branch (229:26): [True: 4.65k, False: 368] ------------------ 230| 4.65k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 4.65k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 612| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 389k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 388k, False: 518] ------------------ 236| 388k| simd8x64 input(reinterpret_cast(in + pos)); 237| 388k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 294k, False: 93.9k] ------------------ 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| 93.9k| static_assert( 245| 93.9k| (simd8x64::NUM_CHUNKS == 2) || 246| 93.9k| (simd8x64::NUM_CHUNKS == 4), 247| 93.9k| "We support either two or four chunks per 64-byte block."); 248| 93.9k| 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| 93.9k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 253| 93.9k| this->check_utf8_bytes(input.chunks[0], zero); 254| 93.9k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 255| 93.9k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 256| 93.9k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 257| 93.9k| } 258| 93.9k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 93.9k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 94, False: 93.8k] | Branch (259:25): [True: 0, False: 93.8k] ------------------ 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| 94| result res = 264| 94| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 94| pos, in + pos, size - pos, utf16_output); 266| 94| res.count += pos; 267| 94| return res; 268| 94| } 269| 93.8k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 93.8k| 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| 93.8k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 926k| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 832k, False: 93.8k] ------------------ 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| 832k| size_t consumed = convert_masked_utf8_to_utf16( 287| 832k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 832k| pos += consumed; 289| 832k| utf8_end_of_code_point_mask >>= consumed; 290| 832k| } 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| 93.8k| } 296| 388k| } 297| 518| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 518] ------------------ 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| 518| if (pos < size) { ------------------ | Branch (307:9): [True: 514, 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| 514| result res = 312| 514| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 514| pos, in + pos, size - pos, utf16_output); 314| 514| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 260, False: 254] ------------------ 315| 260| res.count += pos; 316| 260| return res; 317| 260| } else { // In case of success, we want the number of word written 318| 254| utf16_output += res.count; 319| 254| } 320| 514| } 321| 258| return result(error_code::SUCCESS, utf16_output - start); 322| 518| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKcmPDs: 218| 754| char16_t *utf16_output) { 219| 754| size_t pos = 0; 220| 754| 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| 754| size_t leading_byte = 0; 228| 754| size_t margin = size; 229| 6.69k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 6.44k, False: 248] | Branch (229:26): [True: 5.94k, False: 506] ------------------ 230| 5.94k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 5.94k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 754| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 429k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 429k, False: 662] ------------------ 236| 429k| simd8x64 input(reinterpret_cast(in + pos)); 237| 429k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 335k, False: 94.0k] ------------------ 238| 335k| input.store_ascii_as_utf16(utf16_output); 239| 335k| utf16_output += 64; 240| 335k| pos += 64; 241| 335k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 94.0k| static_assert( 245| 94.0k| (simd8x64::NUM_CHUNKS == 2) || 246| 94.0k| (simd8x64::NUM_CHUNKS == 4), 247| 94.0k| "We support either two or four chunks per 64-byte block."); 248| 94.0k| 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| 94.0k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 253| 94.0k| this->check_utf8_bytes(input.chunks[0], zero); 254| 94.0k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 255| 94.0k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 256| 94.0k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 257| 94.0k| } 258| 94.0k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 94.0k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 92, False: 93.9k] | Branch (259:25): [True: 0, False: 93.9k] ------------------ 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| 92| result res = 264| 92| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 92| pos, in + pos, size - pos, utf16_output); 266| 92| res.count += pos; 267| 92| return res; 268| 92| } 269| 93.9k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 93.9k| 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| 93.9k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 912k| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 818k, False: 93.9k] ------------------ 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| 818k| size_t consumed = convert_masked_utf8_to_utf16( 287| 818k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 818k| pos += consumed; 289| 818k| utf8_end_of_code_point_mask >>= consumed; 290| 818k| } 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| 93.9k| } 296| 429k| } 297| 662| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 662] ------------------ 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| 662| if (pos < size) { ------------------ | Branch (307:9): [True: 658, 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| 658| result res = 312| 658| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 658| pos, in + pos, size - pos, utf16_output); 314| 658| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 336, False: 322] ------------------ 315| 336| res.count += pos; 316| 336| return res; 317| 336| } else { // In case of success, we want the number of word written 318| 322| utf16_output += res.count; 319| 322| } 320| 658| } 321| 326| return result(error_code::SUCCESS, utf16_output - start); 322| 662| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE0EEEmPKcmPDs: 10| 444| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 444| size_t pos = 0; 14| 444| char16_t *start{utf16_output}; 15| 444| const size_t safety_margin = 16; // to avoid overruns! 16| 412k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 411k, False: 444] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 411k| simd8x64 in(reinterpret_cast(input + pos)); 20| 411k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 239k, False: 172k] ------------------ 21| 239k| in.store_ascii_as_utf16(utf16_output); 22| 239k| utf16_output += 64; 23| 239k| pos += 64; 24| 239k| } 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| 172k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 172k| 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| 172k| 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| 172k| 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| 1.70M| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 1.53M, False: 172k] ------------------ 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| 1.53M| size_t consumed = convert_masked_utf8_to_utf16( 56| 1.53M| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 1.53M| pos += consumed; 58| 1.53M| utf8_end_of_code_point_mask >>= consumed; 59| 1.53M| } 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| 172k| } 65| 411k| } 66| 444| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 444| input + pos, size - pos, utf16_output); 68| 444| return utf16_output - start; 69| 444|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE1EEEmPKcmPDs: 10| 476| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 476| size_t pos = 0; 14| 476| char16_t *start{utf16_output}; 15| 476| const size_t safety_margin = 16; // to avoid overruns! 16| 476k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 475k, False: 476] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 475k| simd8x64 in(reinterpret_cast(input + pos)); 20| 475k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 421k, False: 54.9k] ------------------ 21| 421k| in.store_ascii_as_utf16(utf16_output); 22| 421k| utf16_output += 64; 23| 421k| pos += 64; 24| 421k| } 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| 54.9k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 54.9k| 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| 54.9k| 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| 54.9k| 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| 519k| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 464k, False: 54.9k] ------------------ 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| 464k| size_t consumed = convert_masked_utf8_to_utf16( 56| 464k| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 464k| pos += consumed; 58| 464k| utf8_end_of_code_point_mask >>= consumed; 59| 464k| } 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| 54.9k| } 65| 475k| } 66| 476| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 476| input + pos, size - pos, utf16_output); 68| 476| return utf16_output - start; 69| 476|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE0EEEmPKcmPDs: 10| 444| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 444| size_t pos = 0; 14| 444| char16_t *start{utf16_output}; 15| 444| const size_t safety_margin = 16; // to avoid overruns! 16| 412k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 411k, False: 444] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 411k| simd8x64 in(reinterpret_cast(input + pos)); 20| 411k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 239k, False: 172k] ------------------ 21| 239k| in.store_ascii_as_utf16(utf16_output); 22| 239k| utf16_output += 64; 23| 239k| pos += 64; 24| 239k| } 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| 172k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 172k| 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| 172k| 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| 172k| 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| 1.70M| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 1.53M, False: 172k] ------------------ 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| 1.53M| size_t consumed = convert_masked_utf8_to_utf16( 56| 1.53M| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 1.53M| pos += consumed; 58| 1.53M| utf8_end_of_code_point_mask >>= consumed; 59| 1.53M| } 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| 172k| } 65| 411k| } 66| 444| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 444| input + pos, size - pos, utf16_output); 68| 444| return utf16_output - start; 69| 444|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE1EEEmPKcmPDs: 10| 476| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 476| size_t pos = 0; 14| 476| char16_t *start{utf16_output}; 15| 476| const size_t safety_margin = 16; // to avoid overruns! 16| 476k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 475k, False: 476] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 475k| simd8x64 in(reinterpret_cast(input + pos)); 20| 475k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 421k, False: 54.9k] ------------------ 21| 421k| in.store_ascii_as_utf16(utf16_output); 22| 421k| utf16_output += 64; 23| 421k| pos += 64; 24| 421k| } 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| 54.9k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 54.9k| 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| 54.9k| 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| 54.9k| 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| 519k| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 464k, False: 54.9k] ------------------ 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| 464k| size_t consumed = convert_masked_utf8_to_utf16( 56| 464k| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 464k| pos += consumed; 58| 464k| utf8_end_of_code_point_mask >>= consumed; 59| 464k| } 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| 54.9k| } 65| 475k| } 66| 476| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 476| input + pos, size - pos, utf16_output); 68| 476| return utf16_output - start; 69| 476|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoderC2Ev: 114| 1.57k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder7convertEPKcmPDi: 129| 784| char32_t *utf32_output) { 130| 784| size_t pos = 0; 131| 784| 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| 784| size_t leading_byte = 0; 139| 784| size_t margin = size; 140| 7.59k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (140:12): [True: 7.34k, False: 250] | Branch (140:26): [True: 6.80k, False: 534] ------------------ 141| 6.80k| leading_byte += (int8_t(in[margin - 1]) > -65); 142| 6.80k| } 143| | // If the input is long enough, then we have that margin-1 is the fourth 144| | // last leading byte. 145| 784| const size_t safety_margin = size - margin + 1; // to avoid overruns! 146| 665k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (146:12): [True: 664k, False: 764] ------------------ 147| 664k| simd8x64 input(reinterpret_cast(in + pos)); 148| 664k| if (input.is_ascii()) { ------------------ | Branch (148:11): [True: 414k, False: 250k] ------------------ 149| 414k| input.store_ascii_as_utf32(utf32_output); 150| 414k| utf32_output += 64; 151| 414k| pos += 64; 152| 414k| } else { 153| | // you might think that a for-loop would work, but under Visual Studio, 154| | // it is not good enough. 155| 250k| static_assert( 156| 250k| (simd8x64::NUM_CHUNKS == 2) || 157| 250k| (simd8x64::NUM_CHUNKS == 4), 158| 250k| "We support either two or four chunks per 64-byte block."); 159| 250k| auto zero = simd8{uint8_t(0)}; 160| 250k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 161| 250k| this->check_utf8_bytes(input.chunks[0], zero); 162| 250k| 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| 250k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 170| 250k| if (utf8_continuation_mask & 1) { ------------------ | Branch (170:13): [True: 20, False: 250k] ------------------ 171| 20| return 0; // we have an error 172| 20| } 173| 250k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 174| 250k| 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| 250k| size_t max_starting_point = (pos + 64) - 12; 179| | // Next loop is going to run at least five times. 180| 2.49M| while (pos < max_starting_point) { ------------------ | Branch (180:16): [True: 2.24M, False: 250k] ------------------ 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| 2.24M| size_t consumed = convert_masked_utf8_to_utf32( 191| 2.24M| in + pos, utf8_end_of_code_point_mask, utf32_output); 192| 2.24M| pos += consumed; 193| 2.24M| utf8_end_of_code_point_mask >>= consumed; 194| 2.24M| } 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| 250k| } 200| 664k| } 201| 764| if (errors()) { ------------------ | Branch (201:9): [True: 286, False: 478] ------------------ 202| 286| return 0; 203| 286| } 204| 478| if (pos < size) { ------------------ | Branch (204:9): [True: 476, False: 2] ------------------ 205| 476| size_t howmany = 206| 476| scalar::utf8_to_utf32::convert(in + pos, size - pos, utf32_output); 207| 476| if (howmany == 0) { ------------------ | Branch (207:11): [True: 222, False: 254] ------------------ 208| 222| return 0; 209| 222| } 210| 254| utf32_output += howmany; 211| 254| } 212| 256| return utf32_output - start; 213| 478| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 687k| 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| 687k| simd8 prev1 = input.prev<1>(prev_input); 124| 687k| simd8 sc = check_special_cases(input, prev1); 125| 687k| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 687k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3219check_special_casesENS1_4simd5simd8IhEES5_: 8| 687k|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| 687k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 687k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 687k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 687k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 687k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 687k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 687k| 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| 687k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 687k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 687k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 687k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 687k| TOO_LONG, 40| | // 10______ ________ 41| 687k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 687k| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 687k| TOO_SHORT, 46| | // 1110____ ________ 47| 687k| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 687k| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 687k| constexpr const uint8_t CARRY = 51| 687k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 687k| const simd8 byte_1_low = 53| 687k| (prev1 & 0x0F) 54| 687k| .lookup_16( 55| | // ____0000 ________ 56| 687k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 687k| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 687k| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 687k| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 687k| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 687k| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 687k| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 687k| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 687k| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 687k| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 687k| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 687k| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 687k| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 687k| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 687k| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 687k| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 687k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 687k| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 687k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 687k| OVERLONG_4, 88| | // ________ 1001____ 89| 687k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 687k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 687k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 687k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 687k| return (byte_1_high & byte_1_low & byte_2_high); 97| 687k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3223check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 687k| const simd8 sc) { 102| 687k| simd8 prev2 = input.prev<2>(prev_input); 103| 687k| simd8 prev3 = input.prev<3>(prev_input); 104| 687k| simd8 must23 = 105| 687k| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 687k| simd8 must23_80 = must23 & uint8_t(0x80); 107| 687k| return must23_80 ^ sc; 108| 687k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder6errorsEv: 310| 95.1k| simdutf_really_inline bool errors() const { 311| 95.1k| return this->error.any_bits_set_anywhere(); 312| 95.1k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder19convert_with_errorsEPKcmPDi: 216| 794| char32_t *utf32_output) { 217| 794| size_t pos = 0; 218| 794| 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| 794| size_t leading_byte = 0; 226| 794| size_t margin = size; 227| 7.00k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (227:12): [True: 6.73k, False: 270] | Branch (227:26): [True: 6.21k, False: 524] ------------------ 228| 6.21k| leading_byte += (int8_t(in[margin - 1]) > -65); 229| 6.21k| } 230| | // If the input is long enough, then we have that margin-1 is the fourth 231| | // last leading byte. 232| 794| const size_t safety_margin = size - margin + 1; // to avoid overruns! 233| 461k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (233:12): [True: 460k, False: 720] ------------------ 234| 460k| simd8x64 input(reinterpret_cast(in + pos)); 235| 460k| if (input.is_ascii()) { ------------------ | Branch (235:11): [True: 367k, False: 93.6k] ------------------ 236| 367k| input.store_ascii_as_utf32(utf32_output); 237| 367k| utf32_output += 64; 238| 367k| pos += 64; 239| 367k| } else { 240| | // you might think that a for-loop would work, but under Visual Studio, 241| | // it is not good enough. 242| 93.6k| static_assert( 243| 93.6k| (simd8x64::NUM_CHUNKS == 2) || 244| 93.6k| (simd8x64::NUM_CHUNKS == 4), 245| 93.6k| "We support either two or four chunks per 64-byte block."); 246| 93.6k| auto zero = simd8{uint8_t(0)}; 247| 93.6k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 248| 93.6k| this->check_utf8_bytes(input.chunks[0], zero); 249| 93.6k| 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| 93.6k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 257| 93.6k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (257:13): [True: 74, False: 93.5k] | Branch (257:25): [True: 0, False: 93.5k] ------------------ 258| 74| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 259| 74| pos, in + pos, size - pos, utf32_output); 260| 74| res.count += pos; 261| 74| return res; 262| 74| } 263| 93.5k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 264| 93.5k| 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| 93.5k| size_t max_starting_point = (pos + 64) - 12; 269| | // Next loop is going to run at least five times. 270| 911k| while (pos < max_starting_point) { ------------------ | Branch (270:16): [True: 817k, False: 93.5k] ------------------ 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| 817k| size_t consumed = convert_masked_utf8_to_utf32( 281| 817k| in + pos, utf8_end_of_code_point_mask, utf32_output); 282| 817k| pos += consumed; 283| 817k| utf8_end_of_code_point_mask >>= consumed; 284| 817k| } 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| 93.5k| } 290| 460k| } 291| 720| if (errors()) { ------------------ | Branch (291:9): [True: 0, False: 720] ------------------ 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| 720| if (pos < size) { ------------------ | Branch (297:9): [True: 716, False: 4] ------------------ 298| 716| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 299| 716| pos, in + pos, size - pos, utf32_output); 300| 716| if (res.error) { // In case of error, we want the error position ------------------ | Branch (300:11): [True: 360, False: 356] ------------------ 301| 360| res.count += pos; 302| 360| return res; 303| 360| } else { // In case of success, we want the number of word written 304| 356| utf32_output += res.count; 305| 356| } 306| 716| } 307| 360| return result(error_code::SUCCESS, utf32_output - start); 308| 720| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoderC2Ev: 114| 1.57k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder7convertEPKcmPDi: 129| 784| char32_t *utf32_output) { 130| 784| size_t pos = 0; 131| 784| 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| 784| size_t leading_byte = 0; 139| 784| size_t margin = size; 140| 7.59k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (140:12): [True: 7.34k, False: 250] | Branch (140:26): [True: 6.80k, False: 534] ------------------ 141| 6.80k| leading_byte += (int8_t(in[margin - 1]) > -65); 142| 6.80k| } 143| | // If the input is long enough, then we have that margin-1 is the fourth 144| | // last leading byte. 145| 784| const size_t safety_margin = size - margin + 1; // to avoid overruns! 146| 665k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (146:12): [True: 664k, False: 764] ------------------ 147| 664k| simd8x64 input(reinterpret_cast(in + pos)); 148| 664k| if (input.is_ascii()) { ------------------ | Branch (148:11): [True: 414k, False: 250k] ------------------ 149| 414k| input.store_ascii_as_utf32(utf32_output); 150| 414k| utf32_output += 64; 151| 414k| pos += 64; 152| 414k| } else { 153| | // you might think that a for-loop would work, but under Visual Studio, 154| | // it is not good enough. 155| 250k| static_assert( 156| 250k| (simd8x64::NUM_CHUNKS == 2) || 157| 250k| (simd8x64::NUM_CHUNKS == 4), 158| 250k| "We support either two or four chunks per 64-byte block."); 159| 250k| 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| 250k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 164| 250k| this->check_utf8_bytes(input.chunks[0], zero); 165| 250k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 166| 250k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 167| 250k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 168| 250k| } 169| 250k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 170| 250k| if (utf8_continuation_mask & 1) { ------------------ | Branch (170:13): [True: 20, False: 250k] ------------------ 171| 20| return 0; // we have an error 172| 20| } 173| 250k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 174| 250k| 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| 250k| size_t max_starting_point = (pos + 64) - 12; 179| | // Next loop is going to run at least five times. 180| 2.49M| while (pos < max_starting_point) { ------------------ | Branch (180:16): [True: 2.24M, False: 250k] ------------------ 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| 2.24M| size_t consumed = convert_masked_utf8_to_utf32( 191| 2.24M| in + pos, utf8_end_of_code_point_mask, utf32_output); 192| 2.24M| pos += consumed; 193| 2.24M| utf8_end_of_code_point_mask >>= consumed; 194| 2.24M| } 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| 250k| } 200| 664k| } 201| 764| if (errors()) { ------------------ | Branch (201:9): [True: 286, False: 478] ------------------ 202| 286| return 0; 203| 286| } 204| 478| if (pos < size) { ------------------ | Branch (204:9): [True: 476, False: 2] ------------------ 205| 476| size_t howmany = 206| 476| scalar::utf8_to_utf32::convert(in + pos, size - pos, utf32_output); 207| 476| if (howmany == 0) { ------------------ | Branch (207:11): [True: 222, False: 254] ------------------ 208| 222| return 0; 209| 222| } 210| 254| utf32_output += howmany; 211| 254| } 212| 256| return utf32_output - start; 213| 478| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 1.37M| 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| 1.37M| simd8 prev1 = input.prev<1>(prev_input); 124| 1.37M| simd8 sc = check_special_cases(input, prev1); 125| 1.37M| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 1.37M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3219check_special_casesENS1_4simd5simd8IhEES5_: 8| 1.37M|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| 1.37M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 1.37M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 1.37M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 1.37M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 1.37M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 1.37M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 1.37M| 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| 1.37M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 1.37M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 1.37M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 1.37M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 1.37M| TOO_LONG, 40| | // 10______ ________ 41| 1.37M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 1.37M| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 1.37M| TOO_SHORT, 46| | // 1110____ ________ 47| 1.37M| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 1.37M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 1.37M| constexpr const uint8_t CARRY = 51| 1.37M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 1.37M| const simd8 byte_1_low = 53| 1.37M| (prev1 & 0x0F) 54| 1.37M| .lookup_16( 55| | // ____0000 ________ 56| 1.37M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 1.37M| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 1.37M| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 1.37M| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 1.37M| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 1.37M| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 1.37M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 1.37M| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 1.37M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 1.37M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 1.37M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 1.37M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 1.37M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 1.37M| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 1.37M| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 1.37M| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 1.37M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 1.37M| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 1.37M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 1.37M| OVERLONG_4, 88| | // ________ 1001____ 89| 1.37M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 1.37M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 1.37M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 1.37M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 1.37M| return (byte_1_high & byte_1_low & byte_2_high); 97| 1.37M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3223check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 1.37M| const simd8 sc) { 102| 1.37M| simd8 prev2 = input.prev<2>(prev_input); 103| 1.37M| simd8 prev3 = input.prev<3>(prev_input); 104| 1.37M| simd8 must23 = 105| 1.37M| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 1.37M| simd8 must23_80 = must23 & uint8_t(0x80); 107| 1.37M| return must23_80 ^ sc; 108| 1.37M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder6errorsEv: 310| 95.1k| simdutf_really_inline bool errors() const { 311| 95.1k| return this->error.any_bits_set_anywhere(); 312| 95.1k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder19convert_with_errorsEPKcmPDi: 216| 794| char32_t *utf32_output) { 217| 794| size_t pos = 0; 218| 794| 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| 794| size_t leading_byte = 0; 226| 794| size_t margin = size; 227| 7.00k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (227:12): [True: 6.73k, False: 270] | Branch (227:26): [True: 6.21k, False: 524] ------------------ 228| 6.21k| leading_byte += (int8_t(in[margin - 1]) > -65); 229| 6.21k| } 230| | // If the input is long enough, then we have that margin-1 is the fourth 231| | // last leading byte. 232| 794| const size_t safety_margin = size - margin + 1; // to avoid overruns! 233| 461k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (233:12): [True: 460k, False: 720] ------------------ 234| 460k| simd8x64 input(reinterpret_cast(in + pos)); 235| 460k| if (input.is_ascii()) { ------------------ | Branch (235:11): [True: 367k, False: 93.6k] ------------------ 236| 367k| input.store_ascii_as_utf32(utf32_output); 237| 367k| utf32_output += 64; 238| 367k| pos += 64; 239| 367k| } else { 240| | // you might think that a for-loop would work, but under Visual Studio, 241| | // it is not good enough. 242| 93.6k| static_assert( 243| 93.6k| (simd8x64::NUM_CHUNKS == 2) || 244| 93.6k| (simd8x64::NUM_CHUNKS == 4), 245| 93.6k| "We support either two or four chunks per 64-byte block."); 246| 93.6k| 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| 93.6k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 251| 93.6k| this->check_utf8_bytes(input.chunks[0], zero); 252| 93.6k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 253| 93.6k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 254| 93.6k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 255| 93.6k| } 256| 93.6k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 257| 93.6k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (257:13): [True: 74, False: 93.5k] | Branch (257:25): [True: 0, False: 93.5k] ------------------ 258| 74| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 259| 74| pos, in + pos, size - pos, utf32_output); 260| 74| res.count += pos; 261| 74| return res; 262| 74| } 263| 93.5k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 264| 93.5k| 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| 93.5k| size_t max_starting_point = (pos + 64) - 12; 269| | // Next loop is going to run at least five times. 270| 911k| while (pos < max_starting_point) { ------------------ | Branch (270:16): [True: 817k, False: 93.5k] ------------------ 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| 817k| size_t consumed = convert_masked_utf8_to_utf32( 281| 817k| in + pos, utf8_end_of_code_point_mask, utf32_output); 282| 817k| pos += consumed; 283| 817k| utf8_end_of_code_point_mask >>= consumed; 284| 817k| } 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| 93.5k| } 290| 460k| } 291| 720| if (errors()) { ------------------ | Branch (291:9): [True: 0, False: 720] ------------------ 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| 720| if (pos < size) { ------------------ | Branch (297:9): [True: 716, False: 4] ------------------ 298| 716| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 299| 716| pos, in + pos, size - pos, utf32_output); 300| 716| if (res.error) { // In case of error, we want the error position ------------------ | Branch (300:11): [True: 360, False: 356] ------------------ 301| 360| res.count += pos; 302| 360| return res; 303| 360| } else { // In case of success, we want the number of word written 304| 356| utf32_output += res.count; 305| 356| } 306| 716| } 307| 360| return result(error_code::SUCCESS, utf32_output - start); 308| 720| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3213convert_validEPKcmPDi: 9| 450| char32_t *utf32_output) noexcept { 10| 450| size_t pos = 0; 11| 450| char32_t *start{utf32_output}; 12| 450| const size_t safety_margin = 16; // to avoid overruns! 13| 420k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (13:10): [True: 419k, False: 450] ------------------ 14| 419k| simd8x64 in(reinterpret_cast(input + pos)); 15| 419k| if (in.is_ascii()) { ------------------ | Branch (15:9): [True: 356k, False: 63.3k] ------------------ 16| 356k| in.store_ascii_as_utf32(utf32_output); 17| 356k| utf32_output += 64; 18| 356k| pos += 64; 19| 356k| } else { 20| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 21| | // byte 22| 63.3k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 23| 63.3k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 24| 63.3k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 25| 63.3k| size_t max_starting_point = (pos + 64) - 12; 26| 611k| while (pos < max_starting_point) { ------------------ | Branch (26:14): [True: 547k, False: 63.3k] ------------------ 27| 547k| size_t consumed = convert_masked_utf8_to_utf32( 28| 547k| input + pos, utf8_end_of_code_point_mask, utf32_output); 29| 547k| pos += consumed; 30| 547k| utf8_end_of_code_point_mask >>= consumed; 31| 547k| } 32| 63.3k| } 33| 419k| } 34| 450| utf32_output += scalar::utf8_to_utf32::convert_valid(input + pos, size - pos, 35| 450| utf32_output); 36| 450| return utf32_output - start; 37| 450|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3213convert_validEPKcmPDi: 9| 450| char32_t *utf32_output) noexcept { 10| 450| size_t pos = 0; 11| 450| char32_t *start{utf32_output}; 12| 450| const size_t safety_margin = 16; // to avoid overruns! 13| 420k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (13:10): [True: 419k, False: 450] ------------------ 14| 419k| simd8x64 in(reinterpret_cast(input + pos)); 15| 419k| if (in.is_ascii()) { ------------------ | Branch (15:9): [True: 356k, False: 63.3k] ------------------ 16| 356k| in.store_ascii_as_utf32(utf32_output); 17| 356k| utf32_output += 64; 18| 356k| pos += 64; 19| 356k| } else { 20| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 21| | // byte 22| 63.3k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 23| 63.3k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 24| 63.3k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 25| 63.3k| size_t max_starting_point = (pos + 64) - 12; 26| 611k| while (pos < max_starting_point) { ------------------ | Branch (26:14): [True: 547k, False: 63.3k] ------------------ 27| 547k| size_t consumed = convert_masked_utf8_to_utf32( 28| 547k| input + pos, utf8_end_of_code_point_mask, utf32_output); 29| 547k| pos += consumed; 30| 547k| utf8_end_of_code_point_mask >>= consumed; 31| 547k| } 32| 63.3k| } 33| 419k| } 34| 450| utf32_output += scalar::utf8_to_utf32::convert_valid(input + pos, size - pos, 35| 450| utf32_output); 36| 450| return utf32_output - start; 37| 450|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker16check_next_inputERKNS1_4simd8simd8x64IhEE: 187| 4.68M| simdutf_really_inline void check_next_input(const simd8x64 &input) { 188| 4.68M| if (simdutf_likely(is_ascii(input))) { ------------------ | | 90| 4.68M| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.77M, False: 905k] | | ------------------ ------------------ 189| 3.77M| this->error |= this->prev_incomplete; 190| 3.77M| } else { 191| | // you might think that a for-loop would work, but under Visual Studio, it 192| | // is not good enough. 193| 905k| static_assert((simd8x64::NUM_CHUNKS == 2) || 194| 905k| (simd8x64::NUM_CHUNKS == 4), 195| 905k| "We support either two or four chunks per 64-byte block."); 196| 905k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 197| 905k| this->check_utf8_bytes(input.chunks[0], this->prev_input_block); 198| 905k| 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| 905k| this->prev_incomplete = 206| 905k| is_incomplete(input.chunks[simd8x64::NUM_CHUNKS - 1]); 207| 905k| this->prev_input_block = input.chunks[simd8x64::NUM_CHUNKS - 1]; 208| 905k| } 209| 4.68M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker16check_utf8_bytesENS1_4simd5simd8IhEES6_: 169| 1.81M| 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| 1.81M| simd8 prev1 = input.prev<1>(prev_input); 174| 1.81M| simd8 sc = check_special_cases(input, prev1); 175| 1.81M| this->error |= check_multibyte_lengths(input, prev_input, sc); 176| 1.81M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation19check_special_casesENS1_4simd5simd8IhEES5_: 9| 1.81M|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| 1.81M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 17| | // 11______ 11______ 18| 1.81M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 19| 1.81M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 20| 1.81M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 21| 1.81M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 22| 1.81M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 23| 1.81M| 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| 1.81M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 32| | // 11110101 1000____ 33| | // 1111011_ 1000____ 34| | // 11111___ 1000____ 35| 1.81M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 36| | 37| 1.81M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 38| | // 0_______ ________ 39| 1.81M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 40| 1.81M| TOO_LONG, 41| | // 10______ ________ 42| 1.81M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 43| | // 1100____ ________ 44| 1.81M| TOO_SHORT | OVERLONG_2, 45| | // 1101____ ________ 46| 1.81M| TOO_SHORT, 47| | // 1110____ ________ 48| 1.81M| TOO_SHORT | OVERLONG_3 | SURROGATE, 49| | // 1111____ ________ 50| 1.81M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 51| 1.81M| constexpr const uint8_t CARRY = 52| 1.81M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 53| 1.81M| const simd8 byte_1_low = 54| 1.81M| (prev1 & 0x0F) 55| 1.81M| .lookup_16( 56| | // ____0000 ________ 57| 1.81M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 58| | // ____0001 ________ 59| 1.81M| CARRY | OVERLONG_2, 60| | // ____001_ ________ 61| 1.81M| CARRY, CARRY, 62| | 63| | // ____0100 ________ 64| 1.81M| CARRY | TOO_LARGE, 65| | // ____0101 ________ 66| 1.81M| CARRY | TOO_LARGE | TOO_LARGE_1000, 67| | // ____011_ ________ 68| 1.81M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| 1.81M| CARRY | TOO_LARGE | TOO_LARGE_1000, 70| | 71| | // ____1___ ________ 72| 1.81M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 1.81M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 1.81M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 1.81M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| 1.81M| CARRY | TOO_LARGE | TOO_LARGE_1000, 77| | // ____1101 ________ 78| 1.81M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 79| 1.81M| CARRY | TOO_LARGE | TOO_LARGE_1000, 80| 1.81M| CARRY | TOO_LARGE | TOO_LARGE_1000); 81| 1.81M| const simd8 byte_2_high = input.shr<4>().lookup_16( 82| | // ________ 0_______ 83| 1.81M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 84| 1.81M| TOO_SHORT, TOO_SHORT, 85| | 86| | // ________ 1000____ 87| 1.81M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 88| 1.81M| OVERLONG_4, 89| | // ________ 1001____ 90| 1.81M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 91| | // ________ 101_____ 92| 1.81M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| 1.81M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 94| | 95| | // ________ 11______ 96| 1.81M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 97| 1.81M| return (byte_1_high & byte_1_low & byte_2_high); 98| 1.81M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation23check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 102| 1.81M| const simd8 sc) { 103| 1.81M| simd8 prev2 = input.prev<2>(prev_input); 104| 1.81M| simd8 prev3 = input.prev<3>(prev_input); 105| 1.81M| simd8 must23 = 106| 1.81M| simd8(must_be_2_3_continuation(prev2, prev3)); 107| 1.81M| simd8 must23_80 = must23 & uint8_t(0x80); 108| 1.81M| return must23_80 ^ sc; 109| 1.81M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation13is_incompleteENS1_4simd5simd8IhEE: 115| 905k|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| 905k| static const uint8_t max_array[32] = {255, 120| 905k| 255, 121| 905k| 255, 122| 905k| 255, 123| 905k| 255, 124| 905k| 255, 125| 905k| 255, 126| 905k| 255, 127| 905k| 255, 128| 905k| 255, 129| 905k| 255, 130| 905k| 255, 131| 905k| 255, 132| 905k| 255, 133| 905k| 255, 134| 905k| 255, 135| 905k| 255, 136| 905k| 255, 137| 905k| 255, 138| 905k| 255, 139| 905k| 255, 140| 905k| 255, 141| 905k| 255, 142| 905k| 255, 143| 905k| 255, 144| 905k| 255, 145| 905k| 255, 146| 905k| 255, 147| 905k| 255, 148| 905k| 0b11110000u - 1, 149| 905k| 0b11100000u - 1, 150| 905k| 0b11000000u - 1}; 151| 905k| const simd8 max_value( 152| 905k| &max_array[sizeof(max_array) - sizeof(simd8)]); 153| 905k| return input.gt_bits(max_value); 154| 905k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker9check_eofEv: 181| 6.09k| 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| 6.09k| this->error |= this->prev_incomplete; 185| 6.09k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker6errorsEv: 212| 1.97M| simdutf_really_inline bool errors() const { 213| 1.97M| return this->error.any_bits_set_anywhere(); 214| 1.97M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker16check_next_inputERKNS1_4simd8simd8x64IhEE: 187| 4.68M| simdutf_really_inline void check_next_input(const simd8x64 &input) { 188| 4.68M| if (simdutf_likely(is_ascii(input))) { ------------------ | | 90| 4.68M| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.77M, False: 905k] | | ------------------ ------------------ 189| 3.77M| this->error |= this->prev_incomplete; 190| 3.77M| } else { 191| | // you might think that a for-loop would work, but under Visual Studio, it 192| | // is not good enough. 193| 905k| static_assert((simd8x64::NUM_CHUNKS == 2) || 194| 905k| (simd8x64::NUM_CHUNKS == 4), 195| 905k| "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| 905k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 200| 905k| this->check_utf8_bytes(input.chunks[0], this->prev_input_block); 201| 905k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 202| 905k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 203| 905k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 204| 905k| } 205| 905k| this->prev_incomplete = 206| 905k| is_incomplete(input.chunks[simd8x64::NUM_CHUNKS - 1]); 207| 905k| this->prev_input_block = input.chunks[simd8x64::NUM_CHUNKS - 1]; 208| 905k| } 209| 4.68M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker16check_utf8_bytesENS1_4simd5simd8IhEES6_: 169| 3.62M| 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| 3.62M| simd8 prev1 = input.prev<1>(prev_input); 174| 3.62M| simd8 sc = check_special_cases(input, prev1); 175| 3.62M| this->error |= check_multibyte_lengths(input, prev_input, sc); 176| 3.62M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation19check_special_casesENS1_4simd5simd8IhEES5_: 9| 3.62M|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| 3.62M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 17| | // 11______ 11______ 18| 3.62M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 19| 3.62M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 20| 3.62M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 21| 3.62M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 22| 3.62M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 23| 3.62M| 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| 3.62M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 32| | // 11110101 1000____ 33| | // 1111011_ 1000____ 34| | // 11111___ 1000____ 35| 3.62M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 36| | 37| 3.62M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 38| | // 0_______ ________ 39| 3.62M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 40| 3.62M| TOO_LONG, 41| | // 10______ ________ 42| 3.62M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 43| | // 1100____ ________ 44| 3.62M| TOO_SHORT | OVERLONG_2, 45| | // 1101____ ________ 46| 3.62M| TOO_SHORT, 47| | // 1110____ ________ 48| 3.62M| TOO_SHORT | OVERLONG_3 | SURROGATE, 49| | // 1111____ ________ 50| 3.62M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 51| 3.62M| constexpr const uint8_t CARRY = 52| 3.62M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 53| 3.62M| const simd8 byte_1_low = 54| 3.62M| (prev1 & 0x0F) 55| 3.62M| .lookup_16( 56| | // ____0000 ________ 57| 3.62M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 58| | // ____0001 ________ 59| 3.62M| CARRY | OVERLONG_2, 60| | // ____001_ ________ 61| 3.62M| CARRY, CARRY, 62| | 63| | // ____0100 ________ 64| 3.62M| CARRY | TOO_LARGE, 65| | // ____0101 ________ 66| 3.62M| CARRY | TOO_LARGE | TOO_LARGE_1000, 67| | // ____011_ ________ 68| 3.62M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| 3.62M| CARRY | TOO_LARGE | TOO_LARGE_1000, 70| | 71| | // ____1___ ________ 72| 3.62M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 3.62M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 3.62M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 3.62M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| 3.62M| CARRY | TOO_LARGE | TOO_LARGE_1000, 77| | // ____1101 ________ 78| 3.62M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 79| 3.62M| CARRY | TOO_LARGE | TOO_LARGE_1000, 80| 3.62M| CARRY | TOO_LARGE | TOO_LARGE_1000); 81| 3.62M| const simd8 byte_2_high = input.shr<4>().lookup_16( 82| | // ________ 0_______ 83| 3.62M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 84| 3.62M| TOO_SHORT, TOO_SHORT, 85| | 86| | // ________ 1000____ 87| 3.62M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 88| 3.62M| OVERLONG_4, 89| | // ________ 1001____ 90| 3.62M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 91| | // ________ 101_____ 92| 3.62M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| 3.62M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 94| | 95| | // ________ 11______ 96| 3.62M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 97| 3.62M| return (byte_1_high & byte_1_low & byte_2_high); 98| 3.62M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation23check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 102| 3.62M| const simd8 sc) { 103| 3.62M| simd8 prev2 = input.prev<2>(prev_input); 104| 3.62M| simd8 prev3 = input.prev<3>(prev_input); 105| 3.62M| simd8 must23 = 106| 3.62M| simd8(must_be_2_3_continuation(prev2, prev3)); 107| 3.62M| simd8 must23_80 = must23 & uint8_t(0x80); 108| 3.62M| return must23_80 ^ sc; 109| 3.62M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation13is_incompleteENS1_4simd5simd8IhEE: 115| 905k|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| 905k| static const uint8_t max_array[32] = {255, 120| 905k| 255, 121| 905k| 255, 122| 905k| 255, 123| 905k| 255, 124| 905k| 255, 125| 905k| 255, 126| 905k| 255, 127| 905k| 255, 128| 905k| 255, 129| 905k| 255, 130| 905k| 255, 131| 905k| 255, 132| 905k| 255, 133| 905k| 255, 134| 905k| 255, 135| 905k| 255, 136| 905k| 255, 137| 905k| 255, 138| 905k| 255, 139| 905k| 255, 140| 905k| 255, 141| 905k| 255, 142| 905k| 255, 143| 905k| 255, 144| 905k| 255, 145| 905k| 255, 146| 905k| 255, 147| 905k| 255, 148| 905k| 0b11110000u - 1, 149| 905k| 0b11100000u - 1, 150| 905k| 0b11000000u - 1}; 151| 905k| const simd8 max_value( 152| 905k| &max_array[sizeof(max_array) - sizeof(simd8)]); 153| 905k| return input.gt_bits(max_value); 154| 905k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker9check_eofEv: 181| 6.09k| 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| 6.09k| this->error |= this->prev_incomplete; 185| 6.09k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker6errorsEv: 212| 1.97M| simdutf_really_inline bool errors() const { 213| 1.97M| return this->error.any_bits_set_anywhere(); 214| 1.97M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation21generic_validate_utf8EPKcm: 27| 3.47k|bool generic_validate_utf8(const char *input, size_t length) { 28| 3.47k| return generic_validate_utf8( 29| 3.47k| reinterpret_cast(input), length); 30| 3.47k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation21generic_validate_utf8INS2_12utf8_checkerEEEbPKhm: 10| 3.47k|bool generic_validate_utf8(const uint8_t *input, size_t length) { 11| 3.47k| checker c{}; 12| 3.47k| buf_block_reader<64> reader(input, length); 13| 2.71M| while (reader.has_full_block()) { ------------------ | Branch (13:10): [True: 2.70M, False: 3.47k] ------------------ 14| 2.70M| simd::simd8x64 in(reader.full_block()); 15| 2.70M| c.check_next_input(in); 16| 2.70M| reader.advance(); 17| 2.70M| } 18| 3.47k| uint8_t block[64]{}; 19| 3.47k| reader.get_remainder(block); 20| 3.47k| simd::simd8x64 in(block); 21| 3.47k| c.check_next_input(in); 22| 3.47k| reader.advance(); 23| 3.47k| c.check_eof(); 24| 3.47k| return !c.errors(); 25| 3.47k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsEPKcm: 76| 3.47k|result generic_validate_utf8_with_errors(const char *input, size_t length) { 77| 3.47k| return generic_validate_utf8_with_errors( 78| 3.47k| reinterpret_cast(input), length); 79| 3.47k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsINS2_12utf8_checkerEEENS_6resultEPKhm: 36| 3.47k|result generic_validate_utf8_with_errors(const uint8_t *input, size_t length) { 37| 3.47k| checker c{}; 38| 3.47k| buf_block_reader<64> reader(input, length); 39| 3.47k| size_t count{0}; 40| 1.97M| while (reader.has_full_block()) { ------------------ | Branch (40:10): [True: 1.96M, False: 2.62k] ------------------ 41| 1.96M| simd::simd8x64 in(reader.full_block()); 42| 1.96M| c.check_next_input(in); 43| 1.96M| if (c.errors()) { ------------------ | Branch (43:9): [True: 849, False: 1.96M] ------------------ 44| 849| if (count != 0) { ------------------ | Branch (44:11): [True: 199, False: 650] ------------------ 45| 199| count--; 46| 199| } // Sometimes the error is only detected in the next chunk 47| 849| result res = scalar::utf8::rewind_and_validate_with_errors( 48| 849| reinterpret_cast(input), 49| 849| reinterpret_cast(input + count), length - count); 50| 849| res.count += count; 51| 849| return res; 52| 849| } 53| 1.96M| reader.advance(); 54| 1.96M| count += 64; 55| 1.96M| } 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: 907, False: 1.71k] ------------------ 63| 907| if (count != 0) { ------------------ | Branch (63:9): [True: 245, False: 662] ------------------ 64| 245| count--; 65| 245| } // Sometimes the error is only detected in the next chunk 66| 907| result res = scalar::utf8::rewind_and_validate_with_errors( 67| 907| reinterpret_cast(input), 68| 907| reinterpret_cast(input) + count, length - count); 69| 907| res.count += count; 70| 907| return res; 71| 1.71k| } else { 72| 1.71k| return result(error_code::SUCCESS, length); 73| 1.71k| } 74| 2.62k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation21generic_validate_utf8EPKcm: 27| 3.47k|bool generic_validate_utf8(const char *input, size_t length) { 28| 3.47k| return generic_validate_utf8( 29| 3.47k| reinterpret_cast(input), length); 30| 3.47k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation21generic_validate_utf8INS2_12utf8_checkerEEEbPKhm: 10| 3.47k|bool generic_validate_utf8(const uint8_t *input, size_t length) { 11| 3.47k| checker c{}; 12| 3.47k| buf_block_reader<64> reader(input, length); 13| 2.71M| while (reader.has_full_block()) { ------------------ | Branch (13:10): [True: 2.70M, False: 3.47k] ------------------ 14| 2.70M| simd::simd8x64 in(reader.full_block()); 15| 2.70M| c.check_next_input(in); 16| 2.70M| reader.advance(); 17| 2.70M| } 18| 3.47k| uint8_t block[64]{}; 19| 3.47k| reader.get_remainder(block); 20| 3.47k| simd::simd8x64 in(block); 21| 3.47k| c.check_next_input(in); 22| 3.47k| reader.advance(); 23| 3.47k| c.check_eof(); 24| 3.47k| return !c.errors(); 25| 3.47k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsEPKcm: 76| 3.47k|result generic_validate_utf8_with_errors(const char *input, size_t length) { 77| 3.47k| return generic_validate_utf8_with_errors( 78| 3.47k| reinterpret_cast(input), length); 79| 3.47k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsINS2_12utf8_checkerEEENS_6resultEPKhm: 36| 3.47k|result generic_validate_utf8_with_errors(const uint8_t *input, size_t length) { 37| 3.47k| checker c{}; 38| 3.47k| buf_block_reader<64> reader(input, length); 39| 3.47k| size_t count{0}; 40| 1.97M| while (reader.has_full_block()) { ------------------ | Branch (40:10): [True: 1.96M, False: 2.62k] ------------------ 41| 1.96M| simd::simd8x64 in(reader.full_block()); 42| 1.96M| c.check_next_input(in); 43| 1.96M| if (c.errors()) { ------------------ | Branch (43:9): [True: 849, False: 1.96M] ------------------ 44| 849| if (count != 0) { ------------------ | Branch (44:11): [True: 199, False: 650] ------------------ 45| 199| count--; 46| 199| } // Sometimes the error is only detected in the next chunk 47| 849| result res = scalar::utf8::rewind_and_validate_with_errors( 48| 849| reinterpret_cast(input), 49| 849| reinterpret_cast(input + count), length - count); 50| 849| res.count += count; 51| 849| return res; 52| 849| } 53| 1.96M| reader.advance(); 54| 1.96M| count += 64; 55| 1.96M| } 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: 907, False: 1.71k] ------------------ 63| 907| if (count != 0) { ------------------ | Branch (63:9): [True: 245, False: 662] ------------------ 64| 245| count--; 65| 245| } // Sometimes the error is only detected in the next chunk 66| 907| result res = scalar::utf8::rewind_and_validate_with_errors( 67| 907| reinterpret_cast(input), 68| 907| reinterpret_cast(input) + count, length - count); 69| 907| res.count += count; 70| 907| return res; 71| 1.71k| } else { 72| 1.71k| return result(error_code::SUCCESS, length); 73| 1.71k| } 74| 2.62k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE0EEEKNS_6resultEPKDsm: 49| 2.78k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.78k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.78k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 17, False: 2.77k] | | ------------------ ------------------ 51| 17| return result(error_code::SUCCESS, 0); 52| 17| } 53| | 54| 2.77k| const char16_t *start = input; 55| 2.77k| const char16_t *end = input + size; 56| | 57| 2.77k| const auto v_d8 = simd8::splat(0xd8); 58| 2.77k| const auto v_f8 = simd8::splat(0xf8); 59| 2.77k| const auto v_fc = simd8::splat(0xfc); 60| 2.77k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.67M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.67M, False: 2.53k] ------------------ 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| 2.67M| auto in0 = simd16(input); 67| 2.67M| auto in1 = 68| 2.67M| 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| 2.67M| 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| 2.67M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.67M| const uint16_t surrogates_bitmask = 77| 2.67M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.67M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.64M, False: 25.9k] ------------------ 79| 2.64M| input += 16; 80| 2.64M| } 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| 25.9k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 25.9k| const auto vH = (in & v_fc) == v_dc; 93| 25.9k| 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| 25.9k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 25.9k| const uint16_t a = static_cast( 100| 25.9k| 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| 25.9k| const uint16_t b = static_cast( 104| 25.9k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 25.9k| const uint16_t c = static_cast( 107| 25.9k| V | a | b); // Combine all the masks into the final one. 108| | 109| 25.9k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 19.0k, False: 6.88k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 19.0k| input += 16; 113| 19.0k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 6.64k, False: 240] ------------------ 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| 6.64k| input += 15; 119| 6.64k| } else { 120| 240| return result(error_code::SURROGATE, input - start); 121| 240| } 122| 25.9k| } 123| 2.67M| } 124| | 125| 2.53k| return result(error_code::SUCCESS, input - start); 126| 2.77k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE1EEEKNS_6resultEPKDsm: 49| 2.60k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.60k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.60k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 2.58k] | | ------------------ ------------------ 51| 16| return result(error_code::SUCCESS, 0); 52| 16| } 53| | 54| 2.58k| const char16_t *start = input; 55| 2.58k| const char16_t *end = input + size; 56| | 57| 2.58k| const auto v_d8 = simd8::splat(0xd8); 58| 2.58k| const auto v_f8 = simd8::splat(0xf8); 59| 2.58k| const auto v_fc = simd8::splat(0xfc); 60| 2.58k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.67M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.66M, False: 2.40k] ------------------ 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| 2.66M| auto in0 = simd16(input); 67| 2.66M| auto in1 = 68| 2.66M| 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| 2.66M| 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| 2.66M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.66M| const uint16_t surrogates_bitmask = 77| 2.66M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.66M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.63M, False: 36.8k] ------------------ 79| 2.63M| input += 16; 80| 