_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| 8.58k|extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size) { 640| 8.58k| static const auto fptrs = populate_functions(); 641| 8.58k| 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| 8.58k| if (size < 4) { ------------------ | Branch (646:7): [True: 3, False: 8.57k] ------------------ 647| 3| return 0; 648| 3| } 649| | 650| 8.57k| constexpr auto actionmask = std::bit_ceil(Ncases) - 1; 651| 8.57k| const auto action = data[0] & actionmask; 652| 8.57k| data += 4; 653| 8.57k| size -= 4; 654| | 655| 8.57k| if (action >= Ncases) { ------------------ | Branch (655:7): [True: 1, False: 8.57k] ------------------ 656| 1| return 0; 657| 1| } 658| | 659| 8.57k| if constexpr (use_separate_allocation) { 660| | // this is better at exercising null input and catch buffer underflows 661| 8.57k| const std::vector separate{data, data + size}; 662| 8.57k| fptrs[action](std::span(separate)); 663| | } else { 664| | std::span chardata{(const char*)data, size}; 665| | fptrs[action](chardata); 666| | } 667| | 668| 8.57k| return 0; 669| 8.57k|} conversion.cpp:_ZZ18populate_functionsvENK3$_0clENSt3__14spanIKcLm18446744073709551615EEE: 554| 88| +[](std::span chardata) { \ 555| 88| const auto c = \ 556| 88| Conversion{ \ 559| 88| &I::lenfunc, &I::conversionfunc, \ 560| 88| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 88|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 88| ::std::void_t(); \ | | 1180| 88| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 88| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 88| constexpr auto _size = _name.size(); \ | | 1183| 88| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 88| return _nameof; }() ------------------ 561| 88| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 88|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 88| ::std::void_t(); \ | | 1180| 88| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 88| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 88| constexpr auto _size = _name.size(); \ | | 1183| 88| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 88| return _nameof; }() ------------------ 562| 88| c.fuzz(chardata); \ 563| 88| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 201| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 201| const FromSpan from{reinterpret_cast(chardata.data()), 178| 201| chardata.size() / sizeof(FromType)}; 179| | 180| 201| static const bool do_print_testcase = 181| 201| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 201| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 201] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 201| do { 189| | // step 0 - is the input valid? 190| 201| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 201| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 201] | 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| 201| From == UtfEncodings::UTF8) { 198| 201| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 201] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 201| } 201| | 202| | // step 2 - what is the required size of the output? 203| 201| const auto [output_length, length_agree] = 204| 201| calculate_length(from, inputisvalid); 205| 201| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 201] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 201| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 92, False: 109] | Branch (208:28): [True: 18, False: 74] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 18| return; 211| 18| } 212| | 213| | // step 3 - run the conversion 214| 183| const auto [written, outputs_agree] = 215| 183| do_conversion(from, output_length, inputisvalid); 216| 183| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 183] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 183| return; 221| 183| } 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| 201| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 201| validation_result verify_valid_input(FromSpan src) const { 231| 201| validation_result ret{}; 232| | 233| 201| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 201| const auto implementations = get_supported_implementations(); 235| 201| std::vector results; 236| 201| results.reserve(implementations.size()); 237| | 238| 603| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 603, False: 201] ------------------ 239| 603| results.push_back( 240| 603| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 603| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 603| const bool validation2 = 245| 603| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 603| src.data(), src.size()); 247| 603| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 603] ------------------ 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| 603| } 258| | 259| 201| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 201| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 201] ------------------ 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| 201| } else { 273| 201| ret.implementations_agree = true; 274| 201| } 275| 201| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 201| return r.error == simdutf::SUCCESS; 277| 201| }); 278| 201| return ret; 279| 201| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 402| 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| 419| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 419| return r.error == simdutf::SUCCESS; 277| 419| }); _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 201| bool count_the_input(FromSpan src) const { 290| 201| const auto implementations = get_supported_implementations(); 291| 201| std::vector results; 292| 201| results.reserve(implementations.size()); 293| | 294| 603| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 603, False: 201] ------------------ 295| 603| std::size_t ret; 296| 603| if constexpr (From == UtfEncodings::UTF16BE) { 297| 603| 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| 603| results.push_back(ret); 304| 603| } 305| 201| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 201| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 201] ------------------ 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| 201| return true; 321| 201| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 402| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 201| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 201| length_result ret{}; 344| | 345| 201| const auto implementations = get_supported_implementations(); 346| 201| std::vector results; 347| 201| results.reserve(implementations.size()); 348| | 349| 603| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 603, False: 201] ------------------ 350| 603| const auto len = invoke_lengthcalc(impl, src); 351| 603| results.push_back(len); 352| 603| ret.length.push_back(len); 353| 603| } 354| | 355| 201| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 201| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 201] ------------------ 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| 201| } else { 375| 201| ret.implementations_agree = true; 376| 201| } 377| 201| return ret; 378| 201| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 603| FromSpan src) const { 331| 603| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 603| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 402| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 183| const bool inputisvalid) const { 383| 183| conversion_result ret{}; 384| | 385| 183| const auto implementations = get_supported_implementations(); 386| | 387| 183| std::vector> results; 388| 183| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 183| std::vector> outputbuffers; 393| 183| outputbuffers.reserve(implementations.size()); 394| 732| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 549, False: 183] ------------------ 395| 549| auto impl = implementations[i]; 396| 549| const ToType canary1{42}; 397| 549| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 549| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 549| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 549| const auto success = [](const ConversionResult& r) -> bool { 402| 549| if constexpr (std::is_same_v) { 403| 549| return r != 0; 404| 549| } else { 405| 549| return r.error == simdutf::error_code::SUCCESS; 406| 549| } 407| 549| }(implret1); 408| 549| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 549| 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| 549| const ToType canary2{25}; 414| 549| const auto outputbuffer_first_run = outputbuffer; 415| 549| std::ranges::fill(outputbuffer, canary2); 416| 549| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 549| src.size(), outputbuffer.data()); 418| | 419| 549| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 549] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 549| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 327, False: 222] | Branch (423:29): [True: 318, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 318| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 318| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 318] ------------------ 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| 318| } 440| 549| } 441| 549| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 318, False: 231] ------------------ 442| 549| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 183| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 74, False: 109] ------------------ 447| 222| for (auto& e : results) { ------------------ | Branch (447:20): [True: 222, False: 74] ------------------ 448| 222| e.outputhash.clear(); 449| 222| } 450| 74| } 451| | 452| 183| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 183| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 183] ------------------ 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| 183| } else { 474| 183| ret.implementations_agree = true; 475| 183| } 476| 183| return ret; 477| 183| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 549| const auto success = [](const ConversionResult& r) -> bool { 402| 549| if constexpr (std::is_same_v) { 403| 549| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 549| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 366| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_1clENSt3__14spanIKcLm18446744073709551615EEE: 554| 240| +[](std::span chardata) { \ 555| 240| const auto c = \ 556| 240| Conversion{ \ 559| 240| &I::lenfunc, &I::conversionfunc, \ 560| 240| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 240|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 240| ::std::void_t(); \ | | 1180| 240| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 240| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 240| constexpr auto _size = _name.size(); \ | | 1183| 240| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 240| return _nameof; }() ------------------ 561| 240| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 240|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 240| ::std::void_t(); \ | | 1180| 240| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 240| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 240| constexpr auto _size = _name.size(); \ | | 1183| 240| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 240| return _nameof; }() ------------------ 562| 240| c.fuzz(chardata); \ 563| 240| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 403| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 403| const FromSpan from{reinterpret_cast(chardata.data()), 178| 403| chardata.size() / sizeof(FromType)}; 179| | 180| 403| static const bool do_print_testcase = 181| 403| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 403| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 403] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 403| do { 189| | // step 0 - is the input valid? 190| 403| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 403| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 403] | 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| 403| From == UtfEncodings::UTF8) { 198| 403| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 403] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 403| } 201| | 202| | // step 2 - what is the required size of the output? 203| 403| const auto [output_length, length_agree] = 204| 403| calculate_length(from, inputisvalid); 205| 403| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 403] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 403| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 85, False: 318] | Branch (208:28): [True: 19, False: 66] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 19| return; 211| 19| } 212| | 213| | // step 3 - run the conversion 214| 384| const auto [written, outputs_agree] = 215| 384| do_conversion(from, output_length, inputisvalid); 216| 384| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 384] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 384| return; 221| 384| } 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| 403| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 403| validation_result verify_valid_input(FromSpan src) const { 231| 403| validation_result ret{}; 232| | 233| 403| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 403| const auto implementations = get_supported_implementations(); 235| 403| std::vector results; 236| 403| results.reserve(implementations.size()); 237| | 238| 1.20k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.20k, False: 403] ------------------ 239| 1.20k| results.push_back( 240| 1.20k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.20k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.20k| const bool validation2 = 245| 1.20k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.20k| src.data(), src.size()); 247| 1.20k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.20k] ------------------ 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| 1.20k| } 258| | 259| 403| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 403| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 403] ------------------ 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| 403| } else { 273| 403| ret.implementations_agree = true; 274| 403| } 275| 403| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 403| return r.error == simdutf::SUCCESS; 277| 403| }); 278| 403| return ret; 279| 403| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 806| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 1.03k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.03k| return r.error == simdutf::SUCCESS; 277| 1.03k| }); _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 403| bool count_the_input(FromSpan src) const { 290| 403| const auto implementations = get_supported_implementations(); 291| 403| std::vector results; 292| 403| results.reserve(implementations.size()); 293| | 294| 1.20k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.20k, False: 403] ------------------ 295| 1.20k| std::size_t ret; 296| 1.20k| if constexpr (From == UtfEncodings::UTF16BE) { 297| 1.20k| 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| 1.20k| results.push_back(ret); 304| 1.20k| } 305| 403| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 403| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 403] ------------------ 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| 403| return true; 321| 403| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 806| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 403| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 403| length_result ret{}; 344| | 345| 403| const auto implementations = get_supported_implementations(); 346| 403| std::vector results; 347| 403| results.reserve(implementations.size()); 348| | 349| 1.20k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.20k, False: 403] ------------------ 350| 1.20k| const auto len = invoke_lengthcalc(impl, src); 351| 1.20k| results.push_back(len); 352| 1.20k| ret.length.push_back(len); 353| 1.20k| } 354| | 355| 403| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 403| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 403] ------------------ 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| 403| } else { 375| 403| ret.implementations_agree = true; 376| 403| } 377| 403| return ret; 378| 403| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.20k| FromSpan src) const { 331| 1.20k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.20k| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 806| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 384| const bool inputisvalid) const { 383| 384| conversion_result ret{}; 384| | 385| 384| const auto implementations = get_supported_implementations(); 386| | 387| 384| std::vector> results; 388| 384| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 384| std::vector> outputbuffers; 393| 384| outputbuffers.reserve(implementations.size()); 394| 1.53k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.15k, False: 384] ------------------ 395| 1.15k| auto impl = implementations[i]; 396| 1.15k| const ToType canary1{42}; 397| 1.15k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.15k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.15k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.15k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.15k| if constexpr (std::is_same_v) { 403| 1.15k| return r != 0; 404| 1.15k| } else { 405| 1.15k| return r.error == simdutf::error_code::SUCCESS; 406| 1.15k| } 407| 1.15k| }(implret1); 408| 1.15k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.15k| 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| 1.15k| const ToType canary2{25}; 414| 1.15k| const auto outputbuffer_first_run = outputbuffer; 415| 1.15k| std::ranges::fill(outputbuffer, canary2); 416| 1.15k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.15k| src.size(), outputbuffer.data()); 418| | 419| 1.15k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.15k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.15k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 954, False: 198] | Branch (423:29): [True: 945, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 945| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 945| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 945] ------------------ 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| 945| } 440| 1.15k| } 441| 1.15k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 945, False: 207] ------------------ 442| 1.15k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 384| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 66, False: 318] ------------------ 447| 198| for (auto& e : results) { ------------------ | Branch (447:20): [True: 198, False: 66] ------------------ 448| 198| e.outputhash.clear(); 449| 198| } 450| 66| } 451| | 452| 384| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 384| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 384] ------------------ 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| 384| } else { 474| 384| ret.implementations_agree = true; 475| 384| } 476| 384| return ret; 477| 384| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.15k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.15k| if constexpr (std::is_same_v) { 403| 1.15k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.15k| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 768| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_2clENSt3__14spanIKcLm18446744073709551615EEE: 554| 61| +[](std::span chardata) { \ 555| 61| const auto c = \ 556| 61| Conversion{ \ 559| 61| &I::lenfunc, &I::conversionfunc, \ 560| 61| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 61|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 61| ::std::void_t(); \ | | 1180| 61| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 61| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 61| constexpr auto _size = _name.size(); \ | | 1183| 61| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 61| return _nameof; }() ------------------ 561| 61| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 61|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 61| ::std::void_t(); \ | | 1180| 61| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 61| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 61| constexpr auto _size = _name.size(); \ | | 1183| 61| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 61| return _nameof; }() ------------------ 562| 61| c.fuzz(chardata); \ 563| 61| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 197| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 197| const FromSpan from{reinterpret_cast(chardata.data()), 178| 197| chardata.size() / sizeof(FromType)}; 179| | 180| 197| static const bool do_print_testcase = 181| 197| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 197| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 197] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 197| do { 189| | // step 0 - is the input valid? 190| 197| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 197| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 197] | 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| 197| From == UtfEncodings::UTF8) { 198| 197| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 197] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 197| } 201| | 202| | // step 2 - what is the required size of the output? 203| 197| const auto [output_length, length_agree] = 204| 197| calculate_length(from, inputisvalid); 205| 197| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 197] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 197| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 90, False: 107] | Branch (208:28): [True: 5, False: 85] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 5| return; 211| 5| } 212| | 213| | // step 3 - run the conversion 214| 192| const auto [written, outputs_agree] = 215| 192| do_conversion(from, output_length, inputisvalid); 216| 192| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 192] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 192| return; 221| 192| } 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| 197| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 197| validation_result verify_valid_input(FromSpan src) const { 231| 197| validation_result ret{}; 232| | 233| 197| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 197| const auto implementations = get_supported_implementations(); 235| 197| std::vector results; 236| 197| results.reserve(implementations.size()); 237| | 238| 591| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 591, False: 197] ------------------ 239| 591| results.push_back( 240| 591| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 591| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 591| const bool validation2 = 245| 591| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 591| src.data(), src.size()); 247| 591| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 591] ------------------ 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| 591| } 258| | 259| 197| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 197| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 197] ------------------ 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| 197| } else { 273| 197| ret.implementations_agree = true; 274| 197| } 275| 197| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 197| return r.error == simdutf::SUCCESS; 277| 197| }); 278| 197| return ret; 279| 197| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 394| 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| 411| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 411| return r.error == simdutf::SUCCESS; 277| 411| }); _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 197| bool count_the_input(FromSpan src) const { 290| 197| const auto implementations = get_supported_implementations(); 291| 197| std::vector results; 292| 197| results.reserve(implementations.size()); 293| | 294| 591| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 591, False: 197] ------------------ 295| 591| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 591| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 591| 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| 591| results.push_back(ret); 304| 591| } 305| 197| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 197| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 197] ------------------ 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| 197| return true; 321| 197| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 394| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 197| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 197| length_result ret{}; 344| | 345| 197| const auto implementations = get_supported_implementations(); 346| 197| std::vector results; 347| 197| results.reserve(implementations.size()); 348| | 349| 591| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 591, False: 197] ------------------ 350| 591| const auto len = invoke_lengthcalc(impl, src); 351| 591| results.push_back(len); 352| 591| ret.length.push_back(len); 353| 591| } 354| | 355| 197| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 197| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 197] ------------------ 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| 197| } else { 375| 197| ret.implementations_agree = true; 376| 197| } 377| 197| return ret; 378| 197| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 591| FromSpan src) const { 331| 591| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 591| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 394| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 192| const bool inputisvalid) const { 383| 192| conversion_result ret{}; 384| | 385| 192| const auto implementations = get_supported_implementations(); 386| | 387| 192| std::vector> results; 388| 192| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 192| std::vector> outputbuffers; 393| 192| outputbuffers.reserve(implementations.size()); 394| 768| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 576, False: 192] ------------------ 395| 576| auto impl = implementations[i]; 396| 576| const ToType canary1{42}; 397| 576| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 576| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 576| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 576| const auto success = [](const ConversionResult& r) -> bool { 402| 576| if constexpr (std::is_same_v) { 403| 576| return r != 0; 404| 576| } else { 405| 576| return r.error == simdutf::error_code::SUCCESS; 406| 576| } 407| 576| }(implret1); 408| 576| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 576| 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| 576| const ToType canary2{25}; 414| 576| const auto outputbuffer_first_run = outputbuffer; 415| 576| std::ranges::fill(outputbuffer, canary2); 416| 576| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 576| src.size(), outputbuffer.data()); 418| | 419| 576| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 576] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 576| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 321, False: 255] | Branch (423:29): [True: 309, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 309| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 309| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 309] ------------------ 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| 309| } 440| 576| } 441| 576| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 309, False: 267] ------------------ 442| 576| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 192| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 85, False: 107] ------------------ 447| 255| for (auto& e : results) { ------------------ | Branch (447:20): [True: 255, False: 85] ------------------ 448| 255| e.outputhash.clear(); 449| 255| } 450| 85| } 451| | 452| 192| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 192| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 192] ------------------ 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| 192| } else { 474| 192| ret.implementations_agree = true; 475| 192| } 476| 192| return ret; 477| 192| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 576| const auto success = [](const ConversionResult& r) -> bool { 402| 576| if constexpr (std::is_same_v) { 403| 576| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 576| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 384| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_3clENSt3__14spanIKcLm18446744073709551615EEE: 554| 219| +[](std::span chardata) { \ 555| 219| const auto c = \ 556| 219| Conversion{ \ 559| 219| &I::lenfunc, &I::conversionfunc, \ 560| 219| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 219|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 219| ::std::void_t(); \ | | 1180| 219| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 219| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 219| constexpr auto _size = _name.size(); \ | | 1183| 219| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 219| return _nameof; }() ------------------ 561| 219| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 219|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 219| ::std::void_t(); \ | | 1180| 219| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 219| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 219| constexpr auto _size = _name.size(); \ | | 1183| 219| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 219| return _nameof; }() ------------------ 562| 219| c.fuzz(chardata); \ 563| 219| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 412| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 412| const FromSpan from{reinterpret_cast(chardata.data()), 178| 412| chardata.size() / sizeof(FromType)}; 179| | 180| 412| static const bool do_print_testcase = 181| 412| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 412| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 412] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 412| do { 189| | // step 0 - is the input valid? 190| 412| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 412| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 412] | 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| 412| From == UtfEncodings::UTF8) { 198| 412| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 412] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 412| } 201| | 202| | // step 2 - what is the required size of the output? 203| 412| const auto [output_length, length_agree] = 204| 412| calculate_length(from, inputisvalid); 205| 412| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 412] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 412| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 91, False: 321] | Branch (208:28): [True: 11, False: 80] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 11| return; 211| 11| } 212| | 213| | // step 3 - run the conversion 214| 401| const auto [written, outputs_agree] = 215| 401| do_conversion(from, output_length, inputisvalid); 216| 401| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 401] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 401| return; 221| 401| } 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| 412| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 412| validation_result verify_valid_input(FromSpan src) const { 231| 412| validation_result ret{}; 232| | 233| 412| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 412| const auto implementations = get_supported_implementations(); 235| 412| std::vector results; 236| 412| results.reserve(implementations.size()); 237| | 238| 1.23k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.23k, False: 412] ------------------ 239| 1.23k| results.push_back( 240| 1.23k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.23k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.23k| const bool validation2 = 245| 1.23k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.23k| src.data(), src.size()); 247| 1.23k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.23k] ------------------ 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| 1.23k| } 258| | 259| 412| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 412| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 412] ------------------ 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| 412| } else { 273| 412| ret.implementations_agree = true; 274| 412| } 275| 412| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 412| return r.error == simdutf::SUCCESS; 277| 412| }); 278| 412| return ret; 279| 412| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 824| 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| 1.05k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.05k| return r.error == simdutf::SUCCESS; 277| 1.05k| }); _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 412| bool count_the_input(FromSpan src) const { 290| 412| const auto implementations = get_supported_implementations(); 291| 412| std::vector results; 292| 412| results.reserve(implementations.size()); 293| | 294| 1.23k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.23k, False: 412] ------------------ 295| 1.23k| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 1.23k| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 1.23k| 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| 1.23k| results.push_back(ret); 304| 1.23k| } 305| 412| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 412| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 412] ------------------ 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| 412| return true; 321| 412| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 824| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 412| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 412| length_result ret{}; 344| | 345| 412| const auto implementations = get_supported_implementations(); 346| 412| std::vector results; 347| 412| results.reserve(implementations.size()); 348| | 349| 1.23k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.23k, False: 412] ------------------ 350| 1.23k| const auto len = invoke_lengthcalc(impl, src); 351| 1.23k| results.push_back(len); 352| 1.23k| ret.length.push_back(len); 353| 1.23k| } 354| | 355| 412| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 412| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 412] ------------------ 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| 412| } else { 375| 412| ret.implementations_agree = true; 376| 412| } 377| 412| return ret; 378| 412| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.23k| FromSpan src) const { 331| 1.23k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.23k| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 824| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 401| const bool inputisvalid) const { 383| 401| conversion_result ret{}; 384| | 385| 401| const auto implementations = get_supported_implementations(); 386| | 387| 401| std::vector> results; 388| 401| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 401| std::vector> outputbuffers; 393| 401| outputbuffers.reserve(implementations.size()); 394| 1.60k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.20k, False: 401] ------------------ 395| 1.20k| auto impl = implementations[i]; 396| 1.20k| const ToType canary1{42}; 397| 1.20k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.20k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.20k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.20k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.20k| if constexpr (std::is_same_v) { 403| 1.20k| return r != 0; 404| 1.20k| } else { 405| 1.20k| return r.error == simdutf::error_code::SUCCESS; 406| 1.20k| } 407| 1.20k| }(implret1); 408| 1.20k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.20k| 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| 1.20k| const ToType canary2{25}; 414| 1.20k| const auto outputbuffer_first_run = outputbuffer; 415| 1.20k| std::ranges::fill(outputbuffer, canary2); 416| 1.20k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.20k| src.size(), outputbuffer.data()); 418| | 419| 1.20k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.20k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.20k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 963, False: 240] | Branch (423:29): [True: 951, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 951| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 951| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 951] ------------------ 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| 951| } 440| 1.20k| } 441| 1.20k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 951, False: 252] ------------------ 442| 1.20k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 401| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 80, False: 321] ------------------ 447| 240| for (auto& e : results) { ------------------ | Branch (447:20): [True: 240, False: 80] ------------------ 448| 240| e.outputhash.clear(); 449| 240| } 450| 80| } 451| | 452| 401| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 401| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 401] ------------------ 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| 401| } else { 474| 401| ret.implementations_agree = true; 475| 401| } 476| 401| return ret; 477| 401| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.20k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.20k| if constexpr (std::is_same_v) { 403| 1.20k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.20k| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 802| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_4clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 268| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 268| const FromSpan from{reinterpret_cast(chardata.data()), 178| 268| chardata.size() / sizeof(FromType)}; 179| | 180| 268| static const bool do_print_testcase = 181| 268| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 268| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 268] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 268| do { 189| | // step 0 - is the input valid? 