2.63M| } 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| 36.8k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 36.8k| const auto vH = (in & v_fc) == v_dc; 93| 36.8k| 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| 36.8k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 36.8k| const uint16_t a = static_cast( 100| 36.8k| 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| 36.8k| const uint16_t b = static_cast( 104| 36.8k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 36.8k| const uint16_t c = static_cast( 107| 36.8k| V | a | b); // Combine all the masks into the final one. 108| | 109| 36.8k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 31.0k, False: 5.85k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 31.0k| input += 16; 113| 31.0k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 5.66k, False: 188] ------------------ 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| 5.66k| input += 15; 119| 5.66k| } else { 120| 188| return result(error_code::SURROGATE, input - start); 121| 188| } 122| 36.8k| } 123| 2.66M| } 124| | 125| 2.40k| return result(error_code::SUCCESS, input - start); 126| 2.58k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE0EEEKNS_6resultEPKDsm: 49| 2.78k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.78k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.78k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 17, False: 2.77k] | | ------------------ ------------------ 51| 17| return result(error_code::SUCCESS, 0); 52| 17| } 53| | 54| 2.77k| const char16_t *start = input; 55| 2.77k| const char16_t *end = input + size; 56| | 57| 2.77k| const auto v_d8 = simd8::splat(0xd8); 58| 2.77k| const auto v_f8 = simd8::splat(0xf8); 59| 2.77k| const auto v_fc = simd8::splat(0xfc); 60| 2.77k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.67M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.67M, 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| 2.67M| auto in0 = simd16(input); 67| 2.67M| auto in1 = 68| 2.67M| 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| 2.67M| 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| 2.67M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.67M| const uint16_t surrogates_bitmask = 77| 2.67M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.67M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.64M, False: 26.4k] ------------------ 79| 2.64M| input += 16; 80| 2.64M| } 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| 26.4k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 26.4k| const auto vH = (in & v_fc) == v_dc; 93| 26.4k| 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| 26.4k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 26.4k| const uint16_t a = static_cast( 100| 26.4k| 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| 26.4k| const uint16_t b = static_cast( 104| 26.4k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 26.4k| const uint16_t c = static_cast( 107| 26.4k| V | a | b); // Combine all the masks into the final one. 108| | 109| 26.4k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 19.3k, False: 7.05k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 19.3k| input += 16; 113| 19.3k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 6.72k, False: 330] ------------------ 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| 6.72k| input += 15; 119| 6.72k| } else { 120| 330| return result(error_code::SURROGATE, input - start); 121| 330| } 122| 26.4k| } 123| 2.67M| } 124| | 125| 2.44k| return result(error_code::SUCCESS, input - start); 126| 2.77k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE1EEEKNS_6resultEPKDsm: 49| 2.60k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.60k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.60k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 2.58k] | | ------------------ ------------------ 51| 16| return result(error_code::SUCCESS, 0); 52| 16| } 53| | 54| 2.58k| const char16_t *start = input; 55| 2.58k| const char16_t *end = input + size; 56| | 57| 2.58k| const auto v_d8 = simd8::splat(0xd8); 58| 2.58k| const auto v_f8 = simd8::splat(0xf8); 59| 2.58k| const auto v_fc = simd8::splat(0xfc); 60| 2.58k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.67M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.67M, False: 2.35k] ------------------ 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| 2.67M| auto in0 = simd16(input); 67| 2.67M| auto in1 = 68| 2.67M| 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| 2.67M| 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| 2.67M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.67M| const uint16_t surrogates_bitmask = 77| 2.67M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.67M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.63M, False: 37.2k] ------------------ 79| 2.63M| input += 16; 80| 2.63M| } 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| 37.2k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 37.2k| const auto vH = (in & v_fc) == v_dc; 93| 37.2k| 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| 37.2k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 37.2k| const uint16_t a = static_cast( 100| 37.2k| 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| 37.2k| const uint16_t b = static_cast( 104| 37.2k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 37.2k| const uint16_t c = static_cast( 107| 37.2k| V | a | b); // Combine all the masks into the final one. 108| | 109| 37.2k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 31.2k, False: 6.00k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 31.2k| input += 16; 113| 31.2k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 5.76k, False: 238] ------------------ 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| 5.76k| input += 15; 119| 5.76k| } else { 120| 238| return result(error_code::SURROGATE, input - start); 121| 238| } 122| 37.2k| } 123| 2.67M| } 124| | 125| 2.35k| return result(error_code::SUCCESS, input - start); 126| 2.58k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf328validateEPKDim: 6| 2.61k|simdutf_really_inline bool validate(const char32_t *input, size_t size) { 7| 2.61k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.61k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 23, False: 2.58k] | | ------------------ ------------------ 8| | // empty input is valid UTF-32. protect the implementation from 9| | // handling nullptr 10| 23| return true; 11| 23| } 12| | 13| 2.58k| const char32_t *end = input + size; 14| | 15| 2.58k| using vector_u32 = simd32; 16| | 17| 2.58k| const auto standardmax = vector_u32::splat(0x10ffff); 18| 2.58k| const auto offset = vector_u32::splat(0xffff2000); 19| 2.58k| const auto standardoffsetmax = vector_u32::splat(0xfffff7ff); 20| 2.58k| auto currentmax = vector_u32::zero(); 21| 2.58k| auto currentoffsetmax = vector_u32::zero(); 22| | 23| 2.58k| constexpr size_t N = vector_u32::ELEMENTS; 24| | 25| 1.15M| while (input + N < end) { ------------------ | Branch (25:10): [True: 1.14M, False: 2.58k] ------------------ 26| 1.14M| auto in = vector_u32(input); 27| 1.14M| if constexpr (!match_system(endianness::BIG)) { 28| 1.14M| in.swap_bytes(); 29| 1.14M| } 30| | 31| 1.14M| currentmax = max(currentmax, in); 32| 1.14M| currentoffsetmax = max(currentoffsetmax, in + offset); 33| 1.14M| input += N; 34| 1.14M| } 35| | 36| 2.58k| const auto too_large = currentmax > standardmax; 37| 2.58k| if (too_large.any()) { ------------------ | Branch (37:7): [True: 938, False: 1.65k] ------------------ 38| 938| return false; 39| 938| } 40| | 41| 1.65k| const auto surrogate = currentoffsetmax > standardoffsetmax; 42| 1.65k| if (surrogate.any()) { ------------------ | Branch (42:7): [True: 24, False: 1.62k] ------------------ 43| 24| return false; 44| 24| } 45| | 46| 1.62k| return scalar::utf32::validate(input, end - input); 47| 1.65k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf3220validate_with_errorsEPKDim: 50| 2.61k| size_t size) { 51| 2.61k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.61k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 23, False: 2.58k] | | ------------------ ------------------ 52| | // empty input is valid UTF-32. protect the implementation from 53| | // handling nullptr 54| 23| return result(error_code::SUCCESS, 0); 55| 23| } 56| | 57| 2.58k| const char32_t *start = input; 58| 2.58k| const char32_t *end = input + size; 59| | 60| 2.58k| using vector_u32 = simd32; 61| | 62| 2.58k| const auto standardmax = vector_u32::splat(0x10ffff + 1); 63| 2.58k| const auto surrogate_mask = vector_u32::splat(0xfffff800); 64| 2.58k| const auto surrogate_byte = vector_u32::splat(0x0000d800); 65| | 66| 2.58k| constexpr size_t N = vector_u32::ELEMENTS; 67| | 68| 219k| while (input + N < end) { ------------------ | Branch (68:10): [True: 217k, False: 1.62k] ------------------ 69| 217k| auto in = vector_u32(input); 70| 217k| if constexpr (!match_system(endianness::BIG)) { 71| 217k| in.swap_bytes(); 72| 217k| } 73| | 74| 217k| const auto too_large = in >= standardmax; 75| 217k| const auto surrogate = (in & surrogate_mask) == surrogate_byte; 76| | 77| 217k| const auto combined = too_large | surrogate; 78| 217k| if (simdutf_unlikely(combined.any())) { ------------------ | | 93| 217k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 962, False: 216k] | | ------------------ ------------------ 79| 962| const size_t consumed = input - start; 80| 962| auto sr = scalar::utf32::validate_with_errors(input, end - input); 81| 962| sr.count += consumed; 82| | 83| 962| return sr; 84| 962| } 85| | 86| 216k| input += N; 87| 216k| } 88| | 89| 1.62k| const size_t consumed = input - start; 90| 1.62k| auto sr = scalar::utf32::validate_with_errors(input, end - input); 91| 1.62k| sr.count += consumed; 92| | 93| 1.62k| return sr; 94| 2.58k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf328validateEPKDim: 6| 2.61k|simdutf_really_inline bool validate(const char32_t *input, size_t size) { 7| 2.61k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.61k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 23, False: 2.58k] | | ------------------ ------------------ 8| | // empty input is valid UTF-32. protect the implementation from 9| | // handling nullptr 10| 23| return true; 11| 23| } 12| | 13| 2.58k| const char32_t *end = input + size; 14| | 15| 2.58k| using vector_u32 = simd32; 16| | 17| 2.58k| const auto standardmax = vector_u32::splat(0x10ffff); 18| 2.58k| const auto offset = vector_u32::splat(0xffff2000); 19| 2.58k| const auto standardoffsetmax = vector_u32::splat(0xfffff7ff); 20| 2.58k| auto currentmax = vector_u32::zero(); 21| 2.58k| auto currentoffsetmax = vector_u32::zero(); 22| | 23| 2.58k| constexpr size_t N = vector_u32::ELEMENTS; 24| | 25| 2.29M| while (input + N < end) { ------------------ | Branch (25:10): [True: 2.29M, False: 2.58k] ------------------ 26| 2.29M| auto in = vector_u32(input); 27| 2.29M| if constexpr (!match_system(endianness::BIG)) { 28| 2.29M| in.swap_bytes(); 29| 2.29M| } 30| | 31| 2.29M| currentmax = max(currentmax, in); 32| 2.29M| currentoffsetmax = max(currentoffsetmax, in + offset); 33| 2.29M| input += N; 34| 2.29M| } 35| | 36| 2.58k| const auto too_large = currentmax > standardmax; 37| 2.58k| if (too_large.any()) { ------------------ | Branch (37:7): [True: 1.10k, False: 1.48k] ------------------ 38| 1.10k| return false; 39| 1.10k| } 40| | 41| 1.48k| const auto surrogate = currentoffsetmax > standardoffsetmax; 42| 1.48k| if (surrogate.any()) { ------------------ | Branch (42:7): [True: 33, False: 1.44k] ------------------ 43| 33| return false; 44| 33| } 45| | 46| 1.44k| return scalar::utf32::validate(input, end - input); 47| 1.48k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf3220validate_with_errorsEPKDim: 50| 2.61k| size_t size) { 51| 2.61k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.61k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 23, False: 2.58k] | | ------------------ ------------------ 52| | // empty input is valid UTF-32. protect the implementation from 53| | // handling nullptr 54| 23| return result(error_code::SUCCESS, 0); 55| 23| } 56| | 57| 2.58k| const char32_t *start = input; 58| 2.58k| const char32_t *end = input + size; 59| | 60| 2.58k| using vector_u32 = simd32; 61| | 62| 2.58k| const auto standardmax = vector_u32::splat(0x10ffff + 1); 63| 2.58k| const auto surrogate_mask = vector_u32::splat(0xfffff800); 64| 2.58k| const auto surrogate_byte = vector_u32::splat(0x0000d800); 65| | 66| 2.58k| constexpr size_t N = vector_u32::ELEMENTS; 67| | 68| 436k| while (input + N < end) { ------------------ | Branch (68:10): [True: 435k, False: 1.44k] ------------------ 69| 435k| auto in = vector_u32(input); 70| 435k| if constexpr (!match_system(endianness::BIG)) { 71| 435k| in.swap_bytes(); 72| 435k| } 73| | 74| 435k| const auto too_large = in >= standardmax; 75| 435k| const auto surrogate = (in & surrogate_mask) == surrogate_byte; 76| | 77| 435k| const auto combined = too_large | surrogate; 78| 435k| if (simdutf_unlikely(combined.any())) { ------------------ | | 93| 435k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 1.13k, False: 433k] | | ------------------ ------------------ 79| 1.13k| const size_t consumed = input - start; 80| 1.13k| auto sr = scalar::utf32::validate_with_errors(input, end - input); 81| 1.13k| sr.count += consumed; 82| | 83| 1.13k| return sr; 84| 1.13k| } 85| | 86| 433k| input += N; 87| 433k| } 88| | 89| 1.44k| const size_t consumed = input - start; 90| 1.44k| auto sr = scalar::utf32::validate_with_errors(input, end - input); 91| 1.44k| sr.count += consumed; 92| | 93| 1.44k| return sr; 94| 2.58k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_latin1_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 86| char16_t *utf16_output) { 5| 86| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| | 7| 86| size_t i = 0; 8| 106k| for (; i < rounded_len; i += 16) { ------------------ | Branch (8:10): [True: 106k, False: 86] ------------------ 9| | // Load 16 bytes from the address (input + i) into a xmm register 10| 106k| const __m128i latin1 = 11| 106k| _mm_loadu_si128(reinterpret_cast(latin1_input + i)); 12| | 13| | // Zero extend each byte in `in` to word 14| 106k| __m256i utf16 = _mm256_cvtepu8_epi16(latin1); 15| | 16| 106k| if (big_endian) { ------------------ | Branch (16:9): [Folded, False: 106k] ------------------ 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| 106k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output + i), utf16); 25| 106k| } 26| | 27| 86| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 28| 86|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_latin1_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 88| char16_t *utf16_output) { 5| 88| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| | 7| 88| size_t i = 0; 8| 121k| for (; i < rounded_len; i += 16) { ------------------ | Branch (8:10): [True: 121k, False: 88] ------------------ 9| | // Load 16 bytes from the address (input + i) into a xmm register 10| 121k| const __m128i latin1 = 11| 121k| _mm_loadu_si128(reinterpret_cast(latin1_input + i)); 12| | 13| | // Zero extend each byte in `in` to word 14| 121k| __m256i utf16 = _mm256_cvtepu8_epi16(latin1); 15| | 16| 121k| if (big_endian) { ------------------ | Branch (16:9): [True: 121k, Folded] ------------------ 17| 121k| const __m128i swap128 = 18| 121k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 19| 121k| const __m256i swap = _mm256_set_m128i(swap128, swap128); ------------------ | | 4| 121k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 121k| _mm256_castsi128_si256(xmm2), 2) ------------------ 20| 121k| utf16 = _mm256_shuffle_epi8(utf16, swap); 21| 121k| } 22| | 23| | // Store the contents of xmm1 into the address pointed by (output + i) 24| 121k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output + i), utf16); 25| 121k| } 26| | 27| 88| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 28| 88|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_latin1_to_utf32EPKcmPDi: 3| 110| char32_t *utf32_output) { 4| 110| size_t rounded_len = ((len | 7) ^ 7); // Round down to nearest multiple of 8 5| | 6| 1.03M| for (size_t i = 0; i < rounded_len; i += 8) { ------------------ | Branch (6:22): [True: 1.03M, False: 110] ------------------ 7| | // Load 8 Latin1 characters into a 64-bit register 8| 1.03M| __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| 1.03M| __m256i out = _mm256_cvtepu8_epi32(in); 13| | 14| | // Store the results back to memory 15| 1.03M| _mm256_storeu_si256((__m256i *)&utf32_output[i], out); 16| 1.03M| } 17| | 18| | // return pointers pointing to where we left off 19| 110| return std::make_pair(buf + rounded_len, utf32_output + rounded_len); 20| 110|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_latin1_to_utf8EPKcmPc: 3| 606| char *utf8_output) { 4| 606| const char *end = latin1_input + len; 5| 606| const __m256i v_0000 = _mm256_setzero_si256(); 6| 606| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 7| 606| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 8| 606| const size_t safety_margin = 12; 9| | 10| 2.57M| while (end - latin1_input >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (10:10): [True: 2.57M, False: 606] ------------------ 11| 2.57M| __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| 2.57M| const __m128i v_80 = _mm_set1_epi8((char)0x80); 14| 2.57M| if (_mm_testz_si128(in8, v_80)) { // ASCII fast path!!!! ------------------ | Branch (14:9): [True: 2.24M, False: 325k] ------------------ 15| | // 1. store (16 bytes) 16| 2.24M| _mm_storeu_si128((__m128i *)utf8_output, in8); 17| | // 2. adjust pointers 18| 2.24M| latin1_input += 16; 19| 2.24M| utf8_output += 16; 20| 2.24M| continue; // we are done for this round! 21| 2.24M| } 22| | // We proceed only with the first 16 bytes. 23| 325k| 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| 325k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 29| 325k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 30| | 31| | // t0 = [0000|00aa|bbbb|bb00] 32| 325k| const __m256i t0 = _mm256_slli_epi16(in, 2); 33| | // t1 = [0000|00aa|0000|0000] 34| 325k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 35| | // t2 = [0000|0000|00bb|bbbb] 36| 325k| const __m256i t2 = _mm256_and_si256(in, v_003f); 37| | // t3 = [000a|aaaa|00bb|bbbb] 38| 325k| const __m256i t3 = _mm256_or_si256(t1, t2); 39| | // t4 = [1100|00aa|10bb|bbbb] 40| 325k| 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| 325k| const __m256i one_byte_bytemask = 46| 325k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 47| 325k| const uint32_t one_byte_bitmask = 48| 325k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 49| | 50| 325k| const __m256i utf8_unpacked = _mm256_blendv_epi8(t4, in, one_byte_bytemask); 51| | 52| | // 3. prepare bitmask for 8-bit lookup 53| 325k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 54| 325k| const uint32_t M1 = M0 >> 7; 55| 325k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 56| | // 4. pack the bytes 57| | 58| 325k| const uint8_t *row = 59| 325k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 60| 325k| const uint8_t *row_2 = 61| 325k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 16)] 62| 325k| [0]; 63| | 64| 325k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 65| 325k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 66| | 67| 325k| const __m256i utf8_packed = _mm256_shuffle_epi8( 68| 325k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 325k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 325k| _mm256_castsi128_si256(xmm2), 2) ------------------ 69| | // 5. store bytes 70| 325k| _mm_storeu_si128((__m128i *)utf8_output, 71| 325k| _mm256_castsi256_si128(utf8_packed)); 72| 325k| utf8_output += row[0]; 73| 325k| _mm_storeu_si128((__m128i *)utf8_output, 74| 325k| _mm256_extractf128_si256(utf8_packed, 1)); 75| 325k| utf8_output += row_2[0]; 76| | 77| | // 6. adjust pointers 78| 325k| latin1_input += 16; 79| 325k| continue; 80| | 81| 2.57M| } // while 82| 606| return std::make_pair(latin1_input, utf8_output); 83| 606|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_utf16_to_latin1ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 134| char *latin1_output) { 5| 134| const char16_t *end = buf + len; 6| 2.31k| while (end - buf >= 32) { ------------------ | Branch (6:10): [True: 2.22k, False: 92] ------------------ 7| | // Load 16 UTF-16 characters into 256-bit AVX2 register 8| 2.22k| __m256i in0 = _mm256_loadu_si256(reinterpret_cast(buf)); 9| 2.22k| __m256i in1 = 10| 2.22k| _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| 2.22k| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 21| 2.22k| if (_mm256_testz_si256(_mm256_or_si256(in0, in1), high_byte_mask)) { ------------------ | Branch (21:9): [True: 2.17k, False: 42] ------------------ 22| | // Pack 16-bit characters into 8-bit and store in latin1_output 23| 2.17k| const __m256i packed = _mm256_packus_epi16(in0, in1); 24| | 25| 2.17k| const __m256i result = _mm256_permute4x64_epi64(packed, 0b11011000); 26| | 27| 2.17k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(latin1_output), result); 28| | // Adjust pointers for the next iteration 29| 2.17k| buf += 32; 30| 2.17k| latin1_output += 32; 31| 2.17k| } else { 32| 42| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 33| 42| } 34| 2.22k| } // while 35| 92| return std::make_pair(buf, latin1_output); 36| 134|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_utf16_to_latin1ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 94| char *latin1_output) { 5| 94| const char16_t *end = buf + len; 6| 284| while (end - buf >= 32) { ------------------ | Branch (6:10): [True: 206, False: 78] ------------------ 7| | // Load 16 UTF-16 characters into 256-bit AVX2 register 8| 206| __m256i in0 = _mm256_loadu_si256(reinterpret_cast(buf)); 9| 206| __m256i in1 = 10| 206| _mm256_loadu_si256(reinterpret_cast(buf + 16)); 11| | 12| 206| if constexpr (!match_system(big_endian)) { 13| 206| const __m256i swap = _mm256_setr_epi8( 14| 206| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 15| 206| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 16| 206| in0 = _mm256_shuffle_epi8(in0, swap); 17| 206| in1 = _mm256_shuffle_epi8(in1, swap); 18| 206| } 19| | 20| 206| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 21| 206| if (_mm256_testz_si256(_mm256_or_si256(in0, in1), high_byte_mask)) { ------------------ | Branch (21:9): [True: 190, False: 16] ------------------ 22| | // Pack 16-bit characters into 8-bit and store in latin1_output 23| 190| const __m256i packed = _mm256_packus_epi16(in0, in1); 24| | 25| 190| const __m256i result = _mm256_permute4x64_epi64(packed, 0b11011000); 26| | 27| 190| _mm256_storeu_si256(reinterpret_cast<__m256i *>(latin1_output), result); 28| | // Adjust pointers for the next iteration 29| 190| buf += 32; 30| 190| latin1_output += 32; 31| 190| } else { 32| 16| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 33| 16| } 34| 206| } // while 35| 78| return std::make_pair(buf, latin1_output); 36| 94|} 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| 411k| while (end - buf >= 16) { ------------------ | Branch (44:10): [True: 411k, False: 156] ------------------ 45| 411k| __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| 411k| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 55| 411k| if (_mm256_testz_si256(in, high_byte_mask)) { ------------------ | Branch (55:9): [True: 411k, False: 122] ------------------ 56| 411k| __m128i lo = _mm256_extractf128_si256(in, 0); 57| 411k| __m128i hi = _mm256_extractf128_si256(in, 1); 58| 411k| __m128i latin1_packed_lo = _mm_packus_epi16(lo, lo); 59| 411k| __m128i latin1_packed_hi = _mm_packus_epi16(hi, hi); 60| 411k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 61| 411k| latin1_packed_lo); 62| 411k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output + 8), 63| 411k| latin1_packed_hi); 64| 411k| buf += 16; 65| 411k| latin1_output += 16; 66| 411k| } else { 67| | // Fallback to scalar code for handling errors 68| 618| for (int k = 0; k < 16; k++) { ------------------ | Branch (68:23): [True: 618, False: 0] ------------------ 69| 618| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 70| 618| if (word <= 0xff) { ------------------ | Branch (70:13): [True: 496, False: 122] ------------------ 71| 496| *latin1_output++ = char(word); 72| 496| } else { 73| 122| return std::make_pair( 74| 122| result{error_code::TOO_LARGE, (size_t)(buf - start + k)}, 75| 122| latin1_output); 76| 122| } 77| 618| } 78| 0| buf += 16; 79| 0| } 80| 411k| } // while 81| 156| return std::make_pair(result{error_code::SUCCESS, (size_t)(buf - start)}, 82| 156| latin1_output); 83| 278|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_140avx2_convert_utf16_to_latin1_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 41| 250| char *latin1_output) { 42| 250| const char16_t *start = buf; 43| 250| const char16_t *end = buf + len; 44| 463k| while (end - buf >= 16) { ------------------ | Branch (44:10): [True: 463k, False: 140] ------------------ 45| 463k| __m256i in = _mm256_loadu_si256(reinterpret_cast(buf)); 46| | 47| 463k| if constexpr (!match_system(big_endian)) { 48| 463k| const __m256i swap = _mm256_setr_epi8( 49| 463k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 50| 463k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 51| 463k| in = _mm256_shuffle_epi8(in, swap); 52| 463k| } 53| | 54| 463k| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 55| 463k| if (_mm256_testz_si256(in, high_byte_mask)) { ------------------ | Branch (55:9): [True: 463k, False: 110] ------------------ 56| 463k| __m128i lo = _mm256_extractf128_si256(in, 0); 57| 463k| __m128i hi = _mm256_extractf128_si256(in, 1); 58| 463k| __m128i latin1_packed_lo = _mm_packus_epi16(lo, lo); 59| 463k| __m128i latin1_packed_hi = _mm_packus_epi16(hi, hi); 60| 463k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 61| 463k| latin1_packed_lo); 62| 463k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output + 8), 63| 463k| latin1_packed_hi); 64| 463k| buf += 16; 65| 463k| latin1_output += 16; 66| 463k| } else { 67| | // Fallback to scalar code for handling errors 68| 554| for (int k = 0; k < 16; k++) { ------------------ | Branch (68:23): [True: 554, False: 0] ------------------ 69| 554| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 70| 554| if (word <= 0xff) { ------------------ | Branch (70:13): [True: 444, False: 110] ------------------ 71| 444| *latin1_output++ = char(word); 72| 444| } else { 73| 110| return std::make_pair( 74| 110| result{error_code::TOO_LARGE, (size_t)(buf - start + k)}, 75| 110| latin1_output); 76| 110| } 77| 554| } 78| 0| buf += 16; 79| 0| } 80| 463k| } // while 81| 140| return std::make_pair(result{error_code::SUCCESS, (size_t)(buf - start)}, 82| 140| latin1_output); 83| 250|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf16_to_utf32ILNS_10endiannessE0EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 412| char32_t *utf32_output) { 58| 412| const char16_t *end = buf + len; 59| 412| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 412| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| | 62| 146k| while (end - buf >= 16) { ------------------ | Branch (62:10): [True: 146k, False: 310] ------------------ 63| 146k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 64| 146k| if (big_endian) { ------------------ | Branch (64:9): [Folded, False: 146k] ------------------ 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| 146k| const __m256i surrogates_bytemask = 75| 146k| _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| 146k| const uint32_t surrogates_bitmask = 80| 146k| 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| 146k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (83:9): [True: 142k, False: 4.45k] ------------------ 84| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 85| | // units 86| 142k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 87| 142k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 88| 142k| _mm256_storeu_si256( 89| 142k| reinterpret_cast<__m256i *>(utf32_output + 8), 90| 142k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 91| 142k| utf32_output += 16; 92| 142k| buf += 16; 93| | // surrogate pair(s) in a register 94| 142k| } 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| 4.45k| size_t forward = 15; 99| 4.45k| size_t k = 0; 100| 4.45k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (100:11): [True: 0, False: 4.45k] ------------------ 101| 0| forward = size_t(end - buf - 1); 102| 0| } 103| 65.1k| for (; k < forward; k++) { ------------------ | Branch (103:14): [True: 60.7k, False: 4.34k] ------------------ 104| 60.7k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 105| 60.7k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (105:13): [True: 55.3k, False: 5.39k] ------------------ 106| | // No surrogate pair 107| 55.3k| *utf32_output++ = char32_t(word); 108| 55.3k| } else { 109| | // must be a surrogate pair 110| 5.39k| uint16_t diff = uint16_t(word - 0xD800); 111| 5.39k| uint16_t next_word = 112| 5.39k| scalar::utf16::swap_if_needed(buf[k + 1]); 113| 5.39k| k++; 114| 5.39k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 115| 5.39k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (115:15): [True: 102, False: 5.29k] ------------------ 116| 102| return std::make_pair(nullptr, utf32_output); 117| 102| } 118| 5.29k| uint32_t value = (diff << 10) + diff2 + 0x10000; 119| 5.29k| *utf32_output++ = char32_t(value); 120| 5.29k| } 121| 60.7k| } 122| 4.34k| buf += k; 123| 4.34k| } 124| 146k| } // while 125| 310| return std::make_pair(buf, utf32_output); 126| 412|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf16_to_utf32ILNS_10endiannessE1EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 392| char32_t *utf32_output) { 58| 392| const char16_t *end = buf + len; 59| 392| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 392| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| | 62| 312k| while (end - buf >= 16) { ------------------ | Branch (62:10): [True: 311k, False: 338] ------------------ 63| 311k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 64| 311k| if (big_endian) { ------------------ | Branch (64:9): [True: 311k, Folded] ------------------ 65| 311k| const __m256i swap = _mm256_setr_epi8( 66| 311k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 67| 311k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 68| 311k| in = _mm256_shuffle_epi8(in, swap); 69| 311k| } 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| 311k| const __m256i surrogates_bytemask = 75| 311k| _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| 311k| const uint32_t surrogates_bitmask = 80| 311k| 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| 311k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (83:9): [True: 310k, False: 1.68k] ------------------ 84| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 85| | // units 86| 310k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 87| 310k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 88| 310k| _mm256_storeu_si256( 89| 310k| reinterpret_cast<__m256i *>(utf32_output + 8), 90| 310k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 91| 310k| utf32_output += 16; 92| 310k| buf += 16; 93| | // surrogate pair(s) in a register 94| 310k| } 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| 1.68k| size_t forward = 15; 99| 1.68k| size_t k = 0; 100| 1.68k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (100:11): [True: 0, False: 1.68k] ------------------ 101| 0| forward = size_t(end - buf - 1); 102| 0| } 103| 24.2k| for (; k < forward; k++) { ------------------ | Branch (103:14): [True: 22.6k, False: 1.62k] ------------------ 104| 22.6k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 105| 22.6k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (105:13): [True: 20.3k, False: 2.27k] ------------------ 106| | // No surrogate pair 107| 20.3k| *utf32_output++ = char32_t(word); 108| 20.3k| } else { 109| | // must be a surrogate pair 110| 2.27k| uint16_t diff = uint16_t(word - 0xD800); 111| 2.27k| uint16_t next_word = 112| 2.27k| scalar::utf16::swap_if_needed(buf[k + 1]); 113| 2.27k| k++; 114| 2.27k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 115| 2.27k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (115:15): [True: 54, False: 2.22k] ------------------ 116| 54| return std::make_pair(nullptr, utf32_output); 117| 54| } 118| 2.22k| uint32_t value = (diff << 10) + diff2 + 0x10000; 119| 2.22k| *utf32_output++ = char32_t(value); 120| 2.22k| } 121| 22.6k| } 122| 1.62k| buf += k; 123| 1.62k| } 124| 311k| } // while 125| 338| return std::make_pair(buf, utf32_output); 126| 392|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf16_to_utf32_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 138| 418| char32_t *utf32_output) { 139| 418| const char16_t *start = buf; 140| 418| const char16_t *end = buf + len; 141| 418| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 142| 418| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 143| | 144| 269k| while (end - buf >= 16) { ------------------ | Branch (144:10): [True: 269k, False: 272] ------------------ 145| 269k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 146| 269k| if (big_endian) { ------------------ | Branch (146:9): [Folded, False: 269k] ------------------ 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| 269k| const __m256i surrogates_bytemask = 157| 269k| _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| 269k| const uint32_t surrogates_bitmask = 162| 269k| 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| 269k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (165:9): [True: 259k, False: 9.65k] ------------------ 166| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 167| | // units 168| 259k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 169| 259k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 170| 259k| _mm256_storeu_si256( 171| 259k| reinterpret_cast<__m256i *>(utf32_output + 8), 172| 259k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 173| 259k| utf32_output += 16; 174| 259k| buf += 16; 175| | // surrogate pair(s) in a register 176| 259k| } 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| 9.65k| size_t forward = 15; 181| 9.65k| size_t k = 0; 182| 9.65k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (182:11): [True: 0, False: 9.65k] ------------------ 183| 0| forward = size_t(end - buf - 1); 184| 0| } 185| 139k| for (; k < forward; k++) { ------------------ | Branch (185:14): [True: 129k, False: 9.50k] ------------------ 186| 129k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 187| 129k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (187:13): [True: 115k, False: 14.7k] ------------------ 188| | // No surrogate pair 189| 115k| *utf32_output++ = char32_t(word); 190| 115k| } else { 191| | // must be a surrogate pair 192| 14.7k| uint16_t diff = uint16_t(word - 0xD800); 193| 14.7k| uint16_t next_word = 194| 14.7k| scalar::utf16::swap_if_needed(buf[k + 1]); 195| 14.7k| k++; 196| 14.7k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 197| 14.7k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (197:15): [True: 146, False: 14.5k] ------------------ 198| 146| return std::make_pair( 199| 146| result(error_code::SURROGATE, buf - start + k - 1), 200| 146| utf32_output); 201| 146| } 202| 14.5k| uint32_t value = (diff << 10) + diff2 + 0x10000; 203| 14.5k| *utf32_output++ = char32_t(value); 204| 14.5k| } 205| 129k| } 206| 9.50k| buf += k; 207| 9.50k| } 208| 269k| } // while 209| 272| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 210| 418|} 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| 158k| while (end - buf >= 16) { ------------------ | Branch (144:10): [True: 158k, False: 270] ------------------ 145| 158k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 146| 158k| if (big_endian) { ------------------ | Branch (146:9): [True: 158k, Folded] ------------------ 147| 158k| const __m256i swap = _mm256_setr_epi8( 148| 158k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 149| 158k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 150| 158k| in = _mm256_shuffle_epi8(in, swap); 151| 158k| } 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| 158k| const __m256i surrogates_bytemask = 157| 158k| _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| 158k| const uint32_t surrogates_bitmask = 162| 158k| 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| 158k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (165:9): [True: 152k, False: 5.85k] ------------------ 166| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 167| | // units 168| 152k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 169| 152k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 170| 152k| _mm256_storeu_si256( 171| 152k| reinterpret_cast<__m256i *>(utf32_output + 8), 172| 152k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 173| 152k| utf32_output += 16; 174| 152k| buf += 16; 175| | // surrogate pair(s) in a register 176| 152k| } 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| 5.85k| size_t forward = 15; 181| 5.85k| size_t k = 0; 182| 5.85k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (182:11): [True: 0, False: 5.85k] ------------------ 183| 0| forward = size_t(end - buf - 1); 184| 0| } 185| 83.5k| for (; k < forward; k++) { ------------------ | Branch (185:14): [True: 77.7k, False: 5.78k] ------------------ 186| 77.7k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 187| 77.7k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (187:13): [True: 67.4k, False: 10.2k] ------------------ 188| | // No surrogate pair 189| 67.4k| *utf32_output++ = char32_t(word); 190| 67.4k| } else { 191| | // must be a surrogate pair 192| 10.2k| uint16_t diff = uint16_t(word - 0xD800); 193| 10.2k| uint16_t next_word = 194| 10.2k| scalar::utf16::swap_if_needed(buf[k + 1]); 195| 10.2k| k++; 196| 10.2k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 197| 10.2k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (197:15): [True: 76, False: 10.2k] ------------------ 198| 76| return std::make_pair( 199| 76| result(error_code::SURROGATE, buf - start + k - 1), 200| 76| utf32_output); 201| 76| } 202| 10.2k| uint32_t value = (diff << 10) + diff2 + 0x10000; 203| 10.2k| *utf32_output++ = char32_t(value); 204| 10.2k| } 205| 77.7k| } 206| 5.78k| buf += k; 207| 5.78k| } 208| 158k| } // while 209| 270| 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| 856|avx2_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| 856| const char16_t *end = buf + len; 58| 856| const __m256i v_0000 = _mm256_setzero_si256(); 59| 856| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 856| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| 856| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 62| 856| const size_t safety_margin = 63| 856| 12; // to avoid overruns, see issue 64| | // https://github.com/simdutf/simdutf/issues/92 65| | 66| 1.01M| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (66:10): [True: 1.01M, False: 722] ------------------ 67| 1.01M| __m256i in = _mm256_loadu_si256((__m256i *)buf); 68| 1.01M| if (big_endian) { ------------------ | Branch (68:9): [Folded, False: 1.01M] ------------------ 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| 1.01M| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 76| 1.01M| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 648k, False: 364k] ------------------ 77| | // 1. pack the bytes 78| 648k| const __m128i utf8_packed = _mm_packus_epi16( 79| 648k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 80| | // 2. store (16 bytes) 81| 648k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 82| | // 3. adjust pointers 83| 648k| buf += 16; 84| 648k| utf8_output += 16; 85| 648k| continue; // we are done for this round! 86| 648k| } 87| | // no bits set above 7th bit 88| 364k| const __m256i one_byte_bytemask = 89| 364k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 90| 364k| const uint32_t one_byte_bitmask = 91| 364k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 92| | 93| | // no bits set above 11th bit 94| 364k| const __m256i one_or_two_bytes_bytemask = 95| 364k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 96| 364k| const uint32_t one_or_two_bytes_bitmask = 97| 364k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 98| 364k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (98:9): [True: 259k, False: 105k] ------------------ 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| 259k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 104| 259k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 105| | 106| | // t0 = [000a|aaaa|bbbb|bb00] 107| 259k| const __m256i t0 = _mm256_slli_epi16(in, 2); 108| | // t1 = [000a|aaaa|0000|0000] 109| 259k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 110| | // t2 = [0000|0000|00bb|bbbb] 111| 259k| const __m256i t2 = _mm256_and_si256(in, v_003f); 112| | // t3 = [000a|aaaa|00bb|bbbb] 113| 259k| const __m256i t3 = _mm256_or_si256(t1, t2); 114| | // t4 = [110a|aaaa|10bb|bbbb] 115| 259k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 116| | 117| | // 2. merge ASCII and 2-byte codewords 118| 259k| const __m256i utf8_unpacked = 119| 259k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 120| | 121| | // 3. prepare bitmask for 8-bit lookup 122| 259k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 123| 259k| const uint32_t M1 = M0 >> 7; 124| 259k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 125| | // 4. pack the bytes 126| | 127| 259k| const uint8_t *row = 128| 259k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 129| 259k| const uint8_t *row_2 = 130| 259k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 131| 259k| 16)][0]; 132| | 133| 259k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 134| 259k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 135| | 136| 259k| const __m256i utf8_packed = _mm256_shuffle_epi8( 137| 259k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 259k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 259k| _mm256_castsi128_si256(xmm2), 2) ------------------ 138| | // 5. store bytes 139| 259k| _mm_storeu_si128((__m128i *)utf8_output, 140| 259k| _mm256_castsi256_si128(utf8_packed)); 141| 259k| utf8_output += row[0]; 142| 259k| _mm_storeu_si128((__m128i *)utf8_output, 143| 259k| _mm256_extractf128_si256(utf8_packed, 1)); 144| 259k| utf8_output += row_2[0]; 145| | 146| | // 6. adjust pointers 147| 259k| buf += 16; 148| 259k| continue; 149| 259k| } 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| 105k| const __m256i surrogates_bytemask = 154| 105k| _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| 105k| const uint32_t surrogates_bitmask = 159| 105k| 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| 105k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (162:9): [True: 95.1k, False: 10.1k] ------------------ 163| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 164| 95.1k| const __m256i dup_even = _mm256_setr_epi16( 165| 95.1k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 166| 95.1k| 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| 95.1k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 196| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 197| 95.1k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 198| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 199| 95.1k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 195| 95.1k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 200| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 201| 95.1k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 195| 95.1k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 202| | 203| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 204| 95.1k| const __m256i s0 = _mm256_srli_epi16(in, 4); 205| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 206| 95.1k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 195| 95.1k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 207| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 208| 95.1k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 195| 95.1k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 209| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 210| 95.1k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 195| 95.1k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 211| 95.1k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 212| 95.1k| simdutf_vec(0b0100000000000000)); ------------------ | | 195| 95.1k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 213| 95.1k| const __m256i s4 = _mm256_xor_si256(s3, m0); 214| 95.1k|#undef simdutf_vec 215| | 216| | // 4. expand code units 16-bit => 32-bit 217| 95.1k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 218| 95.1k| 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| 95.1k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 222| 95.1k| (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| 95.1k| const uint8_t mask0 = uint8_t(mask); 243| 95.1k| const uint8_t *row0 = 244| 95.1k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 245| 95.1k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 246| 95.1k| const __m128i utf8_0 = 247| 95.1k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 248| | 249| 95.1k| const uint8_t mask1 = static_cast(mask >> 8); 250| 95.1k| const uint8_t *row1 = 251| 95.1k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 252| 95.1k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 253| 95.1k| const __m128i utf8_1 = 254| 95.1k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 255| | 256| 95.1k| const uint8_t mask2 = static_cast(mask >> 16); 257| 95.1k| const uint8_t *row2 = 258| 95.1k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 259| 95.1k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 260| 95.1k| const __m128i utf8_2 = 261| 95.1k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 262| | 263| 95.1k| const uint8_t mask3 = static_cast(mask >> 24); 264| 95.1k| const uint8_t *row3 = 265| 95.1k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 266| 95.1k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 267| 95.1k| const __m128i utf8_3 = 268| 95.1k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 269| | 270| 95.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 271| 95.1k| utf8_output += row0[0]; 272| 95.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 273| 95.1k| utf8_output += row1[0]; 274| 95.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 275| 95.1k| utf8_output += row2[0]; 276| 95.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 277| 95.1k| utf8_output += row3[0]; 278| 95.1k| buf += 16; 279| | // surrogate pair(s) in a register 280| 95.1k| } 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| 10.1k| size_t forward = 15; 285| 10.1k| size_t k = 0; 286| 10.1k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (286:11): [True: 0, False: 10.1k] ------------------ 287| 0| forward = size_t(end - buf - 1); 288| 0| } 289| 150k| for (; k < forward; k++) { ------------------ | Branch (289:14): [True: 140k, False: 10.0k] ------------------ 290| 140k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 291| 140k| if ((word & 0xFF80) == 0) { ------------------ | Branch (291:13): [True: 18.3k, False: 122k] ------------------ 292| 18.3k| *utf8_output++ = char(word); 293| 122k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (293:20): [True: 4.86k, False: 117k] ------------------ 294| 4.86k| *utf8_output++ = char((word >> 6) | 0b11000000); 295| 4.86k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 296| 117k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (296:20): [True: 106k, False: 11.0k] ------------------ 297| 106k| *utf8_output++ = char((word >> 12) | 0b11100000); 298| 106k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 299| 106k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 300| 106k| } else { 301| | // must be a surrogate pair 302| 11.0k| uint16_t diff = uint16_t(word - 0xD800); 303| 11.0k| uint16_t next_word = 304| 11.0k| scalar::utf16::swap_if_needed(buf[k + 1]); 305| 11.0k| k++; 306| 11.0k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 307| 11.0k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (307:15): [True: 134, False: 10.9k] ------------------ 308| 134| return std::make_pair(nullptr, utf8_output); 309| 134| } 310| 10.9k| uint32_t value = (diff << 10) + diff2 + 0x10000; 311| 10.9k| *utf8_output++ = char((value >> 18) | 0b11110000); 312| 10.9k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 313| 10.9k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 314| 10.9k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 315| 10.9k| } 316| 140k| } 317| 10.0k| buf += k; 318| 10.0k| } 319| 105k| } // while 320| 722| return std::make_pair(buf, utf8_output); 321| 856|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_126avx2_convert_utf16_to_utf8ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 750|avx2_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| 750| const char16_t *end = buf + len; 58| 750| const __m256i v_0000 = _mm256_setzero_si256(); 59| 750| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 750| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| 750| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 62| 750| const size_t safety_margin = 63| 750| 12; // to avoid overruns, see issue 64| | // https://github.com/simdutf/simdutf/issues/92 65| | 66| 921k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (66:10): [True: 920k, False: 646] ------------------ 67| 920k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 68| 920k| if (big_endian) { ------------------ | Branch (68:9): [True: 920k, Folded] ------------------ 69| 920k| const __m256i swap = _mm256_setr_epi8( 70| 920k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 71| 920k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 72| 920k| in = _mm256_shuffle_epi8(in, swap); 73| 920k| } 74| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 75| 920k| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 76| 920k| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 672k, False: 247k] ------------------ 77| | // 1. pack the bytes 78| 672k| const __m128i utf8_packed = _mm_packus_epi16( 79| 672k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 80| | // 2. store (16 bytes) 81| 672k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 82| | // 3. adjust pointers 83| 672k| buf += 16; 84| 672k| utf8_output += 16; 85| 672k| continue; // we are done for this round! 86| 672k| } 87| | // no bits set above 7th bit 88| 247k| const __m256i one_byte_bytemask = 89| 247k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 90| 247k| const uint32_t one_byte_bitmask = 91| 247k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 92| | 93| | // no bits set above 11th bit 94| 247k| const __m256i one_or_two_bytes_bytemask = 95| 247k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 96| 247k| const uint32_t one_or_two_bytes_bitmask = 97| 247k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 98| 247k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (98:9): [True: 17.2k, False: 230k] ------------------ 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| 17.2k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 104| 17.2k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 105| | 106| | // t0 = [000a|aaaa|bbbb|bb00] 107| 17.2k| const __m256i t0 = _mm256_slli_epi16(in, 2); 108| | // t1 = [000a|aaaa|0000|0000] 109| 17.2k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 110| | // t2 = [0000|0000|00bb|bbbb] 111| 17.2k| const __m256i t2 = _mm256_and_si256(in, v_003f); 112| | // t3 = [000a|aaaa|00bb|bbbb] 113| 17.2k| const __m256i t3 = _mm256_or_si256(t1, t2); 114| | // t4 = [110a|aaaa|10bb|bbbb] 115| 17.2k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 116| | 117| | // 2. merge ASCII and 2-byte codewords 118| 17.2k| const __m256i utf8_unpacked = 119| 17.2k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 120| | 121| | // 3. prepare bitmask for 8-bit lookup 122| 17.2k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 123| 17.2k| const uint32_t M1 = M0 >> 7; 124| 17.2k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 125| | // 4. pack the bytes 126| | 127| 17.2k| const uint8_t *row = 128| 17.2k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 129| 17.2k| const uint8_t *row_2 = 130| 17.2k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 131| 17.2k| 16)][0]; 132| | 133| 17.2k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 134| 17.2k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 135| | 136| 17.2k| const __m256i utf8_packed = _mm256_shuffle_epi8( 137| 17.2k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 17.2k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 17.2k| _mm256_castsi128_si256(xmm2), 2) ------------------ 138| | // 5. store bytes 139| 17.2k| _mm_storeu_si128((__m128i *)utf8_output, 140| 17.2k| _mm256_castsi256_si128(utf8_packed)); 141| 17.2k| utf8_output += row[0]; 142| 17.2k| _mm_storeu_si128((__m128i *)utf8_output, 143| 17.2k| _mm256_extractf128_si256(utf8_packed, 1)); 144| 17.2k| utf8_output += row_2[0]; 145| | 146| | // 6. adjust pointers 147| 17.2k| buf += 16; 148| 17.2k| continue; 149| 17.2k| } 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| 230k| const __m256i surrogates_bytemask = 154| 230k| _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| 230k| const uint32_t surrogates_bitmask = 159| 230k| 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| 230k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (162:9): [True: 225k, False: 4.83k] ------------------ 163| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 164| 225k| const __m256i dup_even = _mm256_setr_epi16( 165| 225k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 166| 225k| 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| 225k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 196| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 197| 225k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 198| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 199| 225k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 195| 225k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 200| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 201| 225k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 195| 225k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 202| | 203| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 204| 225k| const __m256i s0 = _mm256_srli_epi16(in, 4); 205| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 206| 225k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 195| 225k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 207| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 208| 225k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 195| 225k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 209| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 210| 225k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 195| 225k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 211| 225k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 212| 225k| simdutf_vec(0b0100000000000000)); ------------------ | | 195| 225k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 213| 225k| const __m256i s4 = _mm256_xor_si256(s3, m0); 214| 225k|#undef simdutf_vec 215| | 216| | // 4. expand code units 16-bit => 32-bit 217| 225k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 218| 225k| 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| 225k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 222| 225k| (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| 225k| const uint8_t mask0 = uint8_t(mask); 243| 225k| const uint8_t *row0 = 244| 225k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 245| 225k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 246| 225k| const __m128i utf8_0 = 247| 225k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 248| | 249| 225k| const uint8_t mask1 = static_cast(mask >> 8); 250| 225k| const uint8_t *row1 = 251| 225k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 252| 225k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 253| 225k| const __m128i utf8_1 = 254| 225k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 255| | 256| 225k| const uint8_t mask2 = static_cast(mask >> 16); 257| 225k| const uint8_t *row2 = 258| 225k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 259| 225k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 260| 225k| const __m128i utf8_2 = 261| 225k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 262| | 263| 225k| const uint8_t mask3 = static_cast(mask >> 24); 264| 225k| const uint8_t *row3 = 265| 225k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 266| 225k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 267| 225k| const __m128i utf8_3 = 268| 225k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 269| | 270| 225k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 271| 225k| utf8_output += row0[0]; 272| 225k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 273| 225k| utf8_output += row1[0]; 274| 225k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 275| 225k| utf8_output += row2[0]; 276| 225k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 277| 225k| utf8_output += row3[0]; 278| 225k| buf += 16; 279| | // surrogate pair(s) in a register 280| 225k| } 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| 4.83k| size_t forward = 15; 285| 4.83k| size_t k = 0; 286| 4.83k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (286:11): [True: 0, False: 4.83k] ------------------ 287| 0| forward = size_t(end - buf - 1); 288| 0| } 289| 71.2k| for (; k < forward; k++) { ------------------ | Branch (289:14): [True: 66.5k, False: 4.73k] ------------------ 290| 66.5k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 291| 66.5k| if ((word & 0xFF80) == 0) { ------------------ | Branch (291:13): [True: 11.6k, False: 54.8k] ------------------ 292| 11.6k| *utf8_output++ = char(word); 293| 54.8k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (293:20): [True: 2.18k, False: 52.6k] ------------------ 294| 2.18k| *utf8_output++ = char((word >> 6) | 0b11000000); 295| 2.18k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 296| 52.6k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (296:20): [True: 47.3k, False: 5.27k] ------------------ 297| 47.3k| *utf8_output++ = char((word >> 12) | 0b11100000); 298| 47.3k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 299| 47.3k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 300| 47.3k| } else { 301| | // must be a surrogate pair 302| 5.27k| uint16_t diff = uint16_t(word - 0xD800); 303| 5.27k| uint16_t next_word = 304| 5.27k| scalar::utf16::swap_if_needed(buf[k + 1]); 305| 5.27k| k++; 306| 5.27k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 307| 5.27k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (307:15): [True: 104, False: 5.16k] ------------------ 308| 104| return std::make_pair(nullptr, utf8_output); 309| 104| } 310| 5.16k| uint32_t value = (diff << 10) + diff2 + 0x10000; 311| 5.16k| *utf8_output++ = char((value >> 18) | 0b11110000); 312| 5.16k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 313| 5.16k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 314| 5.16k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 315| 5.16k| } 316| 66.5k| } 317| 4.73k| buf += k; 318| 4.73k| } 319| 230k| } // while 320| 646| return std::make_pair(buf, utf8_output); 321| 750|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_138avx2_convert_utf16_to_utf8_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 333| 664| char *utf8_output) { 334| 664| const char16_t *start = buf; 335| 664| const char16_t *end = buf + len; 336| | 337| 664| const __m256i v_0000 = _mm256_setzero_si256(); 338| 664| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 339| 664| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 340| 664| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 341| 664| const size_t safety_margin = 342| 664| 12; // to avoid overruns, see issue 343| | // https://github.com/simdutf/simdutf/issues/92 344| | 345| 551k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (345:10): [True: 550k, False: 498] ------------------ 346| 550k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 347| 550k| if (big_endian) { ------------------ | Branch (347:9): [Folded, False: 550k] ------------------ 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| 550k| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 355| 550k| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (355:9): [True: 254k, False: 296k] ------------------ 356| | // 1. pack the bytes 357| 254k| const __m128i utf8_packed = _mm_packus_epi16( 358| 254k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 359| | // 2. store (16 bytes) 360| 254k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 361| | // 3. adjust pointers 362| 254k| buf += 16; 363| 254k| utf8_output += 16; 364| 254k| continue; // we are done for this round! 365| 254k| } 366| | // no bits set above 7th bit 367| 296k| const __m256i one_byte_bytemask = 368| 296k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 369| 296k| const uint32_t one_byte_bitmask = 370| 296k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 371| | 372| | // no bits set above 11th bit 373| 296k| const __m256i one_or_two_bytes_bytemask = 374| 296k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 375| 296k| const uint32_t one_or_two_bytes_bitmask = 376| 296k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 377| 296k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (377:9): [True: 87.7k, False: 209k] ------------------ 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| 87.7k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 383| 87.7k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 384| | 385| | // t0 = [000a|aaaa|bbbb|bb00] 386| 87.7k| const __m256i t0 = _mm256_slli_epi16(in, 2); 387| | // t1 = [000a|aaaa|0000|0000] 388| 87.7k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 389| | // t2 = [0000|0000|00bb|bbbb] 390| 87.7k| const __m256i t2 = _mm256_and_si256(in, v_003f); 391| | // t3 = [000a|aaaa|00bb|bbbb] 392| 87.7k| const __m256i t3 = _mm256_or_si256(t1, t2); 393| | // t4 = [110a|aaaa|10bb|bbbb] 394| 87.7k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 395| | 396| | // 2. merge ASCII and 2-byte codewords 397| 87.7k| const __m256i utf8_unpacked = 398| 87.7k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 399| | 400| | // 3. prepare bitmask for 8-bit lookup 401| 87.7k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 402| 87.7k| const uint32_t M1 = M0 >> 7; 403| 87.7k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 404| | // 4. pack the bytes 405| | 406| 87.7k| const uint8_t *row = 407| 87.7k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 408| 87.7k| const uint8_t *row_2 = 409| 87.7k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 410| 87.7k| 16)][0]; 411| | 412| 87.7k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 413| 87.7k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 414| | 415| 87.7k| const __m256i utf8_packed = _mm256_shuffle_epi8( 416| 87.7k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 87.7k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 87.7k| _mm256_castsi128_si256(xmm2), 2) ------------------ 417| | // 5. store bytes 418| 87.7k| _mm_storeu_si128((__m128i *)utf8_output, 419| 87.7k| _mm256_castsi256_si128(utf8_packed)); 420| 87.7k| utf8_output += row[0]; 421| 87.7k| _mm_storeu_si128((__m128i *)utf8_output, 422| 87.7k| _mm256_extractf128_si256(utf8_packed, 1)); 423| 87.7k| utf8_output += row_2[0]; 424| | 425| | // 6. adjust pointers 426| 87.7k| buf += 16; 427| 87.7k| continue; 428| 87.7k| } 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| 209k| const __m256i surrogates_bytemask = 433| 209k| _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| 209k| const uint32_t surrogates_bitmask = 438| 209k| 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| 209k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (441:9): [True: 207k, False: 2.07k] ------------------ 442| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 443| 207k| const __m256i dup_even = _mm256_setr_epi16( 444| 207k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 445| 207k| 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| 207k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 475| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 476| 207k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 477| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 478| 207k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 474| 207k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 479| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 480| 207k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 474| 207k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 481| | 482| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 483| 207k| const __m256i s0 = _mm256_srli_epi16(in, 4); 484| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 485| 207k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 474| 207k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 486| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 487| 207k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 474| 207k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 488| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 489| 207k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 474| 207k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 490| 207k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 491| 207k| simdutf_vec(0b0100000000000000)); ------------------ | | 474| 207k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 492| 207k| const __m256i s4 = _mm256_xor_si256(s3, m0); 493| 207k|#undef simdutf_vec 494| | 495| | // 4. expand code units 16-bit => 32-bit 496| 207k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 497| 207k| 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| 207k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 501| 207k| (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| 207k| const uint8_t mask0 = uint8_t(mask); 522| 207k| const uint8_t *row0 = 523| 207k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 524| 207k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 525| 207k| const __m128i utf8_0 = 526| 207k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 527| | 528| 207k| const uint8_t mask1 = static_cast(mask >> 8); 529| 207k| const uint8_t *row1 = 530| 207k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 531| 207k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 532| 207k| const __m128i utf8_1 = 533| 207k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 534| | 535| 207k| const uint8_t mask2 = static_cast(mask >> 16); 536| 207k| const uint8_t *row2 = 537| 207k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 538| 207k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 539| 207k| const __m128i utf8_2 = 540| 207k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 541| | 542| 207k| const uint8_t mask3 = static_cast(mask >> 24); 543| 207k| const uint8_t *row3 = 544| 207k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 545| 207k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 546| 207k| const __m128i utf8_3 = 547| 207k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 548| | 549| 207k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 550| 207k| utf8_output += row0[0]; 551| 207k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 552| 207k| utf8_output += row1[0]; 553| 207k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 554| 207k| utf8_output += row2[0]; 555| 207k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 556| 207k| utf8_output += row3[0]; 557| 207k| buf += 16; 558| | // surrogate pair(s) in a register 559| 207k| } 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| 2.07k| size_t forward = 15; 564| 2.07k| size_t k = 0; 565| 2.07k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (565:11): [True: 0, False: 2.07k] ------------------ 566| 0| forward = size_t(end - buf - 1); 567| 0| } 568| 29.5k| for (; k < forward; k++) { ------------------ | Branch (568:14): [True: 27.6k, False: 1.91k] ------------------ 569| 27.6k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 570| 27.6k| if ((word & 0xFF80) == 0) { ------------------ | Branch (570:13): [True: 5.53k, False: 22.0k] ------------------ 571| 5.53k| *utf8_output++ = char(word); 572| 22.0k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (572:20): [True: 1.37k, False: 20.7k] ------------------ 573| 1.37k| *utf8_output++ = char((word >> 6) | 0b11000000); 574| 1.37k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 575| 20.7k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (575:20): [True: 18.2k, False: 2.42k] ------------------ 576| 18.2k| *utf8_output++ = char((word >> 12) | 0b11100000); 577| 18.2k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 578| 18.2k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 579| 18.2k| } else { 580| | // must be a surrogate pair 581| 2.42k| uint16_t diff = uint16_t(word - 0xD800); 582| 2.42k| uint16_t next_word = 583| 2.42k| scalar::utf16::swap_if_needed(buf[k + 1]); 584| 2.42k| k++; 585| 2.42k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 586| 2.42k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (586:15): [True: 166, False: 2.25k] ------------------ 587| 166| return std::make_pair( 588| 166| result(error_code::SURROGATE, buf - start + k - 1), 589| 166| utf8_output); 590| 166| } 591| 2.25k| uint32_t value = (diff << 10) + diff2 + 0x10000; 592| 2.25k| *utf8_output++ = char((value >> 18) | 0b11110000); 593| 2.25k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 594| 2.25k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 595| 2.25k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 596| 2.25k| } 597| 27.6k| } 598| 1.91k| buf += k; 599| 1.91k| } 600| 209k| } // while 601| 498| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 602| 664|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_138avx2_convert_utf16_to_utf8_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 333| 728| char *utf8_output) { 334| 728| const char16_t *start = buf; 335| 728| const char16_t *end = buf + len; 336| | 337| 728| const __m256i v_0000 = _mm256_setzero_si256(); 338| 728| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 339| 728| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 340| 728| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 341| 728| const size_t safety_margin = 342| 728| 12; // to avoid overruns, see issue 343| | // https://github.com/simdutf/simdutf/issues/92 344| | 345| 691k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (345:10): [True: 691k, False: 528] ------------------ 346| 691k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 347| 691k| if (big_endian) { ------------------ | Branch (347:9): [True: 691k, Folded] ------------------ 348| 691k| const __m256i swap = _mm256_setr_epi8( 349| 691k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 350| 691k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 351| 691k| in = _mm256_shuffle_epi8(in, swap); 352| 691k| } 353| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 354| 691k| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 355| 691k| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (355:9): [True: 420k, False: 271k] ------------------ 356| | // 1. pack the bytes 357| 420k| const __m128i utf8_packed = _mm_packus_epi16( 358| 420k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 359| | // 2. store (16 bytes) 360| 420k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 361| | // 3. adjust pointers 362| 420k| buf += 16; 363| 420k| utf8_output += 16; 364| 420k| continue; // we are done for this round! 365| 420k| } 366| | // no bits set above 7th bit 367| 271k| const __m256i one_byte_bytemask = 368| 271k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 369| 271k| const uint32_t one_byte_bitmask = 370| 271k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 371| | 372| | // no bits set above 11th bit 373| 271k| const __m256i one_or_two_bytes_bytemask = 374| 271k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 375| 271k| const uint32_t one_or_two_bytes_bitmask = 376| 271k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 377| 271k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (377:9): [True: 68.2k, False: 202k] ------------------ 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| 68.2k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 383| 68.2k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 384| | 385| | // t0 = [000a|aaaa|bbbb|bb00] 386| 68.2k| const __m256i t0 = _mm256_slli_epi16(in, 2); 387| | // t1 = [000a|aaaa|0000|0000] 388| 68.2k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 389| | // t2 = [0000|0000|00bb|bbbb] 390| 68.2k| const __m256i t2 = _mm256_and_si256(in, v_003f); 391| | // t3 = [000a|aaaa|00bb|bbbb] 392| 68.2k| const __m256i t3 = _mm256_or_si256(t1, t2); 393| | // t4 = [110a|aaaa|10bb|bbbb] 394| 68.2k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 395| | 396| | // 2. merge ASCII and 2-byte codewords 397| 68.2k| const __m256i utf8_unpacked = 398| 68.2k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 399| | 400| | // 3. prepare bitmask for 8-bit lookup 401| 68.2k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 402| 68.2k| const uint32_t M1 = M0 >> 7; 403| 68.2k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 404| | // 4. pack the bytes 405| | 406| 68.2k| const uint8_t *row = 407| 68.2k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 408| 68.2k| const uint8_t *row_2 = 409| 68.2k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 410| 68.2k| 16)][0]; 411| | 412| 68.2k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 413| 68.2k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 414| | 415| 68.2k| const __m256i utf8_packed = _mm256_shuffle_epi8( 416| 68.2k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 68.2k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 68.2k| _mm256_castsi128_si256(xmm2), 2) ------------------ 417| | // 5. store bytes 418| 68.2k| _mm_storeu_si128((__m128i *)utf8_output, 419| 68.2k| _mm256_castsi256_si128(utf8_packed)); 420| 68.2k| utf8_output += row[0]; 421| 68.2k| _mm_storeu_si128((__m128i *)utf8_output, 422| 68.2k| _mm256_extractf128_si256(utf8_packed, 1)); 423| 68.2k| utf8_output += row_2[0]; 424| | 425| | // 6. adjust pointers 426| 68.2k| buf += 16; 427| 68.2k| continue; 428| 68.2k| } 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| 202k| const __m256i surrogates_bytemask = 433| 202k| _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| 202k| const uint32_t surrogates_bitmask = 438| 202k| 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| 202k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (441:9): [True: 177k, False: 25.0k] ------------------ 442| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 443| 177k| const __m256i dup_even = _mm256_setr_epi16( 444| 177k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 445| 177k| 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| 177k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 475| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 476| 177k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 477| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 478| 177k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 474| 177k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 479| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 480| 177k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 474| 177k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 481| | 482| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 483| 177k| const __m256i s0 = _mm256_srli_epi16(in, 4); 484| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 485| 177k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 474| 177k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 486| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 487| 177k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 474| 177k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 488| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 489| 177k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 474| 177k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 490| 177k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 491| 177k| simdutf_vec(0b0100000000000000)); ------------------ | | 474| 177k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 492| 177k| const __m256i s4 = _mm256_xor_si256(s3, m0); 493| 177k|#undef simdutf_vec 494| | 495| | // 4. expand code units 16-bit => 32-bit 496| 177k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 497| 177k| 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| 177k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 501| 177k| (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| 177k| const uint8_t mask0 = uint8_t(mask); 522| 177k| const uint8_t *row0 = 523| 177k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 524| 177k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 525| 177k| const __m128i utf8_0 = 526| 177k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 527| | 528| 177k| const uint8_t mask1 = static_cast(mask >> 8); 529| 177k| const uint8_t *row1 = 530| 177k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 531| 177k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 532| 177k| const __m128i utf8_1 = 533| 177k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 534| | 535| 177k| const uint8_t mask2 = static_cast(mask >> 16); 536| 177k| const uint8_t *row2 = 537| 177k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 538| 177k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 539| 177k| const __m128i utf8_2 = 540| 177k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 541| | 542| 177k| const uint8_t mask3 = static_cast(mask >> 24); 543| 177k| const uint8_t *row3 = 544| 177k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 545| 177k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 546| 177k| const __m128i utf8_3 = 547| 177k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 548| | 549| 177k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 550| 177k| utf8_output += row0[0]; 551| 177k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 552| 177k| utf8_output += row1[0]; 553| 177k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 554| 177k| utf8_output += row2[0]; 555| 177k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 556| 177k| utf8_output += row3[0]; 557| 177k| buf += 16; 558| | // surrogate pair(s) in a register 559| 177k| } 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| 25.0k| size_t forward = 15; 564| 25.0k| size_t k = 0; 565| 25.0k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (565:11): [True: 0, False: 25.0k] ------------------ 566| 0| forward = size_t(end - buf - 1); 567| 0| } 568| 374k| for (; k < forward; k++) { ------------------ | Branch (568:14): [True: 349k, False: 24.8k] ------------------ 569| 349k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 570| 349k| if ((word & 0xFF80) == 0) { ------------------ | Branch (570:13): [True: 138k, False: 210k] ------------------ 571| 138k| *utf8_output++ = char(word); 572| 210k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (572:20): [True: 15.3k, False: 195k] ------------------ 573| 15.3k| *utf8_output++ = char((word >> 6) | 0b11000000); 574| 15.3k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 575| 195k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (575:20): [True: 169k, False: 25.8k] ------------------ 576| 169k| *utf8_output++ = char((word >> 12) | 0b11100000); 577| 169k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 578| 169k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 579| 169k| } else { 580| | // must be a surrogate pair 581| 25.8k| uint16_t diff = uint16_t(word - 0xD800); 582| 25.8k| uint16_t next_word = 583| 25.8k| scalar::utf16::swap_if_needed(buf[k + 1]); 584| 25.8k| k++; 585| 25.8k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 586| 25.8k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (586:15): [True: 200, False: 25.6k] ------------------ 587| 200| return std::make_pair( 588| 200| result(error_code::SURROGATE, buf - start + k - 1), 589| 200| utf8_output); 590| 200| } 591| 25.6k| uint32_t value = (diff << 10) + diff2 + 0x10000; 592| 25.6k| *utf8_output++ = char((value >> 18) | 0b11110000); 593| 25.6k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 594| 25.6k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 595| 25.6k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 596| 25.6k| } 597| 349k| } 598| 24.8k| buf += k; 599| 24.8k| } 600| 202k| } // while 601| 528| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 602| 728|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_utf32_to_latin1EPKDimPc: 3| 172| char *latin1_output) { 4| 172| const size_t rounded_len = 5| 172| len & ~0x1F; // Round down to nearest multiple of 32 6| | 7| 172| const __m256i high_bytes_mask = _mm256_set1_epi32(0xFFFFFF00); 8| | 9| 716| for (size_t i = 0; i < rounded_len; i += 4 * 8) { ------------------ | Branch (9:22): [True: 558, False: 158] ------------------ 10| 558| __m256i a = _mm256_loadu_si256((__m256i *)(buf + 0 * 8)); 11| 558| __m256i b = _mm256_loadu_si256((__m256i *)(buf + 1 * 8)); 12| 558| __m256i c = _mm256_loadu_si256((__m256i *)(buf + 2 * 8)); 13| 558| __m256i d = _mm256_loadu_si256((__m256i *)(buf + 3 * 8)); 14| | 15| 558| const __m256i check_combined = 16| 558| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 17| | 18| 558| if (!_mm256_testz_si256(check_combined, high_bytes_mask)) { ------------------ | Branch (18:9): [True: 14, False: 544] ------------------ 19| 14| return std::make_pair(nullptr, latin1_output); 20| 14| } 21| | 22| 544| b = _mm256_slli_epi32(b, 1 * 8); 23| 544| c = _mm256_slli_epi32(c, 2 * 8); 24| 544| 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| 544| const __m256i t0 = 35| 544| _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| 544| const __m256i shuffle_bytes = 40| 544| _mm256_setr_epi8(0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15, 41| 544| 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15); 42| | 43| 544| 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| 544| const __m256i shuffle_dwords = _mm256_setr_epi32(0, 4, 1, 5, 2, 6, 3, 7); 48| 544| const __m256i t2 = _mm256_permutevar8x32_epi32(t1, shuffle_dwords); 49| |// clang format on 50| | 51| 544| _mm256_storeu_si256((__m256i *)latin1_output, t2); 52| | 53| 544| latin1_output += 32; 54| 544| buf += 32; 55| 544| } 56| | 57| 158| return std::make_pair(buf, latin1_output); 58| 172|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_140avx2_convert_utf32_to_latin1_with_errorsEPKDimPc: 62| 408| char *latin1_output) { 63| 408| const size_t rounded_len = 64| 408| len & ~0x1F; // Round down to nearest multiple of 32 65| | 66| 408| const char32_t *start = buf; 67| | 68| 408| const __m256i high_bytes_mask = _mm256_set1_epi32(0xFFFFFF00); 69| | 70| 15.0k| for (size_t i = 0; i < rounded_len; i += 4 * 8) { ------------------ | Branch (70:22): [True: 14.7k, False: 300] ------------------ 71| 14.7k| __m256i a = _mm256_loadu_si256((__m256i *)(buf + 0 * 8)); 72| 14.7k| __m256i b = _mm256_loadu_si256((__m256i *)(buf + 1 * 8)); 73| 14.7k| __m256i c = _mm256_loadu_si256((__m256i *)(buf + 2 * 8)); 74| 14.7k| __m256i d = _mm256_loadu_si256((__m256i *)(buf + 3 * 8)); 75| | 76| 14.7k| const __m256i check_combined = 77| 14.7k| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 78| | 79| 14.7k| if (!_mm256_testz_si256(check_combined, high_bytes_mask)) { ------------------ | Branch (79:9): [True: 108, False: 14.6k] ------------------ 80| | // Fallback to scalar code for handling errors 81| 988| for (int k = 0; k < 4 * 8; k++) { ------------------ | Branch (81:23): [True: 988, False: 0] ------------------ 82| 988| char32_t codepoint = buf[k]; 83| 988| if (codepoint <= 0xFF) { ------------------ | Branch (83:13): [True: 880, False: 108] ------------------ 84| 880| *latin1_output++ = static_cast(codepoint); 85| 880| } else { 86| 108| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 87| 108| latin1_output); 88| 108| } 89| 988| } 90| 108| } 91| | 92| 14.6k| b = _mm256_slli_epi32(b, 1 * 8); 93| 14.6k| c = _mm256_slli_epi32(c, 2 * 8); 94| 14.6k| d = _mm256_slli_epi32(d, 3 * 8); 95| | 96| 14.6k| const __m256i t0 = 97| 14.6k| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 98| | 99| 14.6k| const __m256i shuffle_bytes = 100| 14.6k| _mm256_setr_epi8(0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15, 101| 14.6k| 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15); 102| | 103| 14.6k| const __m256i t1 = _mm256_shuffle_epi8(t0, shuffle_bytes); 104| | 105| 14.6k| const __m256i shuffle_dwords = _mm256_setr_epi32(0, 4, 1, 5, 2, 6, 3, 7); 106| 14.6k| const __m256i t2 = _mm256_permutevar8x32_epi32(t1, shuffle_dwords); 107| | 108| 14.6k| _mm256_storeu_si256((__m256i *)latin1_output, t2); 109| | 110| 14.6k| latin1_output += 32; 111| 14.6k| buf += 32; 112| 14.6k| } 113| | 114| 300| return std::make_pair(result(error_code::SUCCESS, buf - start), 115| 300| latin1_output); 116| 408|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf32_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKDiPDsEES7_mS8_: 4| 774| char16_t *utf16_output) { 5| 774| const char32_t *end = buf + len; 6| | 7| 774| const size_t safety_margin = 8| 774| 12; // to avoid overruns, see issue 9| | // https://github.com/simdutf/simdutf/issues/92 10| 774| __m256i forbidden_bytemask = _mm256_setzero_si256(); 11| | 12| 774| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 13| 774| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 14| 774| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 15| | 16| 133k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (16:10): [True: 132k, False: 624] ------------------ 17| 132k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 18| | 19| 132k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 132k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 130k, False: 2.72k] | | ------------------ ------------------ 20| | // no bits set above 16th bit <=> can pack to UTF16 21| | // without surrogate pairs 22| 130k| forbidden_bytemask = _mm256_or_si256( 23| 130k| forbidden_bytemask, 24| 130k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800)); 25| | 26| 130k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 27| 130k| _mm256_extractf128_si256(in, 1)); 28| 130k| if (big_endian) { ------------------ | Branch (28:11): [Folded, False: 130k] ------------------ 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| 130k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 34| 130k| utf16_output += 8; 35| 130k| buf += 8; 36| 130k| } else { 37| 2.72k| size_t forward = 7; 38| 2.72k| size_t k = 0; 39| 2.72k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (39:11): [True: 0, False: 2.72k] ------------------ 40| 0| forward = size_t(end - buf - 1); 41| 0| } 42| 21.0k| for (; k < forward; k++) { ------------------ | Branch (42:14): [True: 18.4k, False: 2.57k] ------------------ 43| 18.4k| uint32_t word = buf[k]; 44| 18.4k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (44:13): [True: 6.09k, False: 12.3k] ------------------ 45| | // will not generate a surrogate pair 46| 6.09k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (46:15): [True: 814, False: 5.28k] | Branch (46:33): [True: 26, False: 788] ------------------ 47| 26| return std::make_pair(nullptr, utf16_output); 48| 26| } 49| 6.07k| *utf16_output++ = 50| 6.07k| big_endian ------------------ | Branch (50:15): [Folded, False: 6.07k] ------------------ 51| 6.07k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 52| 6.07k| : char16_t(word); 53| 12.3k| } else { 54| | // will generate a surrogate pair 55| 12.3k| if (word > 0x10FFFF) { ------------------ | Branch (55:15): [True: 124, False: 12.2k] ------------------ 56| 124| return std::make_pair(nullptr, utf16_output); 57| 124| } 58| 12.2k| word -= 0x10000; 59| 12.2k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 60| 12.2k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 61| 12.2k| if (big_endian) { ------------------ | Branch (61:15): [Folded, False: 12.2k] ------------------ 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| 12.2k| *utf16_output++ = char16_t(high_surrogate); 68| 12.2k| *utf16_output++ = char16_t(low_surrogate); 69| 12.2k| } 70| 18.4k| } 71| 2.57k| buf += k; 72| 2.57k| } 73| 132k| } 74| | 75| | // check for invalid input 76| 624| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (76:7): [True: 42, False: 582] ------------------ 77| 42| return std::make_pair(nullptr, utf16_output); 78| 42| } 79| | 80| 582| return std::make_pair(buf, utf16_output); 81| 624|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf32_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKDiPDsEES7_mS8_: 4| 662| char16_t *utf16_output) { 5| 662| const char32_t *end = buf + len; 6| | 7| 662| const size_t safety_margin = 8| 662| 12; // to avoid overruns, see issue 9| | // https://github.com/simdutf/simdutf/issues/92 10| 662| __m256i forbidden_bytemask = _mm256_setzero_si256(); 11| | 12| 662| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 13| 662| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 14| 662| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 15| | 16| 26.1k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (16:10): [True: 25.6k, False: 482] ------------------ 17| 25.6k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 18| | 19| 25.6k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 25.6k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 22.4k, False: 3.23k] | | ------------------ ------------------ 20| | // no bits set above 16th bit <=> can pack to UTF16 21| | // without surrogate pairs 22| 22.4k| forbidden_bytemask = _mm256_or_si256( 23| 22.4k| forbidden_bytemask, 24| 22.4k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800)); 25| | 26| 22.4k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 27| 22.4k| _mm256_extractf128_si256(in, 1)); 28| 22.4k| if (big_endian) { ------------------ | Branch (28:11): [True: 22.4k, Folded] ------------------ 29| 22.4k| const __m128i swap = 30| 22.4k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 31| 22.4k| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 32| 22.4k| } 33| 22.4k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 34| 22.4k| utf16_output += 8; 35| 22.4k| buf += 8; 36| 22.4k| } else { 37| 3.23k| size_t forward = 7; 38| 3.23k| size_t k = 0; 39| 3.23k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (39:11): [True: 0, False: 3.23k] ------------------ 40| 0| forward = size_t(end - buf - 1); 41| 0| } 42| 24.9k| for (; k < forward; k++) { ------------------ | Branch (42:14): [True: 21.8k, False: 3.05k] ------------------ 43| 21.8k| uint32_t word = buf[k]; 44| 21.8k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (44:13): [True: 2.70k, False: 19.1k] ------------------ 45| | // will not generate a surrogate pair 46| 2.70k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (46:15): [True: 446, False: 2.25k] | Branch (46:33): [True: 20, False: 426] ------------------ 47| 20| return std::make_pair(nullptr, utf16_output); 48| 20| } 49| 2.68k| *utf16_output++ = 50| 2.68k| big_endian ------------------ | Branch (50:15): [True: 2.68k, Folded] ------------------ 51| 2.68k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 52| 2.68k| : char16_t(word); 53| 19.1k| } else { 54| | // will generate a surrogate pair 55| 19.1k| if (word > 0x10FFFF) { ------------------ | Branch (55:15): [True: 160, False: 19.0k] ------------------ 56| 160| return std::make_pair(nullptr, utf16_output); 57| 160| } 58| 19.0k| word -= 0x10000; 59| 19.0k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 60| 19.0k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 61| 19.0k| if (big_endian) { ------------------ | Branch (61:15): [True: 19.0k, Folded] ------------------ 62| 19.0k| high_surrogate = 63| 19.0k| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 64| 19.0k| low_surrogate = 65| 19.0k| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 66| 19.0k| } 67| 19.0k| *utf16_output++ = char16_t(high_surrogate); 68| 19.0k| *utf16_output++ = char16_t(low_surrogate); 69| 19.0k| } 70| 21.8k| } 71| 3.05k| buf += k; 72| 3.05k| } 73| 25.6k| } 74| | 75| | // check for invalid input 76| 482| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (76:7): [True: 26, False: 456] ------------------ 77| 26| return std::make_pair(nullptr, utf16_output); 78| 26| } 79| | 80| 456| return std::make_pair(buf, utf16_output); 81| 482|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf32_to_utf16_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 86| 684| char16_t *utf16_output) { 87| 684| const char32_t *start = buf; 88| 684| const char32_t *end = buf + len; 89| | 90| 684| const size_t safety_margin = 91| 684| 12; // to avoid overruns, see issue 92| | // https://github.com/simdutf/simdutf/issues/92 93| | 94| 684| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 95| 684| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 96| 684| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 97| | 98| 55.8k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (98:10): [True: 55.3k, False: 514] ------------------ 99| 55.3k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 100| | 101| 55.3k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 55.3k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 53.4k, False: 1.92k] | | ------------------ ------------------ 102| | // no bits set above 16th bit <=> can pack to UTF16 without surrogate 103| | // pairs 104| 53.4k| const __m256i forbidden_bytemask = 105| 53.4k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800); 106| 53.4k| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != ------------------ | Branch (106:11): [True: 30, False: 53.3k] ------------------ 107| 53.4k| 0x0) { 108| 30| return std::make_pair(result(error_code::SURROGATE, buf - start), 109| 30| utf16_output); 110| 30| } 111| | 112| 53.3k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 113| 53.3k| _mm256_extractf128_si256(in, 1)); 114| 53.3k| if (big_endian) { ------------------ | Branch (114:11): [Folded, False: 53.3k] ------------------ 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| 53.3k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 120| 53.3k| utf16_output += 8; 121| 53.3k| buf += 8; 122| 53.3k| } else { 123| 1.92k| size_t forward = 7; 124| 1.92k| size_t k = 0; 125| 1.92k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (125:11): [True: 0, False: 1.92k] ------------------ 126| 0| forward = size_t(end - buf - 1); 127| 0| } 128| 14.6k| for (; k < forward; k++) { ------------------ | Branch (128:14): [True: 12.9k, False: 1.78k] ------------------ 129| 12.9k| uint32_t word = buf[k]; 130| 12.9k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (130:13): [True: 5.98k, False: 6.92k] ------------------ 131| | // will not generate a surrogate pair 132| 5.98k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (132:15): [True: 542, False: 5.44k] | Branch (132:33): [True: 4, False: 538] ------------------ 133| 4| return std::make_pair( 134| 4| result(error_code::SURROGATE, buf - start + k), utf16_output); 135| 4| } 136| 5.98k| *utf16_output++ = 137| 5.98k| big_endian ------------------ | Branch (137:15): [Folded, False: 5.98k] ------------------ 138| 5.98k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 139| 5.98k| : char16_t(word); 140| 6.92k| } else { 141| | // will generate a surrogate pair 142| 6.92k| if (word > 0x10FFFF) { ------------------ | Branch (142:15): [True: 136, False: 6.79k] ------------------ 143| 136| return std::make_pair( 144| 136| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 145| 136| } 146| 6.79k| word -= 0x10000; 147| 6.79k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 148| 6.79k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 149| 6.79k| if (big_endian) { ------------------ | Branch (149:15): [Folded, False: 6.79k] ------------------ 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| 6.79k| *utf16_output++ = char16_t(high_surrogate); 156| 6.79k| *utf16_output++ = char16_t(low_surrogate); 157| 6.79k| } 158| 12.9k| } 159| 1.78k| buf += k; 160| 1.78k| } 161| 55.3k| } 162| | 163| 514| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 164| 684|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf32_to_utf16_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 86| 630| char16_t *utf16_output) { 87| 630| const char32_t *start = buf; 88| 630| const char32_t *end = buf + len; 89| | 90| 630| const size_t safety_margin = 91| 630| 12; // to avoid overruns, see issue 92| | // https://github.com/simdutf/simdutf/issues/92 93| | 94| 630| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 95| 630| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 96| 630| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 97| | 98| 3.55k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (98:10): [True: 3.05k, False: 500] ------------------ 99| 3.05k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 100| | 101| 3.05k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 3.05k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 2.31k, False: 734] | | ------------------ ------------------ 102| | // no bits set above 16th bit <=> can pack to UTF16 without surrogate 103| | // pairs 104| 2.31k| const __m256i forbidden_bytemask = 105| 2.31k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800); 106| 2.31k| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != ------------------ | Branch (106:11): [True: 6, False: 2.31k] ------------------ 107| 2.31k| 0x0) { 108| 6| return std::make_pair(result(error_code::SURROGATE, buf - start), 109| 6| utf16_output); 110| 6| } 111| | 112| 2.31k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 113| 2.31k| _mm256_extractf128_si256(in, 1)); 114| 2.31k| if (big_endian) { ------------------ | Branch (114:11): [True: 2.31k, Folded] ------------------ 115| 2.31k| const __m128i swap = 116| 2.31k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 117| 2.31k| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 118| 2.31k| } 119| 2.31k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 120| 2.31k| utf16_output += 8; 121| 2.31k| buf += 8; 122| 2.31k| } else { 123| 734| size_t forward = 7; 124| 734| size_t k = 0; 125| 734| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (125:11): [True: 0, False: 734] ------------------ 126| 0| forward = size_t(end - buf - 1); 127| 0| } 128| 5.25k| for (; k < forward; k++) { ------------------ | Branch (128:14): [True: 4.64k, False: 610] ------------------ 129| 4.64k| uint32_t word = buf[k]; 130| 4.64k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (130:13): [True: 2.31k, False: 2.33k] ------------------ 131| | // will not generate a surrogate pair 132| 2.31k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (132:15): [True: 554, False: 1.75k] | Branch (132:33): [True: 6, False: 548] ------------------ 133| 6| return std::make_pair( 134| 6| result(error_code::SURROGATE, buf - start + k), utf16_output); 135| 6| } 136| 2.30k| *utf16_output++ = 137| 2.30k| big_endian ------------------ | Branch (137:15): [True: 2.30k, Folded] ------------------ 138| 2.30k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 139| 2.30k| : char16_t(word); 140| 2.33k| } else { 141| | // will generate a surrogate pair 142| 2.33k| if (word > 0x10FFFF) { ------------------ | Branch (142:15): [True: 118, False: 2.21k] ------------------ 143| 118| return std::make_pair( 144| 118| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 145| 118| } 146| 2.21k| word -= 0x10000; 147| 2.21k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 148| 2.21k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 149| 2.21k| if (big_endian) { ------------------ | Branch (149:15): [True: 2.21k, Folded] ------------------ 150| 2.21k| high_surrogate = 151| 2.21k| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 152| 2.21k| low_surrogate = 153| 2.21k| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 154| 2.21k| } 155| 2.21k| *utf16_output++ = char16_t(high_surrogate); 156| 2.21k| *utf16_output++ = char16_t(low_surrogate); 157| 2.21k| } 158| 4.64k| } 159| 610| buf += k; 160| 610| } 161| 3.05k| } 162| | 163| 500| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 164| 630|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_126avx2_convert_utf32_to_utf8EPKDimPc: 2| 916|avx2_convert_utf32_to_utf8(const char32_t *buf, size_t len, char *utf8_output) { 3| 916| const char32_t *end = buf + len; 4| 916| const __m256i v_0000 = _mm256_setzero_si256(); 5| 916| const __m256i v_ffff0000 = _mm256_set1_epi32((uint32_t)0xffff0000); 6| 916| const __m256i v_ff80 = _mm256_set1_epi16((uint16_t)0xff80); 7| 916| const __m256i v_f800 = _mm256_set1_epi16((uint16_t)0xf800); 8| 916| const __m256i v_c080 = _mm256_set1_epi16((uint16_t)0xc080); 9| 916| const __m256i v_7fffffff = _mm256_set1_epi32((uint32_t)0x7fffffff); 10| 916| __m256i running_max = _mm256_setzero_si256(); 11| 916| __m256i forbidden_bytemask = _mm256_setzero_si256(); 12| | 13| 916| const size_t safety_margin = 14| 916| 12; // to avoid overruns, see issue 15| | // https://github.com/simdutf/simdutf/issues/92 16| | 17| 30.6k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (17:10): [True: 29.9k, False: 686] ------------------ 18| 29.9k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 19| 29.9k| __m256i nextin = _mm256_loadu_si256((__m256i *)buf + 1); 20| 29.9k| 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| 29.9k| __m256i in_16 = _mm256_packus_epi32(_mm256_and_si256(in, v_7fffffff), 25| 29.9k| _mm256_and_si256(nextin, v_7fffffff)); 26| 29.9k| 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| 29.9k| if (_mm256_testz_si256(in_16, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (31:9): [True: 24.7k, False: 5.19k] ------------------ 32| | // 1. pack the bytes 33| 24.7k| const __m128i utf8_packed = _mm_packus_epi16( 34| 24.7k| _mm256_castsi256_si128(in_16), _mm256_extractf128_si256(in_16, 1)); 35| | // 2. store (16 bytes) 36| 24.7k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 37| | // 3. adjust pointers 38| 24.7k| buf += 16; 39| 24.7k| utf8_output += 16; 40| 24.7k| continue; // we are done for this round! 41| 24.7k| } 42| | // no bits set above 7th bit 43| 5.19k| const __m256i one_byte_bytemask = 44| 5.19k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_ff80), v_0000); 45| 5.19k| const uint32_t one_byte_bitmask = 46| 5.19k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 47| | 48| | // no bits set above 11th bit 49| 5.19k| const __m256i one_or_two_bytes_bytemask = 50| 5.19k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_0000); 51| 5.19k| const uint32_t one_or_two_bytes_bitmask = 52| 5.19k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 53| 5.19k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (53:9): [True: 2.02k, False: 3.17k] ------------------ 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| 2.02k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 58| 2.02k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 59| | 60| | // t0 = [000a|aaaa|bbbb|bb00] 61| 2.02k| const __m256i t0 = _mm256_slli_epi16(in_16, 2); 62| | // t1 = [000a|aaaa|0000|0000] 63| 2.02k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 64| | // t2 = [0000|0000|00bb|bbbb] 65| 2.02k| const __m256i t2 = _mm256_and_si256(in_16, v_003f); 66| | // t3 = [000a|aaaa|00bb|bbbb] 67| 2.02k| const __m256i t3 = _mm256_or_si256(t1, t2); 68| | // t4 = [110a|aaaa|10bb|bbbb] 69| 2.02k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 70| | 71| | // 2. merge ASCII and 2-byte codewords 72| 2.02k| const __m256i utf8_unpacked = 73| 2.02k| _mm256_blendv_epi8(t4, in_16, one_byte_bytemask); 74| | 75| | // 3. prepare bitmask for 8-bit lookup 76| 2.02k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 77| 2.02k| const uint32_t M1 = M0 >> 7; 78| 2.02k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 79| | // 4. pack the bytes 80| | 81| 2.02k| const uint8_t *row = 82| 2.02k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 83| 2.02k| const uint8_t *row_2 = 84| 2.02k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 85| 2.02k| 16)][0]; 86| | 87| 2.02k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 88| 2.02k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 89| | 90| 2.02k| const __m256i utf8_packed = _mm256_shuffle_epi8( 91| 2.02k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 2.02k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 2.02k| _mm256_castsi128_si256(xmm2), 2) ------------------ 92| | // 5. store bytes 93| 2.02k| _mm_storeu_si128((__m128i *)utf8_output, 94| 2.02k| _mm256_castsi256_si128(utf8_packed)); 95| 2.02k| utf8_output += row[0]; 96| 2.02k| _mm_storeu_si128((__m128i *)utf8_output, 97| 2.02k| _mm256_extractf128_si256(utf8_packed, 1)); 98| 2.02k| utf8_output += row_2[0]; 99| | 100| | // 6. adjust pointers 101| 2.02k| buf += 16; 102| 2.02k| continue; 103| 2.02k| } 104| | // Must check for overflow in packing 105| 3.17k| const __m256i saturation_bytemask = _mm256_cmpeq_epi32( 106| 3.17k| _mm256_and_si256(_mm256_or_si256(in, nextin), v_ffff0000), v_0000); 107| 3.17k| const uint32_t saturation_bitmask = 108| 3.17k| static_cast(_mm256_movemask_epi8(saturation_bytemask)); 109| 3.17k| if (saturation_bitmask == 0xffffffff) { ------------------ | Branch (109:9): [True: 1.12k, False: 2.04k] ------------------ 110| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 111| 1.12k| const __m256i v_d800 = _mm256_set1_epi16((uint16_t)0xd800); 112| 1.12k| forbidden_bytemask = _mm256_or_si256( 113| 1.12k| forbidden_bytemask, 114| 1.12k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_d800)); 115| | 116| 1.12k| const __m256i dup_even = _mm256_setr_epi16( 117| 1.12k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 118| 1.12k| 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| 1.12k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 148| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 149| 1.12k| const __m256i t0 = _mm256_shuffle_epi8(in_16, dup_even); 150| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 151| 1.12k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 147| 1.12k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 152| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 153| 1.12k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 147| 1.12k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 154| | 155| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 156| 1.12k| const __m256i s0 = _mm256_srli_epi16(in_16, 4); 157| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 158| 1.12k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 147| 1.12k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 159| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 160| 1.12k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 147| 1.12k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 161| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 162| 1.12k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 147| 1.12k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 163| 1.12k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 164| 1.12k| simdutf_vec(0b0100000000000000)); ------------------ | | 147| 1.12k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 165| 1.12k| const __m256i s4 = _mm256_xor_si256(s3, m0); 166| 1.12k|#undef simdutf_vec 167| | 168| | // 4. expand code units 16-bit => 32-bit 169| 1.12k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 170| 1.12k| 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| 1.12k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 174| 1.12k| (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| 1.12k| const uint8_t mask0 = uint8_t(mask); 195| 1.12k| const uint8_t *row0 = 196| 1.12k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 197| 1.12k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 198| 1.12k| const __m128i utf8_0 = 199| 1.12k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 200| | 201| 1.12k| const uint8_t mask1 = static_cast(mask >> 8); 202| 1.12k| const uint8_t *row1 = 203| 1.12k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 204| 1.12k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 205| 1.12k| const __m128i utf8_1 = 206| 1.12k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 207| | 208| 1.12k| const uint8_t mask2 = static_cast(mask >> 16); 209| 1.12k| const uint8_t *row2 = 210| 1.12k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 211| 1.12k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 212| 1.12k| const __m128i utf8_2 = 213| 1.12k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 214| | 215| 1.12k| const uint8_t mask3 = static_cast(mask >> 24); 216| 1.12k| const uint8_t *row3 = 217| 1.12k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 218| 1.12k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 219| 1.12k| const __m128i utf8_3 = 220| 1.12k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 221| | 222| 1.12k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 223| 1.12k| utf8_output += row0[0]; 224| 1.12k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 225| 1.12k| utf8_output += row1[0]; 226| 1.12k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 227| 1.12k| utf8_output += row2[0]; 228| 1.12k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 229| 1.12k| utf8_output += row3[0]; 230| 1.12k| buf += 16; 231| 2.04k| } 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| 2.04k| size_t forward = 15; 237| 2.04k| size_t k = 0; 238| 2.04k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (238:11): [True: 0, False: 2.04k] ------------------ 239| 0| forward = size_t(end - buf - 1); 240| 0| } 241| 30.1k| for (; k < forward; k++) { ------------------ | Branch (241:14): [True: 28.3k, False: 1.81k] ------------------ 242| 28.3k| uint32_t word = buf[k]; 243| 28.3k| if ((word & 0xFFFFFF80) == 0) { // 1-byte (ASCII) ------------------ | Branch (243:13): [True: 18.4k, False: 9.86k] ------------------ 244| 18.4k| *utf8_output++ = char(word); 245| 18.4k| } else if ((word & 0xFFFFF800) == 0) { // 2-byte ------------------ | Branch (245:20): [True: 1.41k, False: 8.45k] ------------------ 246| 1.41k| *utf8_output++ = char((word >> 6) | 0b11000000); 247| 1.41k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 248| 8.45k| } else if ((word & 0xFFFF0000) == 0) { // 3-byte ------------------ | Branch (248:20): [True: 3.52k, False: 4.93k] ------------------ 249| 3.52k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (249:15): [True: 1.24k, False: 2.27k] | Branch (249:33): [True: 32, False: 1.21k] ------------------ 250| 32| return std::make_pair(nullptr, utf8_output); 251| 32| } 252| 3.49k| *utf8_output++ = char((word >> 12) | 0b11100000); 253| 3.49k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 254| 3.49k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 255| 4.93k| } else { // 4-byte 256| 4.93k| if (word > 0x10FFFF) { ------------------ | Branch (256:15): [True: 198, False: 4.73k] ------------------ 257| 198| return std::make_pair(nullptr, utf8_output); 258| 198| } 259| 4.73k| *utf8_output++ = char((word >> 18) | 0b11110000); 260| 4.73k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 261| 4.73k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 262| 4.73k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 263| 4.73k| } 264| 28.3k| } 265| 1.81k| buf += k; 266| 1.81k| } 267| 3.17k| } // while 268| | 269| | // check for invalid input 270| 686| const __m256i v_10ffff = _mm256_set1_epi32((uint32_t)0x10ffff); 271| 686| if (static_cast(_mm256_movemask_epi8(_mm256_cmpeq_epi32( ------------------ | Branch (271:7): [True: 36, False: 650] ------------------ 272| 686| _mm256_max_epu32(running_max, v_10ffff), v_10ffff))) != 0xffffffff) { 273| 36| return std::make_pair(nullptr, utf8_output); 274| 36| } 275| | 276| 650| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (276:7): [True: 72, False: 578] ------------------ 277| 72| return std::make_pair(nullptr, utf8_output); 278| 72| } 279| | 280| 578| return std::make_pair(buf, utf8_output); 281| 650|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_138avx2_convert_utf32_to_utf8_with_errorsEPKDimPc: 285| 934| char *utf8_output) { 286| 934| const char32_t *end = buf + len; 287| 934| const char32_t *start = buf; 288| | 289| 934| const __m256i v_0000 = _mm256_setzero_si256(); 290| 934| const __m256i v_ffff0000 = _mm256_set1_epi32((uint32_t)0xffff0000); 291| 934| const __m256i v_ff80 = _mm256_set1_epi16((uint16_t)0xff80); 292| 934| const __m256i v_f800 = _mm256_set1_epi16((uint16_t)0xf800); 293| 934| const __m256i v_c080 = _mm256_set1_epi16((uint16_t)0xc080); 294| 934| const __m256i v_7fffffff = _mm256_set1_epi32((uint32_t)0x7fffffff); 295| 934| const __m256i v_10ffff = _mm256_set1_epi32((uint32_t)0x10ffff); 296| | 297| 934| const size_t safety_margin = 298| 934| 12; // to avoid overruns, see issue 299| | // https://github.com/simdutf/simdutf/issues/92 300| | 301| 49.7k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (301:10): [True: 49.0k, False: 702] ------------------ 302| 49.0k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 303| 49.0k| __m256i nextin = _mm256_loadu_si256((__m256i *)buf + 1); 304| | // Check for too large input 305| 49.0k| const __m256i max_input = 306| 49.0k| _mm256_max_epu32(_mm256_max_epu32(in, nextin), v_10ffff); 307| 49.0k| if (static_cast(_mm256_movemask_epi8( ------------------ | Branch (307:9): [True: 148, False: 48.8k] ------------------ 308| 49.0k| _mm256_cmpeq_epi32(max_input, v_10ffff))) != 0xffffffff) { 309| 148| return std::make_pair(result(error_code::TOO_LARGE, buf - start), 310| 148| utf8_output); 311| 148| } 312| | 313| | // Pack 32-bit UTF-32 code units to 16-bit UTF-16 code units with unsigned 314| | // saturation 315| 48.8k| __m256i in_16 = _mm256_packus_epi32(_mm256_and_si256(in, v_7fffffff), 316| 48.8k| _mm256_and_si256(nextin, v_7fffffff)); 317| 48.8k| 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| 48.8k| if (_mm256_testz_si256(in_16, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (322:9): [True: 44.1k, False: 4.75k] ------------------ 323| | // 1. pack the bytes 324| 44.1k| const __m128i utf8_packed = _mm_packus_epi16( 325| 44.1k| _mm256_castsi256_si128(in_16), _mm256_extractf128_si256(in_16, 1)); 326| | // 2. store (16 bytes) 327| 44.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 328| | // 3. adjust pointers 329| 44.1k| buf += 16; 330| 44.1k| utf8_output += 16; 331| 44.1k| continue; // we are done for this round! 332| 44.1k| } 333| | // no bits set above 7th bit 334| 4.75k| const __m256i one_byte_bytemask = 335| 4.75k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_ff80), v_0000); 336| 4.75k| const uint32_t one_byte_bitmask = 337| 4.75k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 338| | 339| | // no bits set above 11th bit 340| 4.75k| const __m256i one_or_two_bytes_bytemask = 341| 4.75k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_0000); 342| 4.75k| const uint32_t one_or_two_bytes_bitmask = 343| 4.75k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 344| 4.75k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (344:9): [True: 2.35k, False: 2.40k] ------------------ 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| 2.35k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 349| 2.35k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 350| | 351| | // t0 = [000a|aaaa|bbbb|bb00] 352| 2.35k| const __m256i t0 = _mm256_slli_epi16(in_16, 2); 353| | // t1 = [000a|aaaa|0000|0000] 354| 2.35k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 355| | // t2 = [0000|0000|00bb|bbbb] 356| 2.35k| const __m256i t2 = _mm256_and_si256(in_16, v_003f); 357| | // t3 = [000a|aaaa|00bb|bbbb] 358| 2.35k| const __m256i t3 = _mm256_or_si256(t1, t2); 359| | // t4 = [110a|aaaa|10bb|bbbb] 360| 2.35k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 361| | 362| | // 2. merge ASCII and 2-byte codewords 363| 2.35k| const __m256i utf8_unpacked = 364| 2.35k| _mm256_blendv_epi8(t4, in_16, one_byte_bytemask); 365| | 366| | // 3. prepare bitmask for 8-bit lookup 367| 2.35k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 368| 2.35k| const uint32_t M1 = M0 >> 7; 369| 2.35k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 370| | // 4. pack the bytes 371| | 372| 2.35k| const uint8_t *row = 373| 2.35k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 374| 2.35k| const uint8_t *row_2 = 375| 2.35k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 376| 2.35k| 16)][0]; 377| | 378| 2.35k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 379| 2.35k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 380| | 381| 2.35k| const __m256i utf8_packed = _mm256_shuffle_epi8( 382| 2.35k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 2.35k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 2.35k| _mm256_castsi128_si256(xmm2), 2) ------------------ 383| | // 5. store bytes 384| 2.35k| _mm_storeu_si128((__m128i *)utf8_output, 385| 2.35k| _mm256_castsi256_si128(utf8_packed)); 386| 2.35k| utf8_output += row[0]; 387| 2.35k| _mm_storeu_si128((__m128i *)utf8_output, 388| 2.35k| _mm256_extractf128_si256(utf8_packed, 1)); 389| 2.35k| utf8_output += row_2[0]; 390| | 391| | // 6. adjust pointers 392| 2.35k| buf += 16; 393| 2.35k| continue; 394| 2.35k| } 395| | // Must check for overflow in packing 396| 2.40k| const __m256i saturation_bytemask = _mm256_cmpeq_epi32( 397| 2.40k| _mm256_and_si256(_mm256_or_si256(in, nextin), v_ffff0000), v_0000); 398| 2.40k| const uint32_t saturation_bitmask = 399| 2.40k| static_cast(_mm256_movemask_epi8(saturation_bytemask)); 400| 2.40k| if (saturation_bitmask == 0xffffffff) { ------------------ | Branch (400:9): [True: 904, False: 1.49k] ------------------ 401| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 402| | 403| | // Check for illegal surrogate code units 404| 904| const __m256i v_d800 = _mm256_set1_epi16((uint16_t)0xd800); 405| 904| const __m256i forbidden_bytemask = 406| 904| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_d800); 407| 904| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != ------------------ | Branch (407:11): [True: 54, False: 850] ------------------ 408| 904| 0x0) { 409| 54| return std::make_pair(result(error_code::SURROGATE, buf - start), 410| 54| utf8_output); 411| 54| } 412| | 413| 850| const __m256i dup_even = _mm256_setr_epi16( 414| 850| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 415| 850| 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| 850|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 445| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 446| 850| const __m256i t0 = _mm256_shuffle_epi8(in_16, dup_even); 447| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 448| 850| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 444| 850|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 449| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 450| 850| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 444| 850|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 451| | 452| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 453| 850| const __m256i s0 = _mm256_srli_epi16(in_16, 4); 454| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 455| 850| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 444| 850|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 456| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 457| 850| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 444| 850|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 458| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 459| 850| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 444| 850|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 460| 850| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 461| 850| simdutf_vec(0b0100000000000000)); ------------------ | | 444| 850|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 462| 850| const __m256i s4 = _mm256_xor_si256(s3, m0); 463| 850|#undef simdutf_vec 464| | 465| | // 4. expand code units 16-bit => 32-bit 466| 850| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 467| 850| 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| 850| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 471| 850| (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| 850| const uint8_t mask0 = uint8_t(mask); 492| 850| const uint8_t *row0 = 493| 850| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 494| 850| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 495| 850| const __m128i utf8_0 = 496| 850| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 497| | 498| 850| const uint8_t mask1 = static_cast(mask >> 8); 499| 850| const uint8_t *row1 = 500| 850| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 501| 850| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 502| 850| const __m128i utf8_1 = 503| 850| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 504| | 505| 850| const uint8_t mask2 = static_cast(mask >> 16); 506| 850| const uint8_t *row2 = 507| 850| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 508| 850| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 509| 850| const __m128i utf8_2 = 510| 850| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 511| | 512| 850| const uint8_t mask3 = static_cast(mask >> 24); 513| 850| const uint8_t *row3 = 514| 850| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 515| 850| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 516| 850| const __m128i utf8_3 = 517| 850| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 518| | 519| 850| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 520| 850| utf8_output += row0[0]; 521| 850| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 522| 850| utf8_output += row1[0]; 523| 850| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 524| 850| utf8_output += row2[0]; 525| 850| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 526| 850| utf8_output += row3[0]; 527| 850| buf += 16; 528| 1.49k| } 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| 1.49k| size_t forward = 15; 534| 1.49k| size_t k = 0; 535| 1.49k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (535:11): [True: 0, False: 1.49k] ------------------ 536| 0| forward = size_t(end - buf - 1); 537| 0| } 538| 23.6k| for (; k < forward; k++) { ------------------ | Branch (538:14): [True: 22.2k, False: 1.46k] ------------------ 539| 22.2k| uint32_t word = buf[k]; 540| 22.2k| if ((word & 0xFFFFFF80) == 0) { // 1-byte (ASCII) ------------------ | Branch (540:13): [True: 9.01k, False: 13.2k] ------------------ 541| 9.01k| *utf8_output++ = char(word); 542| 13.2k| } else if ((word & 0xFFFFF800) == 0) { // 2-byte ------------------ | Branch (542:20): [True: 1.81k, False: 11.4k] ------------------ 543| 1.81k| *utf8_output++ = char((word >> 6) | 0b11000000); 544| 1.81k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 545| 11.4k| } else if ((word & 0xFFFF0000) == 0) { // 3-byte ------------------ | Branch (545:20): [True: 6.31k, False: 5.09k] ------------------ 546| 6.31k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (546:15): [True: 3.59k, False: 2.71k] | Branch (546:33): [True: 30, False: 3.56k] ------------------ 547| 30| return std::make_pair( 548| 30| result(error_code::SURROGATE, buf - start + k), utf8_output); 549| 30| } 550| 6.28k| *utf8_output++ = char((word >> 12) | 0b11100000); 551| 6.28k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 552| 6.28k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 553| 6.28k| } else { // 4-byte 554| 5.09k| if (word > 0x10FFFF) { ------------------ | Branch (554:15): [True: 0, False: 5.09k] ------------------ 555| 0| return std::make_pair( 556| 0| result(error_code::TOO_LARGE, buf - start + k), utf8_output); 557| 0| } 558| 5.09k| *utf8_output++ = char((word >> 18) | 0b11110000); 559| 5.09k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 560| 5.09k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 561| 5.09k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 562| 5.09k| } 563| 22.2k| } 564| 1.46k| buf += k; 565| 1.46k| } 566| 2.40k| } // while 567| | 568| 702| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 569| 934|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_129convert_masked_utf8_to_latin1EPKcmRPc: 9| 605k| 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| 605k| const __m128i in = _mm_loadu_si128((__m128i *)input); 21| | 22| 605k| const uint16_t input_utf8_end_of_code_point_mask = 23| 605k| utf8_end_of_code_point_mask & 24| 605k| 0xfff; // we are only processing 12 bytes in case it is not all ASCII 25| | 26| 605k| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (26:7): [True: 2.94k, False: 602k] ------------------ 27| | // We process the data in chunks of 12 bytes. 28| 2.94k| _mm_storeu_si128(reinterpret_cast<__m128i *>(latin1_output), in); 29| 2.94k| latin1_output += 12; // We wrote 12 characters. 30| 2.94k| return 12; // We consumed 1 bytes. 31| 2.94k| } 32| | /// We do not have a fast path available, so we fallback. 33| 602k| const uint8_t idx = 34| 602k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 35| 602k| const uint8_t consumed = 36| 602k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 37| | // this indicates an invalid input: 38| 602k| if (idx >= 64) { ------------------ | Branch (38:7): [True: 32.7k, False: 569k] ------------------ 39| 32.7k| return consumed; 40| 32.7k| } 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| 569k| const __m128i sh = 48| 569k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 49| 569k| const __m128i perm = _mm_shuffle_epi8(in, sh); 50| 569k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 51| 569k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 52| 569k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 53| 569k| 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| 569k| _mm_storel_epi64((__m128i *)latin1_output, latin1_packed); 58| 569k| latin1_output += 6; // We wrote 6 bytes. 59| 569k| return consumed; 60| 602k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE0EEEmPKcmRPDs: 10| 3.02M| 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| 3.02M| const __m128i swap = 23| 3.02M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 3.02M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 3.02M| const uint16_t input_utf8_end_of_code_point_mask = 26| 3.02M| utf8_end_of_code_point_mask & 0xfff; 27| 3.02M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 10.8k, False: 3.01M] ------------------ 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| 3.01M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (40:7): [True: 14.8k, False: 2.99M] ------------------ 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| 14.8k| const __m128i sh = 45| 14.8k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 46| 14.8k| const __m128i perm = _mm_shuffle_epi8(in, sh); 47| 14.8k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 48| 14.8k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 49| 14.8k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 50| 14.8k| if (big_endian) ------------------ | Branch (50:9): [Folded, False: 14.8k] ------------------ 51| 0| composed = _mm_shuffle_epi8(composed, swap); 52| 14.8k| _mm_storeu_si128((__m128i *)utf16_output, composed); 53| 14.8k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 54| 14.8k| return 16; 55| 14.8k| } 56| 2.99M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (56:7): [True: 2.78k, False: 2.99M] ------------------ 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| 2.78k| const __m128i sh = 61| 2.78k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 62| 2.78k| const __m128i perm = _mm_shuffle_epi8(in, sh); 63| 2.78k| const __m128i ascii = 64| 2.78k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 65| 2.78k| const __m128i middlebyte = 66| 2.78k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 67| 2.78k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 68| 2.78k| const __m128i highbyte = 69| 2.78k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 70| 2.78k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 71| 2.78k| const __m128i composed = 72| 2.78k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 73| 2.78k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 74| 2.78k| if (big_endian) ------------------ | Branch (74:9): [Folded, False: 2.78k] ------------------ 75| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 76| 2.78k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 77| 2.78k| utf16_output += 4; 78| 2.78k| return 12; 79| 2.78k| } 80| | 81| 2.99M| const uint8_t idx = simdutf::tables::utf8_to_utf16::utf8bigindex 82| 2.99M| [input_utf8_end_of_code_point_mask][0]; 83| 2.99M| const uint8_t consumed = simdutf::tables::utf8_to_utf16::utf8bigindex 84| 2.99M| [input_utf8_end_of_code_point_mask][1]; 85| 2.99M| if (idx < 64) { ------------------ | Branch (85:7): [True: 2.54M, False: 443k] ------------------ 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| 2.54M| const __m128i sh = _mm_loadu_si128( 93| 2.54M| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 94| 2.54M| const __m128i perm = _mm_shuffle_epi8(in, sh); 95| 2.54M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 96| 2.54M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 97| 2.54M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 98| 2.54M| if (big_endian) ------------------ | Branch (98:9): [Folded, False: 2.54M] ------------------ 99| 0| composed = _mm_shuffle_epi8(composed, swap); 100| 2.54M| _mm_storeu_si128((__m128i *)utf16_output, composed); 101| 2.54M| utf16_output += 6; // We wrote 12 bytes, 6 code points. There is a potential 102| | // overflow of 4 bytes. 103| 2.54M| } else if (idx < 145) { ------------------ | Branch (103:14): [True: 369k, False: 73.8k] ------------------ 104| | // FOUR (4) input code-code units 105| 369k| const __m128i sh = _mm_loadu_si128( 106| 369k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 107| 369k| const __m128i perm = _mm_shuffle_epi8(in, sh); 108| 369k| const __m128i ascii = 109| 369k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 110| 369k| const __m128i middlebyte = 111| 369k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 112| 369k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 113| 369k| const __m128i highbyte = 114| 369k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 115| 369k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 116| 369k| const __m128i composed = 117| 369k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 118| 369k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 119| 369k| if (big_endian) ------------------ | Branch (119:9): [Folded, False: 369k] ------------------ 120| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 121| 369k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 122| 369k| utf16_output += 4; // Here we overflow by 8 bytes. 123| 369k| } else if (idx < 209) { ------------------ | Branch (123:14): [True: 72.5k, False: 1.29k] ------------------ 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| 72.5k| const __m128i sh = _mm_loadu_si128( 135| 72.5k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 136| 72.5k| const __m128i perm = _mm_shuffle_epi8(in, sh); 137| 72.5k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 138| 72.5k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 139| 72.5k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 140| 72.5k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 141| | // correct for spurious high bit 142| 72.5k| const __m128i correct = 143| 72.5k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 144| 72.5k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 145| 72.5k| 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| 72.5k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 149| 72.5k| 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| 72.5k| const __m128i composed = 155| 72.5k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 156| 72.5k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 157| 72.5k| const __m128i composedminus = 158| 72.5k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 159| 72.5k| const __m128i lowtenbits = 160| 72.5k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 161| | // Notice the 0x3ff mask: 162| 72.5k| const __m128i hightenbits = 163| 72.5k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 164| 72.5k| const __m128i lowtenbitsadd = 165| 72.5k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 166| 72.5k| const __m128i hightenbitsadd = 167| 72.5k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 168| 72.5k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 169| 72.5k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 170| 72.5k| uint32_t basic_buffer[4]; 171| 72.5k| uint32_t basic_buffer_swap[4]; 172| 72.5k| if (big_endian) { ------------------ | Branch (172:9): [Folded, False: 72.5k] ------------------ 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| 72.5k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 178| 72.5k| uint32_t surrogate_buffer[4]; 179| 72.5k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 180| 290k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (180:24): [True: 217k, False: 72.5k] ------------------ 181| 217k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (181:11): [True: 59.9k, False: 157k] ------------------ 182| 59.9k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 183| 59.9k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 184| 59.9k| utf16_output += 2; 185| 157k| } else { 186| 157k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (186:27): [Folded, False: 157k] ------------------ 187| 157k| : uint16_t(basic_buffer[i]); 188| 157k| utf16_output++; 189| 157k| } 190| 217k| } 191| 72.5k| } else { 192| | // here we know that there is an error but we do not handle errors 193| 1.29k| } 194| 2.99M| return consumed; 195| 2.99M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE1EEEmPKcmRPDs: 10| 3.04M| 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| 3.04M| const __m128i swap = 23| 3.04M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 3.04M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 3.04M| const uint16_t input_utf8_end_of_code_point_mask = 26| 3.04M| utf8_end_of_code_point_mask & 0xfff; 27| 3.04M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 10.8k, False: 3.03M] ------------------ 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): [True: 10.8k, Folded] ------------------ 31| 10.8k| const __m256i swap256 = _mm256_setr_epi8( 32| 10.8k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 33| 10.8k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 34| 10.8k| ascii = _mm256_shuffle_epi8(ascii, swap256); 35| 10.8k| } 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| 3.03M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (40:7): [True: 27.5k, False: 3.00M] ------------------ 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| 27.5k| const __m128i sh = 45| 27.5k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 46| 27.5k| const __m128i perm = _mm_shuffle_epi8(in, sh); 47| 27.5k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 48| 27.5k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 49| 27.5k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 50| 27.5k| if (big_endian) ------------------ | Branch (50:9): [True: 27.5k, Folded] ------------------ 51| 27.5k| composed = _mm_shuffle_epi8(composed, swap); 52| 27.5k| _mm_storeu_si128((__m128i *)utf16_output, composed); 53| 27.5k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 54| 27.5k| return 16; 55| 27.5k| } 56| 3.00M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (56:7): [True: 7.44k, False: 2.99M] ------------------ 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| 7.44k| const __m128i sh = 61| 7.44k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 62| 7.44k| const __m128i perm = _mm_shuffle_epi8(in, sh); 63| 7.44k| const __m128i ascii = 64| 7.44k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 65| 7.44k| const __m128i middlebyte = 66| 7.44k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 67| 7.44k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 68| 7.44k| const __m128i highbyte = 69| 7.44k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 70| 7.44k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 71| 7.44k| const __m128i composed = 72| 7.44k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 73| 7.44k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 74| 7.44k| if (big_endian) ------------------ | Branch (74:9): [True: 7.44k, Folded] ------------------ 75| 7.44k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 76| 7.44k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 77| 7.44k| utf16_output += 4; 78| 7.44k| return 12; 79| 7.44k| } 80| | 81| 2.99M| const uint8_t idx = simdutf::tables::utf8_to_utf16::utf8bigindex 82| 2.99M| [input_utf8_end_of_code_point_mask][0]; 83| 2.99M| const uint8_t consumed = simdutf::tables::utf8_to_utf16::utf8bigindex 84| 2.99M| [input_utf8_end_of_code_point_mask][1]; 85| 2.99M| if (idx < 64) { ------------------ | Branch (85:7): [True: 2.74M, False: 248k] ------------------ 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| 2.74M| const __m128i sh = _mm_loadu_si128( 93| 2.74M| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 94| 2.74M| const __m128i perm = _mm_shuffle_epi8(in, sh); 95| 2.74M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 96| 2.74M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 97| 2.74M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 98| 2.74M| if (big_endian) ------------------ | Branch (98:9): [True: 2.74M, Folded] ------------------ 99| 2.74M| composed = _mm_shuffle_epi8(composed, swap); 100| 2.74M| _mm_storeu_si128((__m128i *)utf16_output, composed); 101| 2.74M| utf16_output += 6; // We wrote 12 bytes, 6 code points. There is a potential 102| | // overflow of 4 bytes. 103| 2.74M| } else if (idx < 145) { ------------------ | Branch (103:14): [True: 164k, False: 83.8k] ------------------ 104| | // FOUR (4) input code-code units 105| 164k| const __m128i sh = _mm_loadu_si128( 106| 164k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 107| 164k| const __m128i perm = _mm_shuffle_epi8(in, sh); 108| 164k| const __m128i ascii = 109| 164k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 110| 164k| const __m128i middlebyte = 111| 164k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 112| 164k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 113| 164k| const __m128i highbyte = 114| 164k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 115| 164k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 116| 164k| const __m128i composed = 117| 164k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 118| 164k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 119| 164k| if (big_endian) ------------------ | Branch (119:9): [True: 164k, Folded] ------------------ 120| 164k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 121| 164k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 122| 164k| utf16_output += 4; // Here we overflow by 8 bytes. 123| 164k| } else if (idx < 209) { ------------------ | Branch (123:14): [True: 82.9k, False: 894] ------------------ 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| 82.9k| const __m128i sh = _mm_loadu_si128( 135| 82.9k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 136| 82.9k| const __m128i perm = _mm_shuffle_epi8(in, sh); 137| 82.9k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 138| 82.9k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 139| 82.9k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 140| 82.9k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 141| | // correct for spurious high bit 142| 82.9k| const __m128i correct = 143| 82.9k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 144| 82.9k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 145| 82.9k| 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| 82.9k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 149| 82.9k| 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| 82.9k| const __m128i composed = 155| 82.9k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 156| 82.9k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 157| 82.9k| const __m128i composedminus = 158| 82.9k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 159| 82.9k| const __m128i lowtenbits = 160| 82.9k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 161| | // Notice the 0x3ff mask: 162| 82.9k| const __m128i hightenbits = 163| 82.9k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 164| 82.9k| const __m128i lowtenbitsadd = 165| 82.9k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 166| 82.9k| const __m128i hightenbitsadd = 167| 82.9k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 168| 82.9k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 169| 82.9k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 170| 82.9k| uint32_t basic_buffer[4]; 171| 82.9k| uint32_t basic_buffer_swap[4]; 172| 82.9k| if (big_endian) { ------------------ | Branch (172:9): [True: 82.9k, Folded] ------------------ 173| 82.9k| _mm_storeu_si128((__m128i *)basic_buffer_swap, 174| 82.9k| _mm_shuffle_epi8(composed, swap)); 175| 82.9k| surrogates = _mm_shuffle_epi8(surrogates, swap); 176| 82.9k| } 177| 82.9k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 178| 82.9k| uint32_t surrogate_buffer[4]; 179| 82.9k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 180| 331k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (180:24): [True: 248k, False: 82.9k] ------------------ 181| 248k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (181:11): [True: 69.3k, False: 179k] ------------------ 182| 69.3k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 183| 69.3k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 184| 69.3k| utf16_output += 2; 185| 179k| } else { 186| 179k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (186:27): [True: 179k, Folded] ------------------ 187| 179k| : uint16_t(basic_buffer[i]); 188| 179k| utf16_output++; 189| 179k| } 190| 248k| } 191| 82.9k| } else { 192| | // here we know that there is an error but we do not handle errors 193| 894| } 194| 2.99M| return consumed; 195| 3.00M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128convert_masked_utf8_to_utf32EPKcmRPDi: 9| 3.60M| 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| 3.60M| const __m128i in = _mm_loadu_si128((__m128i *)input); 22| 3.60M| const uint16_t input_utf8_end_of_code_point_mask = 23| 3.60M| utf8_end_of_code_point_mask & 0xfff; 24| 3.60M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (24:7): [True: 9.31k, False: 3.59M] ------------------ 25| | // We process the data in chunks of 12 bytes. 26| 9.31k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 27| 9.31k| _mm256_cvtepu8_epi32(in)); 28| 9.31k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output + 8), 29| 9.31k| _mm256_cvtepu8_epi32(_mm_srli_si128(in, 8))); 30| 9.31k| utf32_output += 12; // We wrote 12 32-bit characters. 31| 9.31k| return 12; // We consumed 12 bytes. 32| 9.31k| } 33| 3.59M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (33:7): [True: 28.2k, False: 3.56M] ------------------ 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| 28.2k| const __m128i sh = 38| 28.2k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 39| 28.2k| const __m128i perm = _mm_shuffle_epi8(in, sh); 40| 28.2k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 41| 28.2k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 42| 28.2k| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 43| 28.2k| _mm256_storeu_si256((__m256i *)utf32_output, 44| 28.2k| _mm256_cvtepu16_epi32(composed)); 45| 28.2k| utf32_output += 8; // We wrote 16 bytes, 8 code points. 46| 28.2k| return 16; 47| 28.2k| } 48| 3.56M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (48:7): [True: 2.94k, False: 3.56M] ------------------ 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| 2.94k| const __m128i sh = 53| 2.94k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 54| 2.94k| const __m128i perm = _mm_shuffle_epi8(in, sh); 55| 2.94k| const __m128i ascii = 56| 2.94k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 57| 2.94k| const __m128i middlebyte = 58| 2.94k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 59| 2.94k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 60| 2.94k| const __m128i highbyte = 61| 2.94k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 62| 2.94k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 63| 2.94k| const __m128i composed = 64| 2.94k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 65| 2.94k| _mm_storeu_si128((__m128i *)utf32_output, composed); 66| 2.94k| utf32_output += 4; 67| 2.94k| return 12; 68| 2.94k| } 69| | /// We do not have a fast path available, so we fallback. 70| | 71| 3.56M| const uint8_t idx = 72| 3.56M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 73| 3.56M| const uint8_t consumed = 74| 3.56M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 75| 3.56M| if (idx < 64) { ------------------ | Branch (75:7): [True: 3.29M, False: 271k] ------------------ 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| 3.29M| const __m128i sh = 83| 3.29M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 84| 3.29M| const __m128i perm = _mm_shuffle_epi8(in, sh); 85| 3.29M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 86| 3.29M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 87| 3.29M| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 88| 3.29M| _mm256_storeu_si256((__m256i *)utf32_output, 89| 3.29M| _mm256_cvtepu16_epi32(composed)); 90| 3.29M| utf32_output += 6; // We wrote 24 bytes, 6 code points. There is a potential 91| | // overflow of 32 - 24 = 8 bytes. 92| 3.29M| } else if (idx < 145) { ------------------ | Branch (92:14): [True: 187k, False: 83.6k] ------------------ 93| | // FOUR (4) input code-code units 94| 187k| const __m128i sh = 95| 187k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 96| 187k| const __m128i perm = _mm_shuffle_epi8(in, sh); 97| 187k| const __m128i ascii = 98| 187k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 99| 187k| const __m128i middlebyte = 100| 187k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 101| 187k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 102| 187k| const __m128i highbyte = 103| 187k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 104| 187k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 105| 187k| const __m128i composed = 106| 187k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 107| 187k| _mm_storeu_si128((__m128i *)utf32_output, composed); 108| 187k| utf32_output += 4; 109| 187k| } else if (idx < 209) { ------------------ | Branch (109:14): [True: 82.9k, False: 682] ------------------ 110| | // TWO (2) input code-code units 111| 82.9k| const __m128i sh = 112| 82.9k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 113| 82.9k| const __m128i perm = _mm_shuffle_epi8(in, sh); 114| 82.9k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 115| 82.9k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 116| 82.9k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 117| 82.9k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 118| | // correct for spurious high bit 119| 82.9k| const __m128i correct = 120| 82.9k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 121| 82.9k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 122| 82.9k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 123| 82.9k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0x07000000)); 124| 82.9k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 125| 82.9k| const __m128i composed = 126| 82.9k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 127| 82.9k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 128| 82.9k| _mm_storeu_si128((__m128i *)utf32_output, composed); 129| 82.9k| utf32_output += 130| 82.9k| 3; // We wrote 3 * 4 bytes, there is a potential overflow of 4 bytes. 131| 82.9k| } else { 132| | // here we know that there is an error but we do not handle errors 133| 682| } 134| 3.56M| return consumed; 135| 3.56M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf16_gather_high_bytesILNS_10endiannessE0EEENS1_4simd5simd8IhEERKNS5_6simd16ItEESB_: 3| 2.67M| const simd16 &in1) { 4| 2.67M| if (big_endian) { ------------------ | Branch (4:7): [Folded, False: 2.67M] ------------------ 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| 2.67M| } else { 12| 2.67M| const auto t0 = in0.shr<8>(); 13| 2.67M| const auto t1 = in1.shr<8>(); 14| | 15| 2.67M| return simd16::pack(t0, t1); 16| 2.67M| } 17| 2.67M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf16_gather_high_bytesILNS_10endiannessE1EEENS1_4simd5simd8IhEERKNS5_6simd16ItEESB_: 3| 2.66M| const simd16 &in1) { 4| 2.66M| if (big_endian) { ------------------ | Branch (4:7): [True: 2.66M, Folded] ------------------ 5| | // we want lower bytes 6| 2.66M| const auto mask = simd16(0x00ff); 7| 2.66M| const auto t0 = in0 & mask; 8| 2.66M| const auto t1 = in1 & mask; 9| | 10| 2.66M| return simd16::pack(t0, t1); 11| 2.66M| } 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| 2.66M|} _ZNK7simdutf7haswell14implementation13validate_utf8EPKcm: 266| 3.47k|implementation::validate_utf8(const char *buf, size_t len) const noexcept { 267| 3.47k| return haswell::utf8_validation::generic_validate_utf8(buf, len); 268| 3.47k|} _ZNK7simdutf7haswell14implementation25validate_utf8_with_errorsEPKcm: 273| 3.47k| const char *buf, size_t len) const noexcept { 274| 3.47k| return haswell::utf8_validation::generic_validate_utf8_with_errors(buf, len); 275| 3.47k|} _ZNK7simdutf7haswell14implementation16validate_utf16leEPKDsm: 312| 1.40k| size_t len) const noexcept { 313| 1.40k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.40k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 17, False: 1.38k] | | ------------------ ------------------ 314| | // empty input is valid UTF-16. protect the implementation from 315| | // handling nullptr 316| 17| return true; 317| 17| } 318| 1.38k| const auto res = 319| 1.38k| haswell::utf16::validate_utf16_with_errors(buf, len); 320| 1.38k| if (res.is_err()) { ------------------ | Branch (320:7): [True: 120, False: 1.26k] ------------------ 321| 120| return false; 322| 120| } 323| | 324| 1.26k| if (res.count == len) { ------------------ | Branch (324:7): [True: 0, False: 1.26k] ------------------ 325| 0| return true; 326| 0| } 327| | 328| 1.26k| return scalar::utf16::validate(buf + res.count, 329| 1.26k| len - res.count); 330| 1.26k|} _ZNK7simdutf7haswell14implementation16validate_utf16beEPKDsm: 336| 1.31k| size_t len) const noexcept { 337| 1.31k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.31k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 1.29k] | | ------------------ ------------------ 338| | // empty input is valid UTF-16. protect the implementation from 339| | // handling nullptr 340| 16| return true; 341| 16| } 342| 1.29k| const auto res = 343| 1.29k| haswell::utf16::validate_utf16_with_errors(buf, len); 344| 1.29k| if (res.is_err()) { ------------------ | Branch (344:7): [True: 94, False: 1.20k] ------------------ 345| 94| return false; 346| 94| } 347| | 348| 1.20k| if (res.count == len) { ------------------ | Branch (348:7): [True: 0, False: 1.20k] ------------------ 349| 0| return true; 350| 0| } 351| | 352| 1.20k| return scalar::utf16::validate(buf + res.count, 353| 1.20k| len - res.count); 354| 1.20k|} _ZNK7simdutf7haswell14implementation28validate_utf16le_with_errorsEPKDsm: 357| 1.40k| const char16_t *buf, size_t len) const noexcept { 358| | 359| 1.40k| const result res = 360| 1.40k| haswell::utf16::validate_utf16_with_errors(buf, len); 361| 1.40k| if (res.count != len) { ------------------ | Branch (361:7): [True: 1.38k, False: 17] ------------------ 362| 1.38k| const result scalar_res = 363| 1.38k| scalar::utf16::validate_with_errors( 364| 1.38k| buf + res.count, len - res.count); 365| 1.38k| return result(scalar_res.error, res.count + scalar_res.count); 366| 1.38k| } else { 367| 17| return res; 368| 17| } 369| 1.40k|} _ZNK7simdutf7haswell14implementation28validate_utf16be_with_errorsEPKDsm: 372| 1.31k| const char16_t *buf, size_t len) const noexcept { 373| 1.31k| const result res = 374| 1.31k| haswell::utf16::validate_utf16_with_errors(buf, len); 375| 1.31k| if (res.count != len) { ------------------ | Branch (375:7): [True: 1.29k, False: 16] ------------------ 376| 1.29k| const result scalar_res = 377| 1.29k| scalar::utf16::validate_with_errors(buf + res.count, 378| 1.29k| len - res.count); 379| 1.29k| return result(scalar_res.error, res.count + scalar_res.count); 380| 1.29k| } else { 381| 16| return res; 382| 16| } 383| 1.31k|} _ZNK7simdutf7haswell14implementation14validate_utf32EPKDim: 398| 2.61k|implementation::validate_utf32(const char32_t *buf, size_t len) const noexcept { 399| 2.61k| return utf32::validate(buf, len); 400| 2.61k|} _ZNK7simdutf7haswell14implementation26validate_utf32_with_errorsEPKDim: 405| 2.61k| const char32_t *buf, size_t len) const noexcept { 406| 2.61k| return utf32::validate_with_errors(buf, len); 407| 2.61k|} _ZNK7simdutf7haswell14implementation22convert_latin1_to_utf8EPKcmPc: 412| 606| const char *buf, size_t len, char *utf8_output) const noexcept { 413| 606| std::pair ret = 414| 606| avx2_convert_latin1_to_utf8(buf, len, utf8_output); 415| 606| size_t converted_chars = ret.second - utf8_output; 416| | 417| 606| if (ret.first != buf + len) { ------------------ | Branch (417:7): [True: 602, False: 4] ------------------ 418| 602| const size_t scalar_converted_chars = scalar::latin1_to_utf8::convert( 419| 602| ret.first, len - (ret.first - buf), ret.second); 420| 602| converted_chars += scalar_converted_chars; 421| 602| } 422| | 423| 606| return converted_chars; 424| 606|} _ZNK7simdutf7haswell14implementation25convert_latin1_to_utf16leEPKcmPDs: 429| 86| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 430| 86| std::pair ret = 431| 86| avx2_convert_latin1_to_utf16(buf, len, utf16_output); 432| 86| if (ret.first == nullptr) { ------------------ | Branch (432:7): [True: 4, False: 82] ------------------ 433| 4| return 0; 434| 4| } 435| 82| size_t converted_chars = ret.second - utf16_output; 436| 82| if (ret.first != buf + len) { ------------------ | Branch (436:7): [True: 68, False: 14] ------------------ 437| 68| const size_t scalar_converted_chars = 438| 68| scalar::latin1_to_utf16::convert( 439| 68| ret.first, len - (ret.first - buf), ret.second); 440| 68| if (scalar_converted_chars == 0) { ------------------ | Branch (440:9): [True: 0, False: 68] ------------------ 441| 0| return 0; 442| 0| } 443| 68| converted_chars += scalar_converted_chars; 444| 68| } 445| 82| return converted_chars; 446| 82|} _ZNK7simdutf7haswell14implementation25convert_latin1_to_utf16beEPKcmPDs: 449| 88| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 450| 88| std::pair ret = 451| 88| avx2_convert_latin1_to_utf16(buf, len, utf16_output); 452| 88| if (ret.first == nullptr) { ------------------ | Branch (452:7): [True: 4, False: 84] ------------------ 453| 4| return 0; 454| 4| } 455| 84| size_t converted_chars = ret.second - utf16_output; 456| 84| if (ret.first != buf + len) { ------------------ | Branch (456:7): [True: 68, False: 16] ------------------ 457| 68| const size_t scalar_converted_chars = 458| 68| scalar::latin1_to_utf16::convert( 459| 68| ret.first, len - (ret.first - buf), ret.second); 460| 68| if (scalar_converted_chars == 0) { ------------------ | Branch (460:9): [True: 0, False: 68] ------------------ 461| 0| return 0; 462| 0| } 463| 68| converted_chars += scalar_converted_chars; 464| 68| } 465| 84| return converted_chars; 466| 84|} _ZNK7simdutf7haswell14implementation23convert_latin1_to_utf32EPKcmPDi: 471| 110| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 472| 110| std::pair ret = 473| 110| avx2_convert_latin1_to_utf32(buf, len, utf32_output); 474| 110| if (ret.first == nullptr) { ------------------ | Branch (474:7): [True: 4, False: 106] ------------------ 475| 4| return 0; 476| 4| } 477| 106| size_t converted_chars = ret.second - utf32_output; 478| 106| if (ret.first != buf + len) { ------------------ | Branch (478:7): [True: 72, False: 34] ------------------ 479| 72| const size_t scalar_converted_chars = scalar::latin1_to_utf32::convert( 480| 72| ret.first, len - (ret.first - buf), ret.second); 481| 72| if (scalar_converted_chars == 0) { ------------------ | Branch (481:9): [True: 0, False: 72] ------------------ 482| 0| return 0; 483| 0| } 484| 72| converted_chars += scalar_converted_chars; 485| 72| } 486| 106| return converted_chars; 487| 106|} _ZNK7simdutf7haswell14implementation22convert_utf8_to_latin1EPKcmPc: 492| 590| const char *buf, size_t len, char *latin1_output) const noexcept { 493| 590| utf8_to_latin1::validating_transcoder converter; 494| 590| return converter.convert(buf, len, latin1_output); 495| 590|} _ZNK7simdutf7haswell14implementation34convert_utf8_to_latin1_with_errorsEPKcmPc: 498| 502| const char *buf, size_t len, char *latin1_output) const noexcept { 499| 502| utf8_to_latin1::validating_transcoder converter; 500| 502| return converter.convert_with_errors(buf, len, latin1_output); 501| 502|} _ZNK7simdutf7haswell14implementation23convert_utf8_to_utf16leEPKcmPDs: 511| 720| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 512| 720| utf8_to_utf16::validating_transcoder converter; 513| 720| return converter.convert(buf, len, utf16_output); 514| 720|} _ZNK7simdutf7haswell14implementation23convert_utf8_to_utf16beEPKcmPDs: 517| 756| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 518| 756| utf8_to_utf16::validating_transcoder converter; 519| 756| return converter.convert(buf, len, utf16_output); 520| 756|} _ZNK7simdutf7haswell14implementation35convert_utf8_to_utf16le_with_errorsEPKcmPDs: 523| 612| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 524| 612| utf8_to_utf16::validating_transcoder converter; 525| 612| return converter.convert_with_errors(buf, len, 526| 612| utf16_output); 527| 612|} _ZNK7simdutf7haswell14implementation35convert_utf8_to_utf16be_with_errorsEPKcmPDs: 530| 754| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 531| 754| utf8_to_utf16::validating_transcoder converter; 532| 754| return converter.convert_with_errors(buf, len, utf16_output); 533| 754|} _ZNK7simdutf7haswell14implementation29convert_valid_utf8_to_utf16leEPKcmPDs: 536| 444| const char *input, size_t size, char16_t *utf16_output) const noexcept { 537| 444| return utf8_to_utf16::convert_valid(input, size, 538| 444| utf16_output); 539| 444|} _ZNK7simdutf7haswell14implementation29convert_valid_utf8_to_utf16beEPKcmPDs: 542| 476| const char *input, size_t size, char16_t *utf16_output) const noexcept { 543| 476| return utf8_to_utf16::convert_valid(input, size, 544| 476| utf16_output); 545| 476|} _ZNK7simdutf7haswell14implementation21convert_utf8_to_utf32EPKcmPDi: 550| 784| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 551| 784| utf8_to_utf32::validating_transcoder converter; 552| 784| return converter.convert(buf, len, utf32_output); 553| 784|} _ZNK7simdutf7haswell14implementation33convert_utf8_to_utf32_with_errorsEPKcmPDi: 556| 794| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 557| 794| utf8_to_utf32::validating_transcoder converter; 558| 794| return converter.convert_with_errors(buf, len, utf32_output); 559| 794|} _ZNK7simdutf7haswell14implementation27convert_valid_utf8_to_utf32EPKcmPDi: 562| 450| const char *input, size_t size, char32_t *utf32_output) const noexcept { 563| 450| return utf8_to_utf32::convert_valid(input, size, utf32_output); 564| 450|} _ZNK7simdutf7haswell14implementation25convert_utf16le_to_latin1EPKDsmPc: 569| 134| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 570| 134| std::pair ret = 571| 134| haswell::avx2_convert_utf16_to_latin1(buf, len, 572| 134| latin1_output); 573| 134| if (ret.first == nullptr) { ------------------ | Branch (573:7): [True: 46, False: 88] ------------------ 574| 46| return 0; 575| 46| } 576| 88| size_t saved_bytes = ret.second - latin1_output; 577| 88| if (ret.first != buf + len) { ------------------ | Branch (577:7): [True: 74, False: 14] ------------------ 578| 74| const size_t scalar_saved_bytes = 579| 74| scalar::utf16_to_latin1::convert( 580| 74| ret.first, len - (ret.first - buf), ret.second); 581| 74| if (scalar_saved_bytes == 0) { ------------------ | Branch (581:9): [True: 40, False: 34] ------------------ 582| 40| return 0; 583| 40| } 584| 34| saved_bytes += scalar_saved_bytes; 585| 34| } 586| 48| return saved_bytes; 587| 88|} _ZNK7simdutf7haswell14implementation25convert_utf16be_to_latin1EPKDsmPc: 590| 94| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 591| 94| std::pair ret = 592| 94| haswell::avx2_convert_utf16_to_latin1(buf, len, 593| 94| latin1_output); 594| 94| if (ret.first == nullptr) { ------------------ | Branch (594:7): [True: 20, False: 74] ------------------ 595| 20| return 0; 596| 20| } 597| 74| size_t saved_bytes = ret.second - latin1_output; 598| 74| if (ret.first != buf + len) { ------------------ | Branch (598:7): [True: 60, False: 14] ------------------ 599| 60| const size_t scalar_saved_bytes = 600| 60| scalar::utf16_to_latin1::convert( 601| 60| ret.first, len - (ret.first - buf), ret.second); 602| 60| if (scalar_saved_bytes == 0) { ------------------ | Branch (602:9): [True: 32, False: 28] ------------------ 603| 32| return 0; 604| 32| } 605| 28| saved_bytes += scalar_saved_bytes; 606| 28| } 607| 42| return saved_bytes; 608| 74|} _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: 122, False: 156] ------------------ 617| 122| return ret.first; 618| 122| } // Can return directly since scalar fallback already found correct 619| | // ret.first.count 620| 156| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (620:7): [True: 118, False: 38] ------------------ 621| 118| result scalar_res = 622| 118| scalar::utf16_to_latin1::convert_with_errors( 623| 118| buf + ret.first.count, len - ret.first.count, ret.second); 624| 118| if (scalar_res.error) { ------------------ | Branch (624:9): [True: 56, False: 62] ------------------ 625| 56| scalar_res.count += ret.first.count; 626| 56| return scalar_res; 627| 62| } else { 628| 62| ret.second += scalar_res.count; 629| 62| } 630| 118| } 631| 100| ret.first.count = 632| 100| ret.second - 633| 100| latin1_output; // Set count to the number of 8-bit code units written 634| 100| return ret.first; 635| 156|} _ZNK7simdutf7haswell14implementation37convert_utf16be_to_latin1_with_errorsEPKDsmPc: 639| 250| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 640| 250| std::pair ret = 641| 250| avx2_convert_utf16_to_latin1_with_errors(buf, len, 642| 250| latin1_output); 643| 250| if (ret.first.error) { ------------------ | Branch (643:7): [True: 110, False: 140] ------------------ 644| 110| return ret.first; 645| 110| } // Can return directly since scalar fallback already found correct 646| | // ret.first.count 647| 140| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (647:7): [True: 106, False: 34] ------------------ 648| 106| result scalar_res = 649| 106| scalar::utf16_to_latin1::convert_with_errors( 650| 106| buf + ret.first.count, len - ret.first.count, ret.second); 651| 106| if (scalar_res.error) { ------------------ | Branch (651:9): [True: 54, False: 52] ------------------ 652| 54| scalar_res.count += ret.first.count; 653| 54| return scalar_res; 654| 54| } else { 655| 52| ret.second += scalar_res.count; 656| 52| } 657| 106| } 658| 86| ret.first.count = 659| 86| ret.second - 660| 86| latin1_output; // Set count to the number of 8-bit code units written 661| 86| return ret.first; 662| 140|} _ZNK7simdutf7haswell14implementation23convert_utf16le_to_utf8EPKDsmPc: 679| 856| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 680| 856| std::pair ret = 681| 856| haswell::avx2_convert_utf16_to_utf8(buf, len, 682| 856| utf8_output); 683| 856| if (ret.first == nullptr) { ------------------ | Branch (683:7): [True: 140, False: 716] ------------------ 684| 140| return 0; 685| 140| } 686| 716| size_t saved_bytes = ret.second - utf8_output; 687| 716| if (ret.first != buf + len) { ------------------ | Branch (687:7): [True: 714, False: 2] ------------------ 688| 714| const size_t scalar_saved_bytes = 689| 714| scalar::utf16_to_utf8::convert( 690| 714| ret.first, len - (ret.first - buf), ret.second); 691| 714| if (scalar_saved_bytes == 0) { ------------------ | Branch (691:9): [True: 68, False: 646] ------------------ 692| 68| return 0; 693| 68| } 694| 646| saved_bytes += scalar_saved_bytes; 695| 646| } 696| 648| return saved_bytes; 697| 716|} _ZNK7simdutf7haswell14implementation23convert_utf16be_to_utf8EPKDsmPc: 700| 750| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 701| 750| std::pair ret = 702| 750| haswell::avx2_convert_utf16_to_utf8(buf, len, 703| 750| utf8_output); 704| 750| if (ret.first == nullptr) { ------------------ | Branch (704:7): [True: 108, False: 642] ------------------ 705| 108| return 0; 706| 108| } 707| 642| size_t saved_bytes = ret.second - utf8_output; 708| 642| if (ret.first != buf + len) { ------------------ | Branch (708:7): [True: 640, False: 2] ------------------ 709| 640| const size_t scalar_saved_bytes = 710| 640| scalar::utf16_to_utf8::convert( 711| 640| ret.first, len - (ret.first - buf), ret.second); 712| 640| if (scalar_saved_bytes == 0) { ------------------ | Branch (712:9): [True: 50, False: 590] ------------------ 713| 50| return 0; 714| 50| } 715| 590| saved_bytes += scalar_saved_bytes; 716| 590| } 717| 592| return saved_bytes; 718| 642|} _ZNK7simdutf7haswell14implementation35convert_utf16le_to_utf8_with_errorsEPKDsmPc: 721| 664| 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| 664| std::pair ret = 725| 664| haswell::avx2_convert_utf16_to_utf8_with_errors( 726| 664| buf, len, utf8_output); 727| 664| if (ret.first.error) { ------------------ | Branch (727:7): [True: 166, False: 498] ------------------ 728| 166| return ret.first; 729| 166| } // Can return directly since scalar fallback already found correct 730| | // ret.first.count 731| 498| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (731:7): [True: 494, False: 4] ------------------ 732| 494| result scalar_res = 733| 494| scalar::utf16_to_utf8::convert_with_errors( 734| 494| buf + ret.first.count, len - ret.first.count, ret.second); 735| 494| if (scalar_res.error) { ------------------ | Branch (735:9): [True: 104, False: 390] ------------------ 736| 104| scalar_res.count += ret.first.count; 737| 104| return scalar_res; 738| 390| } else { 739| 390| ret.second += scalar_res.count; 740| 390| } 741| 494| } 742| 394| ret.first.count = 743| 394| ret.second - 744| 394| utf8_output; // Set count to the number of 8-bit code units written 745| 394| return ret.first; 746| 498|} _ZNK7simdutf7haswell14implementation35convert_utf16be_to_utf8_with_errorsEPKDsmPc: 749| 728| 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| 728| std::pair ret = 753| 728| haswell::avx2_convert_utf16_to_utf8_with_errors( 754| 728| buf, len, utf8_output); 755| 728| if (ret.first.error) { ------------------ | Branch (755:7): [True: 200, False: 528] ------------------ 756| 200| return ret.first; 757| 200| } // Can return directly since scalar fallback already found correct 758| | // ret.first.count 759| 528| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (759:7): [True: 524, False: 4] ------------------ 760| 524| result scalar_res = 761| 524| scalar::utf16_to_utf8::convert_with_errors( 762| 524| buf + ret.first.count, len - ret.first.count, ret.second); 763| 524| if (scalar_res.error) { ------------------ | Branch (763:9): [True: 106, False: 418] ------------------ 764| 106| scalar_res.count += ret.first.count; 765| 106| return scalar_res; 766| 418| } else { 767| 418| ret.second += scalar_res.count; 768| 418| } 769| 524| } 770| 422| ret.first.count = 771| 422| ret.second - 772| 422| utf8_output; // Set count to the number of 8-bit code units written 773| 422| return ret.first; 774| 528|} _ZNK7simdutf7haswell14implementation29convert_valid_utf16le_to_utf8EPKDsmPc: 777| 440| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 778| 440| return convert_utf16le_to_utf8(buf, len, utf8_output); 779| 440|} _ZNK7simdutf7haswell14implementation29convert_valid_utf16be_to_utf8EPKDsmPc: 782| 398| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 783| 398| return convert_utf16be_to_utf8(buf, len, utf8_output); 784| 398|} _ZNK7simdutf7haswell14implementation21convert_utf32_to_utf8EPKDimPc: 789| 916| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 790| 916| std::pair ret = 791| 916| avx2_convert_utf32_to_utf8(buf, len, utf8_output); 792| 916| if (ret.first == nullptr) { ------------------ | Branch (792:7): [True: 342, False: 574] ------------------ 793| 342| return 0; 794| 342| } 795| 574| size_t saved_bytes = ret.second - utf8_output; 796| 574| if (ret.first != buf + len) { ------------------ | Branch (796:7): [True: 572, False: 2] ------------------ 797| 572| const size_t scalar_saved_bytes = scalar::utf32_to_utf8::convert( 798| 572| ret.first, len - (ret.first - buf), ret.second); 799| 572| if (scalar_saved_bytes == 0) { ------------------ | Branch (799:9): [True: 260, False: 312] ------------------ 800| 260| return 0; 801| 260| } 802| 312| saved_bytes += scalar_saved_bytes; 803| 312| } 804| 314| return saved_bytes; 805| 574|} _ZNK7simdutf7haswell14implementation23convert_utf32_to_latin1EPKDimPc: 810| 172| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 811| 172| std::pair ret = 812| 172| avx2_convert_utf32_to_latin1(buf, len, latin1_output); 813| 172| if (ret.first == nullptr) { ------------------ | Branch (813:7): [True: 18, False: 154] ------------------ 814| 18| return 0; 815| 18| } 816| 154| size_t saved_bytes = ret.second - latin1_output; 817| 154| if (ret.first != buf + len) { ------------------ | Branch (817:7): [True: 134, False: 20] ------------------ 818| 134| const size_t scalar_saved_bytes = scalar::utf32_to_latin1::convert( 819| 134| ret.first, len - (ret.first - buf), ret.second); 820| 134| if (scalar_saved_bytes == 0) { ------------------ | Branch (820:9): [True: 110, False: 24] ------------------ 821| 110| return 0; 822| 110| } 823| 24| saved_bytes += scalar_saved_bytes; 824| 24| } 825| 44| return saved_bytes; 826| 154|} _ZNK7simdutf7haswell14implementation35convert_utf32_to_latin1_with_errorsEPKDimPc: 829| 408| 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| 408| std::pair ret = 833| 408| avx2_convert_utf32_to_latin1_with_errors(buf, len, latin1_output); 834| 408| if (ret.first.count != len) { ------------------ | Branch (834:7): [True: 388, False: 20] ------------------ 835| 388| result scalar_res = scalar::utf32_to_latin1::convert_with_errors( 836| 388| buf + ret.first.count, len - ret.first.count, ret.second); 837| 388| if (scalar_res.error) { ------------------ | Branch (837:9): [True: 362, False: 26] ------------------ 838| 362| scalar_res.count += ret.first.count; 839| 362| return scalar_res; 840| 362| } else { 841| 26| ret.second += scalar_res.count; 842| 26| } 843| 388| } 844| 46| ret.first.count = 845| 46| ret.second - 846| 46| latin1_output; // Set count to the number of 8-bit code units written 847| 46| return ret.first; 848| 408|} _ZNK7simdutf7haswell14implementation33convert_utf32_to_utf8_with_errorsEPKDimPc: 858| 934| 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| 934| std::pair ret = 862| 934| haswell::avx2_convert_utf32_to_utf8_with_errors(buf, len, utf8_output); 863| 934| if (ret.first.count != len) { ------------------ | Branch (863:7): [True: 930, False: 4] ------------------ 864| 930| result scalar_res = scalar::utf32_to_utf8::convert_with_errors( 865| 930| buf + ret.first.count, len - ret.first.count, ret.second); 866| 930| if (scalar_res.error) { ------------------ | Branch (866:9): [True: 700, False: 230] ------------------ 867| 700| scalar_res.count += ret.first.count; 868| 700| return scalar_res; 869| 700| } else { 870| 230| ret.second += scalar_res.count; 871| 230| } 872| 930| } 873| 234| ret.first.count = 874| 234| ret.second - 875| 234| utf8_output; // Set count to the number of 8-bit code units written 876| 234| return ret.first; 877| 934|} _ZNK7simdutf7haswell14implementation24convert_utf16le_to_utf32EPKDsmPDi: 882| 412| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 883| 412| std::pair ret = 884| 412| haswell::avx2_convert_utf16_to_utf32(buf, len, 885| 412| utf32_output); 886| 412| if (ret.first == nullptr) { ------------------ | Branch (886:7): [True: 108, False: 304] ------------------ 887| 108| return 0; 888| 108| } 889| 304| size_t saved_bytes = ret.second - utf32_output; 890| 304| if (ret.first != buf + len) { ------------------ | Branch (890:7): [True: 248, False: 56] ------------------ 891| 248| const size_t scalar_saved_bytes = 892| 248| scalar::utf16_to_utf32::convert( 893| 248| ret.first, len - (ret.first - buf), ret.second); 894| 248| if (scalar_saved_bytes == 0) { ------------------ | Branch (894:9): [True: 70, False: 178] ------------------ 895| 70| return 0; 896| 70| } 897| 178| saved_bytes += scalar_saved_bytes; 898| 178| } 899| 234| return saved_bytes; 900| 304|} _ZNK7simdutf7haswell14implementation24convert_utf16be_to_utf32EPKDsmPDi: 903| 392| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 904| 392| std::pair ret = 905| 392| haswell::avx2_convert_utf16_to_utf32(buf, len, 906| 392| utf32_output); 907| 392| if (ret.first == nullptr) { ------------------ | Branch (907:7): [True: 60, False: 332] ------------------ 908| 60| return 0; 909| 60| } 910| 332| size_t saved_bytes = ret.second - utf32_output; 911| 332| if (ret.first != buf + len) { ------------------ | Branch (911:7): [True: 292, False: 40] ------------------ 912| 292| const size_t scalar_saved_bytes = 913| 292| scalar::utf16_to_utf32::convert( 914| 292| ret.first, len - (ret.first - buf), ret.second); 915| 292| if (scalar_saved_bytes == 0) { ------------------ | Branch (915:9): [True: 86, False: 206] ------------------ 916| 86| return 0; 917| 86| } 918| 206| saved_bytes += scalar_saved_bytes; 919| 206| } 920| 246| return saved_bytes; 921| 332|} _ZNK7simdutf7haswell14implementation36convert_utf16le_to_utf32_with_errorsEPKDsmPDi: 924| 418| 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| 418| std::pair ret = 928| 418| haswell::avx2_convert_utf16_to_utf32_with_errors( 929| 418| buf, len, utf32_output); 930| 418| if (ret.first.error) { ------------------ | Branch (930:7): [True: 146, False: 272] ------------------ 931| 146| return ret.first; 932| 146| } // Can return directly since scalar fallback already found correct 933| | // ret.first.count 934| 272| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (934:7): [True: 238, False: 34] ------------------ 935| 238| result scalar_res = 936| 238| scalar::utf16_to_utf32::convert_with_errors( 937| 238| buf + ret.first.count, len - ret.first.count, ret.second); 938| 238| if (scalar_res.error) { ------------------ | Branch (938:9): [True: 90, False: 148] ------------------ 939| 90| scalar_res.count += ret.first.count; 940| 90| return scalar_res; 941| 148| } else { 942| 148| ret.second += scalar_res.count; 943| 148| } 944| 238| } 945| 182| ret.first.count = 946| 182| ret.second - 947| 182| utf32_output; // Set count to the number of 8-bit code units written 948| 182| return ret.first; 949| 272|} _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: 76, False: 270] ------------------ 959| 76| return ret.first; 960| 76| } // Can return directly since scalar fallback already found correct 961| | // ret.first.count 962| 270| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (962:7): [True: 246, False: 24] ------------------ 963| 246| result scalar_res = 964| 246| scalar::utf16_to_utf32::convert_with_errors( 965| 246| buf + ret.first.count, len - ret.first.count, ret.second); 966| 246| if (scalar_res.error) { ------------------ | Branch (966:9): [True: 104, False: 142] ------------------ 967| 104| scalar_res.count += ret.first.count; 968| 104| return scalar_res; 969| 142| } else { 970| 142| ret.second += scalar_res.count; 971| 142| } 972| 246| } 973| 166| ret.first.count = 974| 166| ret.second - 975| 166| utf32_output; // Set count to the number of 8-bit code units written 976| 166| return ret.first; 977| 270|} _ZNK7simdutf7haswell14implementation27convert_valid_utf32_to_utf8EPKDimPc: 982| 206| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 983| 206| return convert_utf32_to_utf8(buf, len, utf8_output); 984| 206|} _ZNK7simdutf7haswell14implementation24convert_utf32_to_utf16leEPKDimPDs: 989| 774| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 990| 774| std::pair ret = 991| 774| avx2_convert_utf32_to_utf16(buf, len, utf16_output); 992| 774| if (ret.first == nullptr) { ------------------ | Branch (992:7): [True: 198, False: 576] ------------------ 993| 198| return 0; 994| 198| } 995| 576| size_t saved_bytes = ret.second - utf16_output; 996| 576| if (ret.first != buf + len) { ------------------ | Branch (996:7): [True: 572, False: 4] ------------------ 997| 572| const size_t scalar_saved_bytes = 998| 572| scalar::utf32_to_utf16::convert( 999| 572| ret.first, len - (ret.first - buf), ret.second); 1000| 572| if (scalar_saved_bytes == 0) { ------------------ | Branch (1000:9): [True: 180, False: 392] ------------------ 1001| 180| return 0; 1002| 180| } 1003| 392| saved_bytes += scalar_saved_bytes; 1004| 392| } 1005| 396| return saved_bytes; 1006| 576|} _ZNK7simdutf7haswell14implementation24convert_utf32_to_utf16beEPKDimPDs: 1009| 662| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1010| 662| std::pair ret = 1011| 662| avx2_convert_utf32_to_utf16(buf, len, utf16_output); 1012| 662| if (ret.first == nullptr) { ------------------ | Branch (1012:7): [True: 212, False: 450] ------------------ 1013| 212| return 0; 1014| 212| } 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: 170, False: 278] ------------------ 1021| 170| return 0; 1022| 170| } 1023| 278| saved_bytes += scalar_saved_bytes; 1024| 278| } 1025| 280| return saved_bytes; 1026| 450|} _ZNK7simdutf7haswell14implementation36convert_utf32_to_utf16le_with_errorsEPKDimPDs: 1029| 684| 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| 684| std::pair ret = 1033| 684| haswell::avx2_convert_utf32_to_utf16_with_errors( 1034| 684| buf, len, utf16_output); 1035| 684| if (ret.first.count != len) { ------------------ | Branch (1035:7): [True: 680, False: 4] ------------------ 1036| 680| result scalar_res = 1037| 680| scalar::utf32_to_utf16::convert_with_errors( 1038| 680| buf + ret.first.count, len - ret.first.count, ret.second); 1039| 680| if (scalar_res.error) { ------------------ | Branch (1039:9): [True: 504, False: 176] ------------------ 1040| 504| scalar_res.count += ret.first.count; 1041| 504| return scalar_res; 1042| 504| } else { 1043| 176| ret.second += scalar_res.count; 1044| 176| } 1045| 680| } 1046| 180| ret.first.count = 1047| 180| ret.second - 1048| 180| utf16_output; // Set count to the number of 8-bit code units written 1049| 180| return ret.first; 1050| 684|} _ZNK7simdutf7haswell14implementation36convert_utf32_to_utf16be_with_errorsEPKDimPDs: 1053| 630| 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| 630| std::pair ret = 1057| 630| haswell::avx2_convert_utf32_to_utf16_with_errors( 1058| 630| buf, len, utf16_output); 1059| 630| if (ret.first.count != len) { ------------------ | Branch (1059:7): [True: 626, False: 4] ------------------ 1060| 626| result scalar_res = 1061| 626| scalar::utf32_to_utf16::convert_with_errors( 1062| 626| buf + ret.first.count, len - ret.first.count, ret.second); 1063| 626| if (scalar_res.error) { ------------------ | Branch (1063:9): [True: 458, False: 168] ------------------ 1064| 458| scalar_res.count += ret.first.count; 1065| 458| return scalar_res; 1066| 458| } else { 1067| 168| ret.second += scalar_res.count; 1068| 168| } 1069| 626| } 1070| 172| ret.first.count = 1071| 172| ret.second - 1072| 172| utf16_output; // Set count to the number of 8-bit code units written 1073| 172| return ret.first; 1074| 630|} _ZNK7simdutf7haswell14implementation30convert_valid_utf32_to_utf16leEPKDimPDs: 1077| 248| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1078| 248| return convert_utf32_to_utf16le(buf, len, utf16_output); 1079| 248|} _ZNK7simdutf7haswell14implementation30convert_valid_utf32_to_utf16beEPKDimPDs: 1082| 128| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1083| 128| return convert_utf32_to_utf16be(buf, len, utf16_output); 1084| 128|} _ZNK7simdutf7haswell14implementation30convert_valid_utf16le_to_utf32EPKDsmPDi: 1087| 106| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1088| 106| return convert_utf16le_to_utf32(buf, len, utf32_output); 1089| 106|} _ZNK7simdutf7haswell14implementation30convert_valid_utf16be_to_utf32EPKDsmPDi: 1092| 172| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1093| 172| return convert_utf16be_to_utf32(buf, len, utf32_output); 1094| 172|} _ZNK7simdutf7haswell14implementation13count_utf16leEPKDsm: 1105| 1.40k| const char16_t *input, size_t length) const noexcept { 1106| 1.40k| return utf16::count_code_points(input, length); 1107| 1.40k|} _ZNK7simdutf7haswell14implementation13count_utf16beEPKDsm: 1110| 1.31k| const char16_t *input, size_t length) const noexcept { 1111| 1.31k| return utf16::count_code_points(input, length); 1112| 1.31k|} _ZNK7simdutf7haswell14implementation10count_utf8EPKcm: 1117| 4.01k|implementation::count_utf8(const char *in, size_t size) const noexcept { 1118| 4.01k| return utf8::count_code_points_bytemask(in, size); 1119| 4.01k|} _ZNK7simdutf7haswell14implementation23latin1_length_from_utf8EPKcm: 1124| 546| const char *buf, size_t len) const noexcept { 1125| 546| return count_utf8(buf, len); 1126| 546|} _ZNK7simdutf7haswell14implementation24utf8_length_from_utf16leEPKDsm: 1131| 776| const char16_t *input, size_t length) const noexcept { 1132| 776| return utf16::utf8_length_from_utf16_bytemask(input, 1133| 776| length); 1134| 776|} _ZNK7simdutf7haswell14implementation24utf8_length_from_utf16beEPKDsm: 1137| 756| const char16_t *input, size_t length) const noexcept { 1138| 756| return utf16::utf8_length_from_utf16_bytemask(input, length); 1139| 756|} _ZNK7simdutf7haswell14implementation25utf32_length_from_utf16leEPKDsm: 1144| 420| const char16_t *input, size_t length) const noexcept { 1145| 420| return utf16::utf32_length_from_utf16(input, length); 1146| 420|} _ZNK7simdutf7haswell14implementation25utf32_length_from_utf16beEPKDsm: 1149| 382| const char16_t *input, size_t length) const noexcept { 1150| 382| return utf16::utf32_length_from_utf16(input, length); 1151| 382|} _ZNK7simdutf7haswell14implementation22utf16_length_from_utf8EPKcm: 1156| 1.90k| const char *input, size_t length) const noexcept { 1157| 1.90k| return utf8::utf16_length_from_utf8_bytemask(input, length); 1158| 1.90k|} _ZNK7simdutf7haswell14implementation23utf8_length_from_latin1EPKcm: 1203| 303| const char *input, size_t len) const noexcept { 1204| 303| const uint8_t *data = reinterpret_cast(input); 1205| 303| size_t answer = len / sizeof(__m256i) * sizeof(__m256i); 1206| 303| size_t i = 0; 1207| 303| if (answer >= 2048) { // long strings optimization ------------------ | Branch (1207:7): [True: 142, False: 161] ------------------ 1208| 142| __m256i four_64bits = _mm256_setzero_si256(); 1209| 2.76k| while (i + sizeof(__m256i) <= len) { ------------------ | Branch (1209:12): [True: 2.61k, False: 142] ------------------ 1210| 2.61k| __m256i runner = _mm256_setzero_si256(); 1211| | // We can do up to 255 loops without overflow. 