190| 268| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 268| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 268] | 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| 268| const auto [output_length, length_agree] = 204| 268| calculate_length(from, inputisvalid); 205| 268| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 268] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 268| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 168, False: 100] | Branch (208:28): [True: 4, False: 164] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 4| return; 211| 4| } 212| | 213| | // step 3 - run the conversion 214| 264| const auto [written, outputs_agree] = 215| 264| do_conversion(from, output_length, inputisvalid); 216| 264| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 264] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 264| return; 221| 264| } 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| 268| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 268| validation_result verify_valid_input(FromSpan src) const { 231| 268| validation_result ret{}; 232| | 233| 268| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 268| const auto implementations = get_supported_implementations(); 235| 268| std::vector results; 236| 268| results.reserve(implementations.size()); 237| | 238| 804| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 804, False: 268] ------------------ 239| 804| results.push_back( 240| 804| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 804| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 804| const bool validation2 = 245| 804| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 804| src.data(), src.size()); 247| 804| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 804] ------------------ 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| 804| } 258| | 259| 268| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 268| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 268] ------------------ 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| 268| } else { 273| 268| ret.implementations_agree = true; 274| 268| } 275| 268| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 268| return r.error == simdutf::SUCCESS; 277| 268| }); 278| 268| return ret; 279| 268| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 536| 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| 468| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 468| return r.error == simdutf::SUCCESS; 277| 468| }); _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 268| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 268| length_result ret{}; 344| | 345| 268| const auto implementations = get_supported_implementations(); 346| 268| std::vector results; 347| 268| results.reserve(implementations.size()); 348| | 349| 804| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 804, False: 268] ------------------ 350| 804| const auto len = invoke_lengthcalc(impl, src); 351| 804| results.push_back(len); 352| 804| ret.length.push_back(len); 353| 804| } 354| | 355| 268| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 268| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 268] ------------------ 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| 268| } else { 375| 268| ret.implementations_agree = true; 376| 268| } 377| 268| return ret; 378| 268| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 804| FromSpan src) const { 331| 804| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 804| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 536| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 264| const bool inputisvalid) const { 383| 264| conversion_result ret{}; 384| | 385| 264| const auto implementations = get_supported_implementations(); 386| | 387| 264| std::vector> results; 388| 264| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 264| std::vector> outputbuffers; 393| 264| outputbuffers.reserve(implementations.size()); 394| 1.05k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 792, False: 264] ------------------ 395| 792| auto impl = implementations[i]; 396| 792| const ToType canary1{42}; 397| 792| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 792| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 792| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 792| const auto success = [](const ConversionResult& r) -> bool { 402| 792| if constexpr (std::is_same_v) { 403| 792| return r != 0; 404| 792| } else { 405| 792| return r.error == simdutf::error_code::SUCCESS; 406| 792| } 407| 792| }(implret1); 408| 792| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 792| 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| 792| const ToType canary2{25}; 414| 792| const auto outputbuffer_first_run = outputbuffer; 415| 792| std::ranges::fill(outputbuffer, canary2); 416| 792| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 792| src.size(), outputbuffer.data()); 418| | 419| 792| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 792] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 792| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 300, False: 492] | Branch (423:29): [True: 291, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 291| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 291| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 291] ------------------ 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| 291| } 440| 792| } 441| 792| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 291, False: 501] ------------------ 442| 792| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 264| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 164, False: 100] ------------------ 447| 492| for (auto& e : results) { ------------------ | Branch (447:20): [True: 492, False: 164] ------------------ 448| 492| e.outputhash.clear(); 449| 492| } 450| 164| } 451| | 452| 264| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 264| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 264] ------------------ 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| 264| } else { 474| 264| ret.implementations_agree = true; 475| 264| } 476| 264| return ret; 477| 264| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 792| const auto success = [](const ConversionResult& r) -> bool { 402| 792| if constexpr (std::is_same_v) { 403| 792| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 792| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 528| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_5clENSt3__14spanIKcLm18446744073709551615EEE: 554| 150| +[](std::span chardata) { \ 555| 150| const auto c = \ 556| 150| Conversion{ \ 559| 150| &I::lenfunc, &I::conversionfunc, \ 560| 150| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 150|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 150| ::std::void_t(); \ | | 1180| 150| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 150| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 150| constexpr auto _size = _name.size(); \ | | 1183| 150| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 150| return _nameof; }() ------------------ 561| 150| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 150|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 150| ::std::void_t(); \ | | 1180| 150| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 150| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 150| constexpr auto _size = _name.size(); \ | | 1183| 150| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 150| return _nameof; }() ------------------ 562| 150| c.fuzz(chardata); \ 563| 150| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 366| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 366| const FromSpan from{reinterpret_cast(chardata.data()), 178| 366| chardata.size() / sizeof(FromType)}; 179| | 180| 366| static const bool do_print_testcase = 181| 366| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 366| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 366] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 366| do { 189| | // step 0 - is the input valid? 190| 366| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 366| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 366] | 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| 366| const auto [output_length, length_agree] = 204| 366| calculate_length(from, inputisvalid); 205| 366| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 366] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 366| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 146, False: 220] | Branch (208:28): [True: 5, False: 141] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 5| return; 211| 5| } 212| | 213| | // step 3 - run the conversion 214| 361| const auto [written, outputs_agree] = 215| 361| do_conversion(from, output_length, inputisvalid); 216| 361| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 361] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 361| return; 221| 361| } 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| 366| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 366| validation_result verify_valid_input(FromSpan src) const { 231| 366| validation_result ret{}; 232| | 233| 366| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 366| const auto implementations = get_supported_implementations(); 235| 366| std::vector results; 236| 366| results.reserve(implementations.size()); 237| | 238| 1.09k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.09k, False: 366] ------------------ 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| 366| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 366| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 366] ------------------ 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| 366| } else { 273| 366| ret.implementations_agree = true; 274| 366| } 275| 366| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 366| return r.error == simdutf::SUCCESS; 277| 366| }); 278| 366| return ret; 279| 366| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 732| 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| 806| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 806| return r.error == simdutf::SUCCESS; 277| 806| }); _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 366| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 366| length_result ret{}; 344| | 345| 366| const auto implementations = get_supported_implementations(); 346| 366| std::vector results; 347| 366| results.reserve(implementations.size()); 348| | 349| 1.09k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.09k, False: 366] ------------------ 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| 366| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 366| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 366] ------------------ 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| 366| } else { 375| 366| ret.implementations_agree = true; 376| 366| } 377| 366| return ret; 378| 366| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.09k| FromSpan src) const { 331| 1.09k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.09k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 732| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 361| const bool inputisvalid) const { 383| 361| conversion_result ret{}; 384| | 385| 361| const auto implementations = get_supported_implementations(); 386| | 387| 361| std::vector> results; 388| 361| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 361| std::vector> outputbuffers; 393| 361| outputbuffers.reserve(implementations.size()); 394| 1.44k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.08k, False: 361] ------------------ 395| 1.08k| auto impl = implementations[i]; 396| 1.08k| const ToType canary1{42}; 397| 1.08k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.08k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.08k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.08k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.08k| if constexpr (std::is_same_v) { 403| 1.08k| return r != 0; 404| 1.08k| } else { 405| 1.08k| return r.error == simdutf::error_code::SUCCESS; 406| 1.08k| } 407| 1.08k| }(implret1); 408| 1.08k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.08k| 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| 1.08k| const ToType canary2{25}; 414| 1.08k| const auto outputbuffer_first_run = outputbuffer; 415| 1.08k| std::ranges::fill(outputbuffer, canary2); 416| 1.08k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.08k| src.size(), outputbuffer.data()); 418| | 419| 1.08k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.08k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.08k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 660, False: 423] | Branch (423:29): [True: 648, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 648| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 648| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 648] ------------------ 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| 648| } 440| 1.08k| } 441| 1.08k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 648, False: 435] ------------------ 442| 1.08k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 361| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 141, False: 220] ------------------ 447| 423| for (auto& e : results) { ------------------ | Branch (447:20): [True: 423, False: 141] ------------------ 448| 423| e.outputhash.clear(); 449| 423| } 450| 141| } 451| | 452| 361| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 361| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 361] ------------------ 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| 361| } else { 474| 361| ret.implementations_agree = true; 475| 361| } 476| 361| return ret; 477| 361| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.08k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.08k| if constexpr (std::is_same_v) { 403| 1.08k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.08k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 722| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_6clENSt3__14spanIKcLm18446744073709551615EEE: 554| 102| +[](std::span chardata) { \ 555| 102| const auto c = \ 556| 102| Conversion{ \ 559| 102| &I::lenfunc, &I::conversionfunc, \ 560| 102| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 102|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 102| ::std::void_t(); \ | | 1180| 102| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 102| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 102| constexpr auto _size = _name.size(); \ | | 1183| 102| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 102| return _nameof; }() ------------------ 561| 102| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 102|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 102| ::std::void_t(); \ | | 1180| 102| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 102| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 102| constexpr auto _size = _name.size(); \ | | 1183| 102| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 102| return _nameof; }() ------------------ 562| 102| c.fuzz(chardata); \ 563| 102| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 404| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 404| const FromSpan from{reinterpret_cast(chardata.data()), 178| 404| chardata.size() / sizeof(FromType)}; 179| | 180| 404| static const bool do_print_testcase = 181| 404| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 404| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 404] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 404| do { 189| | // step 0 - is the input valid? 190| 404| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 404| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 404] | 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| 404| const auto [output_length, length_agree] = 204| 404| calculate_length(from, inputisvalid); 205| 404| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 404] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 404| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 256, False: 148] | Branch (208:28): [True: 6, False: 250] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 6| return; 211| 6| } 212| | 213| | // step 3 - run the conversion 214| 398| const auto [written, outputs_agree] = 215| 398| do_conversion(from, output_length, inputisvalid); 216| 398| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 398] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 398| return; 221| 398| } 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| 404| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 404| validation_result verify_valid_input(FromSpan src) const { 231| 404| validation_result ret{}; 232| | 233| 404| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 404| const auto implementations = get_supported_implementations(); 235| 404| std::vector results; 236| 404| results.reserve(implementations.size()); 237| | 238| 1.21k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.21k, False: 404] ------------------ 239| 1.21k| results.push_back( 240| 1.21k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.21k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.21k| const bool validation2 = 245| 1.21k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.21k| src.data(), src.size()); 247| 1.21k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.21k] ------------------ 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| 1.21k| } 258| | 259| 404| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 404| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 404] ------------------ 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| 404| } else { 273| 404| ret.implementations_agree = true; 274| 404| } 275| 404| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 404| return r.error == simdutf::SUCCESS; 277| 404| }); 278| 404| return ret; 279| 404| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 808| 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| 700| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 700| return r.error == simdutf::SUCCESS; 277| 700| }); _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 404| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 404| length_result ret{}; 344| | 345| 404| const auto implementations = get_supported_implementations(); 346| 404| std::vector results; 347| 404| results.reserve(implementations.size()); 348| | 349| 1.21k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.21k, False: 404] ------------------ 350| 1.21k| const auto len = invoke_lengthcalc(impl, src); 351| 1.21k| results.push_back(len); 352| 1.21k| ret.length.push_back(len); 353| 1.21k| } 354| | 355| 404| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 404| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 404] ------------------ 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| 404| } else { 375| 404| ret.implementations_agree = true; 376| 404| } 377| 404| return ret; 378| 404| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.21k| FromSpan src) const { 331| 1.21k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.21k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 808| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 398| const bool inputisvalid) const { 383| 398| conversion_result ret{}; 384| | 385| 398| const auto implementations = get_supported_implementations(); 386| | 387| 398| std::vector> results; 388| 398| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 398| std::vector> outputbuffers; 393| 398| outputbuffers.reserve(implementations.size()); 394| 1.59k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.19k, False: 398] ------------------ 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: 444, False: 750] | Branch (423:29): [True: 435, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 435| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 435| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 435] ------------------ 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| 435| } 440| 1.19k| } 441| 1.19k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 435, False: 759] ------------------ 442| 1.19k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 398| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 250, False: 148] ------------------ 447| 750| for (auto& e : results) { ------------------ | Branch (447:20): [True: 750, False: 250] ------------------ 448| 750| e.outputhash.clear(); 449| 750| } 450| 250| } 451| | 452| 398| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 398| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 398] ------------------ 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| 398| } else { 474| 398| ret.implementations_agree = true; 475| 398| } 476| 398| return ret; 477| 398| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.19k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.19k| if constexpr (std::is_same_v) { 403| 1.19k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.19k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 796| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_7clENSt3__14spanIKcLm18446744073709551615EEE: 554| 228| +[](std::span chardata) { \ 555| 228| const auto c = \ 556| 228| Conversion{ \ 559| 228| &I::lenfunc, &I::conversionfunc, \ 560| 228| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 228|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 228| ::std::void_t(); \ | | 1180| 228| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 228| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 228| constexpr auto _size = _name.size(); \ | | 1183| 228| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 228| return _nameof; }() ------------------ 561| 228| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 228|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 228| ::std::void_t(); \ | | 1180| 228| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 228| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 228| constexpr auto _size = _name.size(); \ | | 1183| 228| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 228| return _nameof; }() ------------------ 562| 228| c.fuzz(chardata); \ 563| 228| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 584| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 584| const FromSpan from{reinterpret_cast(chardata.data()), 178| 584| chardata.size() / sizeof(FromType)}; 179| | 180| 584| static const bool do_print_testcase = 181| 584| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 584| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 584] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 584| do { 189| | // step 0 - is the input valid? 190| 584| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 584| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 584] | 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| 584| From == UtfEncodings::UTF8) { 198| 584| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 584] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 584| } 201| | 202| | // step 2 - what is the required size of the output? 203| 584| const auto [output_length, length_agree] = 204| 584| calculate_length(from, inputisvalid); 205| 584| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 584] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 584| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 268, False: 316] | Branch (208:28): [True: 9, False: 259] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 9| return; 211| 9| } 212| | 213| | // step 3 - run the conversion 214| 575| const auto [written, outputs_agree] = 215| 575| do_conversion(from, output_length, inputisvalid); 216| 575| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 575] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 575| return; 221| 575| } 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| 584| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 584| validation_result verify_valid_input(FromSpan src) const { 231| 584| validation_result ret{}; 232| | 233| 584| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 584| const auto implementations = get_supported_implementations(); 235| 584| std::vector results; 236| 584| results.reserve(implementations.size()); 237| | 238| 1.75k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.75k, False: 584] ------------------ 239| 1.75k| results.push_back( 240| 1.75k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.75k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.75k| const bool validation2 = 245| 1.75k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.75k| src.data(), src.size()); 247| 1.75k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.75k] ------------------ 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| 1.75k| } 258| | 259| 584| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 584| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 584] ------------------ 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| 584| } else { 273| 584| ret.implementations_agree = true; 274| 584| } 275| 584| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 584| return r.error == simdutf::SUCCESS; 277| 584| }); 278| 584| return ret; 279| 584| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 1.16k| 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| 1.21k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.21k| return r.error == simdutf::SUCCESS; 277| 1.21k| }); _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 584| bool count_the_input(FromSpan src) const { 290| 584| const auto implementations = get_supported_implementations(); 291| 584| std::vector results; 292| 584| results.reserve(implementations.size()); 293| | 294| 1.75k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.75k, False: 584] ------------------ 295| 1.75k| 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| 1.75k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.75k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.75k| } 303| 1.75k| results.push_back(ret); 304| 1.75k| } 305| 584| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 584| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 584] ------------------ 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| 584| return true; 321| 584| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 1.16k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 584| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 584| length_result ret{}; 344| | 345| 584| const auto implementations = get_supported_implementations(); 346| 584| std::vector results; 347| 584| results.reserve(implementations.size()); 348| | 349| 1.75k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.75k, False: 584] ------------------ 350| 1.75k| const auto len = invoke_lengthcalc(impl, src); 351| 1.75k| results.push_back(len); 352| 1.75k| ret.length.push_back(len); 353| 1.75k| } 354| | 355| 584| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 584| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 584] ------------------ 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| 584| } else { 375| 584| ret.implementations_agree = true; 376| 584| } 377| 584| return ret; 378| 584| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.75k| FromSpan src) const { 331| 1.75k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.75k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 1.16k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 575| const bool inputisvalid) const { 383| 575| conversion_result ret{}; 384| | 385| 575| const auto implementations = get_supported_implementations(); 386| | 387| 575| std::vector> results; 388| 575| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 575| std::vector> outputbuffers; 393| 575| outputbuffers.reserve(implementations.size()); 394| 2.30k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.72k, False: 575] ------------------ 395| 1.72k| auto impl = implementations[i]; 396| 1.72k| const ToType canary1{42}; 397| 1.72k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.72k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.72k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.72k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.72k| if constexpr (std::is_same_v) { 403| 1.72k| return r != 0; 404| 1.72k| } else { 405| 1.72k| return r.error == simdutf::error_code::SUCCESS; 406| 1.72k| } 407| 1.72k| }(implret1); 408| 1.72k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.72k| 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| 1.72k| const ToType canary2{25}; 414| 1.72k| const auto outputbuffer_first_run = outputbuffer; 415| 1.72k| std::ranges::fill(outputbuffer, canary2); 416| 1.72k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.72k| src.size(), outputbuffer.data()); 418| | 419| 1.72k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.72k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.72k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 948, False: 777] | Branch (423:29): [True: 939, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 939| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 939| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 939] ------------------ 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| 939| } 440| 1.72k| } 441| 1.72k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 939, False: 786] ------------------ 442| 1.72k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 575| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 259, False: 316] ------------------ 447| 777| for (auto& e : results) { ------------------ | Branch (447:20): [True: 777, False: 259] ------------------ 448| 777| e.outputhash.clear(); 449| 777| } 450| 259| } 451| | 452| 575| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 575| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 575] ------------------ 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| 575| } else { 474| 575| ret.implementations_agree = true; 475| 575| } 476| 575| return ret; 477| 575| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.72k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.72k| if constexpr (std::is_same_v) { 403| 1.72k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.72k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 1.15k| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_8clENSt3__14spanIKcLm18446744073709551615EEE: 554| 197| +[](std::span chardata) { \ 555| 197| const auto c = \ 556| 197| Conversion{ \ 559| 197| &I::lenfunc, &I::conversionfunc, \ 560| 197| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 197|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 197| ::std::void_t(); \ | | 1180| 197| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 197| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 197| constexpr auto _size = _name.size(); \ | | 1183| 197| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 197| return _nameof; }() ------------------ 561| 197| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 197|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 197| ::std::void_t(); \ | | 1180| 197| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 197| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 197| constexpr auto _size = _name.size(); \ | | 1183| 197| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 197| return _nameof; }() ------------------ 562| 197| c.fuzz(chardata); \ 563| 197| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 527| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 527| const FromSpan from{reinterpret_cast(chardata.data()), 178| 527| chardata.size() / sizeof(FromType)}; 179| | 180| 527| static const bool do_print_testcase = 181| 527| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 527| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 527] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 527| do { 189| | // step 0 - is the input valid? 190| 527| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 527| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 527] | 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| 527| From == UtfEncodings::UTF8) { 198| 527| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 527] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 527| } 201| | 202| | // step 2 - what is the required size of the output? 203| 527| const auto [output_length, length_agree] = 204| 527| calculate_length(from, inputisvalid); 205| 527| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 527] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 527| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 230, False: 297] | Branch (208:28): [True: 13, False: 217] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 13| return; 211| 13| } 212| | 213| | // step 3 - run the conversion 214| 514| const auto [written, outputs_agree] = 215| 514| do_conversion(from, output_length, inputisvalid); 216| 514| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 514] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 514| return; 221| 514| } 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| 527| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 527| validation_result verify_valid_input(FromSpan src) const { 231| 527| validation_result ret{}; 232| | 233| 527| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 527| const auto implementations = get_supported_implementations(); 235| 527| std::vector results; 236| 527| results.reserve(implementations.size()); 237| | 238| 1.58k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.58k, False: 527] ------------------ 239| 1.58k| results.push_back( 240| 1.58k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.58k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.58k| const bool validation2 = 245| 1.58k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.58k| src.data(), src.size()); 247| 1.58k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.58k] ------------------ 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| 1.58k| } 258| | 259| 527| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 527| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 527] ------------------ 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| 527| } else { 273| 527| ret.implementations_agree = true; 274| 527| } 275| 527| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 527| return r.error == simdutf::SUCCESS; 277| 527| }); 278| 527| return ret; 279| 527| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 1.05k| 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| 1.12k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.12k| return r.error == simdutf::SUCCESS; 277| 1.12k| }); _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 527| bool count_the_input(FromSpan src) const { 290| 527| const auto implementations = get_supported_implementations(); 291| 527| std::vector results; 292| 527| results.reserve(implementations.size()); 293| | 294| 1.58k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.58k, False: 527] ------------------ 295| 1.58k| 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| 1.58k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.58k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.58k| } 303| 1.58k| results.push_back(ret); 304| 1.58k| } 305| 527| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 527| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 527] ------------------ 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| 527| return true; 321| 527| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 1.05k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 527| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 527| length_result ret{}; 344| | 345| 527| const auto implementations = get_supported_implementations(); 346| 527| std::vector results; 347| 527| results.reserve(implementations.size()); 348| | 349| 1.58k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.58k, False: 527] ------------------ 350| 1.58k| const auto len = invoke_lengthcalc(impl, src); 351| 1.58k| results.push_back(len); 352| 1.58k| ret.length.push_back(len); 353| 1.58k| } 354| | 355| 527| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 527| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 527] ------------------ 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| 527| } else { 375| 527| ret.implementations_agree = true; 376| 527| } 377| 527| return ret; 378| 527| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.58k| FromSpan src) const { 331| 1.58k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.58k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 1.05k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 514| const bool inputisvalid) const { 383| 514| conversion_result ret{}; 384| | 385| 514| const auto implementations = get_supported_implementations(); 386| | 387| 514| std::vector> results; 388| 514| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 514| std::vector> outputbuffers; 393| 514| outputbuffers.reserve(implementations.size()); 394| 2.05k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.54k, False: 514] ------------------ 395| 1.54k| auto impl = implementations[i]; 396| 1.54k| const ToType canary1{42}; 397| 1.54k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.54k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.54k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.54k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.54k| if constexpr (std::is_same_v) { 403| 1.54k| return r != 0; 404| 1.54k| } else { 405| 1.54k| return r.error == simdutf::error_code::SUCCESS; 406| 1.54k| } 407| 1.54k| }(implret1); 408| 1.54k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.54k| 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| 1.54k| const ToType canary2{25}; 414| 1.54k| const auto outputbuffer_first_run = outputbuffer; 415| 1.54k| std::ranges::fill(outputbuffer, canary2); 416| 1.54k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.54k| src.size(), outputbuffer.data()); 418| | 419| 1.54k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.54k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.54k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 891, False: 651] | Branch (423:29): [True: 882, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 882| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 882| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 882] ------------------ 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| 882| } 440| 1.54k| } 441| 1.54k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 882, False: 660] ------------------ 442| 1.54k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 514| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 217, False: 297] ------------------ 447| 651| for (auto& e : results) { ------------------ | Branch (447:20): [True: 651, False: 217] ------------------ 448| 651| e.outputhash.clear(); 449| 651| } 450| 217| } 451| | 452| 514| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 514| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 514] ------------------ 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| 514| } else { 474| 514| ret.implementations_agree = true; 475| 514| } 476| 514| return ret; 477| 514| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.54k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.54k| if constexpr (std::is_same_v) { 403| 1.54k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.54k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 1.02k| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_9clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 564| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 564| const FromSpan from{reinterpret_cast(chardata.data()), 178| 564| chardata.size() / sizeof(FromType)}; 179| | 180| 564| static const bool do_print_testcase = 181| 564| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 564| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 564] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 564| do { 189| | // step 0 - is the input valid? 190| 564| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 564| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 564] | 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| 564| From == UtfEncodings::UTF8) { 198| 564| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 564] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 564| } 201| | 202| | // step 2 - what is the required size of the output? 