1212| 2.61k| size_t iterations = (len - i) / sizeof(__m256i); 1213| 2.61k| if (iterations > 255) { ------------------ | Branch (1213:11): [True: 2.47k, False: 142] ------------------ 1214| 2.47k| iterations = 255; 1215| 2.47k| } 1216| 2.61k| size_t max_i = i + iterations * sizeof(__m256i) - sizeof(__m256i); 1217| 161k| for (; i + 4 * sizeof(__m256i) <= max_i; i += 4 * sizeof(__m256i)) { ------------------ | Branch (1217:14): [True: 158k, False: 2.61k] ------------------ 1218| 158k| __m256i input1 = _mm256_loadu_si256((const __m256i *)(data + i)); 1219| 158k| __m256i input2 = 1220| 158k| _mm256_loadu_si256((const __m256i *)(data + i + sizeof(__m256i))); 1221| 158k| __m256i input3 = _mm256_loadu_si256( 1222| 158k| (const __m256i *)(data + i + 2 * sizeof(__m256i))); 1223| 158k| __m256i input4 = _mm256_loadu_si256( 1224| 158k| (const __m256i *)(data + i + 3 * sizeof(__m256i))); 1225| 158k| __m256i input12 = 1226| 158k| _mm256_add_epi8(_mm256_cmpgt_epi8(_mm256_setzero_si256(), input1), 1227| 158k| _mm256_cmpgt_epi8(_mm256_setzero_si256(), input2)); 1228| 158k| __m256i input23 = 1229| 158k| _mm256_add_epi8(_mm256_cmpgt_epi8(_mm256_setzero_si256(), input3), 1230| 158k| _mm256_cmpgt_epi8(_mm256_setzero_si256(), input4)); 1231| 158k| __m256i input1234 = _mm256_add_epi8(input12, input23); 1232| 158k| runner = _mm256_sub_epi8(runner, input1234); 1233| 158k| } 1234| 10.3k| for (; i <= max_i; i += sizeof(__m256i)) { ------------------ | Branch (1234:14): [True: 7.78k, False: 2.61k] ------------------ 1235| 7.78k| __m256i input_256_chunk = 1236| 7.78k| _mm256_loadu_si256((const __m256i *)(data + i)); 1237| 7.78k| runner = _mm256_sub_epi8( 1238| 7.78k| runner, _mm256_cmpgt_epi8(_mm256_setzero_si256(), input_256_chunk)); 1239| 7.78k| } 1240| 2.61k| four_64bits = _mm256_add_epi64( 1241| 2.61k| four_64bits, _mm256_sad_epu8(runner, _mm256_setzero_si256())); 1242| 2.61k| } 1243| 142| answer += _mm256_extract_epi64(four_64bits, 0) + 1244| 142| _mm256_extract_epi64(four_64bits, 1) + 1245| 142| _mm256_extract_epi64(four_64bits, 2) + 1246| 142| _mm256_extract_epi64(four_64bits, 3); 1247| 161| } else if (answer > 0) { ------------------ | Branch (1247:14): [True: 83, False: 78] ------------------ 1248| 1.18k| for (; i + sizeof(__m256i) <= len; i += sizeof(__m256i)) { ------------------ | Branch (1248:12): [True: 1.10k, False: 83] ------------------ 1249| 1.10k| __m256i latin = _mm256_loadu_si256((const __m256i *)(data + i)); 1250| 1.10k| uint32_t non_ascii = _mm256_movemask_epi8(latin); 1251| 1.10k| answer += count_ones(non_ascii); 1252| 1.10k| } 1253| 83| } 1254| 303| return answer + scalar::latin1::utf8_length_from_latin1( 1255| 303| reinterpret_cast(data + i), len - i); 1256| 303|} _ZNK7simdutf7haswell14implementation22utf8_length_from_utf32EPKDim: 1261| 937| const char32_t *input, size_t length) const noexcept { 1262| 937| return utf32::utf8_length_from_utf32(input, length); 1263| 937|} _ZNK7simdutf7haswell14implementation23utf16_length_from_utf32EPKDim: 1268| 1.38k| const char32_t *input, size_t length) const noexcept { 1269| 1.38k| const __m256i v_00000000 = _mm256_setzero_si256(); 1270| 1.38k| const __m256i v_ffff0000 = _mm256_set1_epi32((uint32_t)0xffff0000); 1271| 1.38k| size_t pos = 0; 1272| 1.38k| size_t count = 0; 1273| 764k| for (; pos + 8 <= length; pos += 8) { ------------------ | Branch (1273:10): [True: 763k, False: 1.38k] ------------------ 1274| 763k| __m256i in = _mm256_loadu_si256((__m256i *)(input + pos)); 1275| 763k| const __m256i surrogate_bytemask = 1276| 763k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_ffff0000), v_00000000); 1277| 763k| const uint32_t surrogate_bitmask = 1278| 763k| static_cast(_mm256_movemask_epi8(surrogate_bytemask)); 1279| 763k| size_t surrogate_count = (32 - count_ones(surrogate_bitmask)) / 4; 1280| 763k| count += 8 + surrogate_count; 1281| 763k| } 1282| 1.38k| return count + 1283| 1.38k| scalar::utf32::utf16_length_from_utf32(input + pos, length - pos); 1284| 1.38k|} _ZNK7simdutf7haswell14implementation22utf32_length_from_utf8EPKcm: 1289| 1.01k| const char *input, size_t length) const noexcept { 1290| 1.01k| return utf8::count_code_points(input, length); 1291| 1.01k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_18is_asciiERKNS1_4simd8simd8x64IhEE: 13| 4.68M|simdutf_really_inline bool is_ascii(const simd8x64 &input) { 14| 4.68M| return input.reduce_or().is_ascii(); 15| 4.68M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_124must_be_2_3_continuationENS1_4simd5simd8IhEES4_: 22| 3.41M| const simd8 prev3) { 23| 3.41M| simd8 is_third_byte = 24| 3.41M| prev2.saturating_sub(0xe0u - 0x80); // Only 111_____ will be > 0x80 25| 3.41M| simd8 is_fourth_byte = 26| 3.41M| prev3.saturating_sub(0xf0u - 0x80); // Only 1111____ will be > 0x80 27| 3.41M| return simd8(is_third_byte | is_fourth_byte); 28| 3.41M|} _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| 3.63M|simdutf_really_inline long long int count_ones(uint64_t input_num) { 15| | return _popcnt64(input_num); 16| 3.63M|} _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| 20.8k| simdutf_really_inline simd8() : base8_numeric() {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEC2Ev: 124| 20.8k| simdutf_really_inline base8_numeric() : base8() {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2Ev: 71| 20.8k| simdutf_really_inline base8() : base>() {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2Ev: 14| 20.8k| simdutf_really_inline base() : value{__m256i()} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IhE9reduce_orEv: 294| 4.68M| simdutf_really_inline simd8 reduce_or() const { 295| 4.68M| return this->chunks[0] | this->chunks[1]; 296| 4.68M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEorES5_: 50| 15.4M| simdutf_really_inline Child operator|(const Child other) const { 51| 15.4M| return _mm256_or_si256(*this, other); 52| 15.4M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2EDv4_x: 193| 168M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEC2EDv4_x: 126| 168M| : base8(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2EDv4_x: 73| 168M| simdutf_really_inline base8(const __m256i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2EDv4_x: 17| 168M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE8is_asciiEv: 234| 4.68M| simdutf_really_inline bool is_ascii() const { 235| 4.68M| return _mm256_movemask_epi8(*this) == 0; 236| 4.68M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEoRES5_: 59| 7.33M| simdutf_really_inline Child &operator|=(const Child other) { 60| 7.33M| auto this_cast = static_cast(this); 61| 7.33M| *this_cast = *this_cast | other; 62| 7.33M| return *this_cast; 63| 7.33M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi1EEENS4_IhEES8_: 83| 3.55M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 84| | return _mm256_alignr_epi8( 85| 3.55M| *this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N); 86| 3.55M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE3shrILi4EEES4_v: 245| 7.10M| template simdutf_really_inline simd8 shr() const { 246| 7.10M| return simd8(_mm256_srli_epi16(*this, N)) & uint8_t(0xFFu >> N); 247| 7.10M| } 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| 10.6M| L replace15) const { 158| 10.6M| return lookup_16(simd8::repeat_16( 159| 10.6M| replace0, replace1, replace2, replace3, replace4, replace5, replace6, 160| 10.6M| replace7, replace8, replace9, replace10, replace11, replace12, 161| 10.6M| replace13, replace14, replace15)); 162| 10.6M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES8_: 148| 10.6M| simdutf_really_inline simd8 lookup_16(simd8 lookup_table) const { 149| 10.6M| return _mm256_shuffle_epi8(lookup_table, *this); 150| 10.6M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE9repeat_16Ehhhhhhhhhhhhhhhh: 118| 10.6M| T v14, T v15) { 119| 10.6M| return simd8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, 120| 10.6M| v14, v15, v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, 121| 10.6M| v12, v13, v14, v15); 122| 10.6M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2Ehhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh: 207| 10.6M| : simd8(_mm256_setr_epi8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, 208| 10.6M| v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, 209| 10.6M| v22, v23, v24, v25, v26, v27, v28, v29, v30, 210| 10.6M| v31)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2Eh: 195| 24.6M| simdutf_really_inline simd8(uint8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi2EEENS4_IhEES8_: 83| 3.41M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 84| | return _mm256_alignr_epi8( 85| 3.41M| *this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N); 86| 3.41M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi3EEENS4_IhEES8_: 83| 3.41M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 84| | return _mm256_alignr_epi8( 85| 3.41M| *this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N); 86| 3.41M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE14saturating_subES4_: 214| 7.74M| saturating_sub(const simd8 other) const { 215| 7.74M| return _mm256_subs_epu8(*this, other); 216| 7.74M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IbEC2EDv4_x: 95| 24.8M| simdutf_really_inline simd8(const __m256i _value) : base8(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IbNS2_5simd8IbEEEC2EDv4_x: 73| 24.8M| simdutf_really_inline base8(const __m256i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEC2EDv4_x: 17| 24.8M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEcvRKDv4_xEv: 19| 15.5M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEeoES5_: 56| 3.41M| simdutf_really_inline Child operator^(const Child other) const { 57| 3.41M| return _mm256_xor_si256(*this, other); 58| 3.41M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2EPKh: 197| 905k| simdutf_really_inline simd8(const uint8_t values[32]) : simd8(load(values)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE4loadEPKh: 111| 10.2M| static simdutf_really_inline simd8 load(const T values[32]) { 112| 10.2M| return _mm256_loadu_si256(reinterpret_cast(values)); 113| 10.2M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE7gt_bitsES4_: 225| 905k| gt_bits(const simd8 other) const { 226| 905k| return this->saturating_sub(other); 227| 905k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEC2EDv4_x: 17| 73.3M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEcvRKDv4_xEv: 19| 103M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IbE10to_bitmaskEv: 99| 9.27M| simdutf_really_inline uint32_t to_bitmask() const { 100| 9.27M| return uint32_t(_mm256_movemask_epi8(value)); 101| 9.27M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE21any_bits_set_anywhereEv: 241| 2.27M| simdutf_really_inline bool any_bits_set_anywhere() const { 242| 2.27M| return !bits_not_set_anywhere(); 243| 2.27M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE21bits_not_set_anywhereEv: 237| 2.27M| simdutf_really_inline bool bits_not_set_anywhere() const { 238| 2.27M| return _mm256_testz_si256(*this, *this); 239| 2.27M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaEC2EPKa: 274| 5.97M| : chunks{simd8::load(ptr), 275| 5.97M| simd8::load(ptr + sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaE4loadEPKa: 111| 21.2M| static simdutf_really_inline simd8 load(const T values[32]) { 112| 21.2M| return _mm256_loadu_si256(reinterpret_cast(values)); 113| 21.2M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEC2EDv4_x: 169| 37.6M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaEC2EDv4_x: 126| 37.6M| : base8(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IaNS2_5simd8IbEEEC2EDv4_x: 73| 37.6M| simdutf_really_inline base8(const __m256i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEC2EDv4_x: 17| 37.6M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE8is_asciiEv: 298| 5.20M| simdutf_really_inline bool is_ascii() const { 299| 5.20M| return this->reduce_or().is_ascii(); 300| 5.20M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE9reduce_orEv: 294| 5.20M| simdutf_really_inline simd8 reduce_or() const { 295| 5.20M| return this->chunks[0] | this->chunks[1]; 296| 5.20M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEorES5_: 50| 5.20M| simdutf_really_inline Child operator|(const Child other) const { 51| 5.20M| return _mm256_or_si256(*this, other); 52| 5.20M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEcvRKDv4_xEv: 19| 60.9M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaE8is_asciiEv: 177| 5.20M| simdutf_really_inline bool is_ascii() const { 178| 5.20M| return _mm256_movemask_epi8(*this) == 0; 179| 5.20M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE5storeEPa: 277| 853k| simdutf_really_inline void store(T *ptr) const { 278| 853k| this->chunks[0].store(ptr + sizeof(simd8) * 0 / sizeof(T)); 279| 853k| this->chunks[1].store(ptr + sizeof(simd8) * 1 / sizeof(T)); 280| 853k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaE5storeEPa: 129| 1.70M| simdutf_really_inline void store(T dst[32]) const { 130| 1.70M| return _mm256_storeu_si256(reinterpret_cast<__m256i *>(dst), *this); 131| 1.70M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEcvNS3_IhEEEv: 256| 2.61M|simdutf_really_inline simd8::operator simd8() const { 257| 2.61M| return this->value; 258| 2.61M|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE2ltEa: 325| 1.16M| simdutf_really_inline uint64_t lt(const T m) const { 326| 1.16M| const simd8 mask = simd8::splat(m); 327| 1.16M| return simd8x64(this->chunks[0] < mask, this->chunks[1] < mask) 328| 1.16M| .to_bitmask(); 329| 1.16M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaE5splatEa: 105| 11.2M| static simdutf_really_inline simd8 splat(T _value) { 106| 11.2M| return _mm256_set1_epi8(_value); 107| 11.2M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEltES4_: 184| 2.32M| simdutf_really_inline simd8 operator<(const simd8 other) const { 185| 2.32M| return _mm256_cmpgt_epi8(other, *this); 186| 2.32M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IbEC2ENS2_5simd8IbEES6_: 272| 1.93M| : chunks{chunk0, chunk1} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IbE10to_bitmaskEv: 282| 1.93M| simdutf_really_inline uint64_t to_bitmask() const { 283| 1.93M| uint64_t r_lo = uint32_t(this->chunks[0].to_bitmask()); 284| 1.93M| uint64_t r_hi = this->chunks[1].to_bitmask(); 285| 1.93M| return r_lo | (r_hi << 32); 286| 1.93M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf32EPDi: 310| 1.13M| simdutf_really_inline void store_ascii_as_utf32(char32_t *ptr) const { 311| 1.13M| this->chunks[0].store_ascii_as_utf32(ptr + sizeof(simd8) * 0); 312| 1.13M| this->chunks[1].store_ascii_as_utf32(ptr + sizeof(simd8) * 1); 313| 1.13M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf32EPDi: 36| 2.27M| simdutf_really_inline void store_ascii_as_utf32(char32_t *ptr) const { 37| 2.27M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), 38| 2.27M| _mm256_cvtepu8_epi32(_mm256_castsi256_si128(*this))); 39| 2.27M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 8), 40| 2.27M| _mm256_cvtepu8_epi32(_mm256_castsi256_si128( 41| 2.27M| _mm256_srli_si256(*this, 8)))); 42| 2.27M| _mm256_storeu_si256( 43| 2.27M| reinterpret_cast<__m256i *>(ptr + 16), 44| 2.27M| _mm256_cvtepu8_epi32(_mm256_extractf128_si256(*this, 1))); 45| 2.27M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 24), 46| | _mm256_cvtepu8_epi32(_mm_srli_si128( 47| 2.27M| _mm256_extractf128_si256(*this, 1), 8))); 48| 2.27M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE4zeroEv: 108| 100k| static simdutf_really_inline simd8 zero() { 109| 100k| return _mm256_setzero_si256(); 110| 100k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEgtES4_: 181| 10.8M| simdutf_really_inline simd8 operator>(const simd8 other) const { 182| 10.8M| return _mm256_cmpgt_epi8(*this, other); 183| 10.8M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEC2Ea: 172| 9.28M| simdutf_really_inline simd8(int8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEmIENS2_5simd8IhEE: 137| 12.1M| simdutf_really_inline simd8 &operator-=(const simd8 other) { 138| 12.1M| *this = *this - other; 139| 12.1M| return *static_cast *>(this); 140| 12.1M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEmiENS2_5simd8IhEE: 134| 12.1M| simdutf_really_inline simd8 operator-(const simd8 other) const { 135| 12.1M| return _mm256_sub_epi8(*this, other); 136| 12.1M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd10sum_8bytesENS2_5simd8IhEE: 353| 47.5k|simdutf_really_inline simd64 sum_8bytes(const simd8 v) { 354| 47.5k| return _mm256_sad_epu8(v.value, simd8::zero()); 355| 47.5k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE9sum_bytesEv: 249| 2.70k| simdutf_really_inline uint64_t sum_bytes() const { 250| 2.70k| const auto tmp = _mm256_sad_epu8(value, _mm256_setzero_si256()); 251| | 252| 2.70k| return _mm256_extract_epi64(tmp, 0) + _mm256_extract_epi64(tmp, 1) + 253| 2.70k| _mm256_extract_epi64(tmp, 2) + _mm256_extract_epi64(tmp, 3); 254| 2.70k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEgeES4_: 229| 2.82M| operator>=(const simd8 other) const { 230| 2.82M| return other.min_val(*this) == other; 231| 2.82M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE7min_valES4_: 220| 2.82M| min_val(const simd8 other) const { 221| 2.82M| return _mm256_min_epu8(other, *this); 222| 2.82M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE2gtEa: 331| 773k| simdutf_really_inline uint64_t gt(const T m) const { 332| 773k| const simd8 mask = simd8::splat(m); 333| 773k| return simd8x64(this->chunks[0] > mask, this->chunks[1] > mask) 334| 773k| .to_bitmask(); 335| 773k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEC2EDv4_x: 17| 14.6M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEcvRKDv4_xEv: 19| 14.6M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE5splatEh: 105| 24.6M| static simdutf_really_inline simd8 splat(T _value) { 106| 24.6M| return _mm256_set1_epi8(_value); 107| 24.6M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEcvRKDv4_xEv: 19| 225M| 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| 26.5M| simdutf_really_inline Child operator&(const Child other) const { 54| 26.5M| return _mm256_and_si256(*this, other); 55| 26.5M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IhEC2EPKh: 274| 4.68M| : chunks{simd8::load(ptr), 275| 4.68M| simd8::load(ptr + sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEanES5_: 53| 11.6M| simdutf_really_inline Child operator&(const Child other) const { 54| 11.6M| return _mm256_and_si256(*this, other); 55| 11.6M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 303| 947k| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 304| 947k| this->chunks[0].template store_ascii_as_utf16(ptr + 305| 947k| sizeof(simd8) * 0); 306| 947k| this->chunks[1].template store_ascii_as_utf16(ptr + 307| 947k| sizeof(simd8) * 1); 308| 947k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 22| 1.89M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 23| 1.89M| __m256i first = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(*this)); 24| 1.89M| __m256i second = _mm256_cvtepu8_epi16(_mm256_extractf128_si256(*this, 1)); 25| 1.89M| if (big_endian) { ------------------ | Branch (25:9): [Folded, False: 1.89M] ------------------ 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| 1.89M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), first); 33| 1.89M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 16), second); 34| 1.89M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 303| 1.10M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 304| 1.10M| this->chunks[0].template store_ascii_as_utf16(ptr + 305| 1.10M| sizeof(simd8) * 0); 306| 1.10M| this->chunks[1].template store_ascii_as_utf16(ptr + 307| 1.10M| sizeof(simd8) * 1); 308| 1.10M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 22| 2.20M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 23| 2.20M| __m256i first = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(*this)); 24| 2.20M| __m256i second = _mm256_cvtepu8_epi16(_mm256_extractf128_si256(*this, 1)); 25| 2.20M| if (big_endian) { ------------------ | Branch (25:9): [True: 2.20M, Folded] ------------------ 26| 2.20M| const __m256i swap = _mm256_setr_epi8( 27| 2.20M| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 28| 2.20M| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 29| 2.20M| first = _mm256_shuffle_epi8(first, swap); 30| 2.20M| second = _mm256_shuffle_epi8(second, swap); 31| 2.20M| } 32| 2.20M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), first); 33| 2.20M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 16), second); 34| 2.20M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEorES5_: 50| 4.18M| simdutf_really_inline Child operator|(const Child other) const { 51| 4.18M| return _mm256_or_si256(*this, other); 52| 4.18M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEC2EDv4_x: 97| 73.3M| : base16_numeric(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItEC2EDv4_x: 73| 73.3M| : base16(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6base16ItNS2_6simd16IbEEEC2EDv4_x: 20| 73.3M| : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE4packERKS4_S6_: 149| 5.34M| 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| 5.34M| const __m128i lo_0 = _mm256_extracti128_si256(v0, 0); 156| 5.34M| const __m128i lo_1 = _mm256_extracti128_si256(v1, 0); 157| | 158| | // get the 1st lanes 159| 5.34M| const __m128i hi_0 = _mm256_extracti128_si256(v0, 1); 160| 5.34M| const __m128i hi_1 = _mm256_extracti128_si256(v1, 1); 161| | 162| | // build new vectors (shuffle lanes) 163| 5.34M| const __m256i t0 = _mm256_set_m128i(lo_1, lo_0); ------------------ | | 4| 5.34M| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 5.34M| _mm256_castsi128_si256(xmm2), 2) ------------------ 164| 5.34M| const __m256i t1 = _mm256_set_m128i(hi_1, hi_0); ------------------ | | 4| 5.34M| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 5.34M| _mm256_castsi128_si256(xmm2), 2) ------------------ 165| | 166| | // pack code units in linear order from v0 and v1 167| 5.34M| return _mm256_packus_epi16(t0, t1); 168| 5.34M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd9simd16x32ItEC2EPKt: 202| 2.09M| : chunks{simd16::load(ptr), 203| 2.09M| simd16::load(ptr + sizeof(simd16) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItE4loadEPKt: 66| 16.9M| static simdutf_really_inline simd16 load(const T values[8]) { 67| 16.9M| return _mm256_loadu_si256(reinterpret_cast(values)); 68| 16.9M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd9simd16x32ItE10swap_bytesEv: 229| 999k| simdutf_really_inline void swap_bytes() { 230| 999k| this->chunks[0] = this->chunks[0].swap_bytes(); 231| 999k| this->chunks[1] = this->chunks[1].swap_bytes(); 232| 999k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE10swap_bytesEv: 139| 3.03M| simdutf_really_inline simd16 swap_bytes() const { 140| 3.03M| const __m256i swap = _mm256_setr_epi8( 141| 3.03M| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 142| 3.03M| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 143| 3.03M| return _mm256_shuffle_epi8(*this, swap); 144| 3.03M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItE5splatEt: 58| 15.2M| static simdutf_really_inline simd16 splat(T _value) { 59| 15.2M| return _mm256_set1_epi16(_value); 60| 15.2M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdeqENS2_6simd16ItEES4_: 26| 10.4M| const simd16 rhs) { 27| 10.4M| return _mm256_cmpeq_epi16(lhs, rhs); 28| 10.4M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16IbEC2EDv4_x: 45| 14.6M| simdutf_really_inline simd16(const __m256i _value) : base16(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6base16IbNS2_6simd16IbEEEC2EDv4_x: 20| 14.6M| : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd9simd16x32IbEC2ENS2_6simd16IbEES6_: 200| 2.09M| : chunks{chunk0, chunk1} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd9simd16x32IbE10to_bitmaskEv: 210| 2.09M| simdutf_really_inline uint64_t to_bitmask() const { 211| 2.09M| uint64_t r_lo = uint32_t(this->chunks[0].to_bitmask()); 212| 2.09M| uint64_t r_hi = this->chunks[1].to_bitmask(); 213| 2.09M| return r_lo | (r_hi << 32); 214| 2.09M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16IbE10to_bitmaskEv: 50| 4.18M| simdutf_really_inline bitmask_type to_bitmask() const { 51| 4.18M| return _mm256_movemask_epi8(*this); 52| 4.18M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE3shrILi8EEES4_v: 134| 5.34M| template simdutf_really_inline simd16 shr() const { 135| 5.34M| return simd16(_mm256_srli_epi16(*this, N)); 136| 5.34M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEleES4_: 117| 4.18M| operator<=(const simd16 other) const { 118| 4.18M| return other.max_val(*this) == other; 119| 4.18M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE7max_valES4_: 108| 4.18M| max_val(const simd16 other) const { 109| 4.18M| return _mm256_max_epu16(*this, other); 110| 4.18M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEC2EPKDs: 104| 10.6M| : simd16(load(reinterpret_cast(values))) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEC2Et: 100| 11.0M| simdutf_really_inline simd16(uint16_t _value) : simd16(splat(_value)) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd9simd16x32ItE12not_in_rangeEtt: 249| 2.09M| simdutf_really_inline uint64_t not_in_range(const T low, const T high) const { 250| 2.09M| const simd16 mask_low = simd16::splat(static_cast(low - 1)); 251| 2.09M| const simd16 mask_high = simd16::splat(static_cast(high + 1)); 252| 2.09M| return simd16x32( 253| 2.09M| (this->chunks[0] >= mask_high) | (this->chunks[0] <= mask_low), 254| 2.09M| (this->chunks[1] >= mask_high) | (this->chunks[1] <= mask_low)) 255| 2.09M| .to_bitmask(); 256| 2.09M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEgeES4_: 121| 4.18M| operator>=(const simd16 other) const { 122| 4.18M| return other.min_val(*this) == other; 123| 4.18M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE7min_valES4_: 112| 4.18M| min_val(const simd16 other) const { 113| 4.18M| return _mm256_min_epu16(*this, other); 114| 4.18M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItE4zeroEv: 62| 1.56k| static simdutf_really_inline simd16 zero() { 63| 1.56k| return _mm256_setzero_si256(); 64| 1.56k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd3minENS2_6simd16ItEES4_: 259| 4.21M|simd16 min(const simd16 a, simd16 b) { 260| 4.21M| return _mm256_min_epu16(a.value, b.value); 261| 4.21M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItEpLENS2_6simd16ItEE: 87| 6.32M| simdutf_really_inline simd16 &operator+=(const simd16 other) { 88| 6.32M| *this = *this + other; 89| 6.32M| return *static_cast *>(this); 90| 6.32M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItEplENS2_6simd16ItEE: 84| 6.32M| simdutf_really_inline simd16 operator+(const simd16 other) const { 85| 6.32M| return _mm256_add_epi16(*this, other); 86| 6.32M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE3sumEv: 170| 1.56k| simdutf_really_inline uint64_t sum() const { 171| 1.56k| const auto lo_u16 = _mm256_and_si256(value, _mm256_set1_epi32(0x0000ffff)); 172| 1.56k| const auto hi_u16 = _mm256_srli_epi32(value, 16); 173| 1.56k| const auto sum_u32 = _mm256_add_epi32(lo_u16, hi_u16); 174| | 175| 1.56k| const auto lo_u32 = 176| 1.56k| _mm256_and_si256(sum_u32, _mm256_set1_epi64x(0xffffffff)); 177| 1.56k| const auto hi_u32 = _mm256_srli_epi64(sum_u32, 32); 178| 1.56k| const auto sum_u64 = _mm256_add_epi64(lo_u32, hi_u32); 179| | 180| 1.56k| return uint64_t(_mm256_extract_epi64(sum_u64, 0)) + 181| 1.56k| uint64_t(_mm256_extract_epi64(sum_u64, 1)) + 182| 1.56k| uint64_t(_mm256_extract_epi64(sum_u64, 2)) + 183| | uint64_t(_mm256_extract_epi64(sum_u64, 3)); 184| 1.56k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE5splatEj: 46| 19.2k| simdutf_really_inline static simd32 splat(uint32_t v) { 47| 19.2k| return _mm256_set1_epi32(v); 48| 19.2k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjEC2EDv4_x: 9| 6.85M| simdutf_really_inline simd32(const __m256i v) : value(v) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE4zeroEv: 42| 5.88k| simdutf_really_inline static simd32 zero() { 43| 5.88k| return _mm256_setzero_si256(); 44| 5.88k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjEC2IDiEEPKT_: 13| 1.66M| : 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| 2.29M| const simd32 b) { 81| 2.29M| return _mm256_max_epu32(a.value, b.value); 82| 2.29M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdplENS2_6simd32IjEES4_: 90| 1.14M| const simd32 b) { 91| 1.14M| return _mm256_add_epi32(a.value, b.value); 92| 1.14M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdgtENS2_6simd32IjEES4_: 109| 4.23k| const simd32 b) { 110| 4.23k| return !(b >= a); 111| 4.23k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdntENS2_6simd32IbEE: 104| 4.23k|simdutf_really_inline simd32 operator!(const simd32 v) { 105| 4.23k| return _mm256_xor_si256(v.value, _mm256_set1_epi8(-1)); 106| 4.23k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IbEC2EDv4_x: 59| 661k| simdutf_really_inline simd32(const __m256i v) : value(v) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdgeENS2_6simd32IjEES4_: 100| 221k| const simd32 b) { 101| 221k| return _mm256_cmpeq_epi32(_mm256_max_epu32(a.value, b.value), a.value); 102| 221k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd32IbE3anyEv: 61| 221k| simdutf_really_inline bool any() const { 62| 221k| return _mm256_movemask_epi8(value) != 0; 63| 221k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdeqENS2_6simd32IjEES4_: 95| 217k| const simd32 b) { 96| 217k| return _mm256_cmpeq_epi32(a.value, b.value); 97| 217k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdanENS2_6simd32IjEES4_: 85| 1.12M| const simd32 a) { 86| 1.12M| return _mm256_and_si256(a.value, b.value); 87| 1.12M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdorIbEENS2_6simd32IT_EES6_S6_: 70| 217k| const simd32 b) { 71| 217k| return _mm256_or_si256(a.value, b.value); 72| 217k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd3minENS2_6simd32IjEES4_: 75| 902k| const simd32 a) { 76| 902k| return _mm256_min_epu32(a.value, b.value); 77| 902k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjEpLES4_: 36| 902k| simdutf_really_inline simd32 &operator+=(const simd32 other) { 37| 902k| value = _mm256_add_epi32(value, other.value); 38| 902k| return *this; 39| 902k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE3sumEv: 15| 705| simdutf_really_inline uint64_t sum() const { 16| 705| const __m256i mask = _mm256_set1_epi64x(0xffffffff); 17| 705| const __m256i t0 = _mm256_and_si256(value, mask); 18| 705| const __m256i t1 = _mm256_srli_epi64(value, 32); 19| 705| const __m256i t2 = _mm256_add_epi64(t0, t1); 20| | 21| 705| return uint64_t(_mm256_extract_epi64(t2, 0)) + 22| 705| uint64_t(_mm256_extract_epi64(t2, 1)) + 23| 705| uint64_t(_mm256_extract_epi64(t2, 2)) + 24| | uint64_t(_mm256_extract_epi64(t2, 3)); 25| 705| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd64ImE4zeroEv: 27| 5.92k| simdutf_really_inline static simd64 zero() { 28| 5.92k| return _mm256_setzero_si256(); 29| 5.92k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd64ImEC2EDv4_x: 9| 53.4k| simdutf_really_inline simd64(const __m256i v) : value(v) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd64ImEpLES4_: 21| 47.5k| simdutf_really_inline simd64 &operator+=(const simd64 other) { 22| 47.5k| value = _mm256_add_epi64(value, other.value); 23| 47.5k| return *this; 24| 47.5k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd64ImE3sumEv: 15| 5.92k| simdutf_really_inline uint64_t sum() const { 16| 5.92k| return _mm256_extract_epi64(value, 0) + _mm256_extract_epi64(value, 1) + 17| 5.92k| _mm256_extract_epi64(value, 2) + _mm256_extract_epi64(value, 3); 18| 5.92k| } _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| 4.40M|simdutf_really_inline long long int count_ones(uint64_t input_num) { 15| | return _popcnt64(input_num); 16| 4.40M|} _ZN7simdutf8westmere14implementationC2Ev: 18| 1| : simdutf::implementation("westmere", "Intel/AMD SSE4.2", 19| 1| internal::instruction_set::SSE42) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2Ev: 183| 20.8k| simdutf_really_inline simd8() : base8_numeric() {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEC2Ev: 118| 20.8k| simdutf_really_inline base8_numeric() : base8() {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2Ev: 68| 20.8k| simdutf_really_inline base8() : base>() {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2Ev: 13| 20.8k| simdutf_really_inline base() : value{__m128i()} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IhE9reduce_orEv: 290| 4.68M| simdutf_really_inline simd8 reduce_or() const { 291| 4.68M| return (this->chunks[0] | this->chunks[1]) | 292| 4.68M| (this->chunks[2] | this->chunks[3]); 293| 4.68M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEorES5_: 42| 31.7M| simdutf_really_inline Child operator|(const Child other) const { 43| 31.7M| return _mm_or_si128(*this, other); 44| 31.7M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2EDv2_x: 185| 324M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEC2EDv2_x: 120| 324M| : base8(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2EDv2_x: 69| 324M| simdutf_really_inline base8(const __m128i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2EDv2_x: 16| 324M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE8is_asciiEv: 228| 4.68M| simdutf_really_inline bool is_ascii() const { 229| 4.68M| return _mm_movemask_epi8(*this) == 0; 230| 4.68M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEoRES5_: 51| 10.8M| simdutf_really_inline Child &operator|=(const Child other) { 52| 10.8M| auto this_cast = static_cast(this); 53| 10.8M| *this_cast = *this_cast | other; 54| 10.8M| return *this_cast; 55| 10.8M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi1EEENS4_IhEES8_: 79| 7.10M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 80| | return _mm_alignr_epi8(*this, prev_chunk, 16 - N); 81| 7.10M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE3shrILi4EEES4_v: 238| 14.2M| template simdutf_really_inline simd8 shr() const { 239| 14.2M| return simd8(_mm_srli_epi16(*this, N)) & uint8_t(0xFFu >> N); 240| 14.2M| } 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| 21.3M| L replace15) const { 152| 21.3M| return lookup_16(simd8::repeat_16( 153| 21.3M| replace0, replace1, replace2, replace3, replace4, replace5, replace6, 154| 21.3M| replace7, replace8, replace9, replace10, replace11, replace12, 155| 21.3M| replace13, replace14, replace15)); 156| 21.3M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES8_: 142| 21.3M| simdutf_really_inline simd8 lookup_16(simd8 lookup_table) const { 143| 21.3M| return _mm_shuffle_epi8(lookup_table, *this); 144| 21.3M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE9repeat_16Ehhhhhhhhhhhhhhhh: 113| 21.3M| T v14, T v15) { 114| 21.3M| return simd8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, 115| 21.3M| v14, v15); 116| 21.3M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2Ehhhhhhhhhhhhhhhh: 196| 21.3M| : simd8(_mm_setr_epi8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, 197| 21.3M| v12, v13, v14, v15)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2Eh: 188| 48.3M| simdutf_really_inline simd8(uint8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi2EEENS4_IhEES8_: 79| 6.83M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 80| | return _mm_alignr_epi8(*this, prev_chunk, 16 - N); 81| 6.83M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi3EEENS4_IhEES8_: 79| 6.83M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 80| | return _mm_alignr_epi8(*this, prev_chunk, 16 - N); 81| 6.83M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE14saturating_subES4_: 201| 14.5M| saturating_sub(const simd8 other) const { 202| 14.5M| return _mm_subs_epu8(*this, other); 203| 14.5M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IbEC2EDv2_x: 91| 44.2M| simdutf_really_inline simd8(const __m128i _value) : base8(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IbNS2_5simd8IbEEEC2EDv2_x: 69| 44.2M| simdutf_really_inline base8(const __m128i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEC2EDv2_x: 16| 44.2M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEcvRKDv2_xEv: 18| 44.2M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEeoES5_: 48| 6.83M| simdutf_really_inline Child operator^(const Child other) const { 49| 6.83M| return _mm_xor_si128(*this, other); 50| 6.83M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2EPKh: 190| 905k| simdutf_really_inline simd8(const uint8_t *values) : simd8(load(values)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE4loadEPKh: 106| 19.6M| static simdutf_really_inline simd8 load(const T values[16]) { 107| 19.6M| return _mm_loadu_si128(reinterpret_cast(values)); 108| 19.6M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE7gt_bitsES4_: 212| 905k| gt_bits(const simd8 other) const { 213| 905k| return this->saturating_sub(other); 214| 905k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEC2EDv2_x: 16| 118M| 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| 13.1M| simdutf_really_inline int to_bitmask() const { 96| 13.1M| return _mm_movemask_epi8(*this); 97| 13.1M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE21any_bits_set_anywhereEv: 235| 2.27M| simdutf_really_inline bool any_bits_set_anywhere() const { 236| 2.27M| return !bits_not_set_anywhere(); 237| 2.27M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE21bits_not_set_anywhereEv: 232| 2.27M| simdutf_really_inline bool bits_not_set_anywhere() const { 233| 2.27M| return _mm_testz_si128(*this, *this); 234| 2.27M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaEC2EPKa: 270| 5.97M| : chunks{simd8::load(ptr), 271| 5.97M| simd8::load(ptr + sizeof(simd8) / sizeof(T)), 272| 5.97M| simd8::load(ptr + 2 * sizeof(simd8) / sizeof(T)), 273| 5.97M| simd8::load(ptr + 3 * sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaE4loadEPKa: 106| 42.4M| static simdutf_really_inline simd8 load(const T values[16]) { 107| 42.4M| return _mm_loadu_si128(reinterpret_cast(values)); 108| 42.4M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEC2EDv2_x: 163| 78.6M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaEC2EDv2_x: 120| 78.6M| : base8(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IaNS2_5simd8IbEEEC2EDv2_x: 69| 78.6M| simdutf_really_inline base8(const __m128i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEC2EDv2_x: 16| 78.6M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE8is_asciiEv: 295| 5.20M| simdutf_really_inline bool is_ascii() const { 296| 5.20M| return this->reduce_or().is_ascii(); 297| 5.20M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE9reduce_orEv: 290| 5.20M| simdutf_really_inline simd8 reduce_or() const { 291| 5.20M| return (this->chunks[0] | this->chunks[1]) | 292| 5.20M| (this->chunks[2] | this->chunks[3]); 293| 5.20M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEorES5_: 42| 15.6M| simdutf_really_inline Child operator|(const Child other) const { 43| 15.6M| return _mm_or_si128(*this, other); 44| 15.6M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEcvRKDv2_xEv: 18| 127M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaE8is_asciiEv: 168| 5.20M| simdutf_really_inline bool is_ascii() const { 169| 5.20M| return _mm_movemask_epi8(*this) == 0; 170| 5.20M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE5storeEPa: 275| 853k| simdutf_really_inline void store(T *ptr) const { 276| 853k| this->chunks[0].store(ptr + sizeof(simd8) * 0 / sizeof(T)); 277| 853k| this->chunks[1].store(ptr + sizeof(simd8) * 1 / sizeof(T)); 278| 853k| this->chunks[2].store(ptr + sizeof(simd8) * 2 / sizeof(T)); 279| 853k| this->chunks[3].store(ptr + sizeof(simd8) * 3 / sizeof(T)); 280| 853k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaE5storeEPa: 123| 3.41M| simdutf_really_inline void store(T dst[16]) const { 124| 3.41M| return _mm_storeu_si128(reinterpret_cast<__m128i *>(dst), *this); 125| 3.41M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEcvNS3_IhEEEv: 251| 6.10M|simdutf_really_inline simd8::operator simd8() const { 252| 6.10M| return this->value; 253| 6.10M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE2ltEa: 326| 1.16M| simdutf_really_inline uint64_t lt(const T m) const { 327| 1.16M| const simd8 mask = simd8::splat(m); 328| 1.16M| return simd8x64(this->chunks[0] < mask, this->chunks[1] < mask, 329| 1.16M| this->chunks[2] < mask, this->chunks[3] < mask) 330| 1.16M| .to_bitmask(); 331| 1.16M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaE5splatEa: 102| 20.5M| static simdutf_really_inline simd8 splat(T _value) { 103| 20.5M| return _mm_set1_epi8(_value); 104| 20.5M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEltES4_: 176| 4.65M| simdutf_really_inline simd8 operator<(const simd8 other) const { 177| 4.65M| return _mm_cmpgt_epi8(other, *this); 178| 4.65M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IbEC2ENS2_5simd8IbEES6_S6_S6_: 268| 1.93M| : chunks{chunk0, chunk1, chunk2, chunk3} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IbE10to_bitmaskEv: 318| 1.93M| simdutf_really_inline uint64_t to_bitmask() const { 319| 1.93M| uint64_t r0 = uint32_t(this->chunks[0].to_bitmask()); 320| 1.93M| uint64_t r1 = this->chunks[1].to_bitmask(); 321| 1.93M| uint64_t r2 = this->chunks[2].to_bitmask(); 322| 1.93M| uint64_t r3 = this->chunks[3].to_bitmask(); 323| 1.93M| return r0 | (r1 << 16) | (r2 << 32) | (r3 << 48); 324| 1.93M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf32EPDi: 311| 1.13M| simdutf_really_inline void store_ascii_as_utf32(char32_t *ptr) const { 312| 1.13M| this->chunks[0].store_ascii_as_utf32(ptr + sizeof(simd8) * 0); 313| 1.13M| this->chunks[1].store_ascii_as_utf32(ptr + sizeof(simd8) * 1); 314| 1.13M| this->chunks[2].store_ascii_as_utf32(ptr + sizeof(simd8) * 2); 315| 1.13M| this->chunks[3].store_ascii_as_utf32(ptr + sizeof(simd8) * 3); 316| 1.13M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf32EPDi: 32| 4.55M| simdutf_really_inline void store_ascii_as_utf32(char32_t *p) const { 33| 4.55M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p), _mm_cvtepu8_epi32(*this)); 34| 4.55M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 4), 35| 4.55M| _mm_cvtepu8_epi32(_mm_srli_si128(*this, 4))); 36| 4.55M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 8), 37| 4.55M| _mm_cvtepu8_epi32(_mm_srli_si128(*this, 8))); 38| 4.55M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 12), 39| | _mm_cvtepu8_epi32(_mm_srli_si128(*this, 12))); 40| 4.55M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE4zeroEv: 105| 196k| static simdutf_really_inline simd8 zero() { return _mm_setzero_si128(); } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEgtES4_: 173| 21.6M| simdutf_really_inline simd8 operator>(const simd8 other) const { 174| 21.6M| return _mm_cmpgt_epi8(*this, other); 175| 21.6M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEC2Ea: 165| 18.5M| simdutf_really_inline simd8(int8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEmIENS2_5simd8IhEE: 134| 24.2M| simdutf_really_inline simd8 &operator-=(const simd8 other) { 135| 24.2M| *this = *this - other; 136| 24.2M| return *static_cast *>(this); 137| 24.2M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEmiENS2_5simd8IhEE: 131| 24.2M| simdutf_really_inline simd8 operator-(const simd8 other) const { 132| 24.2M| return _mm_sub_epi8(*this, other); 133| 24.2M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd10sum_8bytesENS2_5simd8IhEE: 366| 95.3k|simdutf_really_inline simd64 sum_8bytes(const simd8 v) { 367| 95.3k| return _mm_sad_epu8(v.value, simd8::zero()); 368| 95.3k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE9sum_bytesEv: 245| 3.78k| simdutf_really_inline uint64_t sum_bytes() const { 246| 3.78k| const auto tmp = _mm_sad_epu8(value, _mm_setzero_si128()); 247| 3.78k| return _mm_extract_epi64(tmp, 0) + _mm_extract_epi64(tmp, 1); 248| 3.78k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEgeES4_: 217| 5.65M| operator>=(const simd8 other) const { 218| 5.65M| return other.min_val(*this) == other; 219| 5.65M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE7min_valES4_: 207| 5.65M| min_val(const simd8 other) const { 208| 5.65M| return _mm_min_epu8(*this, other); 209| 5.65M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE2gtEa: 333| 773k| simdutf_really_inline uint64_t gt(const T m) const { 334| 773k| const simd8 mask = simd8::splat(m); 335| 773k| return simd8x64(this->chunks[0] > mask, this->chunks[1] > mask, 336| 773k| this->chunks[2] > mask, this->chunks[3] > mask) 337| 773k| .to_bitmask(); 338| 773k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEC2EDv2_x: 16| 29.3M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEcvRKDv2_xEv: 18| 29.3M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE5splatEh: 102| 48.3M| static simdutf_really_inline simd8 splat(T _value) { 103| 48.3M| return _mm_set1_epi8(_value); 104| 48.3M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEcvRKDv2_xEv: 18| 398M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_5simd8IhEES4_: 72| 11.0M| const simd8 rhs) { 73| 11.0M| return _mm_cmpeq_epi8(lhs, rhs); 74| 11.0M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEanES5_: 45| 47.7M| simdutf_really_inline Child operator&(const Child other) const { 46| 47.7M| return _mm_and_si128(*this, other); 47| 47.7M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IhEC2EPKh: 270| 4.68M| : chunks{simd8::load(ptr), 271| 4.68M| simd8::load(ptr + sizeof(simd8) / sizeof(T)), 272| 4.68M| simd8::load(ptr + 2 * sizeof(simd8) / sizeof(T)), 273| 4.68M| simd8::load(ptr + 3 * sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEanES5_: 45| 17.9M| simdutf_really_inline Child operator&(const Child other) const { 46| 17.9M| return _mm_and_si128(*this, other); 47| 17.9M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 300| 947k| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 301| 947k| this->chunks[0].template store_ascii_as_utf16(ptr + 302| 947k| sizeof(simd8) * 0); 303| 947k| this->chunks[1].template store_ascii_as_utf16(ptr + 304| 947k| sizeof(simd8) * 1); 305| 947k| this->chunks[2].template store_ascii_as_utf16(ptr + 306| 947k| sizeof(simd8) * 2); 307| 947k| this->chunks[3].template store_ascii_as_utf16(ptr + 308| 947k| sizeof(simd8) * 3); 309| 947k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 20| 3.78M| simdutf_really_inline void store_ascii_as_utf16(char16_t *p) const { 21| 3.78M| __m128i first = _mm_cvtepu8_epi16(*this); 22| 3.78M| __m128i second = _mm_cvtepu8_epi16(_mm_srli_si128(*this, 8)); 23| 3.78M| if (big_endian) { ------------------ | Branch (23:9): [Folded, False: 3.78M] ------------------ 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| 3.78M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p), first); 30| 3.78M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 8), second); 31| 3.78M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 300| 1.10M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 301| 1.10M| this->chunks[0].template store_ascii_as_utf16(ptr + 302| 1.10M| sizeof(simd8) * 0); 303| 1.10M| this->chunks[1].template store_ascii_as_utf16(ptr + 304| 1.10M| sizeof(simd8) * 1); 305| 1.10M| this->chunks[2].template store_ascii_as_utf16(ptr + 306| 1.10M| sizeof(simd8) * 2); 307| 1.10M| this->chunks[3].template store_ascii_as_utf16(ptr + 308| 1.10M| sizeof(simd8) * 3); 309| 1.10M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 20| 4.41M| simdutf_really_inline void store_ascii_as_utf16(char16_t *p) const { 21| 4.41M| __m128i first = _mm_cvtepu8_epi16(*this); 22| 4.41M| __m128i second = _mm_cvtepu8_epi16(_mm_srli_si128(*this, 8)); 23| 4.41M| if (big_endian) { ------------------ | Branch (23:9): [True: 4.41M, Folded] ------------------ 24| 4.41M| const __m128i swap = 25| 4.41M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 26| 4.41M| first = _mm_shuffle_epi8(first, swap); 27| 4.41M| second = _mm_shuffle_epi8(second, swap); 28| 4.41M| } 29| 4.41M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p), first); 30| 4.41M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 8), second); 31| 4.41M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd12as_vector_u8ENS2_6simd32IjEE: 370| 10.3k|simdutf_really_inline simd8 as_vector_u8(const simd32 v) { 371| 10.3k| return simd8(v.value); 372| 10.3k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEorES5_: 42| 8.37M| simdutf_really_inline Child operator|(const Child other) const { 43| 8.37M| return _mm_or_si128(*this, other); 44| 8.37M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEC2EDv2_x: 79| 118M| : base16_numeric(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItEC2EDv2_x: 54| 118M| : base16(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6base16ItNS2_6simd16IbEEEC2EDv2_x: 8| 118M| : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE4packERKS4_S6_: 130| 5.34M| const simd16 &v1) { 131| 5.34M| return _mm_packus_epi16(v0, v1); 132| 5.34M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd9simd16x32ItEC2EPKt: 164| 2.09M| : chunks{simd16::load(ptr), 165| 2.09M| simd16::load(ptr + sizeof(simd16) / sizeof(T)), 166| 2.09M| simd16::load(ptr + 2 * sizeof(simd16) / sizeof(T)), 167| 2.09M| simd16::load(ptr + 3 * sizeof(simd16) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItE4loadEPKt: 47| 23.2M| static simdutf_really_inline simd16 load(const T values[8]) { 48| 23.2M| return _mm_loadu_si128(reinterpret_cast(values)); 49| 23.2M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd9simd16x32ItE10swap_bytesEv: 200| 999k| simdutf_really_inline void swap_bytes() { 201| 999k| this->chunks[0] = this->chunks[0].swap_bytes(); 202| 999k| this->chunks[1] = this->chunks[1].swap_bytes(); 203| 999k| this->chunks[2] = this->chunks[2].swap_bytes(); 204| 999k| this->chunks[3] = this->chunks[3].swap_bytes(); 205| 999k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE10swap_bytesEv: 121| 6.06M| simdutf_really_inline simd16 swap_bytes() const { 122| 6.06M| const __m128i swap = 123| 6.06M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 124| 6.06M| return _mm_shuffle_epi8(*this, swap); 125| 6.06M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItE5splatEt: 41| 23.7M| static simdutf_really_inline simd16 splat(T _value) { 42| 23.7M| return _mm_set1_epi16(_value); 43| 23.7M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_6simd16ItEES4_: 11| 20.9M| const simd16 rhs) { 12| 20.9M| return _mm_cmpeq_epi16(lhs, rhs); 13| 20.9M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16IbEC2EDv2_x: 28| 29.3M| simdutf_really_inline simd16(const __m128i _value) : base16(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6base16IbNS2_6simd16IbEEEC2EDv2_x: 8| 29.3M| : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd9simd16x32IbEC2ENS2_6simd16IbEES6_S6_S6_: 162| 2.09M| : chunks{chunk0, chunk1, chunk2, chunk3} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd9simd16x32IbE10to_bitmaskEv: 192| 2.09M| simdutf_really_inline uint64_t to_bitmask() const { 193| 2.09M| uint64_t r0 = uint32_t(this->chunks[0].to_bitmask()); 194| 2.09M| uint64_t r1 = this->chunks[1].to_bitmask(); 195| 2.09M| uint64_t r2 = this->chunks[2].to_bitmask(); 196| 2.09M| uint64_t r3 = this->chunks[3].to_bitmask(); 197| 2.09M| return r0 | (r1 << 16) | (r2 << 32) | (r3 << 48); 198| 2.09M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16IbE10to_bitmaskEv: 33| 8.37M| simdutf_really_inline int to_bitmask() const { 34| 8.37M| return _mm_movemask_epi8(*this); 35| 8.37M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEgeES4_: 104| 8.37M| operator>=(const simd16 other) const { 105| 8.37M| return other.min_val(*this) == other; 106| 8.37M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE7min_valES4_: 95| 8.37M| min_val(const simd16 other) const { 96| 8.37M| return _mm_min_epu16(*this, other); 97| 8.37M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE3shrILi8EEES4_v: 116| 5.34M| template simdutf_really_inline simd16 shr() const { 117| 5.34M| return simd16(_mm_srli_epi16(*this, N)); 118| 5.34M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEleES4_: 100| 8.37M| operator<=(const simd16 other) const { 101| 8.37M| return other.max_val(*this) == other; 102| 8.37M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE7max_valES4_: 90| 8.37M| max_val(const simd16 other) const { 91| 8.37M| return _mm_max_epu16(*this, other); 92| 8.37M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEC2EPKDs: 86| 10.6M| : simd16(load(reinterpret_cast(values))) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEC2Et: 82| 19.5M| simdutf_really_inline simd16(uint16_t _value) : simd16(splat(_value)) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd9simd16x32ItE12not_in_rangeEtt: 228| 2.09M| simdutf_really_inline uint64_t not_in_range(const T low, const T high) const { 229| 2.09M| const simd16 mask_low = simd16::splat(static_cast(low - 1)); 230| 2.09M| const simd16 mask_high = simd16::splat(static_cast(high + 1)); 231| 2.09M| return simd16x32( 232| 2.09M| (this->chunks[0] >= mask_high) | (this->chunks[0] <= mask_low), 233| 2.09M| (this->chunks[1] >= mask_high) | (this->chunks[1] <= mask_low), 234| 2.09M| (this->chunks[2] >= mask_high) | (this->chunks[2] <= mask_low), 235| 2.09M| (this->chunks[3] >= mask_high) | (this->chunks[3] <= mask_low)) 236| 2.09M| .to_bitmask(); 237| 2.09M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItE4zeroEv: 45| 1.62k| static simdutf_really_inline simd16 zero() { return _mm_setzero_si128(); } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd3minENS2_6simd16ItEES4_: 240| 8.42M|simd16 min(const simd16 a, simd16 b) { 241| 8.42M| return _mm_min_epu16(a.value, b.value); 242| 8.42M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItEpLENS2_6simd16ItEE: 68| 12.6M| simdutf_really_inline simd16 &operator+=(const simd16 other) { 69| 12.6M| *this = *this + other; 70| 12.6M| return *static_cast *>(this); 71| 12.6M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItEplENS2_6simd16ItEE: 65| 12.6M| simdutf_really_inline simd16 operator+(const simd16 other) const { 66| 12.6M| return _mm_add_epi16(*this, other); 67| 12.6M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE3sumEv: 134| 1.62k| simdutf_really_inline uint64_t sum() const { 135| 1.62k| const auto lo_u16 = _mm_and_si128(value, _mm_set1_epi32(0x0000ffff)); 136| 1.62k| const auto hi_u16 = _mm_srli_epi32(value, 16); 137| 1.62k| const auto sum_u32 = _mm_add_epi32(lo_u16, hi_u16); 138| | 139| 1.62k| const auto lo_u32 = _mm_and_si128(sum_u32, _mm_set1_epi64x(0xffffffff)); 140| 1.62k| const auto hi_u32 = _mm_srli_epi64(sum_u32, 32); 141| 1.62k| const auto sum_u64 = _mm_add_epi64(lo_u32, hi_u32); 142| | 143| 1.62k| return uint64_t(_mm_extract_epi64(sum_u64, 0)) + 144| | uint64_t(_mm_extract_epi64(sum_u64, 1)); 145| 1.62k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE5splatEj: 56| 35.3k| simdutf_really_inline static simd32 splat(uint32_t v) { 57| 35.3k| return _mm_set1_epi32(v); 58| 35.3k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjEC2EDv2_x: 9| 13.8M| simdutf_really_inline simd32(const __m128i v) : value(v) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE4zeroEv: 52| 6.20k| simdutf_really_inline static simd32 zero() { 53| 6.20k| return _mm_setzero_si128(); 54| 6.20k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjEC2IDiEEPKT_: 13| 3.33M| : 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| 4.59M| const simd32 b) { 95| 4.59M| return _mm_max_epu32(a.value, b.value); 96| 4.59M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdplENS2_6simd32IjEES4_: 119| 2.30M| const simd32 b) { 120| 2.30M| return _mm_add_epi32(a.value, b.value); 121| 2.30M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdgtENS2_6simd32IjEES4_: 143| 20.0k| const simd32 b) { 144| 20.0k| return !