203| 564| const auto [output_length, length_agree] = 204| 564| calculate_length(from, inputisvalid); 205| 564| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 564] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 564| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 229, False: 335] | Branch (208:28): [True: 11, False: 218] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 11| return; 211| 11| } 212| | 213| | // step 3 - run the conversion 214| 553| const auto [written, outputs_agree] = 215| 553| do_conversion(from, output_length, inputisvalid); 216| 553| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 553] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 553| return; 221| 553| } 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| 564| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 564| validation_result verify_valid_input(FromSpan src) const { 231| 564| validation_result ret{}; 232| | 233| 564| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 564| const auto implementations = get_supported_implementations(); 235| 564| std::vector results; 236| 564| results.reserve(implementations.size()); 237| | 238| 1.69k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.69k, False: 564] ------------------ 239| 1.69k| results.push_back( 240| 1.69k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.69k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.69k| const bool validation2 = 245| 1.69k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.69k| src.data(), src.size()); 247| 1.69k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.69k] ------------------ 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| 1.69k| } 258| | 259| 564| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 564| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 564] ------------------ 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| 564| } else { 273| 564| ret.implementations_agree = true; 274| 564| } 275| 564| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 564| return r.error == simdutf::SUCCESS; 277| 564| }); 278| 564| return ret; 279| 564| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 1.12k| 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| 1.23k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.23k| return r.error == simdutf::SUCCESS; 277| 1.23k| }); _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 564| bool count_the_input(FromSpan src) const { 290| 564| const auto implementations = get_supported_implementations(); 291| 564| std::vector results; 292| 564| results.reserve(implementations.size()); 293| | 294| 1.69k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.69k, False: 564] ------------------ 295| 1.69k| 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| 1.69k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.69k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.69k| } 303| 1.69k| results.push_back(ret); 304| 1.69k| } 305| 564| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 564| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 564] ------------------ 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| 564| return true; 321| 564| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 1.12k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 564| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 564| length_result ret{}; 344| | 345| 564| const auto implementations = get_supported_implementations(); 346| 564| std::vector results; 347| 564| results.reserve(implementations.size()); 348| | 349| 1.69k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.69k, False: 564] ------------------ 350| 1.69k| const auto len = invoke_lengthcalc(impl, src); 351| 1.69k| results.push_back(len); 352| 1.69k| ret.length.push_back(len); 353| 1.69k| } 354| | 355| 564| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 564| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 564] ------------------ 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| 564| } else { 375| 564| ret.implementations_agree = true; 376| 564| } 377| 564| return ret; 378| 564| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.69k| FromSpan src) const { 331| 1.69k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.69k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 1.12k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 553| const bool inputisvalid) const { 383| 553| conversion_result ret{}; 384| | 385| 553| const auto implementations = get_supported_implementations(); 386| | 387| 553| std::vector> results; 388| 553| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 553| std::vector> outputbuffers; 393| 553| outputbuffers.reserve(implementations.size()); 394| 2.21k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.65k, False: 553] ------------------ 395| 1.65k| auto impl = implementations[i]; 396| 1.65k| const ToType canary1{42}; 397| 1.65k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.65k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.65k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.65k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.65k| if constexpr (std::is_same_v) { 403| 1.65k| return r != 0; 404| 1.65k| } else { 405| 1.65k| return r.error == simdutf::error_code::SUCCESS; 406| 1.65k| } 407| 1.65k| }(implret1); 408| 1.65k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.65k| 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| 1.65k| const ToType canary2{25}; 414| 1.65k| const auto outputbuffer_first_run = outputbuffer; 415| 1.65k| std::ranges::fill(outputbuffer, canary2); 416| 1.65k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.65k| src.size(), outputbuffer.data()); 418| | 419| 1.65k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.65k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.65k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 1.00k, False: 654] | Branch (423:29): [True: 999, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 999| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 999| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 999] ------------------ 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| 999| } 440| 1.65k| } 441| 1.65k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 999, False: 660] ------------------ 442| 1.65k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 553| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 218, False: 335] ------------------ 447| 654| for (auto& e : results) { ------------------ | Branch (447:20): [True: 654, False: 218] ------------------ 448| 654| e.outputhash.clear(); 449| 654| } 450| 218| } 451| | 452| 553| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 553| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 553] ------------------ 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| 553| } else { 474| 553| ret.implementations_agree = true; 475| 553| } 476| 553| return ret; 477| 553| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.65k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.65k| if constexpr (std::is_same_v) { 403| 1.65k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.65k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 1.10k| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_10clENSt3__14spanIKcLm18446744073709551615EEE: 554| 63| +[](std::span chardata) { \ 555| 63| const auto c = \ 556| 63| Conversion{ \ 559| 63| &I::lenfunc, &I::conversionfunc, \ 560| 63| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 63|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 63| ::std::void_t(); \ | | 1180| 63| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 63| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 63| constexpr auto _size = _name.size(); \ | | 1183| 63| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 63| return _nameof; }() ------------------ 561| 63| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 63|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 63| ::std::void_t(); \ | | 1180| 63| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 63| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 63| constexpr auto _size = _name.size(); \ | | 1183| 63| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 63| return _nameof; }() ------------------ 562| 63| c.fuzz(chardata); \ 563| 63| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 63| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 63| const FromSpan from{reinterpret_cast(chardata.data()), 178| 63| chardata.size() / sizeof(FromType)}; 179| | 180| 63| static const bool do_print_testcase = 181| 63| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 63| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 63] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 63| do { 189| | // step 0 - is the input valid? 190| 63| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 63| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 63] | 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| 63| From == UtfEncodings::UTF8) { 198| 63| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 63] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 63| } 201| | 202| | // step 2 - what is the required size of the output? 203| 63| const auto [output_length, length_agree] = 204| 63| calculate_length(from, inputisvalid); 205| 63| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 63] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 63| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 9, False: 54] | 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| 63| const auto [written, outputs_agree] = 215| 63| do_conversion(from, output_length, inputisvalid); 216| 63| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 63] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 63| return; 221| 63| } 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| 63| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 230| 63| validation_result verify_valid_input(FromSpan src) const { 231| 63| validation_result ret{}; 232| | 233| 63| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 63| const auto implementations = get_supported_implementations(); 235| 63| std::vector results; 236| 63| results.reserve(implementations.size()); 237| | 238| 189| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 189, False: 63] ------------------ 239| 189| results.push_back( 240| 189| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 189| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 189| const bool validation2 = 245| 189| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 189| src.data(), src.size()); 247| 189| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 189] ------------------ 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| 189| } 258| | 259| 63| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 63| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 63] ------------------ 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| 63| } else { 273| 63| ret.implementations_agree = true; 274| 63| } 275| 63| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 63| return r.error == simdutf::SUCCESS; 277| 63| }); 278| 63| return ret; 279| 63| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 126| 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| 171| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 171| return r.error == simdutf::SUCCESS; 277| 171| }); _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEE: 289| 63| bool count_the_input(FromSpan src) const { 290| 63| const auto implementations = get_supported_implementations(); 291| 63| std::vector results; 292| 63| results.reserve(implementations.size()); 293| | 294| 189| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 189, False: 63] ------------------ 295| 189| std::size_t ret; 296| 189| if constexpr (From == UtfEncodings::UTF16BE) { 297| 189| 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| 189| results.push_back(ret); 304| 189| } 305| 63| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 63| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 63] ------------------ 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| 63| return true; 321| 63| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 126| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 63| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 63| length_result ret{}; 344| | 345| 63| const auto implementations = get_supported_implementations(); 346| 63| std::vector results; 347| 63| results.reserve(implementations.size()); 348| | 349| 189| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 189, False: 63] ------------------ 350| 189| const auto len = invoke_lengthcalc(impl, src); 351| 189| results.push_back(len); 352| 189| ret.length.push_back(len); 353| 189| } 354| | 355| 63| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 63| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 63] ------------------ 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| 63| } else { 375| 63| ret.implementations_agree = true; 376| 63| } 377| 63| return ret; 378| 63| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 189| FromSpan src) const { 339| 189| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 189| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 126| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 63| const bool inputisvalid) const { 383| 63| conversion_result ret{}; 384| | 385| 63| const auto implementations = get_supported_implementations(); 386| | 387| 63| std::vector> results; 388| 63| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 63| std::vector> outputbuffers; 393| 63| outputbuffers.reserve(implementations.size()); 394| 252| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 189, False: 63] ------------------ 395| 189| auto impl = implementations[i]; 396| 189| const ToType canary1{42}; 397| 189| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 189| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 189| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 189| const auto success = [](const ConversionResult& r) -> bool { 402| 189| if constexpr (std::is_same_v) { 403| 189| return r != 0; 404| 189| } else { 405| 189| return r.error == simdutf::error_code::SUCCESS; 406| 189| } 407| 189| }(implret1); 408| 189| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 189| 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| 189| const ToType canary2{25}; 414| 189| const auto outputbuffer_first_run = outputbuffer; 415| 189| std::ranges::fill(outputbuffer, canary2); 416| 189| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 189| src.size(), outputbuffer.data()); 418| | 419| 189| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 189] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 189| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 162, False: 27] | Branch (423:29): [True: 93, False: 69] ------------------ 424| | // only care about the output if the input is valid 425| 93| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 93| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 93] ------------------ 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| 93| } 440| 189| } 441| 189| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 93, False: 96] ------------------ 442| 189| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 63| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 9, False: 54] ------------------ 447| 27| for (auto& e : results) { ------------------ | Branch (447:20): [True: 27, False: 9] ------------------ 448| 27| e.outputhash.clear(); 449| 27| } 450| 9| } 451| | 452| 63| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 63| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 63] ------------------ 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| 63| } else { 474| 63| ret.implementations_agree = true; 475| 63| } 476| 63| return ret; 477| 63| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 189| const auto success = [](const ConversionResult& r) -> bool { 402| 189| if constexpr (std::is_same_v) { 403| 189| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 189| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 126| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_11clENSt3__14spanIKcLm18446744073709551615EEE: 554| 113| +[](std::span chardata) { \ 555| 113| const auto c = \ 556| 113| Conversion{ \ 559| 113| &I::lenfunc, &I::conversionfunc, \ 560| 113| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 113|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 113| ::std::void_t(); \ | | 1180| 113| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 113| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 113| constexpr auto _size = _name.size(); \ | | 1183| 113| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 113| return _nameof; }() ------------------ 561| 113| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 113|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 113| ::std::void_t(); \ | | 1180| 113| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 113| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 113| constexpr auto _size = _name.size(); \ | | 1183| 113| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 113| return _nameof; }() ------------------ 562| 113| c.fuzz(chardata); \ 563| 113| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_12clENSt3__14spanIKcLm18446744073709551615EEE: 554| 163| +[](std::span chardata) { \ 555| 163| const auto c = \ 556| 163| Conversion{ \ 559| 163| &I::lenfunc, &I::conversionfunc, \ 560| 163| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 163|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 163| ::std::void_t(); \ | | 1180| 163| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 163| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 163| constexpr auto _size = _name.size(); \ | | 1183| 163| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 163| return _nameof; }() ------------------ 561| 163| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 163|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 163| ::std::void_t(); \ | | 1180| 163| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 163| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 163| constexpr auto _size = _name.size(); \ | | 1183| 163| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 163| return _nameof; }() ------------------ 562| 163| c.fuzz(chardata); \ 563| 163| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_13clENSt3__14spanIKcLm18446744073709551615EEE: 554| 62| +[](std::span chardata) { \ 555| 62| const auto c = \ 556| 62| Conversion{ \ 559| 62| &I::lenfunc, &I::conversionfunc, \ 560| 62| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 62|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 62| ::std::void_t(); \ | | 1180| 62| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 62| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 62| constexpr auto _size = _name.size(); \ | | 1183| 62| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 62| return _nameof; }() ------------------ 561| 62| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 62|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 62| ::std::void_t(); \ | | 1180| 62| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 62| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 62| constexpr auto _size = _name.size(); \ | | 1183| 62| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 62| return _nameof; }() ------------------ 562| 62| c.fuzz(chardata); \ 563| 62| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 62| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 62| const FromSpan from{reinterpret_cast(chardata.data()), 178| 62| chardata.size() / sizeof(FromType)}; 179| | 180| 62| static const bool do_print_testcase = 181| 62| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 62| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 62] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 62| do { 189| | // step 0 - is the input valid? 190| 62| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 62| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 62] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 62| From == UtfEncodings::UTF8) { 198| 62| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 62] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 62| } 201| | 202| | // step 2 - what is the required size of the output? 203| 62| const auto [output_length, length_agree] = 204| 62| calculate_length(from, inputisvalid); 205| 62| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 62] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 62| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 16, False: 46] | Branch (208:28): [True: 0, False: 16] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 62| const auto [written, outputs_agree] = 215| 62| do_conversion(from, output_length, inputisvalid); 216| 62| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 62] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 62| return; 221| 62| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 62| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 230| 62| validation_result verify_valid_input(FromSpan src) const { 231| 62| validation_result ret{}; 232| | 233| 62| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 62| const auto implementations = get_supported_implementations(); 235| 62| std::vector results; 236| 62| results.reserve(implementations.size()); 237| | 238| 186| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 186, False: 62] ------------------ 239| 186| results.push_back( 240| 186| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 186| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 186| const bool validation2 = 245| 186| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 186| src.data(), src.size()); 247| 186| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 186] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 186| } 258| | 259| 62| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 62| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 62] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 62| } else { 273| 62| ret.implementations_agree = true; 274| 62| } 275| 62| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 62| return r.error == simdutf::SUCCESS; 277| 62| }); 278| 62| return ret; 279| 62| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 124| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 154| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 154| return r.error == simdutf::SUCCESS; 277| 154| }); _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEE: 289| 62| bool count_the_input(FromSpan src) const { 290| 62| const auto implementations = get_supported_implementations(); 291| 62| std::vector results; 292| 62| results.reserve(implementations.size()); 293| | 294| 186| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 186, False: 62] ------------------ 295| 186| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 186| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 186| ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 186| results.push_back(ret); 304| 186| } 305| 62| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 62| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 62] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 62| return true; 321| 62| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 124| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 62| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 62| length_result ret{}; 344| | 345| 62| const auto implementations = get_supported_implementations(); 346| 62| std::vector results; 347| 62| results.reserve(implementations.size()); 348| | 349| 186| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 186, False: 62] ------------------ 350| 186| const auto len = invoke_lengthcalc(impl, src); 351| 186| results.push_back(len); 352| 186| ret.length.push_back(len); 353| 186| } 354| | 355| 62| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 62| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 62] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 62| } else { 375| 62| ret.implementations_agree = true; 376| 62| } 377| 62| return ret; 378| 62| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 186| FromSpan src) const { 339| 186| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 186| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 124| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 62| const bool inputisvalid) const { 383| 62| conversion_result ret{}; 384| | 385| 62| const auto implementations = get_supported_implementations(); 386| | 387| 62| std::vector> results; 388| 62| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 62| std::vector> outputbuffers; 393| 62| outputbuffers.reserve(implementations.size()); 394| 248| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 186, False: 62] ------------------ 395| 186| auto impl = implementations[i]; 396| 186| const ToType canary1{42}; 397| 186| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 186| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 186| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 186| const auto success = [](const ConversionResult& r) -> bool { 402| 186| if constexpr (std::is_same_v) { 403| 186| return r != 0; 404| 186| } else { 405| 186| return r.error == simdutf::error_code::SUCCESS; 406| 186| } 407| 186| }(implret1); 408| 186| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 186| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 186| const ToType canary2{25}; 414| 186| const auto outputbuffer_first_run = outputbuffer; 415| 186| std::ranges::fill(outputbuffer, canary2); 416| 186| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 186| src.size(), outputbuffer.data()); 418| | 419| 186| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 186] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 186| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 138, False: 48] | Branch (423:29): [True: 57, False: 81] ------------------ 424| | // only care about the output if the input is valid 425| 57| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 57| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 57] ------------------ 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| 57| } 440| 186| } 441| 186| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 57, False: 129] ------------------ 442| 186| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 62| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 16, False: 46] ------------------ 447| 48| for (auto& e : results) { ------------------ | Branch (447:20): [True: 48, False: 16] ------------------ 448| 48| e.outputhash.clear(); 449| 48| } 450| 16| } 451| | 452| 62| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 62| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 62] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 62| } else { 474| 62| ret.implementations_agree = true; 475| 62| } 476| 62| return ret; 477| 62| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 186| const auto success = [](const ConversionResult& r) -> bool { 402| 186| if constexpr (std::is_same_v) { 403| 186| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 186| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 124| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_14clENSt3__14spanIKcLm18446744073709551615EEE: 554| 136| +[](std::span chardata) { \ 555| 136| const auto c = \ 556| 136| Conversion{ \ 559| 136| &I::lenfunc, &I::conversionfunc, \ 560| 136| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 136|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 136| ::std::void_t(); \ | | 1180| 136| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 136| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 136| constexpr auto _size = _name.size(); \ | | 1183| 136| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 136| return _nameof; }() ------------------ 561| 136| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 136|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 136| ::std::void_t(); \ | | 1180| 136| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 136| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 136| constexpr auto _size = _name.size(); \ | | 1183| 136| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 136| return _nameof; }() ------------------ 562| 136| c.fuzz(chardata); \ 563| 136| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_15clENSt3__14spanIKcLm18446744073709551615EEE: 554| 193| +[](std::span chardata) { \ 555| 193| const auto c = \ 556| 193| Conversion{ \ 559| 193| &I::lenfunc, &I::conversionfunc, \ 560| 193| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 193|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 193| ::std::void_t(); \ | | 1180| 193| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 193| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 193| constexpr auto _size = _name.size(); \ | | 1183| 193| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 193| return _nameof; }() ------------------ 561| 193| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 193|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 193| ::std::void_t(); \ | | 1180| 193| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 193| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 193| constexpr auto _size = _name.size(); \ | | 1183| 193| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 193| return _nameof; }() ------------------ 562| 193| c.fuzz(chardata); \ 563| 193| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_16clENSt3__14spanIKcLm18446744073709551615EEE: 554| 94| +[](std::span chardata) { \ 555| 94| const auto c = \ 556| 94| Conversion{ \ 559| 94| &I::lenfunc, &I::conversionfunc, \ 560| 94| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 94|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 94| ::std::void_t(); \ | | 1180| 94| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 94| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 94| constexpr auto _size = _name.size(); \ | | 1183| 94| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 94| return _nameof; }() ------------------ 561| 94| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 94|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 94| ::std::void_t(); \ | | 1180| 94| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 94| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 94| constexpr auto _size = _name.size(); \ | | 1183| 94| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 94| return _nameof; }() ------------------ 562| 94| c.fuzz(chardata); \ 563| 94| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 94| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 94| const FromSpan from{reinterpret_cast(chardata.data()), 178| 94| chardata.size() / sizeof(FromType)}; 179| | 180| 94| static const bool do_print_testcase = 181| 94| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 94| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 94] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 94| do { 189| | // step 0 - is the input valid? 190| 94| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 94| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 94] | 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| 94| const auto [output_length, length_agree] = 204| 94| calculate_length(from, inputisvalid); 205| 94| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 94] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 94| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 62, False: 32] | Branch (208:28): [True: 0, False: 62] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 94| const auto [written, outputs_agree] = 215| 94| do_conversion(from, output_length, inputisvalid); 216| 94| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 94] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 94| return; 221| 94| } 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| 94| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 230| 94| validation_result verify_valid_input(FromSpan src) const { 231| 94| validation_result ret{}; 232| | 233| 94| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 94| const auto implementations = get_supported_implementations(); 235| 94| std::vector results; 236| 94| results.reserve(implementations.size()); 237| | 238| 282| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 282, False: 94] ------------------ 239| 282| results.push_back( 240| 282| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 282| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 282| const bool validation2 = 245| 282| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 282| src.data(), src.size()); 247| 282| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 282] ------------------ 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| 282| } 258| | 259| 94| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 94| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 94] ------------------ 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| 94| } else { 273| 94| ret.implementations_agree = true; 274| 94| } 275| 94| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 94| return r.error == simdutf::SUCCESS; 277| 94| }); 278| 94| return ret; 279| 94| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 188| 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| 158| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 158| return r.error == simdutf::SUCCESS; 277| 158| }); _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 94| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 94| length_result ret{}; 344| | 345| 94| const auto implementations = get_supported_implementations(); 346| 94| std::vector results; 347| 94| results.reserve(implementations.size()); 348| | 349| 282| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 282, False: 94] ------------------ 350| 282| const auto len = invoke_lengthcalc(impl, src); 351| 282| results.push_back(len); 352| 282| ret.length.push_back(len); 353| 282| } 354| | 355| 94| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 94| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 94] ------------------ 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| 94| } else { 375| 94| ret.implementations_agree = true; 376| 94| } 377| 94| return ret; 378| 94| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 282| FromSpan src) const { 339| 282| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 282| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 188| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 94| const bool inputisvalid) const { 383| 94| conversion_result ret{}; 384| | 385| 94| const auto implementations = get_supported_implementations(); 386| | 387| 94| std::vector> results; 388| 94| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 94| std::vector> outputbuffers; 393| 94| outputbuffers.reserve(implementations.size()); 394| 376| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 282, False: 94] ------------------ 395| 282| auto impl = implementations[i]; 396| 282| const ToType canary1{42}; 397| 282| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 282| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 282| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 282| const auto success = [](const ConversionResult& r) -> bool { 402| 282| if constexpr (std::is_same_v) { 403| 282| return r != 0; 404| 282| } else { 405| 282| return r.error == simdutf::error_code::SUCCESS; 406| 282| } 407| 282| }(implret1); 408| 282| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 282| 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| 282| const ToType canary2{25}; 414| 282| const auto outputbuffer_first_run = outputbuffer; 415| 282| std::ranges::fill(outputbuffer, canary2); 416| 282| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 282| src.size(), outputbuffer.data()); 418| | 419| 282| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 282] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 282| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 96, False: 186] | Branch (423:29): [True: 84, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 84| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 84| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 84] ------------------ 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| 84| } 440| 282| } 441| 282| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 84, False: 198] ------------------ 442| 282| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 94| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 62, False: 32] ------------------ 447| 186| for (auto& e : results) { ------------------ | Branch (447:20): [True: 186, False: 62] ------------------ 448| 186| e.outputhash.clear(); 449| 186| } 450| 62| } 451| | 452| 94| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 94| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 94] ------------------ 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| 94| } else { 474| 94| ret.implementations_agree = true; 475| 94| } 476| 94| return ret; 477| 94| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 282| const auto success = [](const ConversionResult& r) -> bool { 402| 282| if constexpr (std::is_same_v) { 403| 282| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 282| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 188| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_17clENSt3__14spanIKcLm18446744073709551615EEE: 554| 221| +[](std::span chardata) { \ 555| 221| const auto c = \ 556| 221| Conversion{ \ 559| 221| &I::lenfunc, &I::conversionfunc, \ 560| 221| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 221|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 221| ::std::void_t(); \ | | 1180| 221| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 221| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 221| constexpr auto _size = _name.size(); \ | | 1183| 221| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 221| return _nameof; }() ------------------ 561| 221| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 221|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 221| ::std::void_t(); \ | | 1180| 221| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 221| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 221| constexpr auto _size = _name.size(); \ | | 1183| 221| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 221| return _nameof; }() ------------------ 562| 221| c.fuzz(chardata); \ 563| 221| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_18clENSt3__14spanIKcLm18446744073709551615EEE: 554| 216| +[](std::span chardata) { \ 555| 216| const auto c = \ 556| 216| Conversion{ \ 559| 216| &I::lenfunc, &I::conversionfunc, \ 560| 216| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 216|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 216| ::std::void_t(); \ | | 1180| 216| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 216| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 216| constexpr auto _size = _name.size(); \ | | 1183| 216| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 216| return _nameof; }() ------------------ 561| 216| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 216|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 216| ::std::void_t(); \ | | 1180| 216| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 216| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 216| constexpr auto _size = _name.size(); \ | | 1183| 216| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 216| return _nameof; }() ------------------ 562| 216| c.fuzz(chardata); \ 563| 216| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_19clENSt3__14spanIKcLm18446744073709551615EEE: 554| 302| +[](std::span chardata) { \ 555| 302| const auto c = \ 556| 302| Conversion{ \ 559| 302| &I::lenfunc, &I::conversionfunc, \ 560| 302| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 302|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 302| ::std::void_t(); \ | | 1180| 302| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 302| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 302| constexpr auto _size = _name.size(); \ | | 1183| 302| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 302| return _nameof; }() ------------------ 561| 302| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 302|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 302| ::std::void_t(); \ | | 1180| 302| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 302| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 302| constexpr auto _size = _name.size(); \ | | 1183| 302| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 302| return _nameof; }() ------------------ 562| 302| c.fuzz(chardata); \ 563| 302| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_20clENSt3__14spanIKcLm18446744073709551615EEE: 554| 281| +[](std::span chardata) { \ 555| 281| const auto c = \ 556| 281| Conversion{ \ 559| 281| &I::lenfunc, &I::conversionfunc, \ 560| 281| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 281|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 281| ::std::void_t(); \ | | 1180| 281| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 281| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 281| constexpr auto _size = _name.size(); \ | | 1183| 281| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 281| return _nameof; }() ------------------ 561| 281| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 281|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 281| ::std::void_t(); \ | | 1180| 281| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 281| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 281| constexpr auto _size = _name.size(); \ | | 1183| 281| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 281| return _nameof; }() ------------------ 562| 281| c.fuzz(chardata); \ 563| 281| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 281| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 281| const FromSpan from{reinterpret_cast(chardata.data()), 178| 281| chardata.size() / sizeof(FromType)}; 179| | 180| 281| static const bool do_print_testcase = 181| 281| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 281| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 281] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 281| do { 189| | // step 0 - is the input valid? 190| 281| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 281| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 281] | 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| 281| From == UtfEncodings::UTF8) { 198| 281| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 281] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 281| } 201| | 202| | // step 2 - what is the required size of the output? 203| 281| const auto [output_length, length_agree] = 204| 281| calculate_length(from, inputisvalid); 205| 281| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 281] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 281| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 186, False: 95] | Branch (208:28): [True: 0, False: 186] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 281| const auto [written, outputs_agree] = 215| 281| do_conversion(from, output_length, inputisvalid); 216| 281| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 281] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 281| return; 221| 281| } 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| 281| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 281| validation_result verify_valid_input(FromSpan src) const { 231| 281| validation_result ret{}; 232| | 233| 281| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 281| const auto implementations = get_supported_implementations(); 235| 281| std::vector results; 236| 281| results.reserve(implementations.size()); 237| | 238| 843| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 843, False: 281] ------------------ 239| 843| results.push_back( 240| 843| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 843| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 843| const bool validation2 = 245| 843| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 843| src.data(), src.size()); 247| 843| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 843] ------------------ 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| 843| } 258| | 259| 281| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 281| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 281] ------------------ 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| 281| } else { 273| 281| ret.implementations_agree = true; 274| 281| } 275| 281| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 281| return r.error == simdutf::SUCCESS; 277| 281| }); 278| 281| return ret; 279| 281| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 562| 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| 471| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 471| return r.error == simdutf::SUCCESS; 277| 471| }); _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 281| bool count_the_input(FromSpan src) const { 290| 281| const auto implementations = get_supported_implementations(); 291| 281| std::vector results; 292| 281| results.reserve(implementations.size()); 293| | 294| 843| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 843, False: 281] ------------------ 295| 843| 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| 843| } else if constexpr (From == UtfEncodings::UTF8) { 301| 843| ret = impl->count_utf8(src.data(), src.size()); 302| 843| } 303| 843| results.push_back(ret); 304| 843| } 305| 281| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 281| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 281] ------------------ 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| 281| return true; 321| 281| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 562| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 281| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 281| length_result ret{}; 344| | 345| 281| const auto implementations = get_supported_implementations(); 346| 281| std::vector results; 347| 281| results.reserve(implementations.size()); 348| | 349| 843| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 843, False: 281] ------------------ 350| 843| const auto len = invoke_lengthcalc(impl, src); 351| 843| results.push_back(len); 352| 843| ret.length.push_back(len); 353| 843| } 354| | 355| 281| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 281| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 281] ------------------ 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| 281| } else { 375| 281| ret.implementations_agree = true; 376| 281| } 377| 281| return ret; 378| 281| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 843| FromSpan src) const { 331| 843| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 843| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 562| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 281| const bool inputisvalid) const { 383| 281| conversion_result ret{}; 384| | 385| 281| const auto implementations = get_supported_implementations(); 386| | 387| 281| std::vector> results; 388| 281| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 281| std::vector> outputbuffers; 393| 281| outputbuffers.reserve(implementations.size()); 394| 1.12k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 843, False: 281] ------------------ 395| 843| auto impl = implementations[i]; 396| 843| const ToType canary1{42}; 397| 843| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 843| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 843| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 843| const auto success = [](const ConversionResult& r) -> bool { 402| 843| if constexpr (std::is_same_v) { 403| 843| return r != 0; 404| 843| } else { 405| 843| return r.error == simdutf::error_code::SUCCESS; 406| 843| } 407| 843| }(implret1); 408| 843| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 843| 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| 843| const ToType canary2{25}; 414| 843| const auto outputbuffer_first_run = outputbuffer; 415| 843| std::ranges::fill(outputbuffer, canary2); 416| 843| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 843| src.size(), outputbuffer.data()); 418| | 419| 843| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 843] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 843| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 285, False: 558] | Branch (423:29): [True: 258, False: 27] ------------------ 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| 843| } 441| 843| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 258, False: 585] ------------------ 442| 843| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 281| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 186, False: 95] ------------------ 447| 558| for (auto& e : results) { ------------------ | Branch (447:20): [True: 558, False: 186] ------------------ 448| 558| e.outputhash.clear(); 449| 558| } 450| 186| } 451| | 452| 281| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 281| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 281] ------------------ 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| 281| } else { 474| 281| ret.implementations_agree = true; 475| 281| } 476| 281| return ret; 477| 281| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 843| const auto success = [](const ConversionResult& r) -> bool { 402| 843| if constexpr (std::is_same_v) { 403| 843| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 843| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 562| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_21clENSt3__14spanIKcLm18446744073709551615EEE: 554| 356| +[](std::span chardata) { \ 555| 356| const auto c = \ 556| 356| Conversion{ \ 559| 356| &I::lenfunc, &I::conversionfunc, \ 560| 356| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 356|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 356| ::std::void_t(); \ | | 1180| 356| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 356| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 356| constexpr auto _size = _name.size(); \ | | 1183| 356| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 356| return _nameof; }() ------------------ 561| 356| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 356|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 356| ::std::void_t(); \ | | 1180| 356| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 356| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 356| constexpr auto _size = _name.size(); \ | | 1183| 356| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 356| return _nameof; }() ------------------ 562| 356| c.fuzz(chardata); \ 563| 356| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_22clENSt3__14spanIKcLm18446744073709551615EEE: 554| 330| +[](std::span chardata) { \ 555| 330| const auto c = \ 556| 330| Conversion{ \ 559| 330| &I::lenfunc, &I::conversionfunc, \ 560| 330| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 330|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 330| ::std::void_t(); \ | | 1180| 330| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 330| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 330| constexpr auto _size = _name.size(); \ | | 1183| 330| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 330| return _nameof; }() ------------------ 561| 330| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 330|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 330| ::std::void_t(); \ | | 1180| 330| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 330| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 330| constexpr auto _size = _name.size(); \ | | 1183| 330| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 330| return _nameof; }() ------------------ 562| 330| c.fuzz(chardata); \ 563| 330| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_23clENSt3__14spanIKcLm18446744073709551615EEE: 554| 356| +[](std::span chardata) { \ 555| 356| const auto c = \ 556| 356| Conversion{ \ 559| 356| &I::lenfunc, &I::conversionfunc, \ 560| 356| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 356|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 356| ::std::void_t(); \ | | 1180| 356| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 356| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 356| constexpr auto _size = _name.size(); \ | | 1183| 356| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 356| return _nameof; }() ------------------ 561| 356| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 356|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 356| ::std::void_t(); \ | | 1180| 356| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 356| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 356| constexpr auto _size = _name.size(); \ | | 1183| 356| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 356| return _nameof; }() ------------------ 562| 356| c.fuzz(chardata); \ 563| 356| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_24clENSt3__14spanIKcLm18446744073709551615EEE: 554| 126| +[](std::span chardata) { \ 555| 126| const auto c = \ 556| 126| Conversion{ \ 559| 126| &I::lenfunc, &I::conversionfunc, \ 560| 126| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 126|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 126| ::std::void_t(); \ | | 1180| 126| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 126| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 126| constexpr auto _size = _name.size(); \ | | 1183| 126| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 126| return _nameof; }() ------------------ 561| 126| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 126|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 126| ::std::void_t(); \ | | 1180| 126| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 126| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 126| constexpr auto _size = _name.size(); \ | | 1183| 126| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 126| return _nameof; }() ------------------ 562| 126| c.fuzz(chardata); \ 563| 126| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 126| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 126| const FromSpan from{reinterpret_cast(chardata.data()), 178| 126| chardata.size() / sizeof(FromType)}; 179| | 180| 126| static const bool do_print_testcase = 181| 126| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 126| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 126] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 126| do { 189| | // step 0 - is the input valid? 190| 126| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 126| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 126] | 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| 126| From == UtfEncodings::UTF8) { 198| 126| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 126] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 126| } 201| | 202| | // step 2 - what is the required size of the output? 203| 126| const auto [output_length, length_agree] = 204| 126| calculate_length(from, inputisvalid); 205| 126| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 126] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 126| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 21, False: 105] | Branch (208:28): [True: 0, False: 21] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 126| const auto [written, outputs_agree] = 215| 126| do_conversion(from, output_length, inputisvalid); 216| 126| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 126] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 126| return; 221| 126| } 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| 126| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEE: 230| 126| validation_result verify_valid_input(FromSpan src) const { 231| 126| validation_result ret{}; 232| | 233| 126| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 126| const auto implementations = get_supported_implementations(); 235| 126| std::vector results; 236| 126| results.reserve(implementations.size()); 237| | 238| 378| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 378, False: 126] ------------------ 239| 378| results.push_back( 240| 378| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 378| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 378| const bool validation2 = 245| 378| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 378| src.data(), src.size()); 247| 378| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 378] ------------------ 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| 378| } 258| | 259| 126| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 126| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 126] ------------------ 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| 126| } else { 273| 126| ret.implementations_agree = true; 274| 126| } 275| 126| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 126| return r.error == simdutf::SUCCESS; 277| 126| }); 278| 126| return ret; 279| 126| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS5_S5_EEDaSJ_SM_: 259| 252| 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| 336| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 336| return r.error == simdutf::SUCCESS; 277| 336| }); _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEE: 289| 126| bool count_the_input(FromSpan src) const { 290| 126| const auto implementations = get_supported_implementations(); 291| 126| std::vector results; 292| 126| results.reserve(implementations.size()); 293| | 294| 378| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 378, False: 126] ------------------ 295| 378| std::size_t ret; 296| 378| if constexpr (From == UtfEncodings::UTF16BE) { 297| 378| 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| 378| results.push_back(ret); 304| 378| } 305| 126| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 126| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 126] ------------------ 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| 126| return true; 321| 126| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 252| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEb: 342| 126| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 126| length_result ret{}; 344| | 345| 126| const auto implementations = get_supported_implementations(); 346| 126| std::vector results; 347| 126| results.reserve(implementations.size()); 348| | 349| 378| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 378, False: 126] ------------------ 350| 378| const auto len = invoke_lengthcalc(impl, src); 351| 378| results.push_back(len); 352| 378| ret.length.push_back(len); 353| 378| } 354| | 355| 126| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 126| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 126] ------------------ 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| 126| } else { 375| 126| ret.implementations_agree = true; 376| 126| } 377| 126| return ret; 378| 126| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSF_NSt3__14spanIS6_Lm18446744073709551615EEE: 338| 378| FromSpan src) const { 339| 378| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 378| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 252| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 126| const bool inputisvalid) const { 383| 126| conversion_result ret{}; 384| | 385| 126| const auto implementations = get_supported_implementations(); 386| | 387| 126| std::vector> results; 388| 126| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 126| std::vector> outputbuffers; 393| 126| outputbuffers.reserve(implementations.size()); 394| 504| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 378, False: 126] ------------------ 395| 378| auto impl = implementations[i]; 396| 378| const ToType canary1{42}; 397| 378| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 378| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 378| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 378| const auto success = [](const ConversionResult& r) -> bool { 402| 378| if constexpr (std::is_same_v) { 403| 378| return r != 0; 404| 378| } else { 405| 378| return r.error == simdutf::error_code::SUCCESS; 406| 378| } 407| 378| }(implret1); 408| 378| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 378| 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| 378| const ToType canary2{25}; 414| 378| const auto outputbuffer_first_run = outputbuffer; 415| 378| std::ranges::fill(outputbuffer, canary2); 416| 378| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 378| src.size(), outputbuffer.data()); 418| | 419| 378| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 378] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 378| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 315, False: 63] | Branch (423:29): [True: 144, False: 171] ------------------ 424| | // only care about the output if the input is valid 425| 144| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 144| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 144] ------------------ 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| 144| } 440| 378| } 441| 378| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 144, False: 234] ------------------ 442| 378| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 126| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 21, False: 105] ------------------ 447| 63| for (auto& e : results) { ------------------ | Branch (447:20): [True: 63, False: 21] ------------------ 448| 63| e.outputhash.clear(); 449| 63| } 450| 21| } 451| | 452| 126| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 126| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 126] ------------------ 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| 126| } else { 474| 126| ret.implementations_agree = true; 475| 126| } 476| 126| return ret; 477| 126| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS5_E_clESM_: 401| 378| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 378| } else { 405| 378| return r.error == simdutf::error_code::SUCCESS; 406| 378| } 407| 378| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS5_ESU_EEDaSN_SQ_: 452| 252| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_25clENSt3__14spanIKcLm18446744073709551615EEE: 554| 150| +[](std::span chardata) { \ 555| 150| const auto c = \ 556| 150| Conversion{ \ 559| 150| &I::lenfunc, &I::conversionfunc, \ 560| 150| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 150|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 150| ::std::void_t(); \ | | 1180| 150| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 150| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 150| constexpr auto _size = _name.size(); \ | | 1183| 150| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 150| return _nameof; }() ------------------ 561| 150| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 150|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 150| ::std::void_t(); \ | | 1180| 150| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 150| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 150| constexpr auto _size = _name.size(); \ | | 1183| 150| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 150| return _nameof; }() ------------------ 562| 150| c.fuzz(chardata); \ 563| 150| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 150| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 150| const FromSpan from{reinterpret_cast(chardata.data()), 178| 150| chardata.size() / sizeof(FromType)}; 179| | 180| 150| static const bool do_print_testcase = 181| 150| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 150| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 150] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 150| do { 189| | // step 0 - is the input valid? 190| 150| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 150| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 150] | 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| 150| From == UtfEncodings::UTF8) { 198| 150| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 150] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 150| } 201| | 202| | // step 2 - what is the required size of the output? 203| 150| const auto [output_length, length_agree] = 204| 150| calculate_length(from, inputisvalid); 205| 150| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 150] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 150| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 70, False: 80] | Branch (208:28): [True: 0, False: 70] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 150| const auto [written, outputs_agree] = 215| 150| do_conversion(from, output_length, inputisvalid); 216| 150| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 150] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 150| return; 221| 150| } 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| 150| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 150| validation_result verify_valid_input(FromSpan src) const { 231| 150| validation_result ret{}; 232| | 233| 150| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 150| const auto implementations = get_supported_implementations(); 235| 150| std::vector results; 236| 150| results.reserve(implementations.size()); 237| | 238| 450| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 450, False: 150] ------------------ 239| 450| results.push_back( 240| 450| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 450| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 450| const bool validation2 = 245| 450| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 450| src.data(), src.size()); 247| 450| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 450] ------------------ 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| 450| } 258| | 259| 150| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 150| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 150] ------------------ 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| 150| } else { 273| 150| ret.implementations_agree = true; 274| 150| } 275| 150| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 150| return r.error == simdutf::SUCCESS; 277| 150| }); 278| 150| return ret; 279| 150| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 300| 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| 310| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 310| return r.error == simdutf::SUCCESS; 277| 310| }); _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 150| bool count_the_input(FromSpan src) const { 290| 150| const auto implementations = get_supported_implementations(); 291| 150| std::vector results; 292| 150| results.reserve(implementations.size()); 293| | 294| 450| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 450, False: 150] ------------------ 295| 450| std::size_t ret; 296| 450| if constexpr (From == UtfEncodings::UTF16BE) { 297| 450| 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| 450| results.push_back(ret); 304| 450| } 305| 150| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 150| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 150] ------------------ 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| 150| return true; 321| 150| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 300| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 150| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 150| length_result ret{}; 344| | 345| 150| const auto implementations = get_supported_implementations(); 346| 150| std::vector results; 347| 150| results.reserve(implementations.size()); 348| | 349| 450| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 450, False: 150] ------------------ 350| 450| const auto len = invoke_lengthcalc(impl, src); 351| 450| results.push_back(len); 352| 450| ret.length.push_back(len); 353| 450| } 354| | 355| 150| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 150| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 150] ------------------ 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| 150| } else { 375| 150| ret.implementations_agree = true; 376| 150| } 377| 150| return ret; 378| 150| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 450| FromSpan src) const { 331| 450| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 450| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 300| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 150| const bool inputisvalid) const { 383| 150| conversion_result ret{}; 384| | 385| 150| const auto implementations = get_supported_implementations(); 386| | 387| 150| std::vector> results; 388| 150| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 150| std::vector> outputbuffers; 393| 150| outputbuffers.reserve(implementations.size()); 394| 600| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 450, False: 150] ------------------ 395| 450| auto impl = implementations[i]; 396| 450| const ToType canary1{42}; 397| 450| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 450| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 450| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 450| const auto success = [](const ConversionResult& r) -> bool { 402| 450| if constexpr (std::is_same_v) { 403| 450| return r != 0; 404| 450| } else { 405| 450| return r.error == simdutf::error_code::SUCCESS; 406| 450| } 407| 450| }(implret1); 408| 450| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 450| 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| 450| const ToType canary2{25}; 414| 450| const auto outputbuffer_first_run = outputbuffer; 415| 450| std::ranges::fill(outputbuffer, canary2); 416| 450| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 450| src.size(), outputbuffer.data()); 418| | 419| 450| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 450] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 450| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 240, False: 210] | Branch (423:29): [True: 240, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 240| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 240| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 240] ------------------ 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| 240| } 440| 450| } 441| 450| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 240, False: 210] ------------------ 442| 450| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 150| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 70, False: 80] ------------------ 447| 210| for (auto& e : results) { ------------------ | Branch (447:20): [True: 210, False: 70] ------------------ 448| 210| e.outputhash.clear(); 449| 210| } 450| 70| } 451| | 452| 150| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 150| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 150] ------------------ 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| 150| } else { 474| 150| ret.implementations_agree = true; 475| 150| } 476| 150| return ret; 477| 150| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 450| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 450| } else { 405| 450| return r.error == simdutf::error_code::SUCCESS; 406| 450| } 407| 450| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 300| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_26clENSt3__14spanIKcLm18446744073709551615EEE: 554| 293| +[](std::span chardata) { \ 555| 293| const auto c = \ 556| 293| Conversion{ \ 559| 293| &I::lenfunc, &I::conversionfunc, \ 560| 293| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 293|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 293| ::std::void_t(); \ | | 1180| 293| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 293| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 293| constexpr auto _size = _name.size(); \ | | 1183| 293| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 293| return _nameof; }() ------------------ 561| 293| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 293|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 293| ::std::void_t(); \ | | 1180| 293| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 293| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 293| constexpr auto _size = _name.size(); \ | | 1183| 293| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 293| return _nameof; }() ------------------ 562| 293| c.fuzz(chardata); \ 563| 293| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 293| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 293| const FromSpan from{reinterpret_cast(chardata.data()), 178| 293| chardata.size() / sizeof(FromType)}; 179| | 180| 293| static const bool do_print_testcase = 181| 293| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 293| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 293] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 293| do { 189| | // step 0 - is the input valid? 190| 293| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 293| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 293] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 293| From == UtfEncodings::UTF8) { 198| 293| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 293] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 293| } 201| | 202| | // step 2 - what is the required size of the output? 203| 293| const auto [output_length, length_agree] = 204| 293| calculate_length(from, inputisvalid); 205| 293| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 293] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 293| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 125, False: 168] | Branch (208:28): [True: 0, False: 125] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 293| const auto [written, outputs_agree] = 215| 293| do_conversion(from, output_length, inputisvalid); 216| 293| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 293] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 293| return; 221| 293| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 293| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 293| validation_result verify_valid_input(FromSpan src) const { 231| 293| validation_result ret{}; 232| | 233| 293| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 293| const auto implementations = get_supported_implementations(); 235| 293| std::vector results; 236| 293| results.reserve(implementations.size()); 237| | 238| 879| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 879, False: 293] ------------------ 239| 879| results.push_back( 240| 879| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 879| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 879| const bool validation2 = 245| 879| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 879| src.data(), src.size()); 247| 879| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 879] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 879| } 258| | 259| 293| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 293| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 293] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 293| } else { 273| 293| ret.implementations_agree = true; 274| 293| } 275| 293| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 293| return r.error == simdutf::SUCCESS; 277| 293| }); 278| 293| return ret; 279| 293| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 586| 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| 629| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 629| return r.error == simdutf::SUCCESS; 277| 629| }); _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 293| bool count_the_input(FromSpan src) const { 290| 293| const auto implementations = get_supported_implementations(); 291| 293| std::vector results; 292| 293| results.reserve(implementations.size()); 293| | 294| 879| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 879, False: 293] ------------------ 295| 879| std::size_t ret; 296| 879| if constexpr (From == UtfEncodings::UTF16BE) { 297| 879| 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| 879| results.push_back(ret); 304| 879| } 305| 293| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 293| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 293] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 293| return true; 321| 293| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 586| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 293| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 293| length_result ret{}; 344| | 345| 293| const auto implementations = get_supported_implementations(); 346| 293| std::vector results; 347| 293| results.reserve(implementations.size()); 348| | 349| 879| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 879, False: 293] ------------------ 350| 879| const auto len = invoke_lengthcalc(impl, src); 351| 879| results.push_back(len); 352| 879| ret.length.push_back(len); 353| 879| } 354| | 355| 293| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 293| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 293] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 293| } else { 375| 293| ret.implementations_agree = true; 376| 293| } 377| 293| return ret; 378| 293| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 879| FromSpan src) const { 331| 879| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 879| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 586| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 293| const bool inputisvalid) const { 383| 293| conversion_result ret{}; 384| | 385| 293| const auto implementations = get_supported_implementations(); 386| | 387| 293| std::vector> results; 388| 293| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 293| std::vector> outputbuffers; 393| 293| outputbuffers.reserve(implementations.size()); 394| 1.17k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 879, False: 293] ------------------ 395| 879| auto impl = implementations[i]; 396| 879| const ToType canary1{42}; 397| 879| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 879| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 879| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 879| const auto success = [](const ConversionResult& r) -> bool { 402| 879| if constexpr (std::is_same_v) { 403| 879| return r != 0; 404| 879| } else { 405| 879| return r.error == simdutf::error_code::SUCCESS; 406| 879| } 407| 879| }(implret1); 408| 879| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 879| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 879| const ToType canary2{25}; 414| 879| const auto outputbuffer_first_run = outputbuffer; 415| 879| std::ranges::fill(outputbuffer, canary2); 416| 879| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 879| src.size(), outputbuffer.data()); 418| | 419| 879| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 879] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 879| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 504, False: 375] | Branch (423:29): [True: 504, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 504| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 504| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 504] ------------------ 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| 504| } 440| 879| } 441| 879| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 504, False: 375] ------------------ 442| 879| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 293| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 125, False: 168] ------------------ 447| 375| for (auto& e : results) { ------------------ | Branch (447:20): [True: 375, False: 125] ------------------ 448| 375| e.outputhash.clear(); 449| 375| } 450| 125| } 451| | 452| 293| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 293| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 293] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 293| } else { 474| 293| ret.implementations_agree = true; 475| 293| } 476| 293| return ret; 477| 293| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 879| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 879| } else { 405| 879| return r.error == simdutf::error_code::SUCCESS; 406| 879| } 407| 879| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 586| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_27clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 115| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 115| const FromSpan from{reinterpret_cast(chardata.data()), 178| 115| chardata.size() / sizeof(FromType)}; 179| | 180| 115| static const bool do_print_testcase = 181| 115| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 115| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 115] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 115| do { 189| | // step 0 - is the input valid? 190| 115| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 115| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 115] | 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| 115| From == UtfEncodings::UTF8) { 198| 115| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 115] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 115| } 201| | 202| | // step 2 - what is the required size of the output? 203| 115| const auto [output_length, length_agree] = 204| 115| calculate_length(from, inputisvalid); 205| 115| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 115] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 115| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 21, False: 94] | Branch (208:28): [True: 0, False: 21] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 115| const auto [written, outputs_agree] = 215| 115| do_conversion(from, output_length, inputisvalid); 216| 115| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 115] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 115| return; 221| 115| } 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| 115| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEE: 230| 115| validation_result verify_valid_input(FromSpan src) const { 231| 115| validation_result ret{}; 232| | 233| 115| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 115| const auto implementations = get_supported_implementations(); 235| 115| std::vector results; 236| 115| results.reserve(implementations.size()); 237| | 238| 345| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 345, False: 115] ------------------ 239| 345| results.push_back( 240| 345| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 345| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 345| const bool validation2 = 245| 345| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 345| src.data(), src.size()); 247| 345| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 345] ------------------ 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| 345| } 258| | 259| 115| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 115| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 115] ------------------ 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| 115| } else { 273| 115| ret.implementations_agree = true; 274| 115| } 275| 115| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 115| return r.error == simdutf::SUCCESS; 277| 115| }); 278| 115| return ret; 279| 115| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS5_S5_EEDaSJ_SM_: 259| 230| 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| 303| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 303| return r.error == simdutf::SUCCESS; 277| 303| }); _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEE: 289| 115| bool count_the_input(FromSpan src) const { 290| 115| const auto implementations = get_supported_implementations(); 291| 115| std::vector results; 292| 115| results.reserve(implementations.size()); 293| | 294| 345| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 345, False: 115] ------------------ 295| 345| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 345| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 345| 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| 345| results.push_back(ret); 304| 345| } 305| 115| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 115| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 115] ------------------ 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| 115| return true; 321| 115| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 230| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEb: 342| 115| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 115| length_result ret{}; 344| | 345| 115| const auto implementations = get_supported_implementations(); 346| 115| std::vector results; 347| 115| results.reserve(implementations.size()); 348| | 349| 345| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 345, False: 115] ------------------ 350| 345| const auto len = invoke_lengthcalc(impl, src); 351| 345| results.push_back(len); 352| 345| ret.length.push_back(len); 353| 345| } 354| | 355| 115| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 115| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 115] ------------------ 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| 115| } else { 375| 115| ret.implementations_agree = true; 376| 115| } 377| 115| return ret; 378| 115| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSF_NSt3__14spanIS6_Lm18446744073709551615EEE: 338| 345| FromSpan src) const { 339| 345| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 345| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 230| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 115| const bool inputisvalid) const { 383| 115| conversion_result ret{}; 384| | 385| 115| const auto implementations = get_supported_implementations(); 386| | 387| 115| std::vector> results; 388| 115| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 115| std::vector> outputbuffers; 393| 115| outputbuffers.reserve(implementations.size()); 394| 460| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 345, False: 115] ------------------ 395| 345| auto impl = implementations[i]; 396| 345| const ToType canary1{42}; 397| 345| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 345| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 345| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 345| const auto success = [](const ConversionResult& r) -> bool { 402| 345| if constexpr (std::is_same_v) { 403| 345| return r != 0; 404| 345| } else { 405| 345| return r.error == simdutf::error_code::SUCCESS; 406| 345| } 407| 345| }(implret1); 408| 345| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 345| 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| 345| const ToType canary2{25}; 414| 345| const auto outputbuffer_first_run = outputbuffer; 415| 345| std::ranges::fill(outputbuffer, canary2); 416| 345| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 345| src.size(), outputbuffer.data()); 418| | 419| 345| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 345] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 345| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 282, False: 63] | Branch (423:29): [True: 123, False: 159] ------------------ 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| 345| } 441| 345| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 123, False: 222] ------------------ 442| 345| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 115| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 21, False: 94] ------------------ 447| 63| for (auto& e : results) { ------------------ | Branch (447:20): [True: 63, False: 21] ------------------ 448| 63| e.outputhash.clear(); 449| 63| } 450| 21| } 451| | 452| 115| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 115| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 115] ------------------ 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| 115| } else { 474| 115| ret.implementations_agree = true; 475| 115| } 476| 115| return ret; 477| 115| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS5_E_clESM_: 401| 345| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 345| } else { 405| 345| return r.error == simdutf::error_code::SUCCESS; 406| 345| } 407| 345| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS5_ESU_EEDaSN_SQ_: 452| 230| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_28clENSt3__14spanIKcLm18446744073709551615EEE: 554| 207| +[](std::span chardata) { \ 555| 207| const auto c = \ 556| 207| Conversion{ \ 559| 207| &I::lenfunc, &I::conversionfunc, \ 560| 207| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 207|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 207| ::std::void_t(); \ | | 1180| 207| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 207| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 207| constexpr auto _size = _name.size(); \ | | 1183| 207| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 207| return _nameof; }() ------------------ 561| 207| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 207|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 207| ::std::void_t(); \ | | 1180| 207| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 207| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 207| constexpr auto _size = _name.size(); \ | | 1183| 207| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 207| return _nameof; }() ------------------ 562| 207| c.fuzz(chardata); \ 563| 207| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 207| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 207| const FromSpan from{reinterpret_cast(chardata.data()), 178| 207| chardata.size() / sizeof(FromType)}; 179| | 180| 207| static const bool do_print_testcase = 181| 207| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 207| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 207] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 207| do { 189| | // step 0 - is the input valid? 