(b >= a); 145| 20.0k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdntENS2_6simd32IbEE: 138| 20.0k|simdutf_really_inline simd32 operator!(const simd32 v) { 139| 20.0k| return _mm_xor_si128(v.value, _mm_set1_epi8(-1)); 140| 20.0k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IbEC2EDv2_x: 69| 1.36M| simdutf_really_inline simd32(const __m128i v) : value(v) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdgeENS2_6simd32IjEES4_: 134| 455k| const simd32 b) { 135| 455k| return _mm_cmpeq_epi32(_mm_max_epu32(a.value, b.value), a.value); 136| 455k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IbE3anyEv: 71| 444k| simdutf_really_inline bool any() const { 72| 444k| return _mm_movemask_epi8(value) != 0; 73| 444k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_6simd32IjEES4_: 129| 435k| const simd32 b) { 130| 435k| return _mm_cmpeq_epi32(a.value, b.value); 131| 435k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdanENS2_6simd32IjEES4_: 104| 2.24M| const simd32 b) { 105| 2.24M| return _mm_and_si128(a.value, b.value); 106| 2.24M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdorIbEENS2_6simd32IT_EES6_S6_: 84| 445k| const simd32 b) { 85| 445k| return _mm_or_si128(a.value, b.value); 86| 445k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd3minENS2_6simd32IjEES4_: 89| 1.80M| const simd32 b) { 90| 1.80M| return _mm_min_epu32(a.value, b.value); 91| 1.80M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjEpLES4_: 46| 1.80M| simdutf_really_inline simd32 &operator+=(const simd32 other) { 47| 1.80M| value = _mm_add_epi32(value, other.value); 48| 1.80M| return *this; 49| 1.80M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE3sumEv: 15| 1.02k| simdutf_really_inline uint64_t sum() const { 16| 1.02k| return uint64_t(_mm_extract_epi32(value, 0)) + 17| 1.02k| uint64_t(_mm_extract_epi32(value, 1)) + 18| 1.02k| uint64_t(_mm_extract_epi32(value, 2)) + 19| | uint64_t(_mm_extract_epi32(value, 3)); 20| 1.02k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_6simd32IjEEj: 99| 10.3k| uint32_t b) { 100| 10.3k| return _mm_cmpeq_epi32(a.value, _mm_set1_epi32(b)); 101| 10.3k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdanENS2_6simd32IjEEj: 109| 31.1k| uint32_t b) { 110| 31.1k| return _mm_and_si128(a.value, _mm_set1_epi32(b)); 111| 31.1k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IbE15to_4bit_bitmaskEv: 75| 10.3k| simdutf_really_inline uint8_t to_4bit_bitmask() const { 76| 10.3k| return uint8_t(_mm_movemask_ps(_mm_castsi128_ps(value))); 77| 10.3k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdmiENS2_6simd32IjEEj: 124| 10.3k| uint32_t b) { 125| 10.3k| return _mm_sub_epi32(a.value, _mm_set1_epi32(b)); 126| 10.3k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE3shrILi10EEES4_v: 29| 10.3k| template simdutf_really_inline simd32 shr() const { 30| 10.3k| return _mm_srli_epi32(value, N); 31| 10.3k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE3shlILi16EEES4_v: 33| 10.3k| template simdutf_really_inline simd32 shl() const { 34| 10.3k| return _mm_slli_epi32(value, N); 35| 10.3k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdorENS2_6simd32IjEEj: 114| 10.3k| uint32_t b) { 115| 10.3k| return _mm_or_si128(a.value, _mm_set1_epi32(b)); 116| 10.3k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdorIjEENS2_6simd32IT_EES6_S6_: 84| 10.3k| const simd32 b) { 85| 10.3k| return _mm_or_si128(a.value, b.value); 86| 10.3k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6selectENS2_6simd32IbEENS3_IjEES5_: 149| 10.3k| const simd32 v_false) { 150| 10.3k| return _mm_blendv_epi8(v_false.value, v_true.value, cond.value); 151| 10.3k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd64ImE4zeroEv: 26| 5.92k| simdutf_really_inline static simd64 zero() { 27| 5.92k| return _mm_setzero_si128(); 28| 5.92k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd64ImEC2EDv2_x: 9| 101k| simdutf_really_inline simd64(const __m128i v) : value(v) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd64ImEpLES4_: 20| 95.3k| simdutf_really_inline simd64 &operator+=(const simd64 other) { 21| 95.3k| value = _mm_add_epi64(value, other.value); 22| 95.3k| return *this; 23| 95.3k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd64ImE3sumEv: 15| 5.92k| simdutf_really_inline uint64_t sum() const { 16| 5.92k| return _mm_extract_epi64(value, 0) + _mm_extract_epi64(value, 1); 17| 5.92k| } _ZNK7simdutf8westmere14implementation13validate_utf8EPKcm: 299| 3.47k|implementation::validate_utf8(const char *buf, size_t len) const noexcept { 300| 3.47k| return westmere::utf8_validation::generic_validate_utf8(buf, len); 301| 3.47k|} _ZNK7simdutf8westmere14implementation25validate_utf8_with_errorsEPKcm: 306| 3.47k| const char *buf, size_t len) const noexcept { 307| 3.47k| return westmere::utf8_validation::generic_validate_utf8_with_errors(buf, len); 308| 3.47k|} _ZNK7simdutf8westmere14implementation16validate_utf16leEPKDsm: 345| 1.40k| size_t len) const noexcept { 346| 1.40k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.40k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 17, False: 1.38k] | | ------------------ ------------------ 347| | // empty input is valid UTF-16. protect the implementation from 348| | // handling nullptr 349| 17| return true; 350| 17| } 351| 1.38k| const auto res = 352| 1.38k| westmere::utf16::validate_utf16_with_errors(buf, len); 353| 1.38k| if (res.is_err()) { ------------------ | Branch (353:7): [True: 165, False: 1.22k] ------------------ 354| 165| return false; 355| 165| } 356| | 357| 1.22k| if (res.count == len) ------------------ | Branch (357:7): [True: 0, False: 1.22k] ------------------ 358| 0| return true; 359| | 360| 1.22k| return scalar::utf16::validate(buf + res.count, 361| 1.22k| len - res.count); 362| 1.22k|} _ZNK7simdutf8westmere14implementation16validate_utf16beEPKDsm: 368| 1.31k| size_t len) const noexcept { 369| 1.31k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.31k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 1.29k] | | ------------------ ------------------ 370| | // empty input is valid UTF-16. protect the implementation from 371| | // handling nullptr 372| 16| return true; 373| 16| } 374| 1.29k| const auto res = 375| 1.29k| westmere::utf16::validate_utf16_with_errors(buf, len); 376| 1.29k| if (res.is_err()) { ------------------ | Branch (376:7): [True: 119, False: 1.17k] ------------------ 377| 119| return false; 378| 119| } 379| | 380| 1.17k| if (res.count == len) ------------------ | Branch (380:7): [True: 0, False: 1.17k] ------------------ 381| 0| return true; 382| | 383| 1.17k| return scalar::utf16::validate(buf + res.count, 384| 1.17k| len - res.count); 385| 1.17k|} _ZNK7simdutf8westmere14implementation28validate_utf16le_with_errorsEPKDsm: 388| 1.40k| const char16_t *buf, size_t len) const noexcept { 389| 1.40k| const result res = 390| 1.40k| westmere::utf16::validate_utf16_with_errors(buf, len); 391| 1.40k| if (res.count != len) { ------------------ | Branch (391:7): [True: 1.38k, False: 17] ------------------ 392| 1.38k| const result scalar_res = 393| 1.38k| scalar::utf16::validate_with_errors( 394| 1.38k| buf + res.count, len - res.count); 395| 1.38k| return result(scalar_res.error, res.count + scalar_res.count); 396| 1.38k| } else { 397| 17| return res; 398| 17| } 399| 1.40k|} _ZNK7simdutf8westmere14implementation28validate_utf16be_with_errorsEPKDsm: 402| 1.31k| const char16_t *buf, size_t len) const noexcept { 403| 1.31k| const result res = 404| 1.31k| westmere::utf16::validate_utf16_with_errors(buf, len); 405| 1.31k| if (res.count != len) { ------------------ | Branch (405:7): [True: 1.29k, False: 16] ------------------ 406| 1.29k| result scalar_res = scalar::utf16::validate_with_errors( 407| 1.29k| buf + res.count, len - res.count); 408| 1.29k| return result(scalar_res.error, res.count + scalar_res.count); 409| 1.29k| } else { 410| 16| return res; 411| 16| } 412| 1.31k|} _ZNK7simdutf8westmere14implementation14validate_utf32EPKDim: 427| 2.61k|implementation::validate_utf32(const char32_t *buf, size_t len) const noexcept { 428| 2.61k| return utf32::validate(buf, len); 429| 2.61k|} _ZNK7simdutf8westmere14implementation26validate_utf32_with_errorsEPKDim: 434| 2.61k| const char32_t *buf, size_t len) const noexcept { 435| 2.61k| return utf32::validate_with_errors(buf, len); 436| 2.61k|} _ZNK7simdutf8westmere14implementation22convert_latin1_to_utf8EPKcmPc: 441| 606| const char *buf, size_t len, char *utf8_output) const noexcept { 442| | 443| 606| std::pair ret = 444| 606| sse_convert_latin1_to_utf8(buf, len, utf8_output); 445| 606| size_t converted_chars = ret.second - utf8_output; 446| | 447| 606| if (ret.first != buf + len) { ------------------ | Branch (447:7): [True: 504, False: 102] ------------------ 448| 504| const size_t scalar_converted_chars = scalar::latin1_to_utf8::convert( 449| 504| ret.first, len - (ret.first - buf), ret.second); 450| 504| converted_chars += scalar_converted_chars; 451| 504| } 452| | 453| 606| return converted_chars; 454| 606|} _ZNK7simdutf8westmere14implementation25convert_latin1_to_utf16leEPKcmPDs: 459| 86| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 460| 86| std::pair ret = 461| 86| sse_convert_latin1_to_utf16(buf, len, utf16_output); 462| 86| if (ret.first == nullptr) { ------------------ | Branch (462:7): [True: 4, False: 82] ------------------ 463| 4| return 0; 464| 4| } 465| 82| size_t converted_chars = ret.second - utf16_output; 466| 82| if (ret.first != buf + len) { ------------------ | Branch (466:7): [True: 68, False: 14] ------------------ 467| 68| const size_t scalar_converted_chars = 468| 68| scalar::latin1_to_utf16::convert( 469| 68| ret.first, len - (ret.first - buf), ret.second); 470| 68| if (scalar_converted_chars == 0) { ------------------ | Branch (470:9): [True: 0, False: 68] ------------------ 471| 0| return 0; 472| 0| } 473| 68| converted_chars += scalar_converted_chars; 474| 68| } 475| 82| return converted_chars; 476| 82|} _ZNK7simdutf8westmere14implementation25convert_latin1_to_utf16beEPKcmPDs: 479| 88| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 480| 88| std::pair ret = 481| 88| sse_convert_latin1_to_utf16(buf, len, utf16_output); 482| 88| if (ret.first == nullptr) { ------------------ | Branch (482:7): [True: 4, False: 84] ------------------ 483| 4| return 0; 484| 4| } 485| 84| size_t converted_chars = ret.second - utf16_output; 486| 84| if (ret.first != buf + len) { ------------------ | Branch (486:7): [True: 68, False: 16] ------------------ 487| 68| const size_t scalar_converted_chars = 488| 68| scalar::latin1_to_utf16::convert( 489| 68| ret.first, len - (ret.first - buf), ret.second); 490| 68| if (scalar_converted_chars == 0) { ------------------ | Branch (490:9): [True: 0, False: 68] ------------------ 491| 0| return 0; 492| 0| } 493| 68| converted_chars += scalar_converted_chars; 494| 68| } 495| 84| return converted_chars; 496| 84|} _ZNK7simdutf8westmere14implementation23convert_latin1_to_utf32EPKcmPDi: 501| 110| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 502| 110| std::pair ret = 503| 110| sse_convert_latin1_to_utf32(buf, len, utf32_output); 504| 110| if (ret.first == nullptr) { ------------------ | Branch (504:7): [True: 4, False: 106] ------------------ 505| 4| return 0; 506| 4| } 507| 106| size_t converted_chars = ret.second - utf32_output; 508| 106| if (ret.first != buf + len) { ------------------ | Branch (508:7): [True: 90, False: 16] ------------------ 509| 90| const size_t scalar_converted_chars = scalar::latin1_to_utf32::convert( 510| 90| ret.first, len - (ret.first - buf), ret.second); 511| 90| if (scalar_converted_chars == 0) { ------------------ | Branch (511:9): [True: 0, False: 90] ------------------ 512| 0| return 0; 513| 0| } 514| 90| converted_chars += scalar_converted_chars; 515| 90| } 516| 106| return converted_chars; 517| 106|} _ZNK7simdutf8westmere14implementation22convert_utf8_to_latin1EPKcmPc: 522| 590| const char *buf, size_t len, char *latin1_output) const noexcept { 523| 590| utf8_to_latin1::validating_transcoder converter; 524| 590| return converter.convert(buf, len, latin1_output); 525| 590|} _ZNK7simdutf8westmere14implementation34convert_utf8_to_latin1_with_errorsEPKcmPc: 528| 502| const char *buf, size_t len, char *latin1_output) const noexcept { 529| 502| utf8_to_latin1::validating_transcoder converter; 530| 502| return converter.convert_with_errors(buf, len, latin1_output); 531| 502|} _ZNK7simdutf8westmere14implementation23convert_utf8_to_utf16leEPKcmPDs: 541| 720| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 542| 720| utf8_to_utf16::validating_transcoder converter; 543| 720| return converter.convert(buf, len, utf16_output); 544| 720|} _ZNK7simdutf8westmere14implementation23convert_utf8_to_utf16beEPKcmPDs: 547| 756| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 548| 756| utf8_to_utf16::validating_transcoder converter; 549| 756| return converter.convert(buf, len, utf16_output); 550| 756|} _ZNK7simdutf8westmere14implementation35convert_utf8_to_utf16le_with_errorsEPKcmPDs: 553| 612| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 554| 612| utf8_to_utf16::validating_transcoder converter; 555| 612| return converter.convert_with_errors(buf, len, 556| 612| utf16_output); 557| 612|} _ZNK7simdutf8westmere14implementation35convert_utf8_to_utf16be_with_errorsEPKcmPDs: 560| 754| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 561| 754| utf8_to_utf16::validating_transcoder converter; 562| 754| return converter.convert_with_errors(buf, len, utf16_output); 563| 754|} _ZNK7simdutf8westmere14implementation29convert_valid_utf8_to_utf16leEPKcmPDs: 566| 444| const char *input, size_t size, char16_t *utf16_output) const noexcept { 567| 444| return utf8_to_utf16::convert_valid(input, size, 568| 444| utf16_output); 569| 444|} _ZNK7simdutf8westmere14implementation29convert_valid_utf8_to_utf16beEPKcmPDs: 572| 476| const char *input, size_t size, char16_t *utf16_output) const noexcept { 573| 476| return utf8_to_utf16::convert_valid(input, size, 574| 476| utf16_output); 575| 476|} _ZNK7simdutf8westmere14implementation21convert_utf8_to_utf32EPKcmPDi: 580| 784| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 581| 784| utf8_to_utf32::validating_transcoder converter; 582| 784| return converter.convert(buf, len, utf32_output); 583| 784|} _ZNK7simdutf8westmere14implementation33convert_utf8_to_utf32_with_errorsEPKcmPDi: 586| 794| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 587| 794| utf8_to_utf32::validating_transcoder converter; 588| 794| return converter.convert_with_errors(buf, len, utf32_output); 589| 794|} _ZNK7simdutf8westmere14implementation27convert_valid_utf8_to_utf32EPKcmPDi: 592| 450| const char *input, size_t size, char32_t *utf32_output) const noexcept { 593| 450| return utf8_to_utf32::convert_valid(input, size, utf32_output); 594| 450|} _ZNK7simdutf8westmere14implementation25convert_utf16le_to_latin1EPKDsmPc: 599| 134| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 600| 134| std::pair ret = 601| 134| sse_convert_utf16_to_latin1(buf, len, latin1_output); 602| 134| if (ret.first == nullptr) { ------------------ | Branch (602:7): [True: 58, False: 76] ------------------ 603| 58| return 0; 604| 58| } 605| 76| size_t saved_bytes = ret.second - latin1_output; 606| | 607| 76| if (ret.first != buf + len) { ------------------ | Branch (607:7): [True: 58, False: 18] ------------------ 608| 58| const size_t scalar_saved_bytes = 609| 58| scalar::utf16_to_latin1::convert( 610| 58| ret.first, len - (ret.first - buf), ret.second); 611| 58| if (scalar_saved_bytes == 0) { ------------------ | Branch (611:9): [True: 28, False: 30] ------------------ 612| 28| return 0; 613| 28| } 614| 30| saved_bytes += scalar_saved_bytes; 615| 30| } 616| 48| return saved_bytes; 617| 76|} _ZNK7simdutf8westmere14implementation25convert_utf16be_to_latin1EPKDsmPc: 620| 94| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 621| 94| std::pair ret = 622| 94| sse_convert_utf16_to_latin1(buf, len, latin1_output); 623| 94| if (ret.first == nullptr) { ------------------ | Branch (623:7): [True: 28, False: 66] ------------------ 624| 28| return 0; 625| 28| } 626| 66| size_t saved_bytes = ret.second - latin1_output; 627| | 628| 66| if (ret.first != buf + len) { ------------------ | Branch (628:7): [True: 48, False: 18] ------------------ 629| 48| const size_t scalar_saved_bytes = 630| 48| scalar::utf16_to_latin1::convert( 631| 48| ret.first, len - (ret.first - buf), ret.second); 632| 48| if (scalar_saved_bytes == 0) { ------------------ | Branch (632:9): [True: 24, False: 24] ------------------ 633| 24| return 0; 634| 24| } 635| 24| saved_bytes += scalar_saved_bytes; 636| 24| } 637| 42| return saved_bytes; 638| 66|} _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: 142, False: 136] ------------------ 647| 142| return ret.first; 648| 142| } // Can return directly since scalar fallback already found correct 649| | // ret.first.count 650| 136| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (650:7): [True: 80, False: 56] ------------------ 651| 80| result scalar_res = 652| 80| scalar::utf16_to_latin1::convert_with_errors( 653| 80| buf + ret.first.count, len - ret.first.count, ret.second); 654| 80| if (scalar_res.error) { ------------------ | Branch (654:9): [True: 36, False: 44] ------------------ 655| 36| scalar_res.count += ret.first.count; 656| 36| return scalar_res; 657| 44| } else { 658| 44| ret.second += scalar_res.count; 659| 44| } 660| 80| } 661| 100| ret.first.count = 662| 100| ret.second - 663| 100| latin1_output; // Set count to the number of 8-bit code units written 664| 100| return ret.first; 665| 136|} _ZNK7simdutf8westmere14implementation37convert_utf16be_to_latin1_with_errorsEPKDsmPc: 669| 250| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 670| 250| std::pair ret = 671| 250| sse_convert_utf16_to_latin1_with_errors(buf, len, 672| 250| latin1_output); 673| 250| if (ret.first.error) { ------------------ | Branch (673:7): [True: 136, False: 114] ------------------ 674| 136| return ret.first; 675| 136| } // Can return directly since scalar fallback already found correct 676| | // ret.first.count 677| 114| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (677:7): [True: 76, False: 38] ------------------ 678| 76| result scalar_res = 679| 76| scalar::utf16_to_latin1::convert_with_errors( 680| 76| buf + ret.first.count, len - ret.first.count, ret.second); 681| 76| if (scalar_res.error) { ------------------ | Branch (681:9): [True: 28, False: 48] ------------------ 682| 28| scalar_res.count += ret.first.count; 683| 28| return scalar_res; 684| 48| } else { 685| 48| ret.second += scalar_res.count; 686| 48| } 687| 76| } 688| 86| ret.first.count = 689| 86| ret.second - 690| 86| latin1_output; // Set count to the number of 8-bit code units written 691| 86| return ret.first; 692| 114|} _ZNK7simdutf8westmere14implementation23convert_utf16le_to_utf8EPKDsmPc: 709| 856| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 710| 856| std::pair ret = 711| 856| sse_convert_utf16_to_utf8(buf, len, utf8_output); 712| 856| if (ret.first == nullptr) { ------------------ | Branch (712:7): [True: 120, False: 736] ------------------ 713| 120| return 0; 714| 120| } 715| 736| size_t saved_bytes = ret.second - utf8_output; 716| 736| if (ret.first != buf + len) { ------------------ | Branch (716:7): [True: 734, False: 2] ------------------ 717| 734| const size_t scalar_saved_bytes = 718| 734| scalar::utf16_to_utf8::convert( 719| 734| ret.first, len - (ret.first - buf), ret.second); 720| 734| if (scalar_saved_bytes == 0) { ------------------ | Branch (720:9): [True: 88, False: 646] ------------------ 721| 88| return 0; 722| 88| } 723| 646| saved_bytes += scalar_saved_bytes; 724| 646| } 725| 648| return saved_bytes; 726| 736|} _ZNK7simdutf8westmere14implementation23convert_utf16be_to_utf8EPKDsmPc: 729| 750| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 730| 750| std::pair ret = 731| 750| sse_convert_utf16_to_utf8(buf, len, utf8_output); 732| 750| if (ret.first == nullptr) { ------------------ | Branch (732:7): [True: 92, False: 658] ------------------ 733| 92| return 0; 734| 92| } 735| 658| size_t saved_bytes = ret.second - utf8_output; 736| 658| if (ret.first != buf + len) { ------------------ | Branch (736:7): [True: 656, False: 2] ------------------ 737| 656| const size_t scalar_saved_bytes = 738| 656| scalar::utf16_to_utf8::convert( 739| 656| ret.first, len - (ret.first - buf), ret.second); 740| 656| if (scalar_saved_bytes == 0) { ------------------ | Branch (740:9): [True: 66, False: 590] ------------------ 741| 66| return 0; 742| 66| } 743| 590| saved_bytes += scalar_saved_bytes; 744| 590| } 745| 592| return saved_bytes; 746| 658|} _ZNK7simdutf8westmere14implementation35convert_utf16le_to_utf8_with_errorsEPKDsmPc: 749| 664| 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| 664| std::pair ret = 753| 664| westmere::sse_convert_utf16_to_utf8_with_errors( 754| 664| buf, len, utf8_output); 755| 664| if (ret.first.error) { ------------------ | Branch (755:7): [True: 148, False: 516] ------------------ 756| 148| return ret.first; 757| 148| } // Can return directly since scalar fallback already found correct 758| | // ret.first.count 759| 516| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (759:7): [True: 512, False: 4] ------------------ 760| 512| result scalar_res = 761| 512| scalar::utf16_to_utf8::convert_with_errors( 762| 512| buf + ret.first.count, len - ret.first.count, ret.second); 763| 512| if (scalar_res.error) { ------------------ | Branch (763:9): [True: 122, False: 390] ------------------ 764| 122| scalar_res.count += ret.first.count; 765| 122| return scalar_res; 766| 390| } else { 767| 390| ret.second += scalar_res.count; 768| 390| } 769| 512| } 770| 394| ret.first.count = 771| 394| ret.second - 772| 394| utf8_output; // Set count to the number of 8-bit code units written 773| 394| return ret.first; 774| 516|} _ZNK7simdutf8westmere14implementation35convert_utf16be_to_utf8_with_errorsEPKDsmPc: 777| 728| 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| 728| std::pair ret = 781| 728| westmere::sse_convert_utf16_to_utf8_with_errors( 782| 728| buf, len, utf8_output); 783| 728| if (ret.first.error) { ------------------ | Branch (783:7): [True: 174, False: 554] ------------------ 784| 174| return ret.first; 785| 174| } // Can return directly since scalar fallback already found correct 786| | // ret.first.count 787| 554| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (787:7): [True: 550, False: 4] ------------------ 788| 550| result scalar_res = 789| 550| scalar::utf16_to_utf8::convert_with_errors( 790| 550| buf + ret.first.count, len - ret.first.count, ret.second); 791| 550| if (scalar_res.error) { ------------------ | Branch (791:9): [True: 132, False: 418] ------------------ 792| 132| scalar_res.count += ret.first.count; 793| 132| return scalar_res; 794| 418| } else { 795| 418| ret.second += scalar_res.count; 796| 418| } 797| 550| } 798| 422| ret.first.count = 799| 422| ret.second - 800| 422| utf8_output; // Set count to the number of 8-bit code units written 801| 422| return ret.first; 802| 554|} _ZNK7simdutf8westmere14implementation29convert_valid_utf16le_to_utf8EPKDsmPc: 805| 440| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 806| 440| return convert_utf16le_to_utf8(buf, len, utf8_output); 807| 440|} _ZNK7simdutf8westmere14implementation29convert_valid_utf16be_to_utf8EPKDsmPc: 810| 398| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 811| 398| return convert_utf16be_to_utf8(buf, len, utf8_output); 812| 398|} _ZNK7simdutf8westmere14implementation23convert_utf32_to_latin1EPKDimPc: 817| 172| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 818| 172| std::pair ret = 819| 172| sse_convert_utf32_to_latin1(buf, len, latin1_output); 820| 172| if (ret.first == nullptr) { ------------------ | Branch (820:7): [True: 30, False: 142] ------------------ 821| 30| return 0; 822| 30| } 823| 142| size_t saved_bytes = ret.second - latin1_output; 824| | // if (ret.first != buf + len) { 825| 142| if (ret.first < buf + len) { ------------------ | Branch (825:7): [True: 120, False: 22] ------------------ 826| 120| const size_t scalar_saved_bytes = scalar::utf32_to_latin1::convert( 827| 120| ret.first, len - (ret.first - buf), ret.second); 828| 120| if (scalar_saved_bytes == 0) { ------------------ | Branch (828:9): [True: 98, False: 22] ------------------ 829| 98| return 0; 830| 98| } 831| 22| saved_bytes += scalar_saved_bytes; 832| 22| } 833| 44| return saved_bytes; 834| 142|} _ZNK7simdutf8westmere14implementation35convert_utf32_to_latin1_with_errorsEPKDimPc: 837| 408| 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| 408| std::pair ret = 841| 408| westmere::sse_convert_utf32_to_latin1_with_errors(buf, len, 842| 408| latin1_output); 843| 408| if (ret.first.count != len) { ------------------ | Branch (843:7): [True: 384, False: 24] ------------------ 844| 384| result scalar_res = scalar::utf32_to_latin1::convert_with_errors( 845| 384| buf + ret.first.count, len - ret.first.count, ret.second); 846| 384| if (scalar_res.error) { ------------------ | Branch (846:9): [True: 362, False: 22] ------------------ 847| 362| scalar_res.count += ret.first.count; 848| 362| return scalar_res; 849| 362| } else { 850| 22| ret.second += scalar_res.count; 851| 22| } 852| 384| } 853| 46| ret.first.count = 854| 46| ret.second - 855| 46| latin1_output; // Set count to the number of 8-bit code units written 856| 46| return ret.first; 857| 408|} _ZNK7simdutf8westmere14implementation21convert_utf32_to_utf8EPKDimPc: 868| 916| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 869| 916| std::pair ret = 870| 916| sse_convert_utf32_to_utf8(buf, len, utf8_output); 871| 916| if (ret.first == nullptr) { ------------------ | Branch (871:7): [True: 302, False: 614] ------------------ 872| 302| return 0; 873| 302| } 874| 614| size_t saved_bytes = ret.second - utf8_output; 875| 614| if (ret.first != buf + len) { ------------------ | Branch (875:7): [True: 612, False: 2] ------------------ 876| 612| const size_t scalar_saved_bytes = scalar::utf32_to_utf8::convert( 877| 612| ret.first, len - (ret.first - buf), ret.second); 878| 612| if (scalar_saved_bytes == 0) { ------------------ | Branch (878:9): [True: 300, False: 312] ------------------ 879| 300| return 0; 880| 300| } 881| 312| saved_bytes += scalar_saved_bytes; 882| 312| } 883| 314| return saved_bytes; 884| 614|} _ZNK7simdutf8westmere14implementation33convert_utf32_to_utf8_with_errorsEPKDimPc: 887| 934| 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| 934| std::pair ret = 891| 934| westmere::sse_convert_utf32_to_utf8_with_errors(buf, len, utf8_output); 892| 934| if (ret.first.count != len) { ------------------ | Branch (892:7): [True: 930, False: 4] ------------------ 893| 930| result scalar_res = scalar::utf32_to_utf8::convert_with_errors( 894| 930| buf + ret.first.count, len - ret.first.count, ret.second); 895| 930| if (scalar_res.error) { ------------------ | Branch (895:9): [True: 700, False: 230] ------------------ 896| 700| scalar_res.count += ret.first.count; 897| 700| return scalar_res; 898| 700| } else { 899| 230| ret.second += scalar_res.count; 900| 230| } 901| 930| } 902| 234| ret.first.count = 903| 234| ret.second - 904| 234| utf8_output; // Set count to the number of 8-bit code units written 905| 234| return ret.first; 906| 934|} _ZNK7simdutf8westmere14implementation24convert_utf16le_to_utf32EPKDsmPDi: 911| 412| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 912| 412| std::pair ret = 913| 412| sse_convert_utf16_to_utf32(buf, len, utf32_output); 914| 412| if (ret.first == nullptr) { ------------------ | Branch (914:7): [True: 122, False: 290] ------------------ 915| 122| return 0; 916| 122| } 917| 290| size_t saved_bytes = ret.second - utf32_output; 918| 290| if (ret.first != buf + len) { ------------------ | Branch (918:7): [True: 220, False: 70] ------------------ 919| 220| const size_t scalar_saved_bytes = 920| 220| scalar::utf16_to_utf32::convert( 921| 220| ret.first, len - (ret.first - buf), ret.second); 922| 220| if (scalar_saved_bytes == 0) { ------------------ | Branch (922:9): [True: 56, False: 164] ------------------ 923| 56| return 0; 924| 56| } 925| 164| saved_bytes += scalar_saved_bytes; 926| 164| } 927| 234| return saved_bytes; 928| 290|} _ZNK7simdutf8westmere14implementation24convert_utf16be_to_utf32EPKDsmPDi: 931| 392| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 932| 392| std::pair ret = 933| 392| sse_convert_utf16_to_utf32(buf, len, utf32_output); 934| 392| if (ret.first == nullptr) { ------------------ | Branch (934:7): [True: 84, False: 308] ------------------ 935| 84| return 0; 936| 84| } 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| 418| 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| 418| std::pair ret = 955| 418| westmere::sse_convert_utf16_to_utf32_with_errors( 956| 418| buf, len, utf32_output); 957| 418| if (ret.first.error) { ------------------ | Branch (957:7): [True: 168, False: 250] ------------------ 958| 168| return ret.first; 959| 168| } // Can return directly since scalar fallback already found correct 960| | // ret.first.count 961| 250| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (961:7): [True: 202, False: 48] ------------------ 962| 202| result scalar_res = 963| 202| scalar::utf16_to_utf32::convert_with_errors( 964| 202| buf + ret.first.count, len - ret.first.count, ret.second); 965| 202| if (scalar_res.error) { ------------------ | Branch (965:9): [True: 68, False: 134] ------------------ 966| 68| scalar_res.count += ret.first.count; 967| 68| return scalar_res; 968| 134| } else { 969| 134| ret.second += scalar_res.count; 970| 134| } 971| 202| } 972| 182| ret.first.count = 973| 182| ret.second - 974| 182| utf32_output; // Set count to the number of 8-bit code units written 975| 182| return ret.first; 976| 250|} _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: 108, False: 238] ------------------ 986| 108| return ret.first; 987| 108| } // Can return directly since scalar fallback already found correct 988| | // ret.first.count 989| 238| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (989:7): [True: 188, False: 50] ------------------ 990| 188| result scalar_res = 991| 188| scalar::utf16_to_utf32::convert_with_errors( 992| 188| buf + ret.first.count, len - ret.first.count, ret.second); 993| 188| if (scalar_res.error) { ------------------ | Branch (993:9): [True: 72, False: 116] ------------------ 994| 72| scalar_res.count += ret.first.count; 995| 72| return scalar_res; 996| 116| } else { 997| 116| ret.second += scalar_res.count; 998| 116| } 999| 188| } 1000| 166| ret.first.count = 1001| 166| ret.second - 1002| 166| utf32_output; // Set count to the number of 8-bit code units written 1003| 166| return ret.first; 1004| 238|} _ZNK7simdutf8westmere14implementation27convert_valid_utf32_to_utf8EPKDimPc: 1009| 206| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 1010| 206| return convert_utf32_to_utf8(buf, len, utf8_output); 1011| 206|} _ZNK7simdutf8westmere14implementation24convert_utf32_to_utf16leEPKDimPDs: 1016| 774| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1017| 774| std::pair ret = 1018| 774| sse_convert_utf32_to_utf16(buf, len, utf16_output); 1019| 774| if (ret.first == nullptr) { ------------------ | Branch (1019:7): [True: 48, False: 726] ------------------ 1020| 48| return 0; 1021| 48| } 1022| 726| size_t saved_bytes = ret.second - utf16_output; 1023| 726| if (ret.first != buf + len) { ------------------ | Branch (1023:7): [True: 722, False: 4] ------------------ 1024| 722| const size_t scalar_saved_bytes = 1025| 722| scalar::utf32_to_utf16::convert( 1026| 722| ret.first, len - (ret.first - buf), ret.second); 1027| 722| if (scalar_saved_bytes == 0) { ------------------ | Branch (1027:9): [True: 330, False: 392] ------------------ 1028| 330| return 0; 1029| 330| } 1030| 392| saved_bytes += scalar_saved_bytes; 1031| 392| } 1032| 396| return saved_bytes; 1033| 726|} _ZNK7simdutf8westmere14implementation24convert_utf32_to_utf16beEPKDimPDs: 1036| 662| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1037| 662| std::pair ret = 1038| 662| sse_convert_utf32_to_utf16(buf, len, utf16_output); 1039| 662| if (ret.first == nullptr) { ------------------ | Branch (1039:7): [True: 106, False: 556] ------------------ 1040| 106| return 0; 1041| 106| } 1042| 556| size_t saved_bytes = ret.second - utf16_output; 1043| 556| if (ret.first != buf + len) { ------------------ | Branch (1043:7): [True: 554, False: 2] ------------------ 1044| 554| const size_t scalar_saved_bytes = 1045| 554| scalar::utf32_to_utf16::convert( 1046| 554| ret.first, len - (ret.first - buf), ret.second); 1047| 554| if (scalar_saved_bytes == 0) { ------------------ | Branch (1047:9): [True: 276, False: 278] ------------------ 1048| 276| return 0; 1049| 276| } 1050| 278| saved_bytes += scalar_saved_bytes; 1051| 278| } 1052| 280| return saved_bytes; 1053| 556|} _ZNK7simdutf8westmere14implementation36convert_utf32_to_utf16le_with_errorsEPKDimPDs: 1056| 684| 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| 684| std::pair ret = 1060| 684| westmere::sse_convert_utf32_to_utf16_with_errors( 1061| 684| buf, len, utf16_output); 1062| 684| if (ret.first.count != len) { ------------------ | Branch (1062:7): [True: 660, False: 24] ------------------ 1063| 660| result scalar_res = 1064| 660| scalar::utf32_to_utf16::convert_with_errors( 1065| 660| buf + ret.first.count, len - ret.first.count, ret.second); 1066| 660| if (scalar_res.error) { ------------------ | Branch (1066:9): [True: 504, False: 156] ------------------ 1067| 504| scalar_res.count += ret.first.count; 1068| 504| return scalar_res; 1069| 504| } else { 1070| 156| ret.second += scalar_res.count; 1071| 156| } 1072| 660| } 1073| 180| ret.first.count = 1074| 180| ret.second - 1075| 180| utf16_output; // Set count to the number of 8-bit code units written 1076| 180| return ret.first; 1077| 684|} _ZNK7simdutf8westmere14implementation36convert_utf32_to_utf16be_with_errorsEPKDimPDs: 1080| 630| 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| 630| std::pair ret = 1084| 630| westmere::sse_convert_utf32_to_utf16_with_errors( 1085| 630| buf, len, utf16_output); 1086| 630| if (ret.first.count != len) { ------------------ | Branch (1086:7): [True: 608, False: 22] ------------------ 1087| 608| result scalar_res = 1088| 608| scalar::utf32_to_utf16::convert_with_errors( 1089| 608| buf + ret.first.count, len - ret.first.count, ret.second); 1090| 608| if (scalar_res.error) { ------------------ | Branch (1090:9): [True: 458, False: 150] ------------------ 1091| 458| scalar_res.count += ret.first.count; 1092| 458| return scalar_res; 1093| 458| } else { 1094| 150| ret.second += scalar_res.count; 1095| 150| } 1096| 608| } 1097| 172| ret.first.count = 1098| 172| ret.second - 1099| 172| utf16_output; // Set count to the number of 8-bit code units written 1100| 172| return ret.first; 1101| 630|} _ZNK7simdutf8westmere14implementation30convert_valid_utf32_to_utf16leEPKDimPDs: 1104| 248| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1105| 248| return convert_utf32_to_utf16le(buf, len, utf16_output); 1106| 248|} _ZNK7simdutf8westmere14implementation30convert_valid_utf32_to_utf16beEPKDimPDs: 1109| 128| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1110| 128| return convert_utf32_to_utf16be(buf, len, utf16_output); 1111| 128|} _ZNK7simdutf8westmere14implementation30convert_valid_utf16le_to_utf32EPKDsmPDi: 1114| 106| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1115| 106| return convert_utf16le_to_utf32(buf, len, utf32_output); 1116| 106|} _ZNK7simdutf8westmere14implementation30convert_valid_utf16be_to_utf32EPKDsmPDi: 1119| 172| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1120| 172| return convert_utf16be_to_utf32(buf, len, utf32_output); 1121| 172|} _ZNK7simdutf8westmere14implementation13count_utf16leEPKDsm: 1132| 1.40k| const char16_t *input, size_t length) const noexcept { 1133| 1.40k| return utf16::count_code_points(input, length); 1134| 1.40k|} _ZNK7simdutf8westmere14implementation13count_utf16beEPKDsm: 1137| 1.31k| const char16_t *input, size_t length) const noexcept { 1138| 1.31k| return utf16::count_code_points(input, length); 1139| 1.31k|} _ZNK7simdutf8westmere14implementation10count_utf8EPKcm: 1144| 4.01k|implementation::count_utf8(const char *input, size_t length) const noexcept { 1145| 4.01k| return utf8::count_code_points_bytemask(input, length); 1146| 4.01k|} _ZNK7simdutf8westmere14implementation23latin1_length_from_utf8EPKcm: 1151| 546| const char *buf, size_t len) const noexcept { 1152| 546| return count_utf8(buf, len); 1153| 546|} _ZNK7simdutf8westmere14implementation24utf8_length_from_utf16leEPKDsm: 1158| 776| const char16_t *input, size_t length) const noexcept { 1159| 776| return utf16::utf8_length_from_utf16_bytemask(input, 1160| 776| length); 1161| 776|} _ZNK7simdutf8westmere14implementation24utf8_length_from_utf16beEPKDsm: 1164| 756| const char16_t *input, size_t length) const noexcept { 1165| 756| return utf16::utf8_length_from_utf16_bytemask(input, length); 1166| 756|} _ZNK7simdutf8westmere14implementation23utf8_length_from_latin1EPKcm: 1171| 303| const char *input, size_t len) const noexcept { 1172| 303| const uint8_t *str = reinterpret_cast(input); 1173| 303| size_t answer = len / sizeof(__m128i) * sizeof(__m128i); 1174| 303| size_t i = 0; 1175| 303| if (answer >= 2048) { // long strings optimization ------------------ | Branch (1175:7): [True: 142, False: 161] ------------------ 1176| 142| __m128i two_64bits = _mm_setzero_si128(); 1177| 5.28k| while (i + sizeof(__m128i) <= len) { ------------------ | Branch (1177:12): [True: 5.13k, False: 142] ------------------ 1178| 5.13k| __m128i runner = _mm_setzero_si128(); 1179| 5.13k| size_t iterations = (len - i) / sizeof(__m128i); 1180| 5.13k| if (iterations > 255) { ------------------ | Branch (1180:11): [True: 4.99k, False: 142] ------------------ 1181| 4.99k| iterations = 255; 1182| 4.99k| } 1183| 5.13k| size_t max_i = i + iterations * sizeof(__m128i) - sizeof(__m128i); 1184| 322k| for (; i + 4 * sizeof(__m128i) <= max_i; i += 4 * sizeof(__m128i)) { ------------------ | Branch (1184:14): [True: 317k, False: 5.13k] ------------------ 1185| 317k| __m128i input1 = _mm_loadu_si128((const __m128i *)(str + i)); 1186| 317k| __m128i input2 = 1187| 317k| _mm_loadu_si128((const __m128i *)(str + i + sizeof(__m128i))); 1188| 317k| __m128i input3 = 1189| 317k| _mm_loadu_si128((const __m128i *)(str + i + 2 * sizeof(__m128i))); 1190| 317k| __m128i input4 = 1191| 317k| _mm_loadu_si128((const __m128i *)(str + i + 3 * sizeof(__m128i))); 1192| 317k| __m128i input12 = 1193| 317k| _mm_add_epi8(_mm_cmpgt_epi8(_mm_setzero_si128(), input1), 1194| 317k| _mm_cmpgt_epi8(_mm_setzero_si128(), input2)); 1195| 317k| __m128i input34 = 1196| 317k| _mm_add_epi8(_mm_cmpgt_epi8(_mm_setzero_si128(), input3), 1197| 317k| _mm_cmpgt_epi8(_mm_setzero_si128(), input4)); 1198| 317k| __m128i input1234 = _mm_add_epi8(input12, input34); 1199| 317k| runner = _mm_sub_epi8(runner, input1234); 1200| 317k| } 1201| 20.4k| for (; i <= max_i; i += sizeof(__m128i)) { ------------------ | Branch (1201:14): [True: 15.3k, False: 5.13k] ------------------ 1202| 15.3k| __m128i more_input = _mm_loadu_si128((const __m128i *)(str + i)); 1203| 15.3k| runner = _mm_sub_epi8(runner, 1204| 15.3k| _mm_cmpgt_epi8(_mm_setzero_si128(), more_input)); 1205| 15.3k| } 1206| 5.13k| two_64bits = 1207| 5.13k| _mm_add_epi64(two_64bits, _mm_sad_epu8(runner, _mm_setzero_si128())); 1208| 5.13k| } 1209| 142| answer += 1210| 142| _mm_extract_epi64(two_64bits, 0) + _mm_extract_epi64(two_64bits, 1); 1211| 161| } else if (answer > 0) { // short string optimization ------------------ | Branch (1211:14): [True: 107, False: 54] ------------------ 1212| 1.20k| for (; i + 2 * sizeof(__m128i) <= len; i += 2 * sizeof(__m128i)) { ------------------ | Branch (1212:12): [True: 1.10k, False: 107] ------------------ 1213| 1.10k| __m128i latin = _mm_loadu_si128((const __m128i *)(input + i)); 1214| 1.10k| uint16_t non_ascii = (uint16_t)_mm_movemask_epi8(latin); 1215| 1.10k| answer += count_ones(non_ascii); 1216| 1.10k| latin = _mm_loadu_si128((const __m128i *)(input + i) + 1); 1217| 1.10k| non_ascii = (uint16_t)_mm_movemask_epi8(latin); 1218| 1.10k| answer += count_ones(non_ascii); 1219| 1.10k| } 1220| 155| for (; i + sizeof(__m128i) <= len; i += sizeof(__m128i)) { ------------------ | Branch (1220:12): [True: 48, False: 107] ------------------ 1221| 48| __m128i latin = _mm_loadu_si128((const __m128i *)(input + i)); 1222| 48| uint16_t non_ascii = (uint16_t)_mm_movemask_epi8(latin); 1223| 48| answer += count_ones(non_ascii); 1224| 48| } 1225| 107| } 1226| 303| return answer + scalar::latin1::utf8_length_from_latin1( 1227| 303| reinterpret_cast(str + i), len - i); 1228| 303|} _ZNK7simdutf8westmere14implementation25utf32_length_from_utf16leEPKDsm: 1233| 420| const char16_t *input, size_t length) const noexcept { 1234| 420| return utf16::utf32_length_from_utf16(input, length); 1235| 420|} _ZNK7simdutf8westmere14implementation25utf32_length_from_utf16beEPKDsm: 1238| 382| const char16_t *input, size_t length) const noexcept { 1239| 382| return utf16::utf32_length_from_utf16(input, length); 1240| 382|} _ZNK7simdutf8westmere14implementation22utf16_length_from_utf8EPKcm: 1245| 1.90k| const char *input, size_t length) const noexcept { 1246| 1.90k| return utf8::utf16_length_from_utf8_bytemask(input, length); 1247| 1.90k|} _ZNK7simdutf8westmere14implementation22utf8_length_from_utf32EPKDim: 1292| 937| const char32_t *input, size_t length) const noexcept { 1293| 937| return utf32::utf8_length_from_utf32(input, length); 1294| 937|} _ZNK7simdutf8westmere14implementation23utf16_length_from_utf32EPKDim: 1299| 1.38k| const char32_t *input, size_t length) const noexcept { 1300| 1.38k| const __m128i v_00000000 = _mm_setzero_si128(); 1301| 1.38k| const __m128i v_ffff0000 = _mm_set1_epi32((uint32_t)0xffff0000); 1302| 1.38k| size_t pos = 0; 1303| 1.38k| size_t count = 0; 1304| 1.52M| for (; pos + 4 <= length; pos += 4) { ------------------ | Branch (1304:10): [True: 1.52M, False: 1.38k] ------------------ 1305| 1.52M| __m128i in = _mm_loadu_si128((__m128i *)(input + pos)); 1306| 1.52M| const __m128i surrogate_bytemask = 1307| 1.52M| _mm_cmpeq_epi32(_mm_and_si128(in, v_ffff0000), v_00000000); 1308| 1.52M| const uint16_t surrogate_bitmask = 1309| 1.52M| static_cast(_mm_movemask_epi8(surrogate_bytemask)); 1310| 1.52M| size_t surrogate_count = (16 - count_ones(surrogate_bitmask)) / 4; 1311| 1.52M| count += 4 + surrogate_count; 1312| 1.52M| } 1313| 1.38k| return count + 1314| 1.38k| scalar::utf32::utf16_length_from_utf32(input + pos, length - pos); 1315| 1.38k|} _ZNK7simdutf8westmere14implementation22utf32_length_from_utf8EPKcm: 1320| 1.01k| const char *input, size_t length) const noexcept { 1321| 1.01k| return utf8::count_code_points(input, length); 1322| 1.01k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_18is_asciiERKNS1_4simd8simd8x64IhEE: 13| 4.68M|simdutf_really_inline bool is_ascii(const simd8x64 &input) { 14| 4.68M| return input.reduce_or().is_ascii(); 15| 4.68M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_124must_be_2_3_continuationENS1_4simd5simd8IhEES4_: 22| 6.83M| const simd8 prev3) { 23| 6.83M| simd8 is_third_byte = 24| 6.83M| prev2.saturating_sub(0xe0u - 0x80); // Only 111_____ will be >= 0x80 25| 6.83M| simd8 is_fourth_byte = 26| 6.83M| prev3.saturating_sub(0xf0u - 0x80); // Only 1111____ will be >= 0x80 27| 6.83M| return simd8(is_third_byte | is_fourth_byte); 28| 6.83M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_18internal8westmere26write_v_u16_11bits_to_utf8EDv2_xRPcS4_S4_: 57| 651k| const __m128i v_ff80) { 58| | // no bits set above 7th bit 59| 651k| const __m128i one_byte_bytemask = 60| 651k| _mm_cmpeq_epi16(_mm_and_si128(v_u16, v_ff80), v_0000); 61| 651k| const uint16_t one_byte_bitmask = 62| 651k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 63| | 64| 651k| write_v_u16_11bits_to_utf8(v_u16, utf8_output, one_byte_bytemask, 65| 651k| one_byte_bitmask); 66| 651k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_18internal8westmere26write_v_u16_11bits_to_utf8EDv2_xRPcS4_t: 10| 1.46M| const uint16_t one_byte_bitmask) { 11| | // 0b1100_0000_1000_0000 12| 1.46M| const __m128i v_c080 = _mm_set1_epi16((int16_t)0xc080); 13| | // 0b0001_1111_0000_0000 14| 1.46M| const __m128i v_1f00 = _mm_set1_epi16((int16_t)0x1f00); 15| | // 0b0000_0000_0011_1111 16| 1.46M| 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| 1.46M| const __m128i t0 = _mm_slli_epi16(v_u16, 2); 24| | // t1 = [000a|aaaa|0000|0000] 25| 1.46M| const __m128i t1 = _mm_and_si128(t0, v_1f00); 26| | // t2 = [0000|0000|00bb|bbbb] 27| 1.46M| const __m128i t2 = _mm_and_si128(v_u16, v_003f); 28| | // t3 = [000a|aaaa|00bb|bbbb] 29| 1.46M| const __m128i t3 = _mm_or_si128(t1, t2); 30| | // t4 = [110a|aaaa|10bb|bbbb] 31| 1.46M| const __m128i t4 = _mm_or_si128(t3, v_c080); 32| | 33| | // 2. merge ASCII and 2-byte codewords 34| 1.46M| 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| 1.46M| const uint16_t m0 = one_byte_bitmask & 0x5555; // m0 = 0h0g0f0e0d0c0b0a 40| 1.46M| const uint16_t m1 = static_cast(m0 >> 7); // m1 = 00000000h0g0f0e0 41| 1.46M| const uint8_t m2 = static_cast((m0 | m1) & 0xff); // m2 = hdgcfbea 42| | // 4. pack the bytes 43| 1.46M| const uint8_t *row = 44| 1.46M| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[m2][0]; 45| 1.46M| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 46| 1.46M| const __m128i utf8_packed = _mm_shuffle_epi8(utf8_unpacked, shuffle); 47| | 48| | // 5. store bytes 49| 1.46M| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 50| | 51| | // 6. adjust pointers 52| 1.46M| utf8_output += row[0]; 53| 1.46M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_latin1_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 86| char16_t *utf16_output) { 5| 86| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| 106k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (6:22): [True: 106k, False: 86] ------------------ 7| | // Load 16 Latin1 characters into a 128-bit register 8| 106k| __m128i in = 9| 106k| _mm_loadu_si128(reinterpret_cast(&latin1_input[i])); 10| 106k| __m128i out1 = big_endian ? _mm_unpacklo_epi8(_mm_setzero_si128(), in) ------------------ | Branch (10:20): [Folded, False: 106k] ------------------ 11| 106k| : _mm_unpacklo_epi8(in, _mm_setzero_si128()); 12| 106k| __m128i out2 = big_endian ? _mm_unpackhi_epi8(_mm_setzero_si128(), in) ------------------ | Branch (12:20): [Folded, False: 106k] ------------------ 13| 106k| : _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| 106k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i]), out1); 17| 106k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i + 8]), out2); 18| 106k| } 19| | // return pointers pointing to where we left off 20| 86| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 21| 86|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_latin1_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 88| char16_t *utf16_output) { 5| 88| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| 121k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (6:22): [True: 121k, False: 88] ------------------ 7| | // Load 16 Latin1 characters into a 128-bit register 8| 121k| __m128i in = 9| 121k| _mm_loadu_si128(reinterpret_cast(&latin1_input[i])); 10| 121k| __m128i out1 = big_endian ? _mm_unpacklo_epi8(_mm_setzero_si128(), in) ------------------ | Branch (10:20): [True: 121k, Folded] ------------------ 11| 121k| : _mm_unpacklo_epi8(in, _mm_setzero_si128()); 12| 121k| __m128i out2 = big_endian ? _mm_unpackhi_epi8(_mm_setzero_si128(), in) ------------------ | Branch (12:20): [True: 121k, Folded] ------------------ 13| 121k| : _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| 121k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i]), out1); 17| 121k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i + 8]), out2); 18| 121k| } 19| | // return pointers pointing to where we left off 20| 88| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 21| 88|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_latin1_to_utf32EPKcmPDi: 3| 110| char32_t *utf32_output) { 4| 110| const char *end = buf + len; 5| | 6| 518k| while (end - buf >= 16) { ------------------ | Branch (6:10): [True: 518k, False: 110] ------------------ 7| | // Load 16 Latin1 characters (16 bytes) into a 128-bit register 8| 518k| __m128i in = _mm_loadu_si128((__m128i *)buf); 9| | 10| | // Shift input to process next 4 bytes 11| 518k| __m128i in_shifted1 = _mm_srli_si128(in, 4); 12| 518k| __m128i in_shifted2 = _mm_srli_si128(in, 8); 13| 518k| __m128i in_shifted3 = _mm_srli_si128(in, 12); 14| | 15| | // expand 8-bit to 32-bit unit 16| 518k| __m128i out1 = _mm_cvtepu8_epi32(in); 17| 518k| __m128i out2 = _mm_cvtepu8_epi32(in_shifted1); 18| 518k| __m128i out3 = _mm_cvtepu8_epi32(in_shifted2); 19| 518k| __m128i out4 = _mm_cvtepu8_epi32(in_shifted3); 20| | 21| 518k| _mm_storeu_si128((__m128i *)utf32_output, out1); 22| 518k| _mm_storeu_si128((__m128i *)(utf32_output + 4), out2); 23| 518k| _mm_storeu_si128((__m128i *)(utf32_output + 8), out3); 24| 518k| _mm_storeu_si128((__m128i *)(utf32_output + 12), out4); 25| | 26| 518k| utf32_output += 16; 27| 518k| buf += 16; 28| 518k| } 29| | 30| 110| return std::make_pair(buf, utf32_output); 31| 110|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_latin1_to_utf8EPKcmPc: 3| 606| const size_t latin_input_length, char *utf8_output) { 4| 606| const char *end = latin_input + latin_input_length; 5| | 6| 606| const __m128i v_0000 = _mm_setzero_si128(); 7| | // 0b1000_0000 8| 606| const __m128i v_80 = _mm_set1_epi8((uint8_t)0x80); 9| | // 0b1111_1111_1000_0000 10| 606| const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80); 11| | 12| 606| const __m128i latin_1_half_into_u16_byte_mask = 13| 606| _mm_setr_epi8(0, '\x80', 1, '\x80', 2, '\x80', 3, '\x80', 4, '\x80', 5, 14| 606| '\x80', 6, '\x80', 7, '\x80'); 15| | 16| 606| const __m128i latin_2_half_into_u16_byte_mask = 17| 606| _mm_setr_epi8(8, '\x80', 9, '\x80', 10, '\x80', 11, '\x80', 12, '\x80', 18| 606| 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| 2.57M| while (end - latin_input >= 16 + 8) { ------------------ | Branch (25:10): [True: 2.57M, False: 606] ------------------ 26| | // Load 16 Latin1 characters (16 bytes) into a 128-bit register 27| 2.57M| __m128i v_latin = _mm_loadu_si128((__m128i *)latin_input); 28| | 29| 2.57M| if (_mm_testz_si128(v_latin, v_80)) { // ASCII fast path!!!! ------------------ | Branch (29:9): [True: 2.24M, False: 325k] ------------------ 30| 2.24M| _mm_storeu_si128((__m128i *)utf8_output, v_latin); 31| 2.24M| latin_input += 16; 32| 2.24M| utf8_output += 16; 33| 2.24M| continue; 34| 2.24M| } 35| | 36| | // assuming a/b are bytes and A/B are uint16 of the same value 37| | // aaaa_aaaa_bbbb_bbbb -> AAAA_AAAA 38| 325k| __m128i v_u16_latin_1_half = 39| 325k| _mm_shuffle_epi8(v_latin, latin_1_half_into_u16_byte_mask); 40| | // aaaa_aaaa_bbbb_bbbb -> BBBB_BBBB 41| 325k| __m128i v_u16_latin_2_half = 42| 325k| _mm_shuffle_epi8(v_latin, latin_2_half_into_u16_byte_mask); 43| | 44| 325k| internal::westmere::write_v_u16_11bits_to_utf8(v_u16_latin_1_half, 45| 325k| utf8_output, v_0000, v_ff80); 46| 325k| internal::westmere::write_v_u16_11bits_to_utf8(v_u16_latin_2_half, 47| 325k| utf8_output, v_0000, v_ff80); 48| 325k| latin_input += 16; 49| 325k| } 50| | 51| 606| if (end - latin_input >= 16) { ------------------ | Branch (51:7): [True: 332, False: 274] ------------------ 52| | // Load 16 Latin1 characters (16 bytes) into a 128-bit register 53| 332| __m128i v_latin = _mm_loadu_si128((__m128i *)latin_input); 54| | 55| 332| if (_mm_testz_si128(v_latin, v_80)) { // ASCII fast path!!!! ------------------ | Branch (55:9): [True: 128, False: 204] ------------------ 56| 128| _mm_storeu_si128((__m128i *)utf8_output, v_latin); 57| 128| latin_input += 16; 58| 128| utf8_output += 16; 59| 204| } else { 60| | // assuming a/b are bytes and A/B are uint16 of the same value 61| | // aaaa_aaaa_bbbb_bbbb -> AAAA_AAAA 62| 204| __m128i v_u16_latin_1_half = 63| 204| _mm_shuffle_epi8(v_latin, latin_1_half_into_u16_byte_mask); 64| 204| internal::westmere::write_v_u16_11bits_to_utf8( 65| 204| v_u16_latin_1_half, utf8_output, v_0000, v_ff80); 66| 204| latin_input += 8; 67| 204| } 68| 332| } 69| | 70| 606| return std::make_pair(latin_input, utf8_output); 71| 606|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_utf16_to_latin1ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 134| char *latin1_output) { 5| 134| const char16_t *end = buf + len; 6| 8.89k| while (end - buf >= 8) { ------------------ | Branch (6:10): [True: 8.81k, False: 80] ------------------ 7| | // Load 8 UTF-16 characters into 128-bit SSE register 8| 8.81k| __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| 8.81k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 17| 8.81k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (17:9): [True: 8.76k, False: 54] ------------------ 18| | // Pack 16-bit characters into 8-bit and store in latin1_output 19| 8.76k| __m128i latin1_packed = _mm_packus_epi16(in, in); 20| 8.76k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 21| 8.76k| latin1_packed); 22| | // Adjust pointers for next iteration 23| 8.76k| buf += 8; 24| 8.76k| latin1_output += 8; 25| 8.76k| } else { 26| 54| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 27| 54| } 28| 8.81k| } // while 29| 80| return std::make_pair(buf, latin1_output); 30| 134|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_utf16_to_latin1ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 94| char *latin1_output) { 5| 94| const char16_t *end = buf + len; 6| 866| while (end - buf >= 8) { ------------------ | Branch (6:10): [True: 796, False: 70] ------------------ 7| | // Load 8 UTF-16 characters into 128-bit SSE register 8| 796| __m128i in = _mm_loadu_si128(reinterpret_cast(buf)); 9| | 10| 796| if constexpr (!match_system(big_endian)) { 11| 796| const __m128i swap = 12| 796| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 13| 796| in = _mm_shuffle_epi8(in, swap); 14| 796| } 15| | 16| 796| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 17| 796| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (17:9): [True: 772, False: 24] ------------------ 18| | // Pack 16-bit characters into 8-bit and store in latin1_output 19| 772| __m128i latin1_packed = _mm_packus_epi16(in, in); 20| 772| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 21| 772| latin1_packed); 22| | // Adjust pointers for next iteration 23| 772| buf += 8; 24| 772| latin1_output += 8; 25| 772| } else { 26| 24| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 27| 24| } 28| 796| } // while 29| 70| return std::make_pair(buf, latin1_output); 30| 94|} 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| 823k| while (end - buf >= 8) { ------------------ | Branch (38:10): [True: 823k, False: 136] ------------------ 39| 823k| __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| 823k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 48| 823k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (48:9): [True: 823k, False: 142] ------------------ 49| 823k| __m128i latin1_packed = _mm_packus_epi16(in, in); 50| 823k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 51| 823k| latin1_packed); 52| 823k| buf += 8; 53| 823k| latin1_output += 8; 54| 823k| } else { 55| | // Fallback to scalar code for handling errors 56| 462| for (int k = 0; k < 8; k++) { ------------------ | Branch (56:23): [True: 462, False: 0] ------------------ 57| 462| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 58| 462| if (word <= 0xff) { ------------------ | Branch (58:13): [True: 320, False: 142] ------------------ 59| 320| *latin1_output++ = char(word); 60| 320| } else { 61| 142| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 62| 142| latin1_output); 63| 142| } 64| 462| } 65| 0| buf += 8; 66| 0| } 67| 823k| } // while 68| 136| return std::make_pair(result(error_code::SUCCESS, buf - start), 69| 136| latin1_output); 70| 278|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_139sse_convert_utf16_to_latin1_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 35| 250| char *latin1_output) { 36| 250| const char16_t *start = buf; 37| 250| const char16_t *end = buf + len; 38| 927k| while (end - buf >= 8) { ------------------ | Branch (38:10): [True: 927k, False: 114] ------------------ 39| 927k| __m128i in = _mm_loadu_si128(reinterpret_cast(buf)); 40| | 41| 927k| if constexpr (!match_system(big_endian)) { 42| 927k| const __m128i swap = 43| 927k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 44| 927k| in = _mm_shuffle_epi8(in, swap); 45| 927k| } 46| | 47| 927k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 48| 927k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (48:9): [True: 926k, False: 136] ------------------ 49| 926k| __m128i latin1_packed = _mm_packus_epi16(in, in); 50| 926k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 51| 926k| latin1_packed); 52| 926k| buf += 8; 53| 926k| latin1_output += 8; 54| 926k| } else { 55| | // Fallback to scalar code for handling errors 56| 476| for (int k = 0; k < 8; k++) { ------------------ | Branch (56:23): [True: 476, False: 0] ------------------ 57| 476| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 58| 476| if (word <= 0xff) { ------------------ | Branch (58:13): [True: 340, False: 136] ------------------ 59| 340| *latin1_output++ = char(word); 60| 340| } else { 61| 136| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 62| 136| latin1_output); 63| 136| } 64| 476| } 65| 0| buf += 8; 66| 0| } 67| 927k| } // while 68| 114| return std::make_pair(result(error_code::SUCCESS, buf - start), 69| 114| latin1_output); 70| 250|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf16_to_utf32ILNS_10endiannessE0EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 412| char32_t *utf32_output) { 58| 412| const char16_t *end = buf + len; 59| | 60| 412| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 61| 412| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 62| | 63| 289k| while (end - buf >= 8) { ------------------ | Branch (63:10): [True: 289k, False: 296] ------------------ 64| 289k| __m128i in = _mm_loadu_si128((__m128i *)buf); 65| | 66| 289k| if (big_endian) { ------------------ | Branch (66:9): [Folded, False: 289k] ------------------ 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| 289k| const __m128i surrogates_bytemask = 76| 289k| _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| 289k| const uint16_t surrogates_bitmask = 81| 289k| 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| 289k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (84:9): [True: 284k, False: 4.35k] ------------------ 85| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 86| 284k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 87| 284k| _mm_cvtepu16_epi32(in)); 88| 284k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 89| 284k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 90| 284k| utf32_output += 8; 91| 284k| buf += 8; 92| | // surrogate pair(s) in a register 93| 284k| } 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| 4.35k| size_t forward = 15; 98| 4.35k| size_t k = 0; 99| 4.35k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (99:11): [True: 58, False: 4.29k] ------------------ 100| 58| forward = size_t(end - buf - 1); 101| 58| } 102| 62.9k| for (; k < forward; k++) { ------------------ | Branch (102:14): [True: 58.7k, False: 4.24k] ------------------ 103| 58.7k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 104| 58.7k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (104:13): [True: 53.3k, False: 5.44k] ------------------ 105| 53.3k| *utf32_output++ = char32_t(word); 106| 53.3k| } else { 107| | // must be a surrogate pair 108| 5.44k| uint16_t diff = uint16_t(word - 0xD800); 109| 5.44k| uint16_t next_word = 110| 5.44k| scalar::utf16::swap_if_needed(buf[k + 1]); 111| 5.44k| k++; 112| 5.44k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 113| 5.44k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (113:15): [True: 116, False: 5.32k] ------------------ 114| 116| return std::make_pair(nullptr, utf32_output); 115| 116| } 116| 5.32k| uint32_t value = (diff << 10) + diff2 + 0x10000; 117| 5.32k| *utf32_output++ = char32_t(value); 118| 5.32k| } 119| 58.7k| } 120| 4.24k| buf += k; 121| 4.24k| } 122| 289k| } // while 123| 296| return std::make_pair(buf, utf32_output); 124| 412|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf16_to_utf32ILNS_10endiannessE1EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 392| char32_t *utf32_output) { 58| 392| const char16_t *end = buf + len; 59| | 60| 392| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 61| 392| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 62| | 63| 622k| while (end - buf >= 8) { ------------------ | Branch (63:10): [True: 622k, False: 314] ------------------ 64| 622k| __m128i in = _mm_loadu_si128((__m128i *)buf); 65| | 66| 622k| if (big_endian) { ------------------ | Branch (66:9): [True: 622k, Folded] ------------------ 67| 622k| const __m128i swap = 68| 622k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 69| 622k| in = _mm_shuffle_epi8(in, swap); 70| 622k| } 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| 622k| const __m128i surrogates_bytemask = 76| 622k| _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| 622k| const uint16_t surrogates_bitmask = 81| 622k| 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| 622k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (84:9): [True: 620k, False: 1.69k] ------------------ 85| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 86| 620k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 87| 620k| _mm_cvtepu16_epi32(in)); 88| 620k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 89| 620k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 90| 620k| utf32_output += 8; 91| 620k| buf += 8; 92| | // surrogate pair(s) in a register 93| 620k| } 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| 1.69k| size_t forward = 15; 98| 1.69k| size_t k = 0; 99| 1.69k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (99:11): [True: 80, False: 1.61k] ------------------ 100| 80| forward = size_t(end - buf - 1); 101| 80| } 102| 23.6k| for (; k < forward; k++) { ------------------ | Branch (102:14): [True: 22.0k, False: 1.61k] ------------------ 103| 22.0k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 104| 22.0k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (104:13): [True: 19.6k, False: 2.37k] ------------------ 105| 19.6k| *utf32_output++ = char32_t(word); 106| 19.6k| } else { 107| | // must be a surrogate pair 108| 2.37k| uint16_t diff = uint16_t(word - 0xD800); 109| 2.37k| uint16_t next_word = 110| 2.37k| scalar::utf16::swap_if_needed(buf[k + 1]); 111| 2.37k| k++; 112| 2.37k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 113| 2.37k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (113:15): [True: 78, False: 2.29k] ------------------ 114| 78| return std::make_pair(nullptr, utf32_output); 115| 78| } 116| 2.29k| uint32_t value = (diff << 10) + diff2 + 0x10000; 117| 2.29k| *utf32_output++ = char32_t(value); 118| 2.29k| } 119| 22.0k| } 120| 1.61k| buf += k; 121| 1.61k| } 122| 622k| } // while 123| 314| return std::make_pair(buf, utf32_output); 124| 392|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf16_to_utf32_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 136| 418| char32_t *utf32_output) { 137| 418| const char16_t *start = buf; 138| 418| const char16_t *end = buf + len; 139| | 140| 418| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 141| 418| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 142| | 143| 530k| while (end - buf >= 8) { ------------------ | Branch (143:10): [True: 529k, False: 250] ------------------ 144| 529k| __m128i in = _mm_loadu_si128((__m128i *)buf); 145| | 146| 529k| if (big_endian) { ------------------ | Branch (146:9): [Folded, False: 529k] ------------------ 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| 529k| const __m128i surrogates_bytemask = 156| 529k| _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| 529k| const uint16_t surrogates_bitmask = 161| 529k| 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| 529k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (164:9): [True: 520k, False: 9.32k] ------------------ 165| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 166| 520k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 167| 520k| _mm_cvtepu16_epi32(in)); 168| 520k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 169| 520k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 170| 520k| utf32_output += 8; 171| 520k| buf += 8; 172| | // surrogate pair(s) in a register 173| 520k| } 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| 9.32k| size_t forward = 15; 178| 9.32k| size_t k = 0; 179| 9.32k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (179:11): [True: 82, False: 9.24k] ------------------ 180| 82| forward = size_t(end - buf - 1); 181| 82| } 182| 133k| for (; k < forward; k++) { ------------------ | Branch (182:14): [True: 124k, False: 9.15k] ------------------ 183| 124k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 184| 124k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (184:13): [True: 109k, False: 14.7k] ------------------ 185| 109k| *utf32_output++ = char32_t(word); 186| 109k| } else { 187| | // must be a surrogate pair 188| 14.7k| uint16_t diff = uint16_t(word - 0xD800); 189| 14.7k| uint16_t next_word = 190| 14.7k| scalar::utf16::swap_if_needed(buf[k + 1]); 191| 14.7k| k++; 192| 14.7k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 193| 14.7k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (193:15): [True: 168, False: 14.6k] ------------------ 194| 168| return std::make_pair( 195| 168| result(error_code::SURROGATE, buf - start + k - 1), 196| 168| utf32_output); 197| 168| } 198| 14.6k| uint32_t value = (diff << 10) + diff2 + 0x10000; 199| 14.6k| *utf32_output++ = char32_t(value); 200| 14.6k| } 201| 124k| } 202| 9.15k| buf += k; 203| 9.15k| } 204| 529k| } // while 205| 250| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 206| 418|} 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| 311k| while (end - buf >= 8) { ------------------ | Branch (143:10): [True: 311k, False: 238] ------------------ 144| 311k| __m128i in = _mm_loadu_si128((__m128i *)buf); 145| | 146| 311k| if (big_endian) { ------------------ | Branch (146:9): [True: 311k, Folded] ------------------ 147| 311k| const __m128i swap = 148| 311k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 149| 311k| in = _mm_shuffle_epi8(in, swap); 150| 311k| } 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| 311k| const __m128i surrogates_bytemask = 156| 311k| _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| 311k| const uint16_t surrogates_bitmask = 161| 311k| 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| 311k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (164:9): [True: 305k, False: 5.68k] ------------------ 165| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 166| 305k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 167| 305k| _mm_cvtepu16_epi32(in)); 168| 305k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 169| 305k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 170| 305k| utf32_output += 8; 171| 305k| buf += 8; 172| | // surrogate pair(s) in a register 173| 305k| } 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| 5.68k| size_t forward = 15; 178| 5.68k| size_t k = 0; 179| 5.68k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (179:11): [True: 88, False: 5.59k] ------------------ 180| 88| forward = size_t(end - buf - 1); 181| 88| } 182| 79.7k| for (; k < forward; k++) { ------------------ | Branch (182:14): [True: 74.1k, False: 5.57k] ------------------ 183| 74.1k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 184| 74.1k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (184:13): [True: 63.7k, False: 10.4k] ------------------ 185| 63.7k| *utf32_output++ = char32_t(word); 186| 63.7k| } else { 187| | // must be a surrogate pair 188| 10.4k| uint16_t diff = uint16_t(word - 0xD800); 189| 10.4k| uint16_t next_word = 190| 10.4k| scalar::utf16::swap_if_needed(buf[k + 1]); 191| 10.4k| k++; 192| 10.4k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 193| 10.4k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (193:15): [True: 108, False: 10.2k] ------------------ 194| 108| return std::make_pair( 195| 108| result(error_code::SURROGATE, buf - start + k - 1), 196| 108| utf32_output); 197| 108| } 198| 10.2k| uint32_t value = (diff << 10) + diff2 + 0x10000; 199| 10.2k| *utf32_output++ = char32_t(value); 200| 10.2k| } 201| 74.1k| } 202| 5.57k| buf += k; 203| 5.57k| } 204| 311k| } // while 205| 238| 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| 856|sse_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| | 58| 856| const char16_t *end = buf + len; 59| | 60| 856| const __m128i v_0000 = _mm_setzero_si128(); 61| 856| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 62| 856| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 63| 856| const size_t safety_margin = 64| 856| 12; // to avoid overruns, see issue 65| | // https://github.com/simdutf/simdutf/issues/92 66| | 67| 1.30M| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (67:10): [True: 1.30M, False: 742] ------------------ 68| 1.30M| __m128i in = _mm_loadu_si128((__m128i *)buf); 69| 1.30M| if (big_endian) { ------------------ | Branch (69:9): [Folded, False: 1.30M] ------------------ 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| 1.30M| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 76| 1.30M| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 715k, False: 584k] ------------------ 77| 715k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 78| 715k| if (big_endian) { ------------------ | Branch (78:11): [Folded, False: 715k] ------------------ 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| 715k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (83:11): [True: 59.0k, False: 656k] ------------------ 84| | // 1. pack the bytes 85| | // obviously suboptimal. 86| 59.0k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 87| | // 2. store (16 bytes) 88| 59.0k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 89| | // 3. adjust pointers 90| 59.0k| buf += 8; 91| 59.0k| utf8_output += 8; 92| 59.0k| in = nextin; 93| 656k| } else { 94| | // 1. pack the bytes 95| | // obviously suboptimal. 96| 656k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 97| | // 2. store (16 bytes) 98| 656k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 99| | // 3. adjust pointers 100| 656k| buf += 16; 101| 656k| utf8_output += 16; 102| 656k| continue; // we are done for this round! 103| 656k| } 104| 715k| } 105| | 106| | // no bits set above 7th bit 107| 643k| const __m128i one_byte_bytemask = 108| 643k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 109| 643k| const uint16_t one_byte_bitmask = 110| 643k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 111| | 112| | // no bits set above 11th bit 113| 643k| const __m128i one_or_two_bytes_bytemask = 114| 643k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 115| 643k| const uint16_t one_or_two_bytes_bitmask = 116| 643k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 117| | 118| 643k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (118:9): [True: 493k, False: 149k] ------------------ 119| 493k| internal::westmere::write_v_u16_11bits_to_utf8( 120| 493k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 121| 493k| buf += 8; 122| 493k| continue; 123| 493k| } 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| 149k| const __m128i surrogates_bytemask = 129| 149k| _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| 149k| const uint16_t surrogates_bitmask = 134| 149k| 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| 149k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (137:9): [True: 139k, False: 9.90k] ------------------ 138| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 139| 139k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 140| 139k| 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| 139k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 170| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 171| 139k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 172| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 173| 139k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 169| 139k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 174| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 175| 139k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 169| 139k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 176| | 177| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 178| 139k| const __m128i s0 = _mm_srli_epi16(in, 4); 179| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 180| 139k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 169| 139k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 181| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 182| 139k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 169| 139k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 183| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 184| 139k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 169| 139k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 185| 139k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 186| 139k| simdutf_vec(0b0100000000000000)); ------------------ | | 169| 139k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 187| 139k| const __m128i s4 = _mm_xor_si128(s3, m0); 188| 139k|#undef simdutf_vec 189| | 190| | // 4. expand code units 16-bit => 32-bit 191| 139k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 192| 139k| 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| 139k| const uint16_t mask = 196| 139k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 197| 139k| if (mask == 0) { ------------------ | Branch (197:11): [True: 23.1k, False: 116k] ------------------ 198| | // We only have three-byte code units. Use fast path. 199| 23.1k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 200| 23.1k| 15, 13, -1, -1, -1, -1); 201| 23.1k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 202| 23.1k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 203| 23.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 204| 23.1k| utf8_output += 12; 205| 23.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 206| 23.1k| utf8_output += 12; 207| 23.1k| buf += 8; 208| 23.1k| continue; 209| 23.1k| } 210| 116k| const uint8_t mask0 = uint8_t(mask); 211| | 212| 116k| const uint8_t *row0 = 213| 116k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 214| 116k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 215| 116k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 216| | 217| 116k| const uint8_t mask1 = static_cast(mask >> 8); 218| | 219| 116k| const uint8_t *row1 = 220| 116k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 221| 116k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 222| 116k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 223| | 224| 116k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 225| 116k| utf8_output += row0[0]; 226| 116k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 227| 116k| utf8_output += row1[0]; 228| | 229| 116k| buf += 8; 230| | // surrogate pair(s) in a register 231| 116k| } 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| 9.90k| size_t forward = 15; 236| 9.90k| size_t k = 0; 237| 9.90k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (237:11): [True: 0, False: 9.90k] ------------------ 238| 0| forward = size_t(end - buf - 1); 239| 0| } 240| 146k| for (; k < forward; k++) { ------------------ | Branch (240:14): [True: 136k, False: 9.78k] ------------------ 241| 136k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 242| 136k| if ((word & 0xFF80) == 0) { ------------------ | Branch (242:13): [True: 17.0k, False: 119k] ------------------ 243| 17.0k| *utf8_output++ = char(word); 244| 119k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (244:20): [True: 4.32k, False: 115k] ------------------ 245| 4.32k| *utf8_output++ = char((word >> 6) | 0b11000000); 246| 4.32k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 247| 115k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (247:20): [True: 104k, False: 11.0k] ------------------ 248| 104k| *utf8_output++ = char((word >> 12) | 0b11100000); 249| 104k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 250| 104k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 251| 104k| } else { 252| | // must be a surrogate pair 253| 11.0k| uint16_t diff = uint16_t(word - 0xD800); 254| 11.0k| uint16_t next_word = 255| 11.0k| scalar::utf16::swap_if_needed(buf[k + 1]); 256| 11.0k| k++; 257| 11.0k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 258| 11.0k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (258:15): [True: 114, False: 10.9k] ------------------ 259| 114| return std::make_pair(nullptr, utf8_output); 260| 114| } 261| 10.9k| uint32_t value = (diff << 10) + diff2 + 0x10000; 262| 10.9k| *utf8_output++ = char((value >> 18) | 0b11110000); 263| 10.9k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 264| 10.9k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 265| 10.9k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 266| 10.9k| } 267| 136k| } 268| 9.78k| buf += k; 269| 9.78k| } 270| 149k| } // while 271| | 272| 742| return std::make_pair(buf, utf8_output); 273| 856|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_125sse_convert_utf16_to_utf8ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 750|sse_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| | 58| 750| const char16_t *end = buf + len; 59| | 60| 750| const __m128i v_0000 = _mm_setzero_si128(); 61| 750| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 62| 750| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 63| 750| const size_t safety_margin = 64| 750| 12; // to avoid overruns, see issue 65| | // https://github.com/simdutf/simdutf/issues/92 66| | 67| 1.12M| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (67:10): [True: 1.12M, False: 662] ------------------ 68| 1.12M| __m128i in = _mm_loadu_si128((__m128i *)buf); 69| 1.12M| if (big_endian) { ------------------ | Branch (69:9): [True: 1.12M, Folded] ------------------ 70| 1.12M| const __m128i swap = 71| 1.12M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 72| 1.12M| in = _mm_shuffle_epi8(in, swap); 73| 1.12M| } 74| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 75| 1.12M| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 76| 1.12M| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 711k, False: 412k] ------------------ 77| 711k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 78| 711k| if (big_endian) { ------------------ | Branch (78:11): [True: 711k, Folded] ------------------ 79| 711k| const __m128i swap = 80| 711k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 81| 711k| nextin = _mm_shuffle_epi8(nextin, swap); 82| 711k| } 83| 711k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (83:11): [True: 35.8k, False: 675k] ------------------ 84| | // 1. pack the bytes 85| | // obviously suboptimal. 86| 35.8k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 87| | // 2. store (16 bytes) 88| 35.8k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 89| | // 3. adjust pointers 90| 35.8k| buf += 8; 91| 35.8k| utf8_output += 8; 92| 35.8k| in = nextin; 93| 675k| } else { 94| | // 1. pack the bytes 95| | // obviously suboptimal. 96| 675k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 97| | // 2. store (16 bytes) 98| 675k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 99| | // 3. adjust pointers 100| 675k| buf += 16; 101| 675k| utf8_output += 16; 102| 675k| continue; // we are done for this round! 103| 675k| } 104| 711k| } 105| | 106| | // no bits set above 7th bit 107| 448k| const __m128i one_byte_bytemask = 108| 448k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 109| 448k| const uint16_t one_byte_bitmask = 110| 448k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 111| | 112| | // no bits set above 11th bit 113| 448k| const __m128i one_or_two_bytes_bytemask = 114| 448k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 115| 448k| const uint16_t one_or_two_bytes_bitmask = 116| 448k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 117| | 118| 448k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (118:9): [True: 32.1k, False: 416k] ------------------ 119| 32.1k| internal::westmere::write_v_u16_11bits_to_utf8( 120| 32.1k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 121| 32.1k| buf += 8; 122| 32.1k| continue; 123| 32.1k| } 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| 416k| const __m128i surrogates_bytemask = 129| 416k| _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| 416k| const uint16_t surrogates_bitmask = 134| 416k| 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| 416k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (137:9): [True: 411k, False: 4.62k] ------------------ 138| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 139| 411k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 140| 411k| 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| 411k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 170| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 171| 411k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 172| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 173| 411k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 169| 411k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 174| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 175| 411k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 169| 411k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 176| | 177| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 178| 411k| const __m128i s0 = _mm_srli_epi16(in, 4); 179| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 180| 411k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 169| 411k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 181| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 182| 411k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 169| 411k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 183| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 184| 411k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 169| 411k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 185| 411k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 186| 411k| simdutf_vec(0b0100000000000000)); ------------------ | | 169| 411k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 187| 411k| const __m128i s4 = _mm_xor_si128(s3, m0); 188| 411k|#undef simdutf_vec 189| | 190| | // 4. expand code units 16-bit => 32-bit 191| 411k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 192| 411k| 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| 411k| const uint16_t mask = 196| 411k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 197| 411k| if (mask == 0) { ------------------ | Branch (197:11): [True: 269k, False: 142k] ------------------ 198| | // We only have three-byte code units. Use fast path. 199| 269k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 200| 269k| 15, 13, -1, -1, -1, -1); 201| 269k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 202| 269k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 203| 269k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 204| 269k| utf8_output += 12; 205| 269k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 206| 269k| utf8_output += 12; 207| 269k| buf += 8; 208| 269k| continue; 209| 269k| } 210| 142k| const uint8_t mask0 = uint8_t(mask); 211| | 212| 142k| const uint8_t *row0 = 213| 142k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 214| 142k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 215| 142k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 216| | 217| 142k| const uint8_t mask1 = static_cast(mask >> 8); 218| | 219| 142k| const uint8_t *row1 = 220| 142k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 221| 142k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 222| 142k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 223| | 224| 142k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 225| 142k| utf8_output += row0[0]; 226| 142k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 227| 142k| utf8_output += row1[0]; 228| | 229| 142k| buf += 8; 230| | // surrogate pair(s) in a register 231| 142k| } 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| 4.62k| size_t forward = 15; 236| 4.62k| size_t k = 0; 237| 4.62k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (237:11): [True: 0, False: 4.62k] ------------------ 238| 0| forward = size_t(end - buf - 1); 239| 0| } 240| 67.8k| for (; k < forward; k++) { ------------------ | Branch (240:14): [True: 63.3k, False: 4.53k] ------------------ 241| 63.3k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 242| 63.3k| if ((word & 0xFF80) == 0) { ------------------ | Branch (242:13): [True: 10.5k, False: 52.7k] ------------------ 243| 10.5k| *utf8_output++ = char(word); 244| 52.7k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (244:20): [True: 1.84k, False: 50.9k] ------------------ 245| 1.84k| *utf8_output++ = char((word >> 6) | 0b11000000); 246| 1.84k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 247| 50.9k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (247:20): [True: 45.7k, False: 5.24k] ------------------ 248| 45.7k| *utf8_output++ = char((word >> 12) | 0b11100000); 249| 45.7k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 250| 45.7k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 251| 45.7k| } else { 252| | // must be a surrogate pair 253| 5.24k| uint16_t diff = uint16_t(word - 0xD800); 254| 5.24k| uint16_t next_word = 255| 5.24k| scalar::utf16::swap_if_needed(buf[k + 1]); 256| 5.24k| k++; 257| 5.24k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 258| 5.24k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (258:15): [True: 88, False: 5.15k] ------------------ 259| 88| return std::make_pair(nullptr, utf8_output); 260| 88| } 261| 5.15k| uint32_t value = (diff << 10) + diff2 + 0x10000; 262| 5.15k| *utf8_output++ = char((value >> 18) | 0b11110000); 263| 5.15k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 264| 5.15k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 265| 5.15k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 266| 5.15k| } 267| 63.3k| } 268| 4.53k| buf += k; 269| 4.53k| } 270| 416k| } // while 271| | 272| 662| return std::make_pair(buf, utf8_output); 273| 750|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_137sse_convert_utf16_to_utf8_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 285| 664| char *utf8_output) { 286| 664| const char16_t *start = buf; 287| 664| const char16_t *end = buf + len; 288| | 289| 664| const __m128i v_0000 = _mm_setzero_si128(); 290| 664| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 291| 664| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 292| 664| const size_t safety_margin = 293| 664| 12; // to avoid overruns, see issue 294| | // https://github.com/simdutf/simdutf/issues/92 295| | 296| 830k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (296:10): [True: 829k, False: 516] ------------------ 297| 829k| __m128i in = _mm_loadu_si128((__m128i *)buf); 298| 829k| if (big_endian) { ------------------ | Branch (298:9): [Folded, False: 829k] ------------------ 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| 829k| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 305| 829k| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (305:9): [True: 270k, False: 559k] ------------------ 306| 270k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 307| 270k| if (big_endian) { ------------------ | Branch (307:11): [Folded, False: 270k] ------------------ 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| 270k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (312:11): [True: 11.9k, False: 258k] ------------------ 313| | // 1. pack the bytes 314| | // obviously suboptimal. 315| 11.9k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 316| | // 2. store (16 bytes) 317| 11.9k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 318| | // 3. adjust pointers 319| 11.9k| buf += 8; 320| 11.9k| utf8_output += 8; 321| 11.9k| in = nextin; 322| 258k| } else { 323| | // 1. pack the bytes 324| | // obviously suboptimal. 325| 258k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 326| | // 2. store (16 bytes) 327| 258k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 328| | // 3. adjust pointers 329| 258k| buf += 16; 330| 258k| utf8_output += 16; 331| 258k| continue; // we are done for this round! 332| 258k| } 333| 270k| } 334| | 335| | // no bits set above 7th bit 336| 571k| const __m128i one_byte_bytemask = 337| 571k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 338| 571k| const uint16_t one_byte_bitmask = 339| 571k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 340| | 341| | // no bits set above 11th bit 342| 571k| const __m128i one_or_two_bytes_bytemask = 343| 571k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 344| 571k| const uint16_t one_or_two_bytes_bitmask = 345| 571k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 346| | 347| 571k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (347:9): [True: 174k, False: 397k] ------------------ 348| 174k| internal::westmere::write_v_u16_11bits_to_utf8( 349| 174k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 350| 174k| buf += 8; 351| 174k| continue; 352| 174k| } 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| 397k| const __m128i surrogates_bytemask = 358| 397k| _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| 397k| const uint16_t surrogates_bitmask = 363| 397k| 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| 397k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (366:9): [True: 395k, False: 1.97k] ------------------ 367| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 368| 395k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 369| 395k| 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| 395k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 399| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 400| 395k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 401| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 402| 395k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 398| 395k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 403| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 404| 395k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 398| 395k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 405| | 406| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 407| 395k| const __m128i s0 = _mm_srli_epi16(in, 4); 408| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 409| 395k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 398| 395k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 410| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 411| 395k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 398| 395k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 412| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 413| 395k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 398| 395k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 414| 395k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 415| 395k| simdutf_vec(0b0100000000000000)); ------------------ | | 398| 395k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 416| 395k| const __m128i s4 = _mm_xor_si128(s3, m0); 417| 395k|#undef simdutf_vec 418| | 419| | // 4. expand code units 16-bit => 32-bit 420| 395k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 421| 395k| 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| 395k| const uint16_t mask = 425| 395k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 426| 395k| if (mask == 0) { ------------------ | Branch (426:11): [True: 313k, False: 81.4k] ------------------ 427| | // We only have three-byte code units. Use fast path. 428| 313k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 429| 313k| 15, 13, -1, -1, -1, -1); 430| 313k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 431| 313k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 432| 313k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 433| 313k| utf8_output += 12; 434| 313k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 435| 313k| utf8_output += 12; 436| 313k| buf += 8; 437| 313k| continue; 438| 313k| } 439| 81.4k| const uint8_t mask0 = uint8_t(mask); 440| | 441| 81.4k| const uint8_t *row0 = 442| 81.4k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 443| 81.4k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 444| 81.4k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 445| | 446| 81.4k| const uint8_t mask1 = static_cast(mask >> 8); 447| | 448| 81.4k| const uint8_t *row1 = 449| 81.4k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 450| 81.4k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 451| 81.4k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 452| | 453| 81.4k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 454| 81.4k| utf8_output += row0[0]; 455| 81.4k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 456| 81.4k| utf8_output += row1[0]; 457| | 458| 81.4k| buf += 8; 459| | // surrogate pair(s) in a register 460| 81.4k| } 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| 1.97k| size_t forward = 15; 465| 1.97k| size_t k = 0; 466| 1.97k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (466:11): [True: 0, False: 1.97k] ------------------ 467| 0| forward = size_t(end - buf - 1); 468| 0| } 469| 27.6k| for (; k < forward; k++) { ------------------ | Branch (469:14): [True: 25.8k, False: 1.82k] ------------------ 470| 25.8k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 471| 25.8k| if ((word & 0xFF80) == 0) { ------------------ | Branch (471:13): [True: 4.96k, False: 20.8k] ------------------ 472| 4.96k| *utf8_output++ = char(word); 473| 20.8k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (473:20): [True: 1.26k, False: 19.5k] ------------------ 474| 1.26k| *utf8_output++ = char((word >> 6) | 0b11000000); 475| 1.26k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 476| 19.5k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (476:20): [True: 17.1k, False: 2.39k] ------------------ 477| 17.1k| *utf8_output++ = char((word >> 12) | 0b11100000); 478| 17.1k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 479| 17.1k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 480| 17.1k| } else { 481| | // must be a surrogate pair 482| 2.39k| uint16_t diff = uint16_t(word - 0xD800); 483| 2.39k| uint16_t next_word = 484| 2.39k| scalar::utf16::swap_if_needed(buf[k + 1]); 485| 2.39k| k++; 486| 2.39k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 487| 2.39k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (487:15): [True: 148, False: 2.24k] ------------------ 488| 148| return std::make_pair( 489| 148| result(error_code::SURROGATE, buf - start + k - 1), 490| 148| utf8_output); 491| 148| } 492| 2.24k| uint32_t value = (diff << 10) + diff2 + 0x10000; 493| 2.24k| *utf8_output++ = char((value >> 18) | 0b11110000); 494| 2.24k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 495| 2.24k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 496| 2.24k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 497| 2.24k| } 498| 25.8k| } 499| 1.82k| buf += k; 500| 1.82k| } 501| 397k| } // while 502| | 503| 516| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 504| 664|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_137sse_convert_utf16_to_utf8_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 285| 728| char *utf8_output) { 286| 728| const char16_t *start = buf; 287| 728| const char16_t *end = buf + len; 288| | 289| 728| const __m128i v_0000 = _mm_setzero_si128(); 290| 728| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 291| 728| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 292| 728| const size_t safety_margin = 293| 728| 12; // to avoid overruns, see issue 294| | // https://github.com/simdutf/simdutf/issues/92 295| | 296| 837k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (296:10): [True: 837k, False: 554] ------------------ 297| 837k| __m128i in = _mm_loadu_si128((__m128i *)buf); 298| 837k| if (big_endian) { ------------------ | Branch (298:9): [True: 837k, Folded] ------------------ 299| 837k| const __m128i swap = 300| 837k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 301| 837k| in = _mm_shuffle_epi8(in, swap); 302| 837k| } 303| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 304| 837k| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 305| 837k| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (305:9): [True: 521k, False: 316k] ------------------ 306| 521k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 307| 521k| if (big_endian) { ------------------ | Branch (307:11): [True: 521k, Folded] ------------------ 308| 521k| const __m128i swap = 309| 521k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 310| 521k| nextin = _mm_shuffle_epi8(nextin, swap); 311| 521k| } 312| 521k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (312:11): [True: 89.2k, False: 432k] ------------------ 313| | // 1. pack the bytes 314| | // obviously suboptimal. 315| 89.2k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 316| | // 2. store (16 bytes) 317| 89.2k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 318| | // 3. adjust pointers 319| 89.2k| buf += 8; 320| 89.2k| utf8_output += 8; 321| 89.2k| in = nextin; 322| 432k| } else { 323| | // 1. pack the bytes 324| | // obviously suboptimal. 325| 432k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 326| | // 2. store (16 bytes) 327| 432k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 328| | // 3. adjust pointers 329| 432k| buf += 16; 330| 432k| utf8_output += 16; 331| 432k| continue; // we are done for this round! 332| 432k| } 333| 521k| } 334| | 335| | // no bits set above 7th bit 336| 405k| const __m128i one_byte_bytemask = 337| 405k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 338| 405k| const uint16_t one_byte_bitmask = 339| 405k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 340| | 341| | // no bits set above 11th bit 342| 405k| const __m128i one_or_two_bytes_bytemask = 343| 405k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 344| 405k| const uint16_t one_or_two_bytes_bitmask = 345| 405k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 346| | 347| 405k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (347:9): [True: 112k, False: 292k] ------------------ 348| 112k| internal::westmere::write_v_u16_11bits_to_utf8( 349| 112k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 350| 112k| buf += 8; 351| 112k| continue; 352| 112k| } 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| 292k| const __m128i surrogates_bytemask = 358| 292k| _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| 292k| const uint16_t surrogates_bitmask = 363| 292k| 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| 292k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (366:9): [True: 268k, False: 23.8k] ------------------ 367| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 368| 268k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 369| 268k| 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| 268k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 399| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 400| 268k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 401| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 402| 268k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 398| 268k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 403| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 404| 268k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 398| 268k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 405| | 406| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 407| 268k| const __m128i s0 = _mm_srli_epi16(in, 4); 408| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 409| 268k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 398| 268k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 410| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 411| 268k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 398| 268k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 412| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 413| 268k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 398| 268k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 414| 268k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 415| 268k| simdutf_vec(0b0100000000000000)); ------------------ | | 398| 268k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 416| 268k| const __m128i s4 = _mm_xor_si128(s3, m0); 417| 268k|#undef simdutf_vec 418| | 419| | // 4. expand code units 16-bit => 32-bit 420| 268k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 421| 268k| 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| 268k| const uint16_t mask = 425| 268k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 426| 268k| if (mask == 0) { ------------------ | Branch (426:11): [True: 52.3k, False: 216k] ------------------ 427| | // We only have three-byte code units. Use fast path. 428| 52.3k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 429| 52.3k| 15, 13, -1, -1, -1, -1); 430| 52.3k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 431| 52.3k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 432| 52.3k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 433| 52.3k| utf8_output += 12; 434| 52.3k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 435| 52.3k| utf8_output += 12; 436| 52.3k| buf += 8; 437| 52.3k| continue; 438| 52.3k| } 439| 216k| const uint8_t mask0 = uint8_t(mask); 440| | 441| 216k| const uint8_t *row0 = 442| 216k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 443| 216k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 444| 216k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 445| | 446| 216k| const uint8_t mask1 = static_cast(mask >> 8); 447| | 448| 216k| const uint8_t *row1 = 449| 216k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 450| 216k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 451| 216k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 452| | 453| 216k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 454| 216k| utf8_output += row0[0]; 455| 216k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 456| 216k| utf8_output += row1[0]; 457| | 458| 216k| buf += 8; 459| | // surrogate pair(s) in a register 460| 216k| } 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| 23.8k| size_t forward = 15; 465| 23.8k| size_t k = 0; 466| 23.8k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (466:11): [True: 0, False: 23.8k] ------------------ 467| 0| forward = size_t(end - buf - 1); 468| 0| } 469| 353k| for (; k < forward; k++) { ------------------ | Branch (469:14): [True: 330k, False: 23.6k] ------------------ 470| 330k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 471| 330k| if ((word & 0xFF80) == 0) { ------------------ | Branch (471:13): [True: 133k, False: 196k] ------------------ 472| 133k| *utf8_output++ = char(word); 473| 196k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (473:20): [True: 14.7k, False: 181k] ------------------ 474| 14.7k| *utf8_output++ = char((word >> 6) | 0b11000000); 475| 14.7k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 476| 181k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (476:20): [True: 156k, False: 25.8k] ------------------ 477| 156k| *utf8_output++ = char((word >> 12) | 0b11100000); 478| 156k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 479| 156k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 480| 156k| } else { 481| | // must be a surrogate pair 482| 25.8k| uint16_t diff = uint16_t(word - 0xD800); 483| 25.8k| uint16_t next_word = 484| 25.8k| scalar::utf16::swap_if_needed(buf[k + 1]); 485| 25.8k| k++; 486| 25.8k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 487| 25.8k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (487:15): [True: 174, False: 25.6k] ------------------ 488| 174| return std::make_pair( 489| 174| result(error_code::SURROGATE, buf - start + k - 1), 490| 174| utf8_output); 491| 174| } 492| 25.6k| uint32_t value = (diff << 10) + diff2 + 0x10000; 493| 25.6k| *utf8_output++ = char((value >> 18) | 0b11110000); 494| 25.6k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 495| 25.6k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 496| 25.6k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 497| 25.6k| } 498| 330k| } 499| 23.6k| buf += k; 500| 23.6k| } 501| 292k| } // while 502| | 503| 554| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 504| 728|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_utf32_to_latin1EPKDimPc: 3| 172| char *latin1_output) { 4| 172| const size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 5| | 6| 172| __m128i high_bytes_mask = _mm_set1_epi32(0xFFFFFF00); 7| 172| __m128i shufmask = 8| 172| _mm_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 12, 8, 4, 0); 9| | 10| 1.26k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (10:22): [True: 1.12k, False: 146] ------------------ 11| 1.12k| __m128i in1 = _mm_loadu_si128((__m128i *)buf); 12| 1.12k| __m128i in2 = _mm_loadu_si128((__m128i *)(buf + 4)); 13| 1.12k| __m128i in3 = _mm_loadu_si128((__m128i *)(buf + 8)); 14| 1.12k| __m128i in4 = _mm_loadu_si128((__m128i *)(buf + 12)); 15| | 16| 1.12k| __m128i check_combined = _mm_or_si128(in1, in2); 17| 1.12k| check_combined = _mm_or_si128(check_combined, in3); 18| 1.12k| check_combined = _mm_or_si128(check_combined, in4); 19| | 20| 1.12k| if (!_mm_testz_si128(check_combined, high_bytes_mask)) { ------------------ | Branch (20:9): [True: 26, False: 1.09k] ------------------ 21| 26| return std::make_pair(nullptr, latin1_output); 22| 26| } 23| 1.09k| __m128i pack1 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in1, shufmask), 24| 1.09k| _mm_shuffle_epi8(in2, shufmask)); 25| 1.09k| __m128i pack2 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in3, shufmask), 26| 1.09k| _mm_shuffle_epi8(in4, shufmask)); 27| 1.09k| __m128i pack = _mm_unpacklo_epi64(pack1, pack2); 28| 1.09k| _mm_storeu_si128((__m128i *)latin1_output, pack); 29| 1.09k| latin1_output += 16; 30| 1.09k| buf += 16; 31| 1.09k| } 32| | 33| 146| return std::make_pair(buf, latin1_output); 34| 172|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_139sse_convert_utf32_to_latin1_with_errorsEPKDimPc: 38| 408| char *latin1_output) { 39| 408| const char32_t *start = buf; 40| 408| const size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 41| | 42| 408| __m128i high_bytes_mask = _mm_set1_epi32(0xFFFFFF00); 43| 408| __m128i shufmask = 44| 408| _mm_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 12, 8, 4, 0); 45| | 46| 29.7k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (46:22): [True: 29.5k, False: 226] ------------------ 47| 29.5k| __m128i in1 = _mm_loadu_si128((__m128i *)buf); 48| 29.5k| __m128i in2 = _mm_loadu_si128((__m128i *)(buf + 4)); 49| 29.5k| __m128i in3 = _mm_loadu_si128((__m128i *)(buf + 8)); 50| 29.5k| __m128i in4 = _mm_loadu_si128((__m128i *)(buf + 12)); 51| | 52| 29.5k| __m128i check_combined = _mm_or_si128(in1, in2); 53| 29.5k| check_combined = _mm_or_si128(check_combined, in3); 54| 29.5k| check_combined = _mm_or_si128(check_combined, in4); 55| | 56| 29.5k| if (!_mm_testz_si128(check_combined, high_bytes_mask)) { ------------------ | Branch (56:9): [True: 182, False: 29.3k] ------------------ 57| | // Fallback to scalar code for handling errors 58| 876| for (int k = 0; k < 16; k++) { ------------------ | Branch (58:23): [True: 876, False: 0] ------------------ 59| 876| char32_t codepoint = buf[k]; 60| 876| if (codepoint <= 0xff) { ------------------ | Branch (60:13): [True: 694, False: 182] ------------------ 61| 694| *latin1_output++ = char(codepoint); 62| 694| } else { 63| 182| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 64| 182| latin1_output); 65| 182| } 66| 876| } 67| 0| buf += 16; 68| 0| continue; 69| 182| } 70| 29.3k| __m128i pack1 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in1, shufmask), 71| 29.3k| _mm_shuffle_epi8(in2, shufmask)); 72| 29.3k| __m128i pack2 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in3, shufmask), 73| 29.3k| _mm_shuffle_epi8(in4, shufmask)); 74| 29.3k| __m128i pack = _mm_unpacklo_epi64(pack1, pack2); 75| 29.3k| _mm_storeu_si128((__m128i *)latin1_output, pack); 76| 29.3k| latin1_output += 16; 77| 29.3k| buf += 16; 78| 29.3k| } 79| | 80| 226| return std::make_pair(result(error_code::SUCCESS, buf - start), 81| 226| latin1_output); 82| 408|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf32_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKDiPDsEES7_mS8_: 58| 774| char16_t *utf16_output) { 59| | 60| 774| const char32_t *end = buf + len; 61| | 62| 774| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 63| 774| __m128i forbidden_bytemask = _mm_setzero_si128(); 64| | 65| 66.9k| while (end - buf >= 16 + 8) { ------------------ | Branch (65:10): [True: 66.2k, False: 752] ------------------ 66| 66.2k| const __m128i *ptr = reinterpret_cast(buf); 67| 66.2k| const __m128i in0 = _mm_loadu_si128(ptr + 0); 68| 66.2k| const __m128i in1 = _mm_loadu_si128(ptr + 1); 69| 66.2k| const __m128i in2 = _mm_loadu_si128(ptr + 2); 70| 66.2k| const __m128i in3 = _mm_loadu_si128(ptr + 3); 71| | 72| 66.2k| const __m128i combined = 73| 66.2k| _mm_or_si128(_mm_or_si128(in2, in3), _mm_or_si128(in0, in1)); 74| 66.2k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 66.2k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 64.9k, False: 1.26k] | | ------------------ ------------------ 75| | // No bits set above 16th, directly pack UTF-32 to UTF-16 76| 64.9k| __m128i utf16_packed0 = _mm_packus_epi32(in0, in1); 77| 64.9k| __m128i utf16_packed1 = _mm_packus_epi32(in2, in3); 78| | 79| 64.9k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 80| 64.9k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 81| 64.9k| forbidden_bytemask = _mm_or_si128( 82| 64.9k| forbidden_bytemask, 83| 64.9k| _mm_or_si128( 84| 64.9k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed0, v_f800), v_d800), 85| 64.9k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed1, v_f800), v_d800))); 86| | 87| 64.9k| if (big_endian) { ------------------ | Branch (87:11): [Folded, False: 64.9k] ------------------ 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| 64.9k| _mm_storeu_si128((__m128i *)utf16_output + 0, utf16_packed0); 95| 64.9k| _mm_storeu_si128((__m128i *)utf16_output + 1, utf16_packed1); 96| 64.9k| utf16_output += 16; 97| 64.9k| buf += 16; 98| 64.9k| } else { 99| 1.26k| if (!validate_utf32(in0, in1) || !validate_utf32(in2, in3)) { ------------------ | Branch (99:11): [True: 12, False: 1.25k] | Branch (99:40): [True: 10, False: 1.24k] ------------------ 100| 22| return std::make_pair(nullptr, utf16_output); 101| 22| } 102| | 103| 1.24k| const auto ret0 = sse_expand_surrogate(in0); 104| 1.24k| _mm_storeu_si128((__m128i *)utf16_output, ret0.compressed); 105| 1.24k| utf16_output += ret0.u16count; 106| | 107| 1.24k| const auto ret1 = sse_expand_surrogate(in1); 108| 1.24k| _mm_storeu_si128((__m128i *)utf16_output, ret1.compressed); 109| 1.24k| utf16_output += ret1.u16count; 110| | 111| 1.24k| const auto ret2 = sse_expand_surrogate(in2); 112| 1.24k| _mm_storeu_si128((__m128i *)utf16_output, ret2.compressed); 113| 1.24k| utf16_output += ret2.u16count; 114| | 115| 1.24k| const auto ret3 = sse_expand_surrogate(in3); 116| 1.24k| _mm_storeu_si128((__m128i *)utf16_output, ret3.compressed); 117| 1.24k| utf16_output += ret3.u16count; 118| | 119| 1.24k| buf += 16; 120| 1.24k| } 121| 66.2k| } 122| | 123| | // check for invalid input 124| 752| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (124:7): [True: 20, False: 732] ------------------ 125| 20| return std::make_pair(nullptr, utf16_output); 126| 20| } 127| | 128| 732| return std::make_pair(buf, utf16_output); 129| 752|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114validate_utf32EDv2_xS2_: 37| 5.34k|simdutf_really_inline bool validate_utf32(const __m128i a, const __m128i b) { 38| 5.34k| using vector_u32 = simd32; 39| | 40| 5.34k| const auto in0 = vector_u32(a); 41| 5.34k| const auto in1 = vector_u32(b); 42| | 43| 5.34k| const auto standardmax = vector_u32::splat(0x10ffff); 44| 5.34k| const auto offset = vector_u32::splat(0xffff2000); 45| 5.34k| const auto standardoffsetmax = vector_u32::splat(0xfffff7ff); 46| | 47| 5.34k| const auto too_large = max(in0, in1) > standardmax; 48| 5.34k| const auto surrogate0 = (in0 + offset) > standardoffsetmax; 49| 5.34k| const auto surrogate1 = (in1 + offset) > standardoffsetmax; 50| | 51| 5.34k| const auto combined = too_large | surrogate0 | surrogate1; 52| 5.34k| return !combined.any(); 53| 5.34k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_120sse_expand_surrogateILNS_10endiannessE0EEENS1_18expansion_result_tEDv2_x: 9| 4.97k|expansion_result_t sse_expand_surrogate(const __m128i x) { 10| 4.97k| using vector_u32 = simd32; 11| 4.97k| using vector_u8 = simd8; 12| | 13| 4.97k| const auto in = vector_u32(x); 14| | 15| 4.97k| const auto non_surrogate_mask = (in & uint32_t(0xffff0000)) == uint32_t(0); 16| 4.97k| const auto mask = (~non_surrogate_mask.to_4bit_bitmask()) & 0xf; 17| | 18| 4.97k| const auto t0 = in - uint32_t(0x00010000); 19| 4.97k| const auto hi = t0.shr<10>() & uint32_t(0x000003ff); 20| 4.97k| const auto lo = t0.shl<16>() & uint32_t(0x03ff0000); 21| 4.97k| const auto surrogates = (lo | hi) | uint32_t(0xdc00d800); 22| | 23| 4.97k| const auto merged = as_vector_u8(select(non_surrogate_mask, in, surrogates)); 24| | 25| 4.97k| const auto shuffle = vector_u8::load( 26| 4.97k| (byte_order == endianness::LITTLE) ------------------ | Branch (26:7): [True: 4.97k, Folded] ------------------ 27| 4.97k| ? tables::utf32_to_utf16::pack_utf32_to_utf16le[mask] 28| 4.97k| : tables::utf32_to_utf16::pack_utf32_to_utf16be[mask]); 29| | 30| 4.97k| const size_t u16count = (4 + count_ones(mask)); 31| 4.97k| const auto compressed = shuffle.lookup_16(merged); 32| | 33| 4.97k| return {u16count, compressed}; 34| 4.97k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf32_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKDiPDsEES7_mS8_: 58| 662| char16_t *utf16_output) { 59| | 60| 662| const char32_t *end = buf + len; 61| | 62| 662| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 63| 662| __m128i forbidden_bytemask = _mm_setzero_si128(); 64| | 65| 13.1k| while (end - buf >= 16 + 8) { ------------------ | Branch (65:10): [True: 12.6k, False: 568] ------------------ 66| 12.6k| const __m128i *ptr = reinterpret_cast(buf); 67| 12.6k| const __m128i in0 = _mm_loadu_si128(ptr + 0); 68| 12.6k| const __m128i in1 = _mm_loadu_si128(ptr + 1); 69| 12.6k| const __m128i in2 = _mm_loadu_si128(ptr + 2); 70| 12.6k| const __m128i in3 = _mm_loadu_si128(ptr + 3); 71| | 72| 12.6k| const __m128i combined = 73| 12.6k| _mm_or_si128(_mm_or_si128(in2, in3), _mm_or_si128(in0, in1)); 74| 12.6k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 12.6k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 11.1k, False: 1.44k] | | ------------------ ------------------ 75| | // No bits set above 16th, directly pack UTF-32 to UTF-16 76| 11.1k| __m128i utf16_packed0 = _mm_packus_epi32(in0, in1); 77| 11.1k| __m128i utf16_packed1 = _mm_packus_epi32(in2, in3); 78| | 79| 11.1k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 80| 11.1k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 81| 11.1k| forbidden_bytemask = _mm_or_si128( 82| 11.1k| forbidden_bytemask, 83| 11.1k| _mm_or_si128( 84| 11.1k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed0, v_f800), v_d800), 85| 11.1k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed1, v_f800), v_d800))); 86| | 87| 11.1k| if (big_endian) { ------------------ | Branch (87:11): [True: 11.1k, Folded] ------------------ 88| 11.1k| const __m128i swap = 89| 11.1k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 90| 11.1k| utf16_packed0 = _mm_shuffle_epi8(utf16_packed0, swap); 91| 11.1k| utf16_packed1 = _mm_shuffle_epi8(utf16_packed1, swap); 92| 11.1k| } 93| | 94| 11.1k| _mm_storeu_si128((__m128i *)utf16_output + 0, utf16_packed0); 95| 11.1k| _mm_storeu_si128((__m128i *)utf16_output + 1, utf16_packed1); 96| 11.1k| utf16_output += 16; 97| 11.1k| buf += 16; 98| 11.1k| } else { 99| 1.44k| if (!validate_utf32(in0, in1) || !validate_utf32(in2, in3)) { ------------------ | Branch (99:11): [True: 66, False: 1.37k] | Branch (99:40): [True: 28, False: 1.35k] ------------------ 100| 94| return std::make_pair(nullptr, utf16_output); 