190| 207| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 207| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 207] | 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| 207| From == UtfEncodings::UTF8) { 198| 207| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 207] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 207| } 201| | 202| | // step 2 - what is the required size of the output? 203| 207| const auto [output_length, length_agree] = 204| 207| calculate_length(from, inputisvalid); 205| 207| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 207] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 207| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 112, False: 95] | Branch (208:28): [True: 0, False: 112] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 207| const auto [written, outputs_agree] = 215| 207| do_conversion(from, output_length, inputisvalid); 216| 207| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 207] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 207| return; 221| 207| } 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| 207| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 207| validation_result verify_valid_input(FromSpan src) const { 231| 207| validation_result ret{}; 232| | 233| 207| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 207| const auto implementations = get_supported_implementations(); 235| 207| std::vector results; 236| 207| results.reserve(implementations.size()); 237| | 238| 621| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 621, False: 207] ------------------ 239| 621| results.push_back( 240| 621| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 621| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 621| const bool validation2 = 245| 621| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 621| src.data(), src.size()); 247| 621| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 621] ------------------ 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| 621| } 258| | 259| 207| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 207| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 207] ------------------ 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| 207| } else { 273| 207| ret.implementations_agree = true; 274| 207| } 275| 207| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 207| return r.error == simdutf::SUCCESS; 277| 207| }); 278| 207| return ret; 279| 207| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 414| 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| 397| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 397| return r.error == simdutf::SUCCESS; 277| 397| }); _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 207| bool count_the_input(FromSpan src) const { 290| 207| const auto implementations = get_supported_implementations(); 291| 207| std::vector results; 292| 207| results.reserve(implementations.size()); 293| | 294| 621| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 621, False: 207] ------------------ 295| 621| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 621| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 621| 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| 621| results.push_back(ret); 304| 621| } 305| 207| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 207| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 207] ------------------ 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| 207| return true; 321| 207| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 414| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 207| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 207| length_result ret{}; 344| | 345| 207| const auto implementations = get_supported_implementations(); 346| 207| std::vector results; 347| 207| results.reserve(implementations.size()); 348| | 349| 621| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 621, False: 207] ------------------ 350| 621| const auto len = invoke_lengthcalc(impl, src); 351| 621| results.push_back(len); 352| 621| ret.length.push_back(len); 353| 621| } 354| | 355| 207| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 207| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 207] ------------------ 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| 207| } else { 375| 207| ret.implementations_agree = true; 376| 207| } 377| 207| return ret; 378| 207| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 621| FromSpan src) const { 331| 621| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 621| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 414| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 207| const bool inputisvalid) const { 383| 207| conversion_result ret{}; 384| | 385| 207| const auto implementations = get_supported_implementations(); 386| | 387| 207| std::vector> results; 388| 207| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 207| std::vector> outputbuffers; 393| 207| outputbuffers.reserve(implementations.size()); 394| 828| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 621, False: 207] ------------------ 395| 621| auto impl = implementations[i]; 396| 621| const ToType canary1{42}; 397| 621| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 621| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 621| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 621| const auto success = [](const ConversionResult& r) -> bool { 402| 621| if constexpr (std::is_same_v) { 403| 621| return r != 0; 404| 621| } else { 405| 621| return r.error == simdutf::error_code::SUCCESS; 406| 621| } 407| 621| }(implret1); 408| 621| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 621| 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| 621| const ToType canary2{25}; 414| 621| const auto outputbuffer_first_run = outputbuffer; 415| 621| std::ranges::fill(outputbuffer, canary2); 416| 621| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 621| src.size(), outputbuffer.data()); 418| | 419| 621| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 621] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 621| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 285, False: 336] | Branch (423:29): [True: 285, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 285| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 285| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 285] ------------------ 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| 285| } 440| 621| } 441| 621| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 285, False: 336] ------------------ 442| 621| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 207| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 112, False: 95] ------------------ 447| 336| for (auto& e : results) { ------------------ | Branch (447:20): [True: 336, False: 112] ------------------ 448| 336| e.outputhash.clear(); 449| 336| } 450| 112| } 451| | 452| 207| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 207| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 207] ------------------ 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| 207| } else { 474| 207| ret.implementations_agree = true; 475| 207| } 476| 207| return ret; 477| 207| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 621| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 621| } else { 405| 621| return r.error == simdutf::error_code::SUCCESS; 406| 621| } 407| 621| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 414| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_29clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__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| 315| From == UtfEncodings::UTF8) { 198| 315| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 315] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 315| } 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: 108, False: 207] | Branch (208:28): [True: 0, False: 108] ------------------ 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| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_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| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_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; }; _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 729| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 729| return r.error == simdutf::SUCCESS; 277| 729| }); _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 315| bool count_the_input(FromSpan src) const { 290| 315| const auto implementations = get_supported_implementations(); 291| 315| std::vector results; 292| 315| results.reserve(implementations.size()); 293| | 294| 945| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 945, False: 315] ------------------ 295| 945| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 945| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 945| 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| 945| results.push_back(ret); 304| 945| } 305| 315| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 315| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 315] ------------------ 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| 315| return true; 321| 315| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 630| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_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| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE17invoke_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| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 630| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_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: 621, False: 324] | Branch (423:29): [True: 621, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 621| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 621| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 621] ------------------ 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| 621| } 440| 945| } 441| 945| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 621, False: 324] ------------------ 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: 108, False: 207] ------------------ 447| 324| for (auto& e : results) { ------------------ | Branch (447:20): [True: 324, False: 108] ------------------ 448| 324| e.outputhash.clear(); 449| 324| } 450| 108| } 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| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_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); _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_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$_30clENSt3__14spanIKcLm18446744073709551615EEE: 554| 210| +[](std::span chardata) { \ 555| 210| const auto c = \ 556| 210| Conversion{ \ 559| 210| &I::lenfunc, &I::conversionfunc, \ 560| 210| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 210|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 210| ::std::void_t(); \ | | 1180| 210| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 210| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 210| constexpr auto _size = _name.size(); \ | | 1183| 210| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 210| return _nameof; }() ------------------ 561| 210| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 210|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 210| ::std::void_t(); \ | | 1180| 210| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 210| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 210| constexpr auto _size = _name.size(); \ | | 1183| 210| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 210| return _nameof; }() ------------------ 562| 210| c.fuzz(chardata); \ 563| 210| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 210| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 210| const FromSpan from{reinterpret_cast(chardata.data()), 178| 210| chardata.size() / sizeof(FromType)}; 179| | 180| 210| static const bool do_print_testcase = 181| 210| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 210| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 210] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 210| do { 189| | // step 0 - is the input valid? 190| 210| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 210| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 210] | 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| 210| const auto [output_length, length_agree] = 204| 210| calculate_length(from, inputisvalid); 205| 210| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 210] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 210| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 176, False: 34] | Branch (208:28): [True: 0, False: 176] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 210| const auto [written, outputs_agree] = 215| 210| do_conversion(from, output_length, inputisvalid); 216| 210| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 210] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 210| return; 221| 210| } 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| 210| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEE: 230| 210| validation_result verify_valid_input(FromSpan src) const { 231| 210| validation_result ret{}; 232| | 233| 210| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 210| const auto implementations = get_supported_implementations(); 235| 210| std::vector results; 236| 210| results.reserve(implementations.size()); 237| | 238| 630| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 630, False: 210] ------------------ 239| 630| results.push_back( 240| 630| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 630| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 630| const bool validation2 = 245| 630| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 630| src.data(), src.size()); 247| 630| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 630] ------------------ 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| 630| } 258| | 259| 210| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 210| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 210] ------------------ 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| 210| } else { 273| 210| ret.implementations_agree = true; 274| 210| } 275| 210| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 210| return r.error == simdutf::SUCCESS; 277| 210| }); 278| 210| return ret; 279| 210| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS5_S5_EEDaSJ_SM_: 259| 420| 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| 278| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 278| return r.error == simdutf::SUCCESS; 277| 278| }); _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEb: 342| 210| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 210| length_result ret{}; 344| | 345| 210| const auto implementations = get_supported_implementations(); 346| 210| std::vector results; 347| 210| results.reserve(implementations.size()); 348| | 349| 630| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 630, False: 210] ------------------ 350| 630| const auto len = invoke_lengthcalc(impl, src); 351| 630| results.push_back(len); 352| 630| ret.length.push_back(len); 353| 630| } 354| | 355| 210| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 210| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 210] ------------------ 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| 210| } else { 375| 210| ret.implementations_agree = true; 376| 210| } 377| 210| return ret; 378| 210| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSF_NSt3__14spanIS6_Lm18446744073709551615EEE: 338| 630| FromSpan src) const { 339| 630| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 630| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 420| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 210| const bool inputisvalid) const { 383| 210| conversion_result ret{}; 384| | 385| 210| const auto implementations = get_supported_implementations(); 386| | 387| 210| std::vector> results; 388| 210| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 210| std::vector> outputbuffers; 393| 210| outputbuffers.reserve(implementations.size()); 394| 840| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 630, False: 210] ------------------ 395| 630| auto impl = implementations[i]; 396| 630| const ToType canary1{42}; 397| 630| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 630| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 630| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 630| const auto success = [](const ConversionResult& r) -> bool { 402| 630| if constexpr (std::is_same_v) { 403| 630| return r != 0; 404| 630| } else { 405| 630| return r.error == simdutf::error_code::SUCCESS; 406| 630| } 407| 630| }(implret1); 408| 630| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 630| 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| 630| const ToType canary2{25}; 414| 630| const auto outputbuffer_first_run = outputbuffer; 415| 630| std::ranges::fill(outputbuffer, canary2); 416| 630| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 630| src.size(), outputbuffer.data()); 418| | 419| 630| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 630] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 630| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 102, False: 528] | Branch (423:29): [True: 75, False: 27] ------------------ 424| | // only care about the output if the input is valid 425| 75| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 75| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 75] ------------------ 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| 75| } 440| 630| } 441| 630| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 75, False: 555] ------------------ 442| 630| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 210| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 176, False: 34] ------------------ 447| 528| for (auto& e : results) { ------------------ | Branch (447:20): [True: 528, False: 176] ------------------ 448| 528| e.outputhash.clear(); 449| 528| } 450| 176| } 451| | 452| 210| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 210| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 210] ------------------ 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| 210| } else { 474| 210| ret.implementations_agree = true; 475| 210| } 476| 210| return ret; 477| 210| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS5_E_clESM_: 401| 630| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 630| } else { 405| 630| return r.error == simdutf::error_code::SUCCESS; 406| 630| } 407| 630| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS5_ESU_EEDaSN_SQ_: 452| 420| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_31clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 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: 217, False: 86] | 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| 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| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 303| validation_result verify_valid_input(FromSpan src) const { 231| 303| validation_result ret{}; 232| | 233| 303| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 303| const auto implementations = get_supported_implementations(); 235| 303| std::vector results; 236| 303| results.reserve(implementations.size()); 237| | 238| 909| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 909, False: 303] ------------------ 239| 909| results.push_back( 240| 909| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 909| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 909| const bool validation2 = 245| 909| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 909| src.data(), src.size()); 247| 909| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 909] ------------------ 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| 909| } 258| | 259| 303| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 303| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 303] ------------------ 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| 303| } else { 273| 303| ret.implementations_agree = true; 274| 303| } 275| 303| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 303| return r.error == simdutf::SUCCESS; 277| 303| }); 278| 303| return ret; 279| 303| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 606| 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| 475| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 475| return r.error == simdutf::SUCCESS; 277| 475| }); _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_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| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 909| FromSpan src) const { 331| 909| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 909| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 606| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_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: 258, False: 651] | 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| 909| } 441| 909| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 258, False: 651] ------------------ 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: 217, False: 86] ------------------ 447| 651| for (auto& e : results) { ------------------ | Branch (447:20): [True: 651, False: 217] ------------------ 448| 651| e.outputhash.clear(); 449| 651| } 450| 217| } 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| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 909| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 909| } else { 405| 909| return r.error == simdutf::error_code::SUCCESS; 406| 909| } 407| 909| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 606| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_32clENSt3__14spanIKcLm18446744073709551615EEE: 554| 358| +[](std::span chardata) { \ 555| 358| const auto c = \ 556| 358| Conversion{ \ 559| 358| &I::lenfunc, &I::conversionfunc, \ 560| 358| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 358|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 358| ::std::void_t(); \ | | 1180| 358| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 358| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 358| constexpr auto _size = _name.size(); \ | | 1183| 358| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 358| return _nameof; }() ------------------ 561| 358| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 358|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 358| ::std::void_t(); \ | | 1180| 358| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 358| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 358| constexpr auto _size = _name.size(); \ | | 1183| 358| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 358| return _nameof; }() ------------------ 562| 358| c.fuzz(chardata); \ 563| 358| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 358| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 358| const FromSpan from{reinterpret_cast(chardata.data()), 178| 358| chardata.size() / sizeof(FromType)}; 179| | 180| 358| static const bool do_print_testcase = 181| 358| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 358| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 358] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 358| do { 189| | // step 0 - is the input valid? 190| 358| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 358| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 358] | 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| 358| const auto [output_length, length_agree] = 204| 358| calculate_length(from, inputisvalid); 205| 358| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 358] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 358| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 265, False: 93] | Branch (208:28): [True: 0, False: 265] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 358| const auto [written, outputs_agree] = 215| 358| do_conversion(from, output_length, inputisvalid); 216| 358| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 358] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 358| return; 221| 358| } 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| 358| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 358| validation_result verify_valid_input(FromSpan src) const { 231| 358| validation_result ret{}; 232| | 233| 358| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 358| const auto implementations = get_supported_implementations(); 235| 358| std::vector results; 236| 358| results.reserve(implementations.size()); 237| | 238| 1.07k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.07k, False: 358] ------------------ 239| 1.07k| results.push_back( 240| 1.07k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.07k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.07k| const bool validation2 = 245| 1.07k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.07k| src.data(), src.size()); 247| 1.07k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.07k] ------------------ 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| 1.07k| } 258| | 259| 358| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 358| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 358] ------------------ 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| 358| } else { 273| 358| ret.implementations_agree = true; 274| 358| } 275| 358| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 358| return r.error == simdutf::SUCCESS; 277| 358| }); 278| 358| return ret; 279| 358| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 716| 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| 544| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 544| return r.error == simdutf::SUCCESS; 277| 544| }); _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 358| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 358| length_result ret{}; 344| | 345| 358| const auto implementations = get_supported_implementations(); 346| 358| std::vector results; 347| 358| results.reserve(implementations.size()); 348| | 349| 1.07k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.07k, False: 358] ------------------ 350| 1.07k| const auto len = invoke_lengthcalc(impl, src); 351| 1.07k| results.push_back(len); 352| 1.07k| ret.length.push_back(len); 353| 1.07k| } 354| | 355| 358| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 358| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 358] ------------------ 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| 358| } else { 375| 358| ret.implementations_agree = true; 376| 358| } 377| 358| return ret; 378| 358| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.07k| FromSpan src) const { 331| 1.07k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.07k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 716| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 358| const bool inputisvalid) const { 383| 358| conversion_result ret{}; 384| | 385| 358| const auto implementations = get_supported_implementations(); 386| | 387| 358| std::vector> results; 388| 358| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 358| std::vector> outputbuffers; 393| 358| outputbuffers.reserve(implementations.size()); 394| 1.43k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.07k, False: 358] ------------------ 395| 1.07k| auto impl = implementations[i]; 396| 1.07k| const ToType canary1{42}; 397| 1.07k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.07k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.07k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.07k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.07k| if constexpr (std::is_same_v) { 403| 1.07k| return r != 0; 404| 1.07k| } else { 405| 1.07k| return r.error == simdutf::error_code::SUCCESS; 406| 1.07k| } 407| 1.07k| }(implret1); 408| 1.07k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.07k| 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| 1.07k| const ToType canary2{25}; 414| 1.07k| const auto outputbuffer_first_run = outputbuffer; 415| 1.07k| std::ranges::fill(outputbuffer, canary2); 416| 1.07k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.07k| src.size(), outputbuffer.data()); 418| | 419| 1.07k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.07k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.07k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 279, False: 795] | Branch (423:29): [True: 279, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 279| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 279| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 279] ------------------ 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| 279| } 440| 1.07k| } 441| 1.07k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 279, False: 795] ------------------ 442| 1.07k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 358| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 265, False: 93] ------------------ 447| 795| for (auto& e : results) { ------------------ | Branch (447:20): [True: 795, False: 265] ------------------ 448| 795| e.outputhash.clear(); 449| 795| } 450| 265| } 451| | 452| 358| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 358| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 358] ------------------ 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| 358| } else { 474| 358| ret.implementations_agree = true; 475| 358| } 476| 358| return ret; 477| 358| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.07k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.07k| } else { 405| 1.07k| return r.error == simdutf::error_code::SUCCESS; 406| 1.07k| } 407| 1.07k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 716| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_33clENSt3__14spanIKcLm18446744073709551615EEE: 554| 415| +[](std::span chardata) { \ 555| 415| const auto c = \ 556| 415| Conversion{ \ 559| 415| &I::lenfunc, &I::conversionfunc, \ 560| 415| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 415|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 415| ::std::void_t(); \ | | 1180| 415| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 415| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 415| constexpr auto _size = _name.size(); \ | | 1183| 415| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 415| return _nameof; }() ------------------ 561| 415| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 415|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 415| ::std::void_t(); \ | | 1180| 415| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 415| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 415| constexpr auto _size = _name.size(); \ | | 1183| 415| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 415| return _nameof; }() ------------------ 562| 415| c.fuzz(chardata); \ 563| 415| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 415| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 415| const FromSpan from{reinterpret_cast(chardata.data()), 178| 415| chardata.size() / sizeof(FromType)}; 179| | 180| 415| static const bool do_print_testcase = 181| 415| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 415| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 415] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 415| do { 189| | // step 0 - is the input valid? 190| 415| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 415| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 415] | 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| 415| const auto [output_length, length_agree] = 204| 415| calculate_length(from, inputisvalid); 205| 415| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 415] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 415| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 316, False: 99] | Branch (208:28): [True: 0, False: 316] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 415| const auto [written, outputs_agree] = 215| 415| do_conversion(from, output_length, inputisvalid); 216| 415| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 415] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 415| return; 221| 415| } 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| 415| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 415| validation_result verify_valid_input(FromSpan src) const { 231| 415| validation_result ret{}; 232| | 233| 415| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 415| const auto implementations = get_supported_implementations(); 235| 415| std::vector results; 236| 415| results.reserve(implementations.size()); 237| | 238| 1.24k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.24k, False: 415] ------------------ 239| 1.24k| results.push_back( 240| 1.24k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.24k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.24k| const bool validation2 = 245| 1.24k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.24k| src.data(), src.size()); 247| 1.24k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.24k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.24k| } 258| | 259| 415| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 415| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 415] ------------------ 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| 415| } else { 273| 415| ret.implementations_agree = true; 274| 415| } 275| 415| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 415| return r.error == simdutf::SUCCESS; 277| 415| }); 278| 415| return ret; 279| 415| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 830| 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| 613| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 613| return r.error == simdutf::SUCCESS; 277| 613| }); _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 415| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 415| length_result ret{}; 344| | 345| 415| const auto implementations = get_supported_implementations(); 346| 415| std::vector results; 347| 415| results.reserve(implementations.size()); 348| | 349| 1.24k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.24k, False: 415] ------------------ 350| 1.24k| const auto len = invoke_lengthcalc(impl, src); 351| 1.24k| results.push_back(len); 352| 1.24k| ret.length.push_back(len); 353| 1.24k| } 354| | 355| 415| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 415| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 415] ------------------ 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| 415| } else { 375| 415| ret.implementations_agree = true; 376| 415| } 377| 415| return ret; 378| 415| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.24k| FromSpan src) const { 331| 1.24k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.24k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 830| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 415| const bool inputisvalid) const { 383| 415| conversion_result ret{}; 384| | 385| 415| const auto implementations = get_supported_implementations(); 386| | 387| 415| std::vector> results; 388| 415| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 415| std::vector> outputbuffers; 393| 415| outputbuffers.reserve(implementations.size()); 394| 1.66k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.24k, False: 415] ------------------ 395| 1.24k| auto impl = implementations[i]; 396| 1.24k| const ToType canary1{42}; 397| 1.24k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.24k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.24k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.24k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.24k| if constexpr (std::is_same_v) { 403| 1.24k| return r != 0; 404| 1.24k| } else { 405| 1.24k| return r.error == simdutf::error_code::SUCCESS; 406| 1.24k| } 407| 1.24k| }(implret1); 408| 1.24k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.24k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.24k| const ToType canary2{25}; 414| 1.24k| const auto outputbuffer_first_run = outputbuffer; 415| 1.24k| std::ranges::fill(outputbuffer, canary2); 416| 1.24k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.24k| src.size(), outputbuffer.data()); 418| | 419| 1.24k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.24k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.24k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 297, False: 948] | Branch (423:29): [True: 297, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 297| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 297| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 297] ------------------ 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| 297| } 440| 1.24k| } 441| 1.24k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 297, False: 948] ------------------ 442| 1.24k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 415| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 316, False: 99] ------------------ 447| 948| for (auto& e : results) { ------------------ | Branch (447:20): [True: 948, False: 316] ------------------ 448| 948| e.outputhash.clear(); 449| 948| } 450| 316| } 451| | 452| 415| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 415| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 415] ------------------ 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| 415| } else { 474| 415| ret.implementations_agree = true; 475| 415| } 476| 415| return ret; 477| 415| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.24k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.24k| } else { 405| 1.24k| return r.error == simdutf::error_code::SUCCESS; 406| 1.24k| } 407| 1.24k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 830| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_34clENSt3__14spanIKcLm18446744073709551615EEE: 554| 258| +[](std::span chardata) { \ 555| 258| const auto c = \ 556| 258| Conversion{ \ 559| 258| &I::lenfunc, &I::conversionfunc, \ 560| 258| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 258|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 258| ::std::void_t(); \ | | 1180| 258| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 258| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 258| constexpr auto _size = _name.size(); \ | | 1183| 258| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 258| return _nameof; }() ------------------ 561| 258| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 258|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 258| ::std::void_t(); \ | | 1180| 258| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 258| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 258| constexpr auto _size = _name.size(); \ | | 1183| 258| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 258| return _nameof; }() ------------------ 562| 258| c.fuzz(chardata); \ 563| 258| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 258| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 258| const FromSpan from{reinterpret_cast(chardata.data()), 178| 258| chardata.size() / sizeof(FromType)}; 179| | 180| 258| static const bool do_print_testcase = 181| 258| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 258| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 258] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 258| do { 189| | // step 0 - is the input valid? 