101| 94| } 102| | 103| 1.35k| const auto ret0 = sse_expand_surrogate(in0); 104| 1.35k| _mm_storeu_si128((__m128i *)utf16_output, ret0.compressed); 105| 1.35k| utf16_output += ret0.u16count; 106| | 107| 1.35k| const auto ret1 = sse_expand_surrogate(in1); 108| 1.35k| _mm_storeu_si128((__m128i *)utf16_output, ret1.compressed); 109| 1.35k| utf16_output += ret1.u16count; 110| | 111| 1.35k| const auto ret2 = sse_expand_surrogate(in2); 112| 1.35k| _mm_storeu_si128((__m128i *)utf16_output, ret2.compressed); 113| 1.35k| utf16_output += ret2.u16count; 114| | 115| 1.35k| const auto ret3 = sse_expand_surrogate(in3); 116| 1.35k| _mm_storeu_si128((__m128i *)utf16_output, ret3.compressed); 117| 1.35k| utf16_output += ret3.u16count; 118| | 119| 1.35k| buf += 16; 120| 1.35k| } 121| 12.6k| } 122| | 123| | // check for invalid input 124| 568| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (124:7): [True: 6, False: 562] ------------------ 125| 6| return std::make_pair(nullptr, utf16_output); 126| 6| } 127| | 128| 562| return std::make_pair(buf, utf16_output); 129| 568|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_120sse_expand_surrogateILNS_10endiannessE1EEENS1_18expansion_result_tEDv2_x: 9| 5.40k|expansion_result_t sse_expand_surrogate(const __m128i x) { 10| 5.40k| using vector_u32 = simd32; 11| 5.40k| using vector_u8 = simd8; 12| | 13| 5.40k| const auto in = vector_u32(x); 14| | 15| 5.40k| const auto non_surrogate_mask = (in & uint32_t(0xffff0000)) == uint32_t(0); 16| 5.40k| const auto mask = (~non_surrogate_mask.to_4bit_bitmask()) & 0xf; 17| | 18| 5.40k| const auto t0 = in - uint32_t(0x00010000); 19| 5.40k| const auto hi = t0.shr<10>() & uint32_t(0x000003ff); 20| 5.40k| const auto lo = t0.shl<16>() & uint32_t(0x03ff0000); 21| 5.40k| const auto surrogates = (lo | hi) | uint32_t(0xdc00d800); 22| | 23| 5.40k| const auto merged = as_vector_u8(select(non_surrogate_mask, in, surrogates)); 24| | 25| 5.40k| const auto shuffle = vector_u8::load( 26| 5.40k| (byte_order == endianness::LITTLE) ------------------ | Branch (26:7): [Folded, False: 5.40k] ------------------ 27| 5.40k| ? tables::utf32_to_utf16::pack_utf32_to_utf16le[mask] 28| 5.40k| : tables::utf32_to_utf16::pack_utf32_to_utf16be[mask]); 29| | 30| 5.40k| const size_t u16count = (4 + count_ones(mask)); 31| 5.40k| const auto compressed = shuffle.lookup_16(merged); 32| | 33| 5.40k| return {u16count, compressed}; 34| 5.40k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf32_to_utf16_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 134| 684| char16_t *utf16_output) { 135| 684| const char32_t *start = buf; 136| 684| const char32_t *end = buf + len; 137| | 138| 684| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 139| | 140| 55.9k| while (end - buf >= 8) { ------------------ | Branch (140:10): [True: 55.6k, False: 316] ------------------ 141| 55.6k| const __m128i in = _mm_loadu_si128((__m128i *)buf); 142| 55.6k| const __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 143| | 144| 55.6k| const __m128i combined = _mm_or_si128(in, nextin); 145| 55.6k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 55.6k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 53.4k, False: 2.17k] | | ------------------ ------------------ 146| | // No bits set above 16th, directly pack UTF-32 to UTF-16 147| 53.4k| __m128i utf16_packed = _mm_packus_epi32(in, nextin); 148| | 149| 53.4k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 150| 53.4k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 151| 53.4k| const __m128i forbidden_bytemask = 152| 53.4k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed, v_f800), v_d800); 153| 53.4k| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (153:11): [True: 44, False: 53.4k] ------------------ 154| 44| return std::make_pair(result(error_code::SURROGATE, buf - start), 155| 44| utf16_output); 156| 44| } 157| | 158| 53.4k| if (big_endian) { ------------------ | Branch (158:11): [Folded, False: 53.4k] ------------------ 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| 53.4k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 165| 53.4k| utf16_output += 8; 166| 53.4k| buf += 8; 167| 53.4k| } else { 168| 2.17k| size_t forward = 7; 169| 2.17k| size_t k = 0; 170| 2.17k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (170:11): [True: 0, False: 2.17k] ------------------ 171| 0| forward = size_t(end - buf - 1); 172| 0| } 173| 15.8k| for (; k < forward; k++) { ------------------ | Branch (173:14): [True: 14.0k, False: 1.85k] ------------------ 174| 14.0k| uint32_t word = buf[k]; 175| 14.0k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (175:13): [True: 6.58k, False: 7.44k] ------------------ 176| | // will not generate a surrogate pair 177| 6.58k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (177:15): [True: 702, False: 5.88k] | Branch (177:33): [True: 30, False: 672] ------------------ 178| 30| return std::make_pair( 179| 30| result(error_code::SURROGATE, buf - start + k), utf16_output); 180| 30| } 181| 6.55k| *utf16_output++ = 182| 6.55k| big_endian ------------------ | Branch (182:15): [Folded, False: 6.55k] ------------------ 183| 6.55k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 184| 6.55k| : char16_t(word); 185| 7.44k| } else { 186| | // will generate a surrogate pair 187| 7.44k| if (word > 0x10FFFF) { ------------------ | Branch (187:15): [True: 294, False: 7.14k] ------------------ 188| 294| return std::make_pair( 189| 294| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 190| 294| } 191| 7.14k| word -= 0x10000; 192| 7.14k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 193| 7.14k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 194| 7.14k| if (big_endian) { ------------------ | Branch (194:15): [Folded, False: 7.14k] ------------------ 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| 7.14k| *utf16_output++ = char16_t(high_surrogate); 201| 7.14k| *utf16_output++ = char16_t(low_surrogate); 202| 7.14k| } 203| 14.0k| } 204| 1.85k| buf += k; 205| 1.85k| } 206| 55.6k| } 207| | 208| 316| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 209| 684|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf32_to_utf16_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 134| 630| char16_t *utf16_output) { 135| 630| const char32_t *start = buf; 136| 630| const char32_t *end = buf + len; 137| | 138| 630| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 139| | 140| 3.68k| while (end - buf >= 8) { ------------------ | Branch (140:10): [True: 3.37k, False: 302] ------------------ 141| 3.37k| const __m128i in = _mm_loadu_si128((__m128i *)buf); 142| 3.37k| const __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 143| | 144| 3.37k| const __m128i combined = _mm_or_si128(in, nextin); 145| 3.37k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 3.37k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 2.37k, False: 1.00k] | | ------------------ ------------------ 146| | // No bits set above 16th, directly pack UTF-32 to UTF-16 147| 2.37k| __m128i utf16_packed = _mm_packus_epi32(in, nextin); 148| | 149| 2.37k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 150| 2.37k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 151| 2.37k| const __m128i forbidden_bytemask = 152| 2.37k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed, v_f800), v_d800); 153| 2.37k| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (153:11): [True: 14, False: 2.36k] ------------------ 154| 14| return std::make_pair(result(error_code::SURROGATE, buf - start), 155| 14| utf16_output); 156| 14| } 157| | 158| 2.36k| if (big_endian) { ------------------ | Branch (158:11): [True: 2.36k, Folded] ------------------ 159| 2.36k| const __m128i swap = 160| 2.36k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 161| 2.36k| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 162| 2.36k| } 163| | 164| 2.36k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 165| 2.36k| utf16_output += 8; 166| 2.36k| buf += 8; 167| 2.36k| } else { 168| 1.00k| size_t forward = 7; 169| 1.00k| size_t k = 0; 170| 1.00k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (170:11): [True: 0, False: 1.00k] ------------------ 171| 0| forward = size_t(end - buf - 1); 172| 0| } 173| 6.43k| for (; k < forward; k++) { ------------------ | Branch (173:14): [True: 5.74k, False: 686] ------------------ 174| 5.74k| uint32_t word = buf[k]; 175| 5.74k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (175:13): [True: 2.98k, False: 2.76k] ------------------ 176| | // will not generate a surrogate pair 177| 2.98k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (177:15): [True: 732, False: 2.25k] | Branch (177:33): [True: 28, False: 704] ------------------ 178| 28| return std::make_pair( 179| 28| result(error_code::SURROGATE, buf - start + k), utf16_output); 180| 28| } 181| 2.95k| *utf16_output++ = 182| 2.95k| big_endian ------------------ | Branch (182:15): [True: 2.95k, Folded] ------------------ 183| 2.95k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 184| 2.95k| : char16_t(word); 185| 2.95k| } else { 186| | // will generate a surrogate pair 187| 2.76k| if (word > 0x10FFFF) { ------------------ | Branch (187:15): [True: 286, False: 2.47k] ------------------ 188| 286| return std::make_pair( 189| 286| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 190| 286| } 191| 2.47k| word -= 0x10000; 192| 2.47k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 193| 2.47k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 194| 2.47k| if (big_endian) { ------------------ | Branch (194:15): [True: 2.47k, Folded] ------------------ 195| 2.47k| high_surrogate = 196| 2.47k| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 197| 2.47k| low_surrogate = 198| 2.47k| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 199| 2.47k| } 200| 2.47k| *utf16_output++ = char16_t(high_surrogate); 201| 2.47k| *utf16_output++ = char16_t(low_surrogate); 202| 2.47k| } 203| 5.74k| } 204| 686| buf += k; 205| 686| } 206| 3.37k| } 207| | 208| 302| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 209| 630|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_125sse_convert_utf32_to_utf8EPKDimPc: 2| 916|sse_convert_utf32_to_utf8(const char32_t *buf, size_t len, char *utf8_output) { 3| 916| const char32_t *end = buf + len; 4| | 5| 916| const __m128i v_0000 = _mm_setzero_si128(); //__m128 = 128 bits 6| 916| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); // 1111 1000 0000 7| | // 0000 8| 916| const __m128i v_c080 = _mm_set1_epi16((uint16_t)0xc080); // 1100 0000 1000 9| | // 0000 10| 916| const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80); // 1111 1111 1000 11| | // 0000 12| 916| const __m128i v_ffff0000 = _mm_set1_epi32( 13| 916| (uint32_t)0xffff0000); // 1111 1111 1111 1111 0000 0000 0000 0000 14| 916| const __m128i v_7fffffff = _mm_set1_epi32( 15| 916| (uint32_t)0x7fffffff); // 0111 1111 1111 1111 1111 1111 1111 1111 16| 916| __m128i running_max = _mm_setzero_si128(); 17| 916| __m128i forbidden_bytemask = _mm_setzero_si128(); 18| 916| const size_t safety_margin = 19| 916| 12; // to avoid overruns, see issue 20| | // https://github.com/simdutf/simdutf/issues/92 21| | 22| 32.2k| while (end - buf >= ------------------ | Branch (22:10): [True: 31.5k, False: 704] ------------------ 23| 32.2k| std::ptrdiff_t( 24| 32.2k| 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| 31.5k| __m128i in = _mm_loadu_si128((__m128i *)buf); 29| 31.5k| __m128i nextin = _mm_loadu_si128( 30| 31.5k| (__m128i *)buf + 1); // These two values can hold only 8 UTF32 chars 31| 31.5k| running_max = _mm_max_epu32( 32| 31.5k| _mm_max_epu32(in, running_max), // take element-wise max char32_t from 33| | // in and running_max vector 34| 31.5k| 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| 31.5k| __m128i in_16 = _mm_packus_epi32( 40| 31.5k| _mm_and_si128(in, v_7fffffff), 41| 31.5k| _mm_and_si128( 42| 31.5k| nextin, 43| 31.5k| 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| 31.5k| if (_mm_testz_si128(in_16, v_ff80)) { // if the first two blocks are ASCII ------------------ | Branch (60:9): [True: 26.2k, False: 5.35k] ------------------ 61| 26.2k| __m128i thirdin = _mm_loadu_si128((__m128i *)buf + 2); 62| 26.2k| __m128i fourthin = _mm_loadu_si128((__m128i *)buf + 3); 63| 26.2k| running_max = _mm_max_epu32( 64| 26.2k| _mm_max_epu32(thirdin, running_max), 65| 26.2k| fourthin); // take the running max of all 4 vectors thus far 66| 26.2k| __m128i nextin_16 = _mm_packus_epi32( 67| 26.2k| _mm_and_si128(thirdin, v_7fffffff), 68| 26.2k| _mm_and_si128(fourthin, 69| 26.2k| v_7fffffff)); // pack into 1 vector, now you have two 70| 26.2k| if (!_mm_testz_si128( ------------------ | Branch (70:11): [True: 1.34k, False: 24.8k] ------------------ 71| 26.2k| nextin_16, 72| 26.2k| v_ff80)) { // checks if the second packed vector is ASCII, if not: 73| | // 1. pack the bytes 74| | // obviously suboptimal. 75| 1.34k| const __m128i utf8_packed = _mm_packus_epi16( 76| 1.34k| in_16, in_16); // creates two copy of in_16 in 1 vector 77| | // 2. store (16 bytes) 78| 1.34k| _mm_storeu_si128((__m128i *)utf8_output, 79| 1.34k| utf8_packed); // put them into the output 80| | // 3. adjust pointers 81| 1.34k| buf += 8; // the char32_t buffer pointer goes up 8 char32_t chars* 32 82| | // bits = 256 bits 83| 1.34k| utf8_output += 84| 1.34k| 8; // same with output, e.g. lift the first two blocks alone. 85| | // Proceed with next input 86| 1.34k| in_16 = nextin_16; 87| | // We need to update in and nextin because they are used later. 88| 1.34k| in = thirdin; 89| 1.34k| nextin = fourthin; 90| 24.8k| } else { 91| | // 1. pack the bytes 92| 24.8k| const __m128i utf8_packed = _mm_packus_epi16(in_16, nextin_16); 93| | // 2. store (16 bytes) 94| 24.8k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 95| | // 3. adjust pointers 96| 24.8k| buf += 16; 97| 24.8k| utf8_output += 16; 98| 24.8k| continue; // we are done for this round! 99| 24.8k| } 100| 26.2k| } 101| | 102| | // no bits set above 7th bit -- find out all the ASCII characters 103| 6.69k| const __m128i one_byte_bytemask = 104| 6.69k| _mm_cmpeq_epi16( // this takes four bytes at a time and compares: 105| 6.69k| _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| 6.69k| v_0000 // 108| 6.69k| ); // 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| 6.69k| const uint16_t one_byte_bitmask = static_cast(_mm_movemask_epi8( 113| 6.69k| 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| 6.69k| const __m128i one_or_two_bytes_bytemask = 118| 6.69k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_0000); 119| 6.69k| const uint16_t one_or_two_bytes_bitmask = 120| 6.69k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 121| | 122| 6.69k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (122:9): [True: 3.16k, False: 3.52k] ------------------ 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| 3.16k| const __m128i v_1f00 = 129| 3.16k| _mm_set1_epi16((int16_t)0x1f00); // 0001 1111 0000 0000 130| 3.16k| const __m128i v_003f = 131| 3.16k| _mm_set1_epi16((int16_t)0x003f); // 0000 0000 0011 1111 132| | 133| | // t0 = [000a|aaaa|bbbb|bb00] 134| 3.16k| const __m128i t0 = _mm_slli_epi16(in_16, 2); // shift packed vector by two 135| | // t1 = [000a|aaaa|0000|0000] 136| 3.16k| const __m128i t1 = _mm_and_si128(t0, v_1f00); // potential first utf8 byte 137| | // t2 = [0000|0000|00bb|bbbb] 138| 3.16k| const __m128i t2 = 139| 3.16k| _mm_and_si128(in_16, v_003f); // potential second utf8 byte 140| | // t3 = [000a|aaaa|00bb|bbbb] 141| 3.16k| const __m128i t3 = 142| 3.16k| _mm_or_si128(t1, t2); // first and second potential utf8 byte together 143| | // t4 = [110a|aaaa|10bb|bbbb] 144| 3.16k| const __m128i t4 = _mm_or_si128( 145| 3.16k| t3, 146| 3.16k| v_c080); // t3 | 1100 0000 1000 0000 = full potential 2-byte utf8 unit 147| | 148| | // 2. merge ASCII and 2-byte codewords 149| 3.16k| const __m128i utf8_unpacked = 150| 3.16k| _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| 3.16k| const uint16_t m0 = one_byte_bitmask & 0x5555; // m0 = 0h0g0f0e0d0c0b0a 156| 3.16k| const uint16_t m1 = 157| 3.16k| static_cast(m0 >> 7); // m1 = 00000000h0g0f0e0 158| 3.16k| const uint8_t m2 = 159| 3.16k| static_cast((m0 | m1) & 0xff); // m2 = hdgcfbea 160| | // 4. pack the bytes 161| 3.16k| const uint8_t *row = 162| 3.16k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[m2][0]; 163| 3.16k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 164| 3.16k| const __m128i utf8_packed = _mm_shuffle_epi8(utf8_unpacked, shuffle); 165| | 166| | // 5. store bytes 167| 3.16k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 168| | 169| | // 6. adjust pointers 170| 3.16k| buf += 8; 171| 3.16k| utf8_output += row[0]; 172| 3.16k| continue; 173| 3.16k| } 174| | 175| | // Check for overflow in packing 176| | 177| 3.52k| const __m128i saturation_bytemask = _mm_cmpeq_epi32( 178| 3.52k| _mm_and_si128(_mm_or_si128(in, nextin), v_ffff0000), v_0000); 179| 3.52k| const uint32_t saturation_bitmask = 180| 3.52k| static_cast(_mm_movemask_epi8(saturation_bytemask)); 181| 3.52k| if (saturation_bitmask == 0xffff) { ------------------ | Branch (181:9): [True: 1.62k, False: 1.90k] ------------------ 182| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 183| 1.62k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 184| 1.62k| forbidden_bytemask = 185| 1.62k| _mm_or_si128(forbidden_bytemask, 186| 1.62k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_d800)); 187| | 188| 1.62k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 189| 1.62k| 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| 1.62k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 219| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 220| 1.62k| const __m128i t0 = _mm_shuffle_epi8(in_16, dup_even); 221| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 222| 1.62k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 218| 1.62k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 223| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 224| 1.62k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 218| 1.62k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 225| | 226| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 227| 1.62k| const __m128i s0 = _mm_srli_epi16(in_16, 4); 228| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 229| 1.62k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 218| 1.62k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 230| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 231| 1.62k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 218| 1.62k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 232| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 233| 1.62k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 218| 1.62k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 234| 1.62k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 235| 1.62k| simdutf_vec(0b0100000000000000)); ------------------ | | 218| 1.62k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 236| 1.62k| const __m128i s4 = _mm_xor_si128(s3, m0); 237| 1.62k|#undef simdutf_vec 238| | 239| | // 4. expand code units 16-bit => 32-bit 240| 1.62k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 241| 1.62k| 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| 1.62k| const uint16_t mask = 245| 1.62k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 246| 1.62k| if (mask == 0) { ------------------ | Branch (246:11): [True: 652, False: 974] ------------------ 247| | // We only have three-byte code units. Use fast path. 248| 652| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 249| 652| 15, 13, -1, -1, -1, -1); 250| 652| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 251| 652| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 252| 652| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 253| 652| utf8_output += 12; 254| 652| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 255| 652| utf8_output += 12; 256| 652| buf += 8; 257| 652| continue; 258| 652| } 259| 974| const uint8_t mask0 = uint8_t(mask); 260| | 261| 974| const uint8_t *row0 = 262| 974| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 263| 974| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 264| 974| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 265| | 266| 974| const uint8_t mask1 = static_cast(mask >> 8); 267| | 268| 974| const uint8_t *row1 = 269| 974| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 270| 974| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 271| 974| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 272| | 273| 974| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 274| 974| utf8_output += row0[0]; 275| 974| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 276| 974| utf8_output += row1[0]; 277| | 278| 974| buf += 8; 279| 1.90k| } 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| 1.90k| size_t forward = 15; 285| 1.90k| size_t k = 0; 286| 1.90k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (286:11): [True: 0, False: 1.90k] ------------------ 287| 0| forward = size_t(end - buf - 1); 288| 0| } 289| 27.9k| for (; k < forward; k++) { ------------------ | Branch (289:14): [True: 26.2k, False: 1.69k] ------------------ 290| 26.2k| uint32_t word = buf[k]; 291| 26.2k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (291:13): [True: 16.9k, False: 9.33k] ------------------ 292| 16.9k| *utf8_output++ = char(word); 293| 16.9k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (293:20): [True: 1.20k, False: 8.13k] ------------------ 294| 1.20k| *utf8_output++ = char((word >> 6) | 0b11000000); 295| 1.20k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 296| 8.13k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (296:20): [True: 3.22k, False: 4.90k] ------------------ 297| 3.22k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (297:15): [True: 1.02k, False: 2.20k] | Branch (297:33): [True: 24, False: 1.00k] ------------------ 298| 24| return std::make_pair(nullptr, utf8_output); 299| 24| } 300| 3.20k| *utf8_output++ = char((word >> 12) | 0b11100000); 301| 3.20k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 302| 3.20k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 303| 4.90k| } else { 304| 4.90k| if (word > 0x10FFFF) { ------------------ | Branch (304:15): [True: 188, False: 4.71k] ------------------ 305| 188| return std::make_pair(nullptr, utf8_output); 306| 188| } 307| 4.71k| *utf8_output++ = char((word >> 18) | 0b11110000); 308| 4.71k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 309| 4.71k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 310| 4.71k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 311| 4.71k| } 312| 26.2k| } 313| 1.69k| buf += k; 314| 1.69k| } 315| 3.52k| } // while 316| | 317| | // check for invalid input 318| 704| const __m128i v_10ffff = _mm_set1_epi32((uint32_t)0x10ffff); 319| 704| if (static_cast(_mm_movemask_epi8(_mm_cmpeq_epi32( ------------------ | Branch (319:7): [True: 22, False: 682] ------------------ 320| 704| _mm_max_epu32(running_max, v_10ffff), v_10ffff))) != 0xffff) { 321| 22| return std::make_pair(nullptr, utf8_output); 322| 22| } 323| | 324| 682| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (324:7): [True: 64, False: 618] ------------------ 325| 64| return std::make_pair(nullptr, utf8_output); 326| 64| } 327| | 328| 618| return std::make_pair(buf, utf8_output); 329| 682|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_137sse_convert_utf32_to_utf8_with_errorsEPKDimPc: 333| 934| char *utf8_output) { 334| 934| const char32_t *end = buf + len; 335| 934| const char32_t *start = buf; 336| | 337| 934| const __m128i v_0000 = _mm_setzero_si128(); 338| 934| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 339| 934| const __m128i v_c080 = _mm_set1_epi16((uint16_t)0xc080); 340| 934| const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80); 341| 934| const __m128i v_ffff0000 = _mm_set1_epi32((uint32_t)0xffff0000); 342| 934| const __m128i v_7fffffff = _mm_set1_epi32((uint32_t)0x7fffffff); 343| 934| const __m128i v_10ffff = _mm_set1_epi32((uint32_t)0x10ffff); 344| | 345| 934| const size_t safety_margin = 346| 934| 12; // to avoid overruns, see issue 347| | // https://github.com/simdutf/simdutf/issues/92 348| | 349| 97.1k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (349:10): [True: 96.3k, False: 734] ------------------ 350| | // We load two 16 bytes registers for a total of 32 bytes or 8 characters. 351| 96.3k| __m128i in = _mm_loadu_si128((__m128i *)buf); 352| 96.3k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 353| | // Check for too large input 354| 96.3k| __m128i max_input = _mm_max_epu32(_mm_max_epu32(in, nextin), v_10ffff); 355| 96.3k| if (static_cast(_mm_movemask_epi8( ------------------ | Branch (355:9): [True: 58, False: 96.3k] ------------------ 356| 96.3k| _mm_cmpeq_epi32(max_input, v_10ffff))) != 0xffff) { 357| 58| return std::make_pair(result(error_code::TOO_LARGE, buf - start), 358| 58| utf8_output); 359| 58| } 360| | 361| | // Pack 32-bit UTF-32 code units to 16-bit UTF-16 code units with unsigned 362| | // saturation 363| 96.3k| __m128i in_16 = _mm_packus_epi32(_mm_and_si128(in, v_7fffffff), 364| 96.3k| _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| 96.3k| if (_mm_testz_si128(in_16, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (369:9): [True: 88.8k, False: 7.45k] ------------------ 370| | // 1. pack the bytes 371| | // obviously suboptimal. 372| 88.8k| const __m128i utf8_packed = _mm_packus_epi16(in_16, in_16); 373| | // 2. store (16 bytes) 374| 88.8k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 375| | // 3. adjust pointers 376| 88.8k| buf += 8; 377| 88.8k| utf8_output += 8; 378| 88.8k| continue; 379| 88.8k| } 380| | 381| | // no bits set above 7th bit 382| 7.45k| const __m128i one_byte_bytemask = 383| 7.45k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_ff80), v_0000); 384| 7.45k| const uint16_t one_byte_bitmask = 385| 7.45k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 386| | 387| | // no bits set above 11th bit 388| 7.45k| const __m128i one_or_two_bytes_bytemask = 389| 7.45k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_0000); 390| 7.45k| const uint16_t one_or_two_bytes_bitmask = 391| 7.45k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 392| | 393| 7.45k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (393:9): [True: 4.35k, False: 3.09k] ------------------ 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| 4.35k| const __m128i v_1f00 = _mm_set1_epi16((int16_t)0x1f00); 400| 4.35k| const __m128i v_003f = _mm_set1_epi16((int16_t)0x003f); 401| | 402| | // t0 = [000a|aaaa|bbbb|bb00] 403| 4.35k| const __m128i t0 = _mm_slli_epi16(in_16, 2); 404| | // t1 = [000a|aaaa|0000|0000] 405| 4.35k| const __m128i t1 = _mm_and_si128(t0, v_1f00); 406| | // t2 = [0000|0000|00bb|bbbb] 407| 4.35k| const __m128i t2 = _mm_and_si128(in_16, v_003f); 408| | // t3 = [000a|aaaa|00bb|bbbb] 409| 4.35k| const __m128i t3 = _mm_or_si128(t1, t2); 410| | // t4 = [110a|aaaa|10bb|bbbb] 411| 4.35k| const __m128i t4 = _mm_or_si128(t3, v_c080); 412| | 413| | // 2. merge ASCII and 2-byte codewords 414| 4.35k| const __m128i utf8_unpacked = 415| 4.35k| _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| 4.35k| const uint16_t m0 = one_byte_bitmask & 0x5555; // m0 = 0h0g0f0e0d0c0b0a 421| 4.35k| const uint16_t m1 = 422| 4.35k| static_cast(m0 >> 7); // m1 = 00000000h0g0f0e0 423| 4.35k| const uint8_t m2 = 424| 4.35k| static_cast((m0 | m1) & 0xff); // m2 = hdgcfbea 425| | // 4. pack the bytes 426| 4.35k| const uint8_t *row = 427| 4.35k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[m2][0]; 428| 4.35k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 429| 4.35k| const __m128i utf8_packed = _mm_shuffle_epi8(utf8_unpacked, shuffle); 430| | 431| | // 5. store bytes 432| 4.35k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 433| | 434| | // 6. adjust pointers 435| 4.35k| buf += 8; 436| 4.35k| utf8_output += row[0]; 437| 4.35k| continue; 438| 4.35k| } 439| | 440| | // Check for overflow in packing 441| 3.09k| const __m128i saturation_bytemask = _mm_cmpeq_epi32( 442| 3.09k| _mm_and_si128(_mm_or_si128(in, nextin), v_ffff0000), v_0000); 443| 3.09k| const uint32_t saturation_bitmask = 444| 3.09k| static_cast(_mm_movemask_epi8(saturation_bytemask)); 445| | 446| 3.09k| if (saturation_bitmask == 0xffff) { ------------------ | Branch (446:9): [True: 1.61k, False: 1.47k] ------------------ 447| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 448| | 449| | // Check for illegal surrogate code units 450| 1.61k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 451| 1.61k| const __m128i forbidden_bytemask = 452| 1.61k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_d800); 453| 1.61k| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (453:11): [True: 34, False: 1.58k] ------------------ 454| 34| return std::make_pair(result(error_code::SURROGATE, buf - start), 455| 34| utf8_output); 456| 34| } 457| | 458| 1.58k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 459| 1.58k| 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| 1.58k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 489| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 490| 1.58k| const __m128i t0 = _mm_shuffle_epi8(in_16, dup_even); 491| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 492| 1.58k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 488| 1.58k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 493| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 494| 1.58k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 488| 1.58k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 495| | 496| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 497| 1.58k| const __m128i s0 = _mm_srli_epi16(in_16, 4); 498| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 499| 1.58k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 488| 1.58k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 500| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 501| 1.58k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 488| 1.58k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 502| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 503| 1.58k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 488| 1.58k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 504| 1.58k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 505| 1.58k| simdutf_vec(0b0100000000000000)); ------------------ | | 488| 1.58k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 506| 1.58k| const __m128i s4 = _mm_xor_si128(s3, m0); 507| 1.58k|#undef simdutf_vec 508| | 509| | // 4. expand code units 16-bit => 32-bit 510| 1.58k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 511| 1.58k| 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| 1.58k| const uint16_t mask = 515| 1.58k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 516| 1.58k| if (mask == 0) { ------------------ | Branch (516:11): [True: 862, False: 722] ------------------ 517| | // We only have three-byte code units. Use fast path. 518| 862| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 519| 862| 15, 13, -1, -1, -1, -1); 520| 862| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 521| 862| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 522| 862| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 523| 862| utf8_output += 12; 524| 862| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 525| 862| utf8_output += 12; 526| 862| buf += 8; 527| 862| continue; 528| 862| } 529| 722| const uint8_t mask0 = uint8_t(mask); 530| | 531| 722| const uint8_t *row0 = 532| 722| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 533| 722| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 534| 722| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 535| | 536| 722| const uint8_t mask1 = static_cast(mask >> 8); 537| | 538| 722| const uint8_t *row1 = 539| 722| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 540| 722| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 541| 722| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 542| | 543| 722| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 544| 722| utf8_output += row0[0]; 545| 722| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 546| 722| utf8_output += row1[0]; 547| | 548| 722| buf += 8; 549| 1.47k| } 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| 1.47k| size_t forward = 15; 555| 1.47k| size_t k = 0; 556| 1.47k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (556:11): [True: 0, False: 1.47k] ------------------ 557| 0| forward = size_t(end - buf - 1); 558| 0| } 559| 23.0k| for (; k < forward; k++) { ------------------ | Branch (559:14): [True: 21.6k, False: 1.36k] ------------------ 560| 21.6k| uint32_t word = buf[k]; 561| 21.6k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (561:13): [True: 8.55k, False: 13.1k] ------------------ 562| 8.55k| *utf8_output++ = char(word); 563| 13.1k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (563:20): [True: 1.60k, False: 11.5k] ------------------ 564| 1.60k| *utf8_output++ = char((word >> 6) | 0b11000000); 565| 1.60k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 566| 11.5k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (566:20): [True: 6.20k, False: 5.31k] ------------------ 567| 6.20k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (567:15): [True: 3.57k, False: 2.62k] | Branch (567:33): [True: 26, False: 3.55k] ------------------ 568| 26| return std::make_pair( 569| 26| result(error_code::SURROGATE, buf - start + k), utf8_output); 570| 26| } 571| 6.17k| *utf8_output++ = char((word >> 12) | 0b11100000); 572| 6.17k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 573| 6.17k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 574| 6.17k| } else { 575| 5.31k| if (word > 0x10FFFF) { ------------------ | Branch (575:15): [True: 82, False: 5.23k] ------------------ 576| 82| return std::make_pair( 577| 82| result(error_code::TOO_LARGE, buf - start + k), utf8_output); 578| 82| } 579| 5.23k| *utf8_output++ = char((word >> 18) | 0b11110000); 580| 5.23k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 581| 5.23k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 582| 5.23k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 583| 5.23k| } 584| 21.6k| } 585| 1.36k| buf += k; 586| 1.36k| } 587| 3.09k| } // while 588| 734| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 589| 934|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_129convert_masked_utf8_to_latin1EPKcmRPc: 9| 605k| 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| 605k| const __m128i in = _mm_loadu_si128((__m128i *)input); 21| 605k| const uint16_t input_utf8_end_of_code_point_mask = 22| 605k| utf8_end_of_code_point_mask & 23| 605k| 0xfff; // we are only processing 12 bytes in case it is not all ASCII 24| 605k| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (24:7): [True: 2.94k, False: 602k] ------------------ 25| | // We process the data in chunks of 12 bytes. 26| 2.94k| _mm_storeu_si128(reinterpret_cast<__m128i *>(latin1_output), in); 27| 2.94k| latin1_output += 12; // We wrote 12 characters. 28| 2.94k| return 12; // We consumed 12 bytes. 29| 2.94k| } 30| | /// We do not have a fast path available, so we fallback. 31| 602k| const uint8_t idx = 32| 602k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 33| 602k| const uint8_t consumed = 34| 602k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 35| | // this indicates an invalid input: 36| 602k| if (idx >= 64) { ------------------ | Branch (36:7): [True: 32.7k, False: 569k] ------------------ 37| 32.7k| return consumed; 38| 32.7k| } 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| 569k| const __m128i sh = 46| 569k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 47| 569k| const __m128i perm = _mm_shuffle_epi8(in, sh); 48| 569k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 49| 569k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 50| 569k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 51| 569k| 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| 569k| _mm_storel_epi64((__m128i *)latin1_output, latin1_packed); 56| 569k| latin1_output += 6; // We wrote 6 bytes. 57| 569k| return consumed; 58| 602k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE0EEEmPKcmRPDs: 10| 3.02M| 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| 3.02M| const __m128i swap = 23| 3.02M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 3.02M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 3.02M| const uint16_t input_utf8_end_of_code_point_mask = 26| 3.02M| utf8_end_of_code_point_mask & 0xfff; 27| 3.02M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 10.8k, False: 3.01M] ------------------ 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| 3.01M| if (((utf8_end_of_code_point_mask & 0xFFFF) == 0xaaaa)) { ------------------ | Branch (42:7): [True: 14.8k, False: 2.99M] ------------------ 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| 14.8k| const __m128i sh = 47| 14.8k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 48| 14.8k| const __m128i perm = _mm_shuffle_epi8(in, sh); 49| 14.8k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 50| 14.8k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 51| 14.8k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 52| 14.8k| if (big_endian) ------------------ | Branch (52:9): [Folded, False: 14.8k] ------------------ 53| 0| composed = _mm_shuffle_epi8(composed, swap); 54| 14.8k| _mm_storeu_si128((__m128i *)utf16_output, composed); 55| 14.8k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 56| 14.8k| return 16; 57| 14.8k| } 58| 2.99M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (58:7): [True: 2.78k, False: 2.99M] ------------------ 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| 2.78k| const __m128i sh = 63| 2.78k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 64| 2.78k| const __m128i perm = _mm_shuffle_epi8(in, sh); 65| 2.78k| const __m128i ascii = 66| 2.78k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 67| 2.78k| const __m128i middlebyte = 68| 2.78k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 69| 2.78k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 70| 2.78k| const __m128i highbyte = 71| 2.78k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 72| 2.78k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 73| 2.78k| const __m128i composed = 74| 2.78k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 75| 2.78k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 76| 2.78k| if (big_endian) ------------------ | Branch (76:9): [Folded, False: 2.78k] ------------------ 77| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 78| 2.78k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 79| 2.78k| utf16_output += 4; 80| 2.78k| return 12; 81| 2.78k| } 82| | /// We do not have a fast path available, so we fallback. 83| | 84| 2.99M| const uint8_t idx = 85| 2.99M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 86| 2.99M| const uint8_t consumed = 87| 2.99M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 88| 2.99M| if (idx < 64) { ------------------ | Branch (88:7): [True: 2.54M, False: 443k] ------------------ 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| 2.54M| const __m128i sh = 96| 2.54M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 97| 2.54M| const __m128i perm = _mm_shuffle_epi8(in, sh); 98| 2.54M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 99| 2.54M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 100| 2.54M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 101| 2.54M| if (big_endian) ------------------ | Branch (101:9): [Folded, False: 2.54M] ------------------ 102| 0| composed = _mm_shuffle_epi8(composed, swap); 103| 2.54M| _mm_storeu_si128((__m128i *)utf16_output, composed); 104| 2.54M| utf16_output += 6; // We wrote 12 bytes, 6 code points. 105| 2.54M| } else if (idx < 145) { ------------------ | Branch (105:14): [True: 369k, False: 73.8k] ------------------ 106| | // FOUR (4) input code-code units 107| 369k| const __m128i sh = 108| 369k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 109| 369k| const __m128i perm = _mm_shuffle_epi8(in, sh); 110| 369k| const __m128i ascii = 111| 369k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 112| 369k| const __m128i middlebyte = 113| 369k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 114| 369k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 115| 369k| const __m128i highbyte = 116| 369k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 117| 369k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 118| 369k| const __m128i composed = 119| 369k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 120| 369k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 121| 369k| if (big_endian) ------------------ | Branch (121:9): [Folded, False: 369k] ------------------ 122| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 123| 369k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 124| 369k| utf16_output += 4; 125| 369k| } else if (idx < 209) { ------------------ | Branch (125:14): [True: 72.5k, False: 1.29k] ------------------ 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| 72.5k| const __m128i sh = 137| 72.5k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 138| 72.5k| const __m128i perm = _mm_shuffle_epi8(in, sh); 139| 72.5k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 140| 72.5k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 141| 72.5k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 142| 72.5k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 143| | // correct for spurious high bit 144| 72.5k| const __m128i correct = 145| 72.5k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 146| 72.5k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 147| 72.5k| 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| 72.5k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 151| 72.5k| 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| 72.5k| const __m128i composed = 157| 72.5k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 158| 72.5k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 159| 72.5k| const __m128i composedminus = 160| 72.5k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 161| 72.5k| const __m128i lowtenbits = 162| 72.5k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 163| | // Notice the 0x3ff mask: 164| 72.5k| const __m128i hightenbits = 165| 72.5k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 166| 72.5k| const __m128i lowtenbitsadd = 167| 72.5k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 168| 72.5k| const __m128i hightenbitsadd = 169| 72.5k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 170| 72.5k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 171| 72.5k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 172| 72.5k| uint32_t basic_buffer[4]; 173| 72.5k| uint32_t basic_buffer_swap[4]; 174| 72.5k| if (big_endian) { ------------------ | Branch (174:9): [Folded, False: 72.5k] ------------------ 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| 72.5k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 180| 72.5k| uint32_t surrogate_buffer[4]; 181| 72.5k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 182| 290k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (182:24): [True: 217k, False: 72.5k] ------------------ 183| 217k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (183:11): [True: 59.9k, False: 157k] ------------------ 184| 59.9k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 185| 59.9k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 186| 59.9k| utf16_output += 2; 187| 157k| } else { 188| 157k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (188:27): [Folded, False: 157k] ------------------ 189| 157k| : uint16_t(basic_buffer[i]); 190| 157k| utf16_output++; 191| 157k| } 192| 217k| } 193| 72.5k| } else { 194| | // here we know that there is an error but we do not handle errors 195| 1.29k| } 196| 2.99M| return consumed; 197| 2.99M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE1EEEmPKcmRPDs: 10| 3.04M| 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| 3.04M| const __m128i swap = 23| 3.04M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 3.04M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 3.04M| const uint16_t input_utf8_end_of_code_point_mask = 26| 3.04M| utf8_end_of_code_point_mask & 0xfff; 27| 3.04M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 10.8k, False: 3.03M] ------------------ 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): [True: 10.8k, Folded] ------------------ 33| 10.8k| ascii_first = _mm_shuffle_epi8(ascii_first, swap); 34| 10.8k| ascii_second = _mm_shuffle_epi8(ascii_second, swap); 35| 10.8k| } 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| 3.03M| if (((utf8_end_of_code_point_mask & 0xFFFF) == 0xaaaa)) { ------------------ | Branch (42:7): [True: 27.5k, False: 3.00M] ------------------ 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| 27.5k| const __m128i sh = 47| 27.5k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 48| 27.5k| const __m128i perm = _mm_shuffle_epi8(in, sh); 49| 27.5k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 50| 27.5k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 51| 27.5k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 52| 27.5k| if (big_endian) ------------------ | Branch (52:9): [True: 27.5k, Folded] ------------------ 53| 27.5k| composed = _mm_shuffle_epi8(composed, swap); 54| 27.5k| _mm_storeu_si128((__m128i *)utf16_output, composed); 55| 27.5k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 56| 27.5k| return 16; 57| 27.5k| } 58| 3.00M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (58:7): [True: 7.44k, False: 2.99M] ------------------ 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| 7.44k| const __m128i sh = 63| 7.44k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 64| 7.44k| const __m128i perm = _mm_shuffle_epi8(in, sh); 65| 7.44k| const __m128i ascii = 66| 7.44k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 67| 7.44k| const __m128i middlebyte = 68| 7.44k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 69| 7.44k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 70| 7.44k| const __m128i highbyte = 71| 7.44k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 72| 7.44k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 73| 7.44k| const __m128i composed = 74| 7.44k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 75| 7.44k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 76| 7.44k| if (big_endian) ------------------ | Branch (76:9): [True: 7.44k, Folded] ------------------ 77| 7.44k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 78| 7.44k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 79| 7.44k| utf16_output += 4; 80| 7.44k| return 12; 81| 7.44k| } 82| | /// We do not have a fast path available, so we fallback. 83| | 84| 2.99M| const uint8_t idx = 85| 2.99M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 86| 2.99M| const uint8_t consumed = 87| 2.99M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 88| 2.99M| if (idx < 64) { ------------------ | Branch (88:7): [True: 2.74M, False: 248k] ------------------ 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| 2.74M| const __m128i sh = 96| 2.74M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 97| 2.74M| const __m128i perm = _mm_shuffle_epi8(in, sh); 98| 2.74M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 99| 2.74M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 100| 2.74M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 101| 2.74M| if (big_endian) ------------------ | Branch (101:9): [True: 2.74M, Folded] ------------------ 102| 2.74M| composed = _mm_shuffle_epi8(composed, swap); 103| 2.74M| _mm_storeu_si128((__m128i *)utf16_output, composed); 104| 2.74M| utf16_output += 6; // We wrote 12 bytes, 6 code points. 105| 2.74M| } else if (idx < 145) { ------------------ | Branch (105:14): [True: 164k, False: 83.8k] ------------------ 106| | // FOUR (4) input code-code units 107| 164k| const __m128i sh = 108| 164k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 109| 164k| const __m128i perm = _mm_shuffle_epi8(in, sh); 110| 164k| const __m128i ascii = 111| 164k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 112| 164k| const __m128i middlebyte = 113| 164k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 114| 164k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 115| 164k| const __m128i highbyte = 116| 164k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 117| 164k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 118| 164k| const __m128i composed = 119| 164k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 120| 164k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 121| 164k| if (big_endian) ------------------ | Branch (121:9): [True: 164k, Folded] ------------------ 122| 164k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 123| 164k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 124| 164k| utf16_output += 4; 125| 164k| } else if (idx < 209) { ------------------ | Branch (125:14): [True: 82.9k, False: 894] ------------------ 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| 82.9k| const __m128i sh = 137| 82.9k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 138| 82.9k| const __m128i perm = _mm_shuffle_epi8(in, sh); 139| 82.9k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 140| 82.9k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 141| 82.9k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 142| 82.9k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 143| | // correct for spurious high bit 144| 82.9k| const __m128i correct = 145| 82.9k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 146| 82.9k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 147| 82.9k| 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| 82.9k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 151| 82.9k| 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| 82.9k| const __m128i composed = 157| 82.9k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 158| 82.9k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 159| 82.9k| const __m128i composedminus = 160| 82.9k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 161| 82.9k| const __m128i lowtenbits = 162| 82.9k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 163| | // Notice the 0x3ff mask: 164| 82.9k| const __m128i hightenbits = 165| 82.9k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 166| 82.9k| const __m128i lowtenbitsadd = 167| 82.9k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 168| 82.9k| const __m128i hightenbitsadd = 169| 82.9k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 170| 82.9k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 171| 82.9k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 172| 82.9k| uint32_t basic_buffer[4]; 173| 82.9k| uint32_t basic_buffer_swap[4]; 174| 82.9k| if (big_endian) { ------------------ | Branch (174:9): [True: 82.9k, Folded] ------------------ 175| 82.9k| _mm_storeu_si128((__m128i *)basic_buffer_swap, 176| 82.9k| _mm_shuffle_epi8(composed, swap)); 177| 82.9k| surrogates = _mm_shuffle_epi8(surrogates, swap); 178| 82.9k| } 179| 82.9k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 180| 82.9k| uint32_t surrogate_buffer[4]; 181| 82.9k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 182| 331k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (182:24): [True: 248k, False: 82.9k] ------------------ 183| 248k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (183:11): [True: 69.3k, False: 179k] ------------------ 184| 69.3k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 185| 69.3k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 186| 69.3k| utf16_output += 2; 187| 179k| } else { 188| 179k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (188:27): [True: 179k, Folded] ------------------ 189| 179k| : uint16_t(basic_buffer[i]); 190| 179k| utf16_output++; 191| 179k| } 192| 248k| } 193| 82.9k| } else { 194| | // here we know that there is an error but we do not handle errors 195| 894| } 196| 2.99M| return consumed; 197| 3.00M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_128convert_masked_utf8_to_utf32EPKcmRPDi: 9| 3.60M| 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| 3.60M| const __m128i in = _mm_loadu_si128((__m128i *)input); 22| 3.60M| const uint16_t input_utf8_end_of_code_point_mask = 23| 3.60M| utf8_end_of_code_point_mask & 0xfff; 24| 3.60M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (24:7): [True: 9.31k, False: 3.59M] ------------------ 25| | // We process the data in chunks of 12 bytes. 26| 9.31k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 27| 9.31k| _mm_cvtepu8_epi32(in)); 28| 9.31k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 29| 9.31k| _mm_cvtepu8_epi32(_mm_srli_si128(in, 4))); 30| 9.31k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 8), 31| 9.31k| _mm_cvtepu8_epi32(_mm_srli_si128(in, 8))); 32| 9.31k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 12), 33| 9.31k| _mm_cvtepu8_epi32(_mm_srli_si128(in, 12))); 34| 9.31k| utf32_output += 12; // We wrote 12 32-bit characters. 35| 9.31k| return 12; // We consumed 12 bytes. 36| 9.31k| } 37| 3.59M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (37:7): [True: 28.2k, False: 3.56M] ------------------ 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| 28.2k| const __m128i sh = 42| 28.2k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 43| 28.2k| const __m128i perm = _mm_shuffle_epi8(in, sh); 44| 28.2k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 45| 28.2k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 46| 28.2k| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 47| 28.2k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 48| 28.2k| _mm_cvtepu16_epi32(composed)); 49| 28.2k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 50| 28.2k| _mm_cvtepu16_epi32(_mm_srli_si128(composed, 8))); 51| 28.2k| utf32_output += 8; // We wrote 32 bytes, 8 code points. 52| 28.2k| return 16; 53| 28.2k| } 54| 3.56M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (54:7): [True: 2.94k, False: 3.56M] ------------------ 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| 2.94k| const __m128i sh = 59| 2.94k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 60| 2.94k| const __m128i perm = _mm_shuffle_epi8(in, sh); 61| 2.94k| const __m128i ascii = 62| 2.94k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 63| 2.94k| const __m128i middlebyte = 64| 2.94k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 65| 2.94k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 66| 2.94k| const __m128i highbyte = 67| 2.94k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 68| 2.94k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 69| 2.94k| const __m128i composed = 70| 2.94k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 71| 2.94k| _mm_storeu_si128((__m128i *)utf32_output, composed); 72| 2.94k| utf32_output += 4; 73| 2.94k| return 12; 74| 2.94k| } 75| | /// We do not have a fast path available, so we fallback. 76| | 77| 3.56M| const uint8_t idx = 78| 3.56M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 79| 3.56M| const uint8_t consumed = 80| 3.56M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 81| 3.56M| if (idx < 64) { ------------------ | Branch (81:7): [True: 3.29M, False: 271k] ------------------ 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| 3.29M| const __m128i sh = 89| 3.29M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 90| 3.29M| const __m128i perm = _mm_shuffle_epi8(in, sh); 91| 3.29M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 92| 3.29M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 93| 3.29M| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 94| 3.29M| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 95| 3.29M| _mm_cvtepu16_epi32(composed)); 96| 3.29M| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 97| 3.29M| _mm_cvtepu16_epi32(_mm_srli_si128(composed, 8))); 98| 3.29M| utf32_output += 6; // We wrote 12 bytes, 6 code points. 99| 3.29M| } else if (idx < 145) { ------------------ | Branch (99:14): [True: 187k, False: 83.6k] ------------------ 100| | // FOUR (4) input code-code units 101| 187k| const __m128i sh = 102| 187k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 103| 187k| const __m128i perm = _mm_shuffle_epi8(in, sh); 104| 187k| const __m128i ascii = 105| 187k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 106| 187k| const __m128i middlebyte = 107| 187k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 108| 187k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 109| 187k| const __m128i highbyte = 110| 187k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 111| 187k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 112| 187k| const __m128i composed = 113| 187k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 114| 187k| _mm_storeu_si128((__m128i *)utf32_output, composed); 115| 187k| utf32_output += 4; 116| 187k| } else if (idx < 209) { ------------------ | Branch (116:14): [True: 82.9k, False: 682] ------------------ 117| | // TWO (2) input code-code units 118| 82.9k| const __m128i sh = 119| 82.9k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 120| 82.9k| const __m128i perm = _mm_shuffle_epi8(in, sh); 121| 82.9k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 122| 82.9k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 123| 82.9k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 124| 82.9k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 125| | // correct for spurious high bit 126| 82.9k| const __m128i correct = 127| 82.9k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 128| 82.9k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 129| 82.9k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 130| 82.9k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0x07000000)); 131| 82.9k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 132| 82.9k| const __m128i composed = 133| 82.9k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 134| 82.9k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 135| 82.9k| _mm_storeu_si128((__m128i *)utf32_output, composed); 136| 82.9k| utf32_output += 3; 137| 82.9k| } else { 138| | // here we know that there is an error but we do not handle errors 139| 682| } 140| 3.56M| return consumed; 141| 3.56M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_123utf16_gather_high_bytesILNS_10endiannessE0EEENS1_4simd5simd8IhEENS4_6simd16ItEES8_: 3| 2.67M| const simd16 in1) { 4| 2.67M| if (big_endian) { ------------------ | Branch (4:7): [Folded, False: 2.67M] ------------------ 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| 2.67M| } else { 12| 2.67M| const auto t0 = in0.shr<8>(); 13| 2.67M| const auto t1 = in1.shr<8>(); 14| | 15| 2.67M| return simd16::pack(t0, t1); 16| 2.67M| } 17| 2.67M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_123utf16_gather_high_bytesILNS_10endiannessE1EEENS1_4simd5simd8IhEENS4_6simd16ItEES8_: 3| 2.67M| const simd16 in1) { 4| 2.67M| if (big_endian) { ------------------ | Branch (4:7): [True: 2.67M, Folded] ------------------ 5| | // we want lower bytes 6| 2.67M| const auto mask = simd16(0x00ff); 7| 2.67M| const auto t0 = in0 & mask; 8| 2.67M| const auto t1 = in1 & mask; 9| | 10| 2.67M| return simd16::pack(t0, t1); 11| 2.67M| } 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| 2.67M|}