190| 258| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 258| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 258] | 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| 258| From == UtfEncodings::UTF8) { 198| 258| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 258] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 258| } 201| | 202| | // step 2 - what is the required size of the output? 203| 258| const auto [output_length, length_agree] = 204| 258| calculate_length(from, inputisvalid); 205| 258| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 258] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 258| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 135, False: 123] | 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| 258| const auto [written, outputs_agree] = 215| 258| do_conversion(from, output_length, inputisvalid); 216| 258| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 258] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 258| return; 221| 258| } 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| 258| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 258| validation_result verify_valid_input(FromSpan src) const { 231| 258| validation_result ret{}; 232| | 233| 258| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 258| const auto implementations = get_supported_implementations(); 235| 258| std::vector results; 236| 258| results.reserve(implementations.size()); 237| | 238| 774| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 774, False: 258] ------------------ 239| 774| results.push_back( 240| 774| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 774| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 774| const bool validation2 = 245| 774| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 774| src.data(), src.size()); 247| 774| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 774] ------------------ 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| 774| } 258| | 259| 258| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 258| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 258] ------------------ 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| 258| } else { 273| 258| ret.implementations_agree = true; 274| 258| } 275| 258| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 258| return r.error == simdutf::SUCCESS; 277| 258| }); 278| 258| return ret; 279| 258| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 516| 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| 504| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 504| return r.error == simdutf::SUCCESS; 277| 504| }); _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 258| bool count_the_input(FromSpan src) const { 290| 258| const auto implementations = get_supported_implementations(); 291| 258| std::vector results; 292| 258| results.reserve(implementations.size()); 293| | 294| 774| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 774, False: 258] ------------------ 295| 774| 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| 774| } else if constexpr (From == UtfEncodings::UTF8) { 301| 774| ret = impl->count_utf8(src.data(), src.size()); 302| 774| } 303| 774| results.push_back(ret); 304| 774| } 305| 258| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 258| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 258] ------------------ 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| 258| return true; 321| 258| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 516| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 258| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 258| length_result ret{}; 344| | 345| 258| const auto implementations = get_supported_implementations(); 346| 258| std::vector results; 347| 258| results.reserve(implementations.size()); 348| | 349| 774| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 774, False: 258] ------------------ 350| 774| const auto len = invoke_lengthcalc(impl, src); 351| 774| results.push_back(len); 352| 774| ret.length.push_back(len); 353| 774| } 354| | 355| 258| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 258| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 258] ------------------ 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| 258| } else { 375| 258| ret.implementations_agree = true; 376| 258| } 377| 258| return ret; 378| 258| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 774| FromSpan src) const { 331| 774| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 774| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 516| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 258| const bool inputisvalid) const { 383| 258| conversion_result ret{}; 384| | 385| 258| const auto implementations = get_supported_implementations(); 386| | 387| 258| std::vector> results; 388| 258| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 258| std::vector> outputbuffers; 393| 258| outputbuffers.reserve(implementations.size()); 394| 1.03k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 774, False: 258] ------------------ 395| 774| auto impl = implementations[i]; 396| 774| const ToType canary1{42}; 397| 774| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 774| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 774| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 774| const auto success = [](const ConversionResult& r) -> bool { 402| 774| if constexpr (std::is_same_v) { 403| 774| return r != 0; 404| 774| } else { 405| 774| return r.error == simdutf::error_code::SUCCESS; 406| 774| } 407| 774| }(implret1); 408| 774| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 774| 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| 774| const ToType canary2{25}; 414| 774| const auto outputbuffer_first_run = outputbuffer; 415| 774| std::ranges::fill(outputbuffer, canary2); 416| 774| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 774| src.size(), outputbuffer.data()); 418| | 419| 774| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 774] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 774| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 369, False: 405] | Branch (423:29): [True: 345, False: 24] ------------------ 424| | // only care about the output if the input is valid 425| 345| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 345| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 345] ------------------ 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| 345| } 440| 774| } 441| 774| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 345, False: 429] ------------------ 442| 774| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 258| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 135, False: 123] ------------------ 447| 405| for (auto& e : results) { ------------------ | Branch (447:20): [True: 405, False: 135] ------------------ 448| 405| e.outputhash.clear(); 449| 405| } 450| 135| } 451| | 452| 258| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 258| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 258] ------------------ 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| 258| } else { 474| 258| ret.implementations_agree = true; 475| 258| } 476| 258| return ret; 477| 258| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 774| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 774| } else { 405| 774| return r.error == simdutf::error_code::SUCCESS; 406| 774| } 407| 774| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 516| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_35clENSt3__14spanIKcLm18446744073709551615EEE: 554| 351| +[](std::span chardata) { \ 555| 351| const auto c = \ 556| 351| Conversion{ \ 559| 351| &I::lenfunc, &I::conversionfunc, \ 560| 351| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 351|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 351| ::std::void_t(); \ | | 1180| 351| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 351| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 351| constexpr auto _size = _name.size(); \ | | 1183| 351| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 351| return _nameof; }() ------------------ 561| 351| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 351|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 351| ::std::void_t(); \ | | 1180| 351| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 351| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 351| constexpr auto _size = _name.size(); \ | | 1183| 351| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 351| return _nameof; }() ------------------ 562| 351| c.fuzz(chardata); \ 563| 351| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 351| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 351| const FromSpan from{reinterpret_cast(chardata.data()), 178| 351| chardata.size() / sizeof(FromType)}; 179| | 180| 351| static const bool do_print_testcase = 181| 351| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 351| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 351] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 351| do { 189| | // step 0 - is the input valid? 190| 351| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 351| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 351] | 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| 351| From == UtfEncodings::UTF8) { 198| 351| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 351] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 351| } 201| | 202| | // step 2 - what is the required size of the output? 203| 351| const auto [output_length, length_agree] = 204| 351| calculate_length(from, inputisvalid); 205| 351| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 351] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 351| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 194, False: 157] | Branch (208:28): [True: 0, False: 194] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 351| const auto [written, outputs_agree] = 215| 351| do_conversion(from, output_length, inputisvalid); 216| 351| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 351] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 351| return; 221| 351| } 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| 351| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 351| validation_result verify_valid_input(FromSpan src) const { 231| 351| validation_result ret{}; 232| | 233| 351| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 351| const auto implementations = get_supported_implementations(); 235| 351| std::vector results; 236| 351| results.reserve(implementations.size()); 237| | 238| 1.05k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.05k, False: 351] ------------------ 239| 1.05k| results.push_back( 240| 1.05k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.05k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.05k| const bool validation2 = 245| 1.05k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.05k| src.data(), src.size()); 247| 1.05k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.05k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.05k| } 258| | 259| 351| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 351| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 351] ------------------ 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| 351| } else { 273| 351| ret.implementations_agree = true; 274| 351| } 275| 351| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 351| return r.error == simdutf::SUCCESS; 277| 351| }); 278| 351| return ret; 279| 351| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 702| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 665| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 665| return r.error == simdutf::SUCCESS; 277| 665| }); _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 351| bool count_the_input(FromSpan src) const { 290| 351| const auto implementations = get_supported_implementations(); 291| 351| std::vector results; 292| 351| results.reserve(implementations.size()); 293| | 294| 1.05k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.05k, False: 351] ------------------ 295| 1.05k| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| 1.05k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.05k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.05k| } 303| 1.05k| results.push_back(ret); 304| 1.05k| } 305| 351| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 351| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 351] ------------------ 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| 351| return true; 321| 351| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 702| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 351| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 351| length_result ret{}; 344| | 345| 351| const auto implementations = get_supported_implementations(); 346| 351| std::vector results; 347| 351| results.reserve(implementations.size()); 348| | 349| 1.05k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.05k, False: 351] ------------------ 350| 1.05k| const auto len = invoke_lengthcalc(impl, src); 351| 1.05k| results.push_back(len); 352| 1.05k| ret.length.push_back(len); 353| 1.05k| } 354| | 355| 351| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 351| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 351] ------------------ 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| 351| } else { 375| 351| ret.implementations_agree = true; 376| 351| } 377| 351| return ret; 378| 351| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.05k| FromSpan src) const { 331| 1.05k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.05k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 702| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 351| const bool inputisvalid) const { 383| 351| conversion_result ret{}; 384| | 385| 351| const auto implementations = get_supported_implementations(); 386| | 387| 351| std::vector> results; 388| 351| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 351| std::vector> outputbuffers; 393| 351| outputbuffers.reserve(implementations.size()); 394| 1.40k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.05k, False: 351] ------------------ 395| 1.05k| auto impl = implementations[i]; 396| 1.05k| const ToType canary1{42}; 397| 1.05k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.05k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.05k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.05k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.05k| if constexpr (std::is_same_v) { 403| 1.05k| return r != 0; 404| 1.05k| } else { 405| 1.05k| return r.error == simdutf::error_code::SUCCESS; 406| 1.05k| } 407| 1.05k| }(implret1); 408| 1.05k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.05k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.05k| const ToType canary2{25}; 414| 1.05k| const auto outputbuffer_first_run = outputbuffer; 415| 1.05k| std::ranges::fill(outputbuffer, canary2); 416| 1.05k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.05k| src.size(), outputbuffer.data()); 418| | 419| 1.05k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.05k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.05k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 471, False: 582] | Branch (423:29): [True: 471, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 471| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 471| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 471] ------------------ 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| 471| } 440| 1.05k| } 441| 1.05k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 471, False: 582] ------------------ 442| 1.05k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 351| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 194, False: 157] ------------------ 447| 582| for (auto& e : results) { ------------------ | Branch (447:20): [True: 582, False: 194] ------------------ 448| 582| e.outputhash.clear(); 449| 582| } 450| 194| } 451| | 452| 351| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 351| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 351] ------------------ 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| 351| } else { 474| 351| ret.implementations_agree = true; 475| 351| } 476| 351| return ret; 477| 351| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.05k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.05k| } else { 405| 1.05k| return r.error == simdutf::error_code::SUCCESS; 406| 1.05k| } 407| 1.05k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 702| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_36clENSt3__14spanIKcLm18446744073709551615EEE: 554| 287| +[](std::span chardata) { \ 555| 287| const auto c = \ 556| 287| Conversion{ \ 559| 287| &I::lenfunc, &I::conversionfunc, \ 560| 287| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 287|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 287| ::std::void_t(); \ | | 1180| 287| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 287| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 287| constexpr auto _size = _name.size(); \ | | 1183| 287| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 287| return _nameof; }() ------------------ 561| 287| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 287|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 287| ::std::void_t(); \ | | 1180| 287| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 287| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 287| constexpr auto _size = _name.size(); \ | | 1183| 287| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 287| return _nameof; }() ------------------ 562| 287| c.fuzz(chardata); \ 563| 287| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 287| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 287| const FromSpan from{reinterpret_cast(chardata.data()), 178| 287| chardata.size() / sizeof(FromType)}; 179| | 180| 287| static const bool do_print_testcase = 181| 287| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 287| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 287] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 287| do { 189| | // step 0 - is the input valid? 190| 287| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 287| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 287] | 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| 287| From == UtfEncodings::UTF8) { 198| 287| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 287] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 287| } 201| | 202| | // step 2 - what is the required size of the output? 203| 287| const auto [output_length, length_agree] = 204| 287| calculate_length(from, inputisvalid); 205| 287| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 287] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 287| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 164, False: 123] | Branch (208:28): [True: 0, False: 164] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 287| const auto [written, outputs_agree] = 215| 287| do_conversion(from, output_length, inputisvalid); 216| 287| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 287] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 287| return; 221| 287| } 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| 287| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 287| validation_result verify_valid_input(FromSpan src) const { 231| 287| validation_result ret{}; 232| | 233| 287| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 287| const auto implementations = get_supported_implementations(); 235| 287| std::vector results; 236| 287| results.reserve(implementations.size()); 237| | 238| 861| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 861, False: 287] ------------------ 239| 861| results.push_back( 240| 861| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 861| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 861| const bool validation2 = 245| 861| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 861| src.data(), src.size()); 247| 861| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 861] ------------------ 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| 861| } 258| | 259| 287| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 287| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 287] ------------------ 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| 287| } else { 273| 287| ret.implementations_agree = true; 274| 287| } 275| 287| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 287| return r.error == simdutf::SUCCESS; 277| 287| }); 278| 287| return ret; 279| 287| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 574| 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| 533| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 533| return r.error == simdutf::SUCCESS; 277| 533| }); _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 287| bool count_the_input(FromSpan src) const { 290| 287| const auto implementations = get_supported_implementations(); 291| 287| std::vector results; 292| 287| results.reserve(implementations.size()); 293| | 294| 861| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 861, False: 287] ------------------ 295| 861| 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| 861| } else if constexpr (From == UtfEncodings::UTF8) { 301| 861| ret = impl->count_utf8(src.data(), src.size()); 302| 861| } 303| 861| results.push_back(ret); 304| 861| } 305| 287| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 287| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 287] ------------------ 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| 287| return true; 321| 287| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 574| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 287| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 287| length_result ret{}; 344| | 345| 287| const auto implementations = get_supported_implementations(); 346| 287| std::vector results; 347| 287| results.reserve(implementations.size()); 348| | 349| 861| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 861, False: 287] ------------------ 350| 861| const auto len = invoke_lengthcalc(impl, src); 351| 861| results.push_back(len); 352| 861| ret.length.push_back(len); 353| 861| } 354| | 355| 287| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 287| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 287] ------------------ 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| 287| } else { 375| 287| ret.implementations_agree = true; 376| 287| } 377| 287| return ret; 378| 287| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 861| FromSpan src) const { 331| 861| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 861| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 574| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 287| const bool inputisvalid) const { 383| 287| conversion_result ret{}; 384| | 385| 287| const auto implementations = get_supported_implementations(); 386| | 387| 287| std::vector> results; 388| 287| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 287| std::vector> outputbuffers; 393| 287| outputbuffers.reserve(implementations.size()); 394| 1.14k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 861, False: 287] ------------------ 395| 861| auto impl = implementations[i]; 396| 861| const ToType canary1{42}; 397| 861| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 861| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 861| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 861| const auto success = [](const ConversionResult& r) -> bool { 402| 861| if constexpr (std::is_same_v) { 403| 861| return r != 0; 404| 861| } else { 405| 861| return r.error == simdutf::error_code::SUCCESS; 406| 861| } 407| 861| }(implret1); 408| 861| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 861| 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| 861| const ToType canary2{25}; 414| 861| const auto outputbuffer_first_run = outputbuffer; 415| 861| std::ranges::fill(outputbuffer, canary2); 416| 861| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 861| src.size(), outputbuffer.data()); 418| | 419| 861| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 861] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 861| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 369, False: 492] | Branch (423:29): [True: 369, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 369| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 369| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 369] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 369| } 440| 861| } 441| 861| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 369, False: 492] ------------------ 442| 861| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 287| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 164, False: 123] ------------------ 447| 492| for (auto& e : results) { ------------------ | Branch (447:20): [True: 492, False: 164] ------------------ 448| 492| e.outputhash.clear(); 449| 492| } 450| 164| } 451| | 452| 287| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 287| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 287] ------------------ 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| 287| } else { 474| 287| ret.implementations_agree = true; 475| 287| } 476| 287| return ret; 477| 287| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 861| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 861| } else { 405| 861| return r.error == simdutf::error_code::SUCCESS; 406| 861| } 407| 861| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 574| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_37clENSt3__14spanIKcLm18446744073709551615EEE: 554| 369| +[](std::span chardata) { \ 555| 369| const auto c = \ 556| 369| Conversion{ \ 559| 369| &I::lenfunc, &I::conversionfunc, \ 560| 369| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 369|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 369| ::std::void_t(); \ | | 1180| 369| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 369| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 369| constexpr auto _size = _name.size(); \ | | 1183| 369| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 369| return _nameof; }() ------------------ 561| 369| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 369|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 369| ::std::void_t(); \ | | 1180| 369| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 369| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 369| constexpr auto _size = _name.size(); \ | | 1183| 369| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 369| return _nameof; }() ------------------ 562| 369| c.fuzz(chardata); \ 563| 369| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 369| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 369| const FromSpan from{reinterpret_cast(chardata.data()), 178| 369| chardata.size() / sizeof(FromType)}; 179| | 180| 369| static const bool do_print_testcase = 181| 369| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 369| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 369] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 369| do { 189| | // step 0 - is the input valid? 190| 369| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 369| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 369] | 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| 369| From == UtfEncodings::UTF8) { 198| 369| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 369] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 369| } 201| | 202| | // step 2 - what is the required size of the output? 203| 369| const auto [output_length, length_agree] = 204| 369| calculate_length(from, inputisvalid); 205| 369| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 369] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 369| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 199, False: 170] | Branch (208:28): [True: 0, False: 199] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 369| const auto [written, outputs_agree] = 215| 369| do_conversion(from, output_length, inputisvalid); 216| 369| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 369] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 369| return; 221| 369| } 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| 369| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 369| validation_result verify_valid_input(FromSpan src) const { 231| 369| validation_result ret{}; 232| | 233| 369| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 369| const auto implementations = get_supported_implementations(); 235| 369| std::vector results; 236| 369| results.reserve(implementations.size()); 237| | 238| 1.10k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.10k, False: 369] ------------------ 239| 1.10k| results.push_back( 240| 1.10k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.10k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.10k| const bool validation2 = 245| 1.10k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.10k| src.data(), src.size()); 247| 1.10k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.10k] ------------------ 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| 1.10k| } 258| | 259| 369| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 369| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 369] ------------------ 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| 369| } else { 273| 369| ret.implementations_agree = true; 274| 369| } 275| 369| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 369| return r.error == simdutf::SUCCESS; 277| 369| }); 278| 369| return ret; 279| 369| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 738| 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| 709| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 709| return r.error == simdutf::SUCCESS; 277| 709| }); _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 369| bool count_the_input(FromSpan src) const { 290| 369| const auto implementations = get_supported_implementations(); 291| 369| std::vector results; 292| 369| results.reserve(implementations.size()); 293| | 294| 1.10k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.10k, False: 369] ------------------ 295| 1.10k| 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| 1.10k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.10k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.10k| } 303| 1.10k| results.push_back(ret); 304| 1.10k| } 305| 369| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 369| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 369] ------------------ 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| 369| return true; 321| 369| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 738| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 369| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 369| length_result ret{}; 344| | 345| 369| const auto implementations = get_supported_implementations(); 346| 369| std::vector results; 347| 369| results.reserve(implementations.size()); 348| | 349| 1.10k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.10k, False: 369] ------------------ 350| 1.10k| const auto len = invoke_lengthcalc(impl, src); 351| 1.10k| results.push_back(len); 352| 1.10k| ret.length.push_back(len); 353| 1.10k| } 354| | 355| 369| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 369| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 369] ------------------ 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| 369| } else { 375| 369| ret.implementations_agree = true; 376| 369| } 377| 369| return ret; 378| 369| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.10k| FromSpan src) const { 331| 1.10k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.10k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 738| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 369| const bool inputisvalid) const { 383| 369| conversion_result ret{}; 384| | 385| 369| const auto implementations = get_supported_implementations(); 386| | 387| 369| std::vector> results; 388| 369| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 369| std::vector> outputbuffers; 393| 369| outputbuffers.reserve(implementations.size()); 394| 1.47k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.10k, False: 369] ------------------ 395| 1.10k| auto impl = implementations[i]; 396| 1.10k| const ToType canary1{42}; 397| 1.10k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.10k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.10k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.10k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.10k| if constexpr (std::is_same_v) { 403| 1.10k| return r != 0; 404| 1.10k| } else { 405| 1.10k| return r.error == simdutf::error_code::SUCCESS; 406| 1.10k| } 407| 1.10k| }(implret1); 408| 1.10k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.10k| 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| 1.10k| const ToType canary2{25}; 414| 1.10k| const auto outputbuffer_first_run = outputbuffer; 415| 1.10k| std::ranges::fill(outputbuffer, canary2); 416| 1.10k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.10k| src.size(), outputbuffer.data()); 418| | 419| 1.10k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.10k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.10k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 510, False: 597] | Branch (423:29): [True: 510, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 510| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 510| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 510] ------------------ 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| 510| } 440| 1.10k| } 441| 1.10k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 510, False: 597] ------------------ 442| 1.10k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 369| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 199, False: 170] ------------------ 447| 597| for (auto& e : results) { ------------------ | Branch (447:20): [True: 597, False: 199] ------------------ 448| 597| e.outputhash.clear(); 449| 597| } 450| 199| } 451| | 452| 369| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 369| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 369] ------------------ 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| 369| } else { 474| 369| ret.implementations_agree = true; 475| 369| } 476| 369| return ret; 477| 369| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.10k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.10k| } else { 405| 1.10k| return r.error == simdutf::error_code::SUCCESS; 406| 1.10k| } 407| 1.10k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 738| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_38clENSt3__14spanIKcLm18446744073709551615EEE: 554| 45| +[](std::span chardata) { \ 555| 45| const auto c = \ 556| 45| Conversion{ \ 559| 45| &I::lenfunc, &I::conversionfunc, \ 560| 45| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 45|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 45| ::std::void_t(); \ | | 1180| 45| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 45| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 45| constexpr auto _size = _name.size(); \ | | 1183| 45| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 45| return _nameof; }() ------------------ 561| 45| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 45|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 45| ::std::void_t(); \ | | 1180| 45| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 45| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 45| constexpr auto _size = _name.size(); \ | | 1183| 45| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 45| return _nameof; }() ------------------ 562| 45| c.fuzz(chardata); \ 563| 45| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE4fuzzENSt3__14spanIS5_Lm18446744073709551615EEE: 175| 45| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 45| const FromSpan from{reinterpret_cast(chardata.data()), 178| 45| chardata.size() / sizeof(FromType)}; 179| | 180| 45| static const bool do_print_testcase = 181| 45| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 45| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 45] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 45| do { 189| | // step 0 - is the input valid? 190| 45| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 45| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 45] | 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| 45| const auto [output_length, length_agree] = 204| 45| calculate_length(from, inputisvalid); 205| 45| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 45] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 45| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 45] | 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| 45| const auto [written, outputs_agree] = 215| 45| do_conversion(from, output_length, inputisvalid); 216| 45| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 45] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 45| return; 221| 45| } 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| 45| } _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 283| 45| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 45| return validation_result{.valid = true, .implementations_agree = true}; 287| 45| } _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 45| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 45| length_result ret{}; 344| | 345| 45| const auto implementations = get_supported_implementations(); 346| 45| std::vector results; 347| 45| results.reserve(implementations.size()); 348| | 349| 135| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 135, False: 45] ------------------ 350| 135| const auto len = invoke_lengthcalc(impl, src); 351| 135| results.push_back(len); 352| 135| ret.length.push_back(len); 353| 135| } 354| | 355| 45| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 45| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 45] ------------------ 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| 45| } else { 375| 45| ret.implementations_agree = true; 376| 45| } 377| 45| return ret; 378| 45| } _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 135| FromSpan src) const { 339| 135| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 135| } _ZZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 90| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 45| const bool inputisvalid) const { 383| 45| conversion_result ret{}; 384| | 385| 45| const auto implementations = get_supported_implementations(); 386| | 387| 45| std::vector> results; 388| 45| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 45| std::vector> outputbuffers; 393| 45| outputbuffers.reserve(implementations.size()); 394| 180| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 135, False: 45] ------------------ 395| 135| auto impl = implementations[i]; 396| 135| const ToType canary1{42}; 397| 135| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 135| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 135| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 135| const auto success = [](const ConversionResult& r) -> bool { 402| 135| if constexpr (std::is_same_v) { 403| 135| return r != 0; 404| 135| } else { 405| 135| return r.error == simdutf::error_code::SUCCESS; 406| 135| } 407| 135| }(implret1); 408| 135| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 135| 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| 135| const ToType canary2{25}; 414| 135| const auto outputbuffer_first_run = outputbuffer; 415| 135| std::ranges::fill(outputbuffer, canary2); 416| 135| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 135| src.size(), outputbuffer.data()); 418| | 419| 135| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 135] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 135| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 135, False: 0] | Branch (423:29): [True: 129, False: 6] ------------------ 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| 135| } 441| 135| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 129, False: 6] ------------------ 442| 135| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 45| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 45] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 45| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 45| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 45] ------------------ 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| 45| } else { 474| 45| ret.implementations_agree = true; 475| 45| } 476| 45| return ret; 477| 45| } _ZZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 135| const auto success = [](const ConversionResult& r) -> bool { 402| 135| if constexpr (std::is_same_v) { 403| 135| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 135| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 90| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_39clENSt3__14spanIKcLm18446744073709551615EEE: 554| 40| +[](std::span chardata) { \ 555| 40| const auto c = \ 556| 40| Conversion{ \ 559| 40| &I::lenfunc, &I::conversionfunc, \ 560| 40| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 40|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 40| ::std::void_t(); \ | | 1180| 40| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 40| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 40| constexpr auto _size = _name.size(); \ | | 1183| 40| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 40| return _nameof; }() ------------------ 561| 40| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 40|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 40| ::std::void_t(); \ | | 1180| 40| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 40| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 40| constexpr auto _size = _name.size(); \ | | 1183| 40| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 40| return _nameof; }() ------------------ 562| 40| c.fuzz(chardata); \ 563| 40| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE4fuzzENSt3__14spanIS5_Lm18446744073709551615EEE: 175| 40| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 40| const FromSpan from{reinterpret_cast(chardata.data()), 178| 40| chardata.size() / sizeof(FromType)}; 179| | 180| 40| static const bool do_print_testcase = 181| 40| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 40| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 40] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 40| do { 189| | // step 0 - is the input valid? 190| 40| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 40| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 40] | 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| 40| const auto [output_length, length_agree] = 204| 40| calculate_length(from, inputisvalid); 205| 40| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 40] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 40| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 40] | 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| 40| const auto [written, outputs_agree] = 215| 40| do_conversion(from, output_length, inputisvalid); 216| 40| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 40] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 40| return; 221| 40| } 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| 40| } _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 283| 40| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 40| return validation_result{.valid = true, .implementations_agree = true}; 287| 40| } _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 40| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 40| length_result ret{}; 344| | 345| 40| const auto implementations = get_supported_implementations(); 346| 40| std::vector results; 347| 40| results.reserve(implementations.size()); 348| | 349| 120| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 120, False: 40] ------------------ 350| 120| const auto len = invoke_lengthcalc(impl, src); 351| 120| results.push_back(len); 352| 120| ret.length.push_back(len); 353| 120| } 354| | 355| 40| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 40| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 40] ------------------ 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| 40| } else { 375| 40| ret.implementations_agree = true; 376| 40| } 377| 40| return ret; 378| 40| } _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 120| FromSpan src) const { 339| 120| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 120| } _ZZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 80| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 40| const bool inputisvalid) const { 383| 40| conversion_result ret{}; 384| | 385| 40| const auto implementations = get_supported_implementations(); 386| | 387| 40| std::vector> results; 388| 40| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 40| std::vector> outputbuffers; 393| 40| outputbuffers.reserve(implementations.size()); 394| 160| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 120, False: 40] ------------------ 395| 120| auto impl = implementations[i]; 396| 120| const ToType canary1{42}; 397| 120| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 120| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 120| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 120| const auto success = [](const ConversionResult& r) -> bool { 402| 120| if constexpr (std::is_same_v) { 403| 120| return r != 0; 404| 120| } else { 405| 120| return r.error == simdutf::error_code::SUCCESS; 406| 120| } 407| 120| }(implret1); 408| 120| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 120| 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| 120| const ToType canary2{25}; 414| 120| const auto outputbuffer_first_run = outputbuffer; 415| 120| std::ranges::fill(outputbuffer, canary2); 416| 120| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 120| src.size(), outputbuffer.data()); 418| | 419| 120| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 120] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 120| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 120, False: 0] | Branch (423:29): [True: 114, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 114| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 114| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 114] ------------------ 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| 114| } 440| 120| } 441| 120| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 114, False: 6] ------------------ 442| 120| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 40| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 40] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 40| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 40| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 40] ------------------ 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| 40| } else { 474| 40| ret.implementations_agree = true; 475| 40| } 476| 40| return ret; 477| 40| } _ZZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 120| const auto success = [](const ConversionResult& r) -> bool { 402| 120| if constexpr (std::is_same_v) { 403| 120| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 120| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 80| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_40clENSt3__14spanIKcLm18446744073709551615EEE: 554| 40| +[](std::span chardata) { \ 555| 40| const auto c = \ 556| 40| Conversion{ \ 559| 40| &I::lenfunc, &I::conversionfunc, \ 560| 40| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 40|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 40| ::std::void_t(); \ | | 1180| 40| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 40| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 40| constexpr auto _size = _name.size(); \ | | 1183| 40| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 40| return _nameof; }() ------------------ 561| 40| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 40|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 40| ::std::void_t(); \ | | 1180| 40| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 40| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 40| constexpr auto _size = _name.size(); \ | | 1183| 40| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 40| return _nameof; }() ------------------ 562| 40| c.fuzz(chardata); \ 563| 40| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE4fuzzENSt3__14spanIS5_Lm18446744073709551615EEE: 175| 40| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 40| const FromSpan from{reinterpret_cast(chardata.data()), 178| 40| chardata.size() / sizeof(FromType)}; 179| | 180| 40| static const bool do_print_testcase = 181| 40| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 40| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 40] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 40| do { 189| | // step 0 - is the input valid? 190| 40| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 40| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 40] | 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| 40| const auto [output_length, length_agree] = 204| 40| calculate_length(from, inputisvalid); 205| 40| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 40] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 40| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 40] | 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| 40| const auto [written, outputs_agree] = 215| 40| do_conversion(from, output_length, inputisvalid); 216| 40| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 40] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 40| return; 221| 40| } 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| 40| } _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 283| 40| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 40| return validation_result{.valid = true, .implementations_agree = true}; 287| 40| } _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 40| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 40| length_result ret{}; 344| | 345| 40| const auto implementations = get_supported_implementations(); 346| 40| std::vector results; 347| 40| results.reserve(implementations.size()); 348| | 349| 120| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 120, False: 40] ------------------ 350| 120| const auto len = invoke_lengthcalc(impl, src); 351| 120| results.push_back(len); 352| 120| ret.length.push_back(len); 353| 120| } 354| | 355| 40| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 40| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 40] ------------------ 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| 40| } else { 375| 40| ret.implementations_agree = true; 376| 40| } 377| 40| return ret; 378| 40| } _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 120| FromSpan src) const { 339| 120| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 120| } _ZZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 80| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 40| const bool inputisvalid) const { 383| 40| conversion_result ret{}; 384| | 385| 40| const auto implementations = get_supported_implementations(); 386| | 387| 40| std::vector> results; 388| 40| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 40| std::vector> outputbuffers; 393| 40| outputbuffers.reserve(implementations.size()); 394| 160| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 120, False: 40] ------------------ 395| 120| auto impl = implementations[i]; 396| 120| const ToType canary1{42}; 397| 120| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 120| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 120| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 120| const auto success = [](const ConversionResult& r) -> bool { 402| 120| if constexpr (std::is_same_v) { 403| 120| return r != 0; 404| 120| } else { 405| 120| return r.error == simdutf::error_code::SUCCESS; 406| 120| } 407| 120| }(implret1); 408| 120| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 120| 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| 120| const ToType canary2{25}; 414| 120| const auto outputbuffer_first_run = outputbuffer; 415| 120| std::ranges::fill(outputbuffer, canary2); 416| 120| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 120| src.size(), outputbuffer.data()); 418| | 419| 120| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 120] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 120| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 120, False: 0] | Branch (423:29): [True: 114, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 114| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 114| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 114] ------------------ 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| 114| } 440| 120| } 441| 120| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 114, False: 6] ------------------ 442| 120| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 40| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 40] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 40| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 40| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 40] ------------------ 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| 40| } else { 474| 40| ret.implementations_agree = true; 475| 40| } 476| 40| return ret; 477| 40| } _ZZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 120| const auto success = [](const ConversionResult& r) -> bool { 402| 120| if constexpr (std::is_same_v) { 403| 120| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 120| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 80| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_41clENSt3__14spanIKcLm18446744073709551615EEE: 554| 269| +[](std::span chardata) { \ 555| 269| const auto c = \ 556| 269| Conversion{ \ 559| 269| &I::lenfunc, &I::conversionfunc, \ 560| 269| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 269|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 269| ::std::void_t(); \ | | 1180| 269| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 269| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 269| constexpr auto _size = _name.size(); \ | | 1183| 269| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 269| return _nameof; }() ------------------ 561| 269| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 269|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 269| ::std::void_t(); \ | | 1180| 269| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 269| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 269| constexpr auto _size = _name.size(); \ | | 1183| 269| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 269| return _nameof; }() ------------------ 562| 269| c.fuzz(chardata); \ 563| 269| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 269| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 269| const FromSpan from{reinterpret_cast(chardata.data()), 178| 269| chardata.size() / sizeof(FromType)}; 179| | 180| 269| static const bool do_print_testcase = 181| 269| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 269| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 269] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 269| do { 189| | // step 0 - is the input valid? 190| 269| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 269| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 269] | 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| 269| const auto [output_length, length_agree] = 204| 269| calculate_length(from, inputisvalid); 205| 269| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 269] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 269| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 269] | 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| 269| const auto [written, outputs_agree] = 215| 269| do_conversion(from, output_length, inputisvalid); 216| 269| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 269] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 269| return; 221| 269| } 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| 269| } _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 283| 269| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 269| return validation_result{.valid = true, .implementations_agree = true}; 287| 269| } _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 269| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 269| length_result ret{}; 344| | 345| 269| const auto implementations = get_supported_implementations(); 346| 269| std::vector results; 347| 269| results.reserve(implementations.size()); 348| | 349| 807| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 807, False: 269] ------------------ 350| 807| const auto len = invoke_lengthcalc(impl, src); 351| 807| results.push_back(len); 352| 807| ret.length.push_back(len); 353| 807| } 354| | 355| 269| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 269| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 269] ------------------ 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| 269| } else { 375| 269| ret.implementations_agree = true; 376| 269| } 377| 269| return ret; 378| 269| } _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 807| FromSpan src) const { 331| 807| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 807| } _ZZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 538| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 269| const bool inputisvalid) const { 383| 269| conversion_result ret{}; 384| | 385| 269| const auto implementations = get_supported_implementations(); 386| | 387| 269| std::vector> results; 388| 269| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 269| std::vector> outputbuffers; 393| 269| outputbuffers.reserve(implementations.size()); 394| 1.07k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 807, False: 269] ------------------ 395| 807| auto impl = implementations[i]; 396| 807| const ToType canary1{42}; 397| 807| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 807| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 807| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 807| const auto success = [](const ConversionResult& r) -> bool { 402| 807| if constexpr (std::is_same_v) { 403| 807| return r != 0; 404| 807| } else { 405| 807| return r.error == simdutf::error_code::SUCCESS; 406| 807| } 407| 807| }(implret1); 408| 807| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 807| 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| 807| const ToType canary2{25}; 414| 807| const auto outputbuffer_first_run = outputbuffer; 415| 807| std::ranges::fill(outputbuffer, canary2); 416| 807| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 807| src.size(), outputbuffer.data()); 418| | 419| 807| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 807] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 807| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 807, False: 0] | Branch (423:29): [True: 801, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 801| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 801| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 801] ------------------ 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| 801| } 440| 807| } 441| 807| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 801, False: 6] ------------------ 442| 807| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 269| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 269] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 269| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 269| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 269] ------------------ 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| 269| } else { 474| 269| ret.implementations_agree = true; 475| 269| } 476| 269| return ret; 477| 269| } _ZZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 807| const auto success = [](const ConversionResult& r) -> bool { 402| 807| if constexpr (std::is_same_v) { 403| 807| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 807| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 538| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZN10FNV1A_hash14fnv1ahash_implENSt3__14spanIKcLm18446744073709551615EEE: 75| 16.7k| static constexpr std::uint64_t fnv1ahash_impl(std::span bytes) { 76| 16.7k| auto hash = offset; 77| | 78| 624M| for (auto byte : bytes) { ------------------ | Branch (78:20): [True: 624M, False: 16.7k] ------------------ 79| 624M| hash ^= static_cast(byte); 80| 624M| hash *= prime; 81| 624M| } 82| | 83| 16.7k| return hash; 84| 16.7k| } _ZN10FNV1A_hash14fnv1ahash_implIKDsLm18446744073709551615EQaaaa11is_hashableIT_Entsr3stdE9is_same_vIS2_cEntsr3stdE9is_same_vIS2_hEEEmNSt3__14spanIS2_XT0_EEE: 89| 13.6k| static constexpr std::uint64_t fnv1ahash_impl(std::span data) { 90| 13.6k| return fnv1ahash_impl({reinterpret_cast(data.data()), 91| 13.6k| data.size_bytes()}); 92| 13.6k| } _ZN10FNV1A_hash14fnv1ahash_implENSt3__14spanIKhLm18446744073709551615EEE: 65| 21.5k| fnv1ahash_impl(std::span bytes) { 66| 21.5k| auto hash = offset; 67| | 68| 2.04G| for (std::uint64_t byte : bytes) { ------------------ | Branch (68:29): [True: 2.04G, False: 21.5k] ------------------ 69| 2.04G| hash ^= byte; 70| 2.04G| hash *= prime; 71| 2.04G| } 72| | 73| 21.5k| return hash; 74| 21.5k| } _Z29get_supported_implementationsv: 26| 31.0k|get_supported_implementations() { 27| 31.0k| static const auto impl = []() -> auto { 28| 31.0k| std::vector ret; 29| 31.0k| for (auto e : simdutf::get_available_implementations()) { 30| 31.0k| std::cerr << "implementation " << e->name() << " is available? " 31| 31.0k| << e->supported_by_runtime_system() << '\n'; 32| 31.0k| if (e->supported_by_runtime_system()) { 33| 31.0k| ret.push_back(e); 34| 31.0k| } 35| 31.0k| } 36| 31.0k| return ret; 37| 31.0k| }(); 38| 31.0k| return {impl.data(), impl.size()}; 39| 31.0k|} _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| 27.6k|inline bool operator!=(const simdutf::result& a, const simdutf::result& b) { 43| 27.6k| return a.count != b.count || a.error != b.error; ------------------ | Branch (43:10): [True: 0, False: 27.6k] | Branch (43:32): [True: 0, False: 27.6k] ------------------ 44| 27.6k|} _ZN10FNV1A_hash6as_strIJNSt3__16vectorIDiNS1_9allocatorIDiEEEEEEENS1_12basic_stringIcNS1_11char_traitsIcEENS3_IcEEEEDpRKT_: 100| 7.88k| template static std::string as_str(const Data&... data) { 101| 7.88k| static_assert(sizeof...(Data) > 0, "must hash with at least one argument"); 102| 7.88k| 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| 7.88k| } else { 108| 7.88k| h = fnv1ahash_impl(data...); 109| 7.88k| } 110| 7.88k| constexpr std::size_t expected_chars = 16; 111| 7.88k| std::string ret(expected_chars, '0'); 112| 7.88k| auto c = std::to_chars(ret.data(), ret.data() + ret.size(), h, 16); 113| 7.88k| assert(c.ec == std::errc{}); ------------------ | Branch (113:5): [True: 7.88k, False: 0] ------------------ 114| 7.88k| auto nwritten = c.ptr - ret.data(); 115| 7.88k| assert(nwritten <= expected_chars); ------------------ | Branch (115:5): [True: 7.88k, False: 0] ------------------ 116| 7.88k| std::rotate(ret.data(), c.ptr, ret.data() + expected_chars); 117| 7.88k| return ret; 118| 7.88k| } _ZN10FNV1A_hash14fnv1ahash_implIDiQ11is_hashableIT_EEEmRKNSt3__16vectorIS1_NS2_9allocatorIS1_EEEE: 96| 7.88k| static constexpr std::uint64_t fnv1ahash_impl(const std::vector& data) { 97| 7.88k| return fnv1ahash_impl(std::span(data)); 98| 7.88k| } _ZN10FNV1A_hash14fnv1ahash_implIKDiLm18446744073709551615EQaaaa11is_hashableIT_Entsr3stdE9is_same_vIS2_cEntsr3stdE9is_same_vIS2_hEEEmNSt3__14spanIS2_XT0_EEE: 89| 7.88k| static constexpr std::uint64_t fnv1ahash_impl(std::span data) { 90| 7.88k| return fnv1ahash_impl({reinterpret_cast(data.data()), 91| 7.88k| data.size_bytes()}); 92| 7.88k| } _ZN10FNV1A_hash6as_strIJNSt3__16vectorIcNS1_9allocatorIcEEEEEEENS1_12basic_stringIcNS1_11char_traitsIcEES4_EEDpRKT_: 100| 16.7k| template static std::string as_str(const Data&... data) { 101| 16.7k| static_assert(sizeof...(Data) > 0, "must hash with at least one argument"); 102| 16.7k| 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| 16.7k| } else { 108| 16.7k| h = fnv1ahash_impl(data...); 109| 16.7k| } 110| 16.7k| constexpr std::size_t expected_chars = 16; 111| 16.7k| std::string ret(expected_chars, '0'); 112| 16.7k| auto c = std::to_chars(ret.data(), ret.data() + ret.size(), h, 16); 113| 16.7k| assert(c.ec == std::errc{}); ------------------ | Branch (113:5): [True: 16.7k, False: 0] ------------------ 114| 16.7k| auto nwritten = c.ptr - ret.data(); 115| 16.7k| assert(nwritten <= expected_chars); ------------------ | Branch (115:5): [True: 16.7k, False: 0] ------------------ 116| 16.7k| std::rotate(ret.data(), c.ptr, ret.data() + expected_chars); 117| 16.7k| return ret; 118| 16.7k| } _ZN10FNV1A_hash14fnv1ahash_implIcQ11is_hashableIT_EEEmRKNSt3__16vectorIS1_NS2_9allocatorIS1_EEEE: 96| 16.7k| static constexpr std::uint64_t fnv1ahash_impl(const std::vector& data) { 97| 16.7k| return fnv1ahash_impl(std::span(data)); 98| 16.7k| } _ZN10FNV1A_hash6as_strIJNSt3__16vectorIDsNS1_9allocatorIDsEEEEEEENS1_12basic_stringIcNS1_11char_traitsIcEENS3_IcEEEEDpRKT_: 100| 13.6k| template static std::string as_str(const Data&... data) { 101| 13.6k| static_assert(sizeof...(Data) > 0, "must hash with at least one argument"); 102| 13.6k| 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| 13.6k| } else { 108| 13.6k| h = fnv1ahash_impl(data...); 109| 13.6k| } 110| 13.6k| constexpr std::size_t expected_chars = 16; 111| 13.6k| std::string ret(expected_chars, '0'); 112| 13.6k| auto c = std::to_chars(ret.data(), ret.data() + ret.size(), h, 16); 113| 13.6k| assert(c.ec == std::errc{}); ------------------ | Branch (113:5): [True: 13.6k, False: 0] ------------------ 114| 13.6k| auto nwritten = c.ptr - ret.data(); 115| 13.6k| assert(nwritten <= expected_chars); ------------------ | Branch (115:5): [True: 13.6k, False: 0] ------------------ 116| 13.6k| std::rotate(ret.data(), c.ptr, ret.data() + expected_chars); 117| 13.6k| return ret; 118| 13.6k| } _ZN10FNV1A_hash14fnv1ahash_implIDsQ11is_hashableIT_EEEmRKNSt3__16vectorIS1_NS2_9allocatorIS1_EEEE: 96| 13.6k| static constexpr std::uint64_t fnv1ahash_impl(const std::vector& data) { 97| 13.6k| return fnv1ahash_impl(std::span(data)); 98| 13.6k| } _ZeqRKN7simdutf6resultES2_: 45| 7.51k|inline bool operator==(const simdutf::result& a, const simdutf::result& b) { 46| 7.51k| return a.count == b.count && a.error == b.error; ------------------ | Branch (46:10): [True: 7.51k, False: 0] | Branch (46:32): [True: 7.51k, False: 0] ------------------ 47| 7.51k|} _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| 962| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt25EE4dataEv: 233| 962| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt25EE4sizeEv: 235| 962| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_0clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 88|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 88| ::std::void_t(); \ 1180| 88| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 88| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 88| constexpr auto _size = _name.size(); \ 1183| 88| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 88| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_0clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 88|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 88| ::std::void_t(); \ 1180| 88| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 88| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 88| constexpr auto _size = _name.size(); \ 1183| 88| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 88| return _nameof; }() _ZNK6nameof7cstringILt30EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 346| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt30EE4dataEv: 233| 346| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt30EE4sizeEv: 235| 346| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_1clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 240|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 240| ::std::void_t(); \ 1180| 240| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 240| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 240| constexpr auto _size = _name.size(); \ 1183| 240| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 240| return _nameof; }() _ZNK6nameof7cstringILt24EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 2.90k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt24EE4dataEv: 233| 2.90k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt24EE4sizeEv: 235| 2.90k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_1clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 240|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 240| ::std::void_t(); \ 1180| 240| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 240| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 240| constexpr auto _size = _name.size(); \ 1183| 240| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 240| return _nameof; }() _ZNK6nameof7cstringILt29EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 884| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt29EE4dataEv: 233| 884| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt29EE4sizeEv: 235| 884| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_2clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 61|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 61| ::std::void_t(); \ 1180| 61| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 61| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 61| constexpr auto _size = _name.size(); \ 1183| 61| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 61| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_2clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 61|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 61| ::std::void_t(); \ 1180| 61| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 61| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 61| constexpr auto _size = _name.size(); \ 1183| 61| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 61| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_3clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 219|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 219| ::std::void_t(); \ 1180| 219| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 219| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 219| constexpr auto _size = _name.size(); \ 1183| 219| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 219| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_3clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 219|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 219| ::std::void_t(); \ 1180| 219| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 219| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 219| constexpr auto _size = _name.size(); \ 1183| 219| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 219| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_4clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 47|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 47| ::std::void_t(); \ 1180| 47| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 47| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 47| constexpr auto _size = _name.size(); \ 1183| 47| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 47| return _nameof; }() _ZNK6nameof7cstringILt23EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 3.28k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt23EE4dataEv: 233| 3.28k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt23EE4sizeEv: 235| 3.28k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_4clENSt3__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_functionsvENK3$_5clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 150|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 150| ::std::void_t(); \ 1180| 150| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 150| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 150| constexpr auto _size = _name.size(); \ 1183| 150| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 150| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_5clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 150|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 150| ::std::void_t(); \ 1180| 150| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 150| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 150| constexpr auto _size = _name.size(); \ 1183| 150| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 150| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_6clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 102|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 102| ::std::void_t(); \ 1180| 102| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 102| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 102| constexpr auto _size = _name.size(); \ 1183| 102| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 102| return _nameof; }() _ZNK6nameof7cstringILt22EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 4.05k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt22EE4dataEv: 233| 4.05k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt22EE4sizeEv: 235| 4.05k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_6clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 102|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 102| ::std::void_t(); \ 1180| 102| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 102| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 102| constexpr auto _size = _name.size(); \ 1183| 102| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 102| return _nameof; }() _ZNK6nameof7cstringILt27EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 310| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt27EE4dataEv: 233| 310| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt27EE4sizeEv: 235| 310| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_7clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 228|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 228| ::std::void_t(); \ 1180| 228| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 228| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 228| constexpr auto _size = _name.size(); \ 1183| 228| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 228| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_7clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 228|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 228| ::std::void_t(); \ 1180| 228| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 228| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 228| constexpr auto _size = _name.size(); \ 1183| 228| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 228| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_8clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 197|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 197| ::std::void_t(); \ 1180| 197| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 197| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 197| constexpr auto _size = _name.size(); \ 1183| 197| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 197| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_8clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 197|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 197| ::std::void_t(); \ 1180| 197| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 197| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 197| constexpr auto _size = _name.size(); \ 1183| 197| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 197| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_9clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 208|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 208| ::std::void_t(); \ 1180| 208| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 208| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 208| constexpr auto _size = _name.size(); \ 1183| 208| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 208| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_9clENSt3__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$_10clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 63|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 63| ::std::void_t(); \ 1180| 63| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 63| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 63| constexpr auto _size = _name.size(); \ 1183| 63| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 63| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_10clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 63|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 63| ::std::void_t(); \ 1180| 63| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 63| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 63| constexpr auto _size = _name.size(); \ 1183| 63| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 63| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_11clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 113|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 113| ::std::void_t(); \ 1180| 113| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 113| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 113| constexpr auto _size = _name.size(); \ 1183| 113| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 113| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_11clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 113|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 113| ::std::void_t(); \ 1180| 113| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 113| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 113| constexpr auto _size = _name.size(); \ 1183| 113| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 113| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_12clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 163|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 163| ::std::void_t(); \ 1180| 163| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 163| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 163| constexpr auto _size = _name.size(); \ 1183| 163| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 163| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_12clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 163|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 163| ::std::void_t(); \ 1180| 163| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 163| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 163| constexpr auto _size = _name.size(); \ 1183| 163| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 163| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_13clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 62|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 62| ::std::void_t(); \ 1180| 62| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 62| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 62| constexpr auto _size = _name.size(); \ 1183| 62| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 62| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_13clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 62|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 62| ::std::void_t(); \ 1180| 62| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 62| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 62| constexpr auto _size = _name.size(); \ 1183| 62| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 62| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_14clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 136|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 136| ::std::void_t(); \ 1180| 136| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 136| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 136| constexpr auto _size = _name.size(); \ 1183| 136| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 136| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_14clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 136|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 136| ::std::void_t(); \ 1180| 136| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 136| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 136| constexpr auto _size = _name.size(); \ 1183| 136| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 136| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_15clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 193|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 193| ::std::void_t(); \ 1180| 193| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 193| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 193| constexpr auto _size = _name.size(); \ 1183| 193| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 193| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_15clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 193|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 193| ::std::void_t(); \ 1180| 193| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 193| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 193| constexpr auto _size = _name.size(); \ 1183| 193| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 193| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_16clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 94|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 94| ::std::void_t(); \ 1180| 94| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 94| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 94| constexpr auto _size = _name.size(); \ 1183| 94| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 94| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_16clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 94|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 94| ::std::void_t(); \ 1180| 94| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 94| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 94| constexpr auto _size = _name.size(); \ 1183| 94| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 94| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_17clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 221|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 221| ::std::void_t(); \ 1180| 221| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 221| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 221| constexpr auto _size = _name.size(); \ 1183| 221| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 221| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_17clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 221|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 221| ::std::void_t(); \ 1180| 221| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 221| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 221| constexpr auto _size = _name.size(); \ 1183| 221| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 221| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_18clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 216|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 216| ::std::void_t(); \ 1180| 216| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 216| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 216| constexpr auto _size = _name.size(); \ 1183| 216| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 216| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_18clENSt3__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; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_19clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 302|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 302| ::std::void_t(); \ 1180| 302| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 302| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 302| constexpr auto _size = _name.size(); \ 1183| 302| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 302| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_19clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 302|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 302| ::std::void_t(); \ 1180| 302| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 302| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 302| constexpr auto _size = _name.size(); \ 1183| 302| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 302| return _nameof; }() _ZNK6nameof7cstringILt21EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 658| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt21EE4dataEv: 233| 658| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt21EE4sizeEv: 235| 658| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_20clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 281|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 281| ::std::void_t(); \ 1180| 281| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 281| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 281| constexpr auto _size = _name.size(); \ 1183| 281| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 281| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_20clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 281|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 281| ::std::void_t(); \ 1180| 281| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 281| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 281| constexpr auto _size = _name.size(); \ 1183| 281| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 281| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_21clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 356|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 356| ::std::void_t(); \ 1180| 356| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 356| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 356| constexpr auto _size = _name.size(); \ 1183| 356| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 356| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_21clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 356|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 356| ::std::void_t(); \ 1180| 356| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 356| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 356| constexpr auto _size = _name.size(); \ 1183| 356| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 356| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_22clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 330|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 330| ::std::void_t(); \ 1180| 330| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 330| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 330| constexpr auto _size = _name.size(); \ 1183| 330| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 330| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_22clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 330|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 330| ::std::void_t(); \ 1180| 330| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 330| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 330| constexpr auto _size = _name.size(); \ 1183| 330| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 330| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_23clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 356|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 356| ::std::void_t(); \ 1180| 356| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 356| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 356| constexpr auto _size = _name.size(); \ 1183| 356| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 356| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_23clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 356|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 356| ::std::void_t(); \ 1180| 356| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 356| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 356| constexpr auto _size = _name.size(); \ 1183| 356| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 356| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_24clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 126|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 126| ::std::void_t(); \ 1180| 126| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 126| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 126| constexpr auto _size = _name.size(); \ 1183| 126| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 126| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_24clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 126|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 126| ::std::void_t(); \ 1180| 126| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 126| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 126| constexpr auto _size = _name.size(); \ 1183| 126| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 126| return _nameof; }() _ZNK6nameof7cstringILt37EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 241| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt37EE4dataEv: 233| 241| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt37EE4sizeEv: 235| 241| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_25clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 150|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 150| ::std::void_t(); \ 1180| 150| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 150| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 150| constexpr auto _size = _name.size(); \ 1183| 150| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 150| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_25clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 150|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 150| ::std::void_t(); \ 1180| 150| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 150| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 150| constexpr auto _size = _name.size(); \ 1183| 150| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 150| return _nameof; }() _ZNK6nameof7cstringILt36EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 1.01k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt36EE4dataEv: 233| 1.01k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt36EE4sizeEv: 235| 1.01k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_26clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 293|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 293| ::std::void_t(); \ 1180| 293| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 293| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 293| constexpr auto _size = _name.size(); \ 1183| 293| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 293| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_26clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 293|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 293| ::std::void_t(); \ 1180| 293| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 293| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 293| constexpr auto _size = _name.size(); \ 1183| 293| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 293| return _nameof; }() _ZNK6nameof7cstringILt35EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 1.45k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt35EE4dataEv: 233| 1.45k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt35EE4sizeEv: 235| 1.45k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_27clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 115|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 115| ::std::void_t(); \ 1180| 115| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 115| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 115| constexpr auto _size = _name.size(); \ 1183| 115| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 115| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_27clENSt3__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; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_28clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 207|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 207| ::std::void_t(); \ 1180| 207| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 207| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 207| constexpr auto _size = _name.size(); \ 1183| 207| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 207| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_28clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 207|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 207| ::std::void_t(); \ 1180| 207| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 207| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 207| constexpr auto _size = _name.size(); \ 1183| 207| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 207| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_29clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 315|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 315| ::std::void_t(); \ 1180| 315| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 315| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 315| constexpr auto _size = _name.size(); \ 1183| 315| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 315| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_29clENSt3__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$_30clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 210|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 210| ::std::void_t(); \ 1180| 210| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 210| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 210| constexpr auto _size = _name.size(); \ 1183| 210| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 210| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_30clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 210|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 210| ::std::void_t(); \ 1180| 210| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 210| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 210| constexpr auto _size = _name.size(); \ 1183| 210| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 210| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_31clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 303|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 303| ::std::void_t(); \ 1180| 303| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 303| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 303| constexpr auto _size = _name.size(); \ 1183| 303| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 303| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_31clENSt3__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; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_32clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 358|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 358| ::std::void_t(); \ 1180| 358| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 358| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 358| constexpr auto _size = _name.size(); \ 1183| 358| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 358| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_32clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 358|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 358| ::std::void_t(); \ 1180| 358| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 358| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 358| constexpr auto _size = _name.size(); \ 1183| 358| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 358| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_33clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 415|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 415| ::std::void_t(); \ 1180| 415| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 415| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 415| constexpr auto _size = _name.size(); \ 1183| 415| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 415| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_33clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 415|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 415| ::std::void_t(); \ 1180| 415| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 415| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 415| constexpr auto _size = _name.size(); \ 1183| 415| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 415| return _nameof; }() _ZNK6nameof7cstringILt33EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 784| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt33EE4dataEv: 233| 784| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt33EE4sizeEv: 235| 784| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_34clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 258|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 258| ::std::void_t(); \ 1180| 258| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 258| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 258| constexpr auto _size = _name.size(); \ 1183| 258| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 258| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_34clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 258|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 258| ::std::void_t(); \ 1180| 258| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 258| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 258| constexpr auto _size = _name.size(); \ 1183| 258| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 258| return _nameof; }() _ZNK6nameof7cstringILt34EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 258| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt34EE4dataEv: 233| 258| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt34EE4sizeEv: 235| 258| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_35clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 351|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 351| ::std::void_t(); \ 1180| 351| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 351| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 351| constexpr auto _size = _name.size(); \ 1183| 351| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 351| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_35clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 351|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 351| ::std::void_t(); \ 1180| 351| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 351| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 351| constexpr auto _size = _name.size(); \ 1183| 351| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 351| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_36clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 287|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 287| ::std::void_t(); \ 1180| 287| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 287| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 287| constexpr auto _size = _name.size(); \ 1183| 287| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 287| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_36clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 287|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 287| ::std::void_t(); \ 1180| 287| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 287| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 287| constexpr auto _size = _name.size(); \ 1183| 287| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 287| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_37clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 369|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 369| ::std::void_t(); \ 1180| 369| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 369| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 369| constexpr auto _size = _name.size(); \ 1183| 369| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 369| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_37clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 369|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 369| ::std::void_t(); \ 1180| 369| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 369| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 369| constexpr auto _size = _name.size(); \ 1183| 369| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 369| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_38clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 45|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 45| ::std::void_t(); \ 1180| 45| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 45| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 45| constexpr auto _size = _name.size(); \ 1183| 45| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 45| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_38clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 45|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 45| ::std::void_t(); \ 1180| 45| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 45| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 45| constexpr auto _size = _name.size(); \ 1183| 45| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 45| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_39clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 40|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 40| ::std::void_t(); \ 1180| 40| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 40| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 40| constexpr auto _size = _name.size(); \ 1183| 40| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 40| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_39clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 40|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 40| ::std::void_t(); \ 1180| 40| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 40| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 40| constexpr auto _size = _name.size(); \ 1183| 40| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 40| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_40clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 40|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 40| ::std::void_t(); \ 1180| 40| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 40| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 40| constexpr auto _size = _name.size(); \ 1183| 40| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 40| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_40clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 40|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 40| ::std::void_t(); \ 1180| 40| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 40| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 40| constexpr auto _size = _name.size(); \ 1183| 40| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 40| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_41clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 269|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 269| ::std::void_t(); \ 1180| 269| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 269| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 269| constexpr auto _size = _name.size(); \ 1183| 269| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 269| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_41clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 269|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 269| ::std::void_t(); \ 1180| 269| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 269| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 269| constexpr auto _size = _name.size(); \ 1183| 269| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 269| return _nameof; }() _ZN7simdutf6resultC2ENS_10error_codeEm: 87| 73.1k| : error{err}, count{pos} {} _ZNK7simdutf6result6is_errEv: 93| 5.02k| simdutf_really_inline simdutf_constexpr23 bool is_err() const noexcept { 94| 5.02k| return error != error_code::SUCCESS; 95| 5.02k| } _ZN7simdutf11full_resultC2ENS_10error_codeEmm: 111| 5.51k| : error{err}, input_count{pos_in}, output_count{pos_out} {} _ZNK7simdutf11full_resultcvNS_6resultEEv: 117| 3.64k| simdutf_really_inline simdutf_constexpr23 operator result() const noexcept { 118| 3.64k| if (error == error_code::SUCCESS) { ------------------ | Branch (118:9): [True: 2.25k, False: 1.39k] ------------------ 119| 2.25k| return result{error, output_count}; 120| 2.25k| } else { 121| 1.39k| return result{error, input_count}; 122| 1.39k| } 123| 3.64k| } _ZNK7simdutf14implementation4nameEv: 5121| 4| virtual std::string_view name() const noexcept { return _name; } _ZNK7simdutf14implementation25required_instruction_setsEv: 5172| 8| virtual uint32_t required_instruction_sets() const { 5173| 8| return _required_instruction_sets; 5174| 8| } _ZNK7simdutf14implementation24utf16_length_from_latin1Em: 6362| 240| utf16_length_from_latin1(size_t length) const noexcept { 6363| 240| return length; 6364| 240| } _ZNK7simdutf14implementation24latin1_length_from_utf32Em: 6550| 912| latin1_length_from_utf32(size_t length) const noexcept { 6551| 912| return length; 6552| 912| } _ZNK7simdutf14implementation24latin1_length_from_utf16Em: 6587| 1.09k| latin1_length_from_utf16(size_t length) const noexcept { 6588| 1.09k| return length; 6589| 1.09k| } _ZNK7simdutf14implementation24utf32_length_from_latin1Em: 6620| 135| utf32_length_from_latin1(size_t length) const noexcept { 6621| 135| return length; 6622| 135| } _ZN7simdutf14implementationC2EPKcS2_j: 7058| 4| : _name(name), _description(description), 7059| 4| _required_instruction_sets(required_instruction_sets) {} _ZN7simdutf8internal29available_implementation_listC2Ev: 7090| 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| 538| size_t len) { 11| 538| const uint8_t *c = reinterpret_cast(buf); 12| 538| size_t answer = 0; 13| 4.23k| for (size_t i = 0; i < len; i++) { ------------------ | Branch (13:22): [True: 3.69k, False: 538] ------------------ 14| 3.69k| if ((c[i] >> 7)) { ------------------ | Branch (14:9): [True: 1.04k, False: 2.64k] ------------------ 15| 1.04k| answer++; 16| 1.04k| } 17| 3.69k| } 18| 538| return answer + len; 19| 538|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115latin1_to_utf167convertILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 14| 204| char16_t *utf16_output) { 15| 204| size_t pos = 0; 16| 204| char16_t *start{utf16_output}; 17| | 18| 2.56M| while (pos < len) { ------------------ | Branch (18:10): [True: 2.56M, False: 204] ------------------ 19| 2.56M| uint16_t word = 20| 2.56M| uint8_t(data[pos]); // extend Latin-1 char to 16-bit Unicode code point 21| 2.56M| *utf16_output++ = 22| 2.56M| char16_t(match_system(big_endian) ? word : u16_swap_bytes(word)); ------------------ | Branch (22:18): [True: 2.56M, Folded] ------------------ 23| 2.56M| pos++; 24| 2.56M| } 25| | 26| 204| return utf16_output - start; 27| 204|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115latin1_to_utf167convertILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 14| 204| char16_t *utf16_output) { 15| 204| size_t pos = 0; 16| 204| char16_t *start{utf16_output}; 17| | 18| 2.36M| while (pos < len) { ------------------ | Branch (18:10): [True: 2.36M, False: 204] ------------------ 19| 2.36M| uint16_t word = 20| 2.36M| uint8_t(data[pos]); // extend Latin-1 char to 16-bit Unicode code point 21| 2.36M| *utf16_output++ = 22| 2.36M| char16_t(match_system(big_endian) ? word : u16_swap_bytes(word)); ------------------ | Branch (22:18): [Folded, False: 2.36M] ------------------ 23| 2.36M| pos++; 24| 2.36M| } 25| | 26| 204| return utf16_output - start; 27| 204|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115latin1_to_utf327convertIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mPDi: 14| 218| char32_t *utf32_output) { 15| 218| char32_t *start{utf32_output}; 16| 2.69M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (16:22): [True: 2.69M, False: 218] ------------------ 17| 2.69M| *utf32_output++ = uint8_t(data[i]); 18| 2.69M| } 19| 218| return utf32_output - start; 20| 218|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf87convertEPKcmPc: 66| 1.52k| char *utf8_output) { 67| 1.52k| return convert(reinterpret_cast(buf), len, 68| 1.52k| utf8_output); 69| 1.52k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf87convertIPKhPcQaasr7simdutf6detailE22indexes_into_byte_likeIT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEEmS7_mS8_: 17| 1.52k| OutputPtr utf8_output) { 18| | // const unsigned char *data = reinterpret_cast(buf); 19| 1.52k| size_t pos = 0; 20| 1.52k| size_t utf8_pos = 0; 21| | 22| 4.70M| while (pos < len) { ------------------ | Branch (22:10): [True: 4.69M, False: 1.52k] ------------------ 23| |#if SIMDUTF_CPLUSPLUS23 24| | if !consteval 25| |#endif 26| 4.69M| { 27| | // try to convert the next block of 16 ASCII bytes 28| 4.69M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (28:11): [True: 4.68M, False: 13.6k] ------------------ 29| | // they are ascii 30| 4.68M| uint64_t v1; 31| 4.68M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 32| 4.68M| uint64_t v2; 33| 4.68M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 34| 4.68M| uint64_t v{v1 | 35| 4.68M| v2}; // We are only interested in these bits: 1000 1000 1000 36| | // 1000, so it makes sense to concatenate everything 37| 4.68M| if ((v & 0x8080808080808080) == ------------------ | Branch (37:13): [True: 2.32M, False: 2.35M] ------------------ 38| 4.68M| 0) { // if NONE of these are set, e.g. all of them are zero, then 39| | // everything is ASCII 40| 2.32M| size_t final_pos = pos + 16; 41| 39.5M| while (pos < final_pos) { ------------------ | Branch (41:18): [True: 37.2M, False: 2.32M] ------------------ 42| 37.2M| utf8_output[utf8_pos++] = char(data[pos]); 43| 37.2M| pos++; 44| 37.2M| } 45| 2.32M| continue; 46| 2.32M| } 47| 4.68M| } // if (pos + 16 <= len) 48| 4.69M| } // !consteval scope 49| | 50| 2.37M| unsigned char byte = data[pos]; 51| 2.37M| if ((byte & 0x80) == 0) { // if ASCII ------------------ | Branch (51:9): [True: 1.46M, False: 909k] ------------------ 52| | // will generate one UTF-8 bytes 53| 1.46M| utf8_output[utf8_pos++] = char(byte); 54| 1.46M| pos++; 55| 1.46M| } else { 56| | // will generate two UTF-8 bytes 57| 909k| utf8_output[utf8_pos++] = char((byte >> 6) | 0b11000000); 58| 909k| utf8_output[utf8_pos++] = char((byte & 0b111111) | 0b10000000); 59| 909k| pos++; 60| 909k| } 61| 2.37M| } // while 62| 1.52k| return utf8_pos; 63| 1.52k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf823utf8_length_from_latin1IPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_m: 152| 269|utf8_length_from_latin1(InputPtr input, size_t length) noexcept { 153| 269| size_t answer = length; 154| 269| size_t i = 0; 155| | 156| |#if SIMDUTF_CPLUSPLUS23 157| | if !consteval 158| |#endif 159| 269| { 160| 269| auto pop = [](uint64_t v) { 161| 269| return (size_t)(((v >> 7) & UINT64_C(0x0101010101010101)) * 162| 269| UINT64_C(0x0101010101010101) >> 163| 269| 56); 164| 269| }; 165| 618k| for (; i + 32 <= length; i += 32) { ------------------ | Branch (165:12): [True: 618k, False: 269] ------------------ 166| 618k| uint64_t v; 167| 618k| memcpy(&v, input + i, 8); 168| 618k| answer += pop(v); 169| 618k| memcpy(&v, input + i + 8, sizeof(v)); 170| 618k| answer += pop(v); 171| 618k| memcpy(&v, input + i + 16, sizeof(v)); 172| 618k| answer += pop(v); 173| 618k| memcpy(&v, input + i + 24, sizeof(v)); 174| 618k| answer += pop(v); 175| 618k| } 176| 538| for (; i + 8 <= length; i += 8) { ------------------ | Branch (176:12): [True: 269, False: 269] ------------------ 177| 269| uint64_t v; 178| 269| memcpy(&v, input + i, sizeof(v)); 179| 269| answer += pop(v); 180| 269| } 181| 269| } // !consteval scope 182| 723| for (; i + 1 <= length; i += 1) { ------------------ | Branch (182:10): [True: 454, False: 269] ------------------ 183| 454| answer += static_cast(input[i]) >> 7; 184| 454| } 185| 269| return answer; 186| 269|} simdutf.cpp:_ZZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf823utf8_length_from_latin1IPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mENKUlmE_clEm: 160| 2.47M| auto pop = [](uint64_t v) { 161| 2.47M| return (size_t)(((v >> 7) & UINT64_C(0x0101010101010101)) * 162| | UINT64_C(0x0101010101010101) >> 163| 2.47M| 56); 164| 2.47M| }; _ZN7simdutf6scalar14u16_swap_bytesEt: 8| 212M|u16_swap_bytes(const uint16_t word) { 9| 212M| return uint16_t((word >> 8) | (word << 8)); 10| 212M|} _ZN7simdutf6scalar5utf1614swap_if_neededILNS_10endiannessE1EEEtt: 27| 101M|template constexpr uint16_t swap_if_needed(uint16_t c) { 28| 101M| return !match_system(big_endian) ? scalar::u16_swap_bytes(c) : c; ------------------ | Branch (28:10): [True: 101M, Folded] ------------------ 29| 101M|} _ZN7simdutf6scalar5utf1614swap_if_neededILNS_10endiannessE0EEEtt: 27| 98.6M|template constexpr uint16_t swap_if_needed(uint16_t c) { 28| 98.6M| return !match_system(big_endian) ? scalar::u16_swap_bytes(c) : c; ------------------ | Branch (28:10): [Folded, False: 98.6M] ------------------ 29| 98.6M|} _ZN7simdutf6scalar5utf168validateILNS_10endiannessE0EEEbPKDsm: 22| 3.68k|validate(const char16_t *data, size_t len) noexcept { 23| 3.68k| uint64_t pos = 0; 24| 21.6M| while (pos < len) { ------------------ | Branch (24:10): [True: 21.6M, False: 2.57k] ------------------ 25| 21.6M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 26| 21.6M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (26:9): [True: 17.6k, False: 21.6M] ------------------ 27| 17.6k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 234, False: 17.3k] ------------------ 28| 234| return false; 29| 234| } 30| 17.3k| char16_t diff = char16_t(word - 0xD800); 31| 17.3k| if (diff > 0x3FF) { ------------------ | Branch (31:11): [True: 304, False: 17.0k] ------------------ 32| 304| return false; 33| 304| } 34| 17.0k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (34:28): [Folded, False: 17.0k] ------------------ 35| 17.0k| ? u16_swap_bytes(data[pos + 1]) 36| 17.0k| : data[pos + 1]; 37| 17.0k| char16_t diff2 = char16_t(next_word - 0xDC00); 38| 17.0k| if (diff2 > 0x3FF) { ------------------ | Branch (38:11): [True: 572, False: 16.5k] ------------------ 39| 572| return false; 40| 572| } 41| 16.5k| pos += 2; 42| 21.6M| } else { 43| 21.6M| pos++; 44| 21.6M| } 45| 21.6M| } 46| 2.57k| return true; 47| 3.68k|} _ZN7simdutf6scalar5utf168validateILNS_10endiannessE1EEEbPKDsm: 22| 3.44k|validate(const char16_t *data, size_t len) noexcept { 23| 3.44k| uint64_t pos = 0; 24| 22.4M| while (pos < len) { ------------------ | Branch (24:10): [True: 22.4M, False: 2.47k] ------------------ 25| 22.4M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 26| 22.4M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (26:9): [True: 17.4k, False: 22.4M] ------------------ 27| 17.4k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 213, False: 17.2k] ------------------ 28| 213| return false; 29| 213| } 30| 17.2k| char16_t diff = char16_t(word - 0xD800); 31| 17.2k| if (diff > 0x3FF) { ------------------ | Branch (31:11): [True: 276, False: 17.0k] ------------------ 32| 276| return false; 33| 276| } 34| 17.0k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (34:28): [True: 17.0k, Folded] ------------------ 35| 17.0k| ? u16_swap_bytes(data[pos + 1]) 36| 17.0k| : data[pos + 1]; 37| 17.0k| char16_t diff2 = char16_t(next_word - 0xDC00); 38| 17.0k| if (diff2 > 0x3FF) { ------------------ | Branch (38:11): [True: 488, False: 16.5k] ------------------ 39| 488| return false; 40| 488| } 41| 16.5k| pos += 2; 42| 22.4M| } else { 43| 22.4M| pos++; 44| 22.4M| } 45| 22.4M| } 46| 2.47k| return true; 47| 3.44k|} _ZN7simdutf6scalar5utf1620validate_with_errorsILNS_10endiannessE0EEENS_6resultEPKDsm: 51| 3.89k|validate_with_errors(const char16_t *data, size_t len) noexcept { 52| 3.89k| size_t pos = 0; 53| 21.6M| while (pos < len) { ------------------ | Branch (53:10): [True: 21.6M, False: 2.57k] ------------------ 54| 21.6M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 55| 21.6M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (55:9): [True: 17.8k, False: 21.6M] ------------------ 56| 17.8k| if (pos + 1 >= len) { ------------------ | Branch (56:11): [True: 234, False: 17.6k] ------------------ 57| 234| return result(error_code::SURROGATE, pos); 58| 234| } 59| 17.6k| char16_t diff = char16_t(word - 0xD800); 60| 17.6k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 408, False: 17.2k] ------------------ 61| 408| return result(error_code::SURROGATE, pos); 62| 408| } 63| 17.2k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (63:28): [Folded, False: 17.2k] ------------------ 64| 17.2k| ? u16_swap_bytes(data[pos + 1]) 65| 17.2k| : data[pos + 1]; 66| 17.2k| char16_t diff2 = uint16_t(next_word - 0xDC00); 67| 17.2k| if (diff2 > 0x3FF) { ------------------ | Branch (67:11): [True: 672, False: 16.5k] ------------------ 68| 672| return result(error_code::SURROGATE, pos); 69| 672| } 70| 16.5k| pos += 2; 71| 21.6M| } else { 72| 21.6M| pos++; 73| 21.6M| } 74| 21.6M| } 75| 2.57k| return result(error_code::SUCCESS, pos); 76| 3.89k|} _ZN7simdutf6scalar5utf1620validate_with_errorsILNS_10endiannessE1EEENS_6resultEPKDsm: 51| 3.67k|validate_with_errors(const char16_t *data, size_t len) noexcept { 52| 3.67k| size_t pos = 0; 53| 22.4M| while (pos < len) { ------------------ | Branch (53:10): [True: 22.4M, False: 2.47k] ------------------ 54| 22.4M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 55| 22.4M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (55:9): [True: 17.7k, False: 22.4M] ------------------ 56| 17.7k| if (pos + 1 >= len) { ------------------ | Branch (56:11): [True: 213, False: 17.5k] ------------------ 57| 213| return result(error_code::SURROGATE, pos); 58| 213| } 59| 17.5k| char16_t diff = char16_t(word - 0xD800); 60| 17.5k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 381, False: 17.2k] ------------------ 61| 381| return result(error_code::SURROGATE, pos); 62| 381| } 63| 17.2k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (63:28): [True: 17.2k, Folded] ------------------ 64| 17.2k| ? u16_swap_bytes(data[pos + 1]) 65| 17.2k| : data[pos + 1]; 66| 17.2k| char16_t diff2 = uint16_t(next_word - 0xDC00); 67| 17.2k| if (diff2 > 0x3FF) { ------------------ | Branch (67:11): [True: 612, False: 16.5k] ------------------ 68| 612| return result(error_code::SURROGATE, pos); 69| 612| } 70| 16.5k| pos += 2; 71| 22.4M| } else { 72| 22.4M| pos++; 73| 22.4M| } 74| 22.4M| } 75| 2.47k| return result(error_code::SUCCESS, pos); 76| 3.67k|} _ZN7simdutf6scalar5utf1617count_code_pointsILNS_10endiannessE0EEEmPKDsm: 79| 4.73k|simdutf_constexpr23 size_t count_code_points(const char16_t *p, size_t len) { 80| | // We are not BOM aware. 81| 4.73k| size_t counter{0}; 82| 29.5M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (82:22): [True: 29.5M, False: 4.73k] ------------------ 83| 29.5M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 84| 29.5M| counter += ((word & 0xFC00) != 0xDC00); 85| 29.5M| } 86| 4.73k| return counter; 87| 4.73k|} _ZN7simdutf6scalar5utf1617count_code_pointsILNS_10endiannessE1EEEmPKDsm: 79| 4.41k|simdutf_constexpr23 size_t count_code_points(const char16_t *p, size_t len) { 80| | // We are not BOM aware. 81| 4.41k| size_t counter{0}; 82| 31.5M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (82:22): [True: 31.5M, False: 4.41k] ------------------ 83| 31.5M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 84| 31.5M| counter += ((word & 0xFC00) != 0xDC00); 85| 31.5M| } 86| 4.41k| return counter; 87| 4.41k|} _ZN7simdutf6scalar5utf1622utf8_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 91| 2.18k| size_t len) { 92| | // We are not BOM aware. 93| 2.18k| size_t counter{0}; 94| 19.1M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (94:22): [True: 19.0M, False: 2.18k] ------------------ 95| 19.0M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 96| 19.0M| counter++; // ASCII 97| 19.0M| counter += static_cast( 98| 19.0M| word > 99| 19.0M| 0x7F); // non-ASCII is at least 2 bytes, surrogates are 2*2 == 4 bytes 100| 19.0M| counter += static_cast((word > 0x7FF && word <= 0xD7FF) || ------------------ | Branch (100:37): [True: 3.17M, False: 15.9M] | Branch (100:53): [True: 2.90M, False: 273k] ------------------ 101| 16.1M| (word >= 0xE000)); // three-byte ------------------ | Branch (101:36): [True: 216k, False: 15.9M] ------------------ 102| 19.0M| } 103| 2.18k| return counter; 104| 2.18k|} _ZN7simdutf6scalar5utf1622utf8_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 91| 2.08k| size_t len) { 92| | // We are not BOM aware. 93| 2.08k| size_t counter{0}; 94| 17.8M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (94:22): [True: 17.8M, False: 2.08k] ------------------ 95| 17.8M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 96| 17.8M| counter++; // ASCII 97| 17.8M| counter += static_cast( 98| 17.8M| word > 99| 17.8M| 0x7F); // non-ASCII is at least 2 bytes, surrogates are 2*2 == 4 bytes 100| 17.8M| counter += static_cast((word > 0x7FF && word <= 0xD7FF) || ------------------ | Branch (100:37): [True: 3.81M, False: 14.0M] | Branch (100:53): [True: 3.66M, False: 156k] ------------------ 101| 14.1M| (word >= 0xE000)); // three-byte ------------------ | Branch (101:36): [True: 132k, False: 14.0M] ------------------ 102| 17.8M| } 103| 2.08k| return counter; 104| 2.08k|} _ZN7simdutf6scalar5utf1623utf32_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 108| 404| size_t len) { 109| | // We are not BOM aware. 110| 404| size_t counter{0}; 111| 6.01M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (111:22): [True: 6.01M, False: 404] ------------------ 112| 6.01M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 113| 6.01M| counter += ((word & 0xFC00) != 0xDC00); 114| 6.01M| } 115| 404| return counter; 116| 404|} _ZN7simdutf6scalar5utf1623utf32_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 108| 351| size_t len) { 109| | // We are not BOM aware. 110| 351| size_t counter{0}; 111| 6.99M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (111:22): [True: 6.99M, False: 351] ------------------ 112| 6.99M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 113| 6.99M| counter += ((word & 0xFC00) != 0xDC00); 114| 6.99M| } 115| 351| return counter; 116| 351|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin17convertILNS_10endiannessE0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 252| OutputPtr latin_output) { 18| 252| if (len == 0) { ------------------ | Branch (18:7): [True: 6, False: 246] ------------------ 19| 6| return 0; 20| 6| } 21| 246| size_t pos = 0; 22| 246| const auto latin_output_start = latin_output; 23| 246| uint16_t word = 0; 24| 246| uint16_t too_large = 0; 25| | 26| 589k| while (pos < len) { ------------------ | Branch (26:10): [True: 589k, False: 246] ------------------ 27| 589k| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (27:12): [Folded, False: 589k] ------------------ 28| 589k| too_large |= word; 29| 589k| *latin_output++ = char(word & 0xFF); 30| 589k| pos++; 31| 589k| } 32| 246| if ((too_large & 0xFF00) != 0) { ------------------ | Branch (32:7): [True: 160, False: 86] ------------------ 33| 160| return 0; 34| 160| } 35| | 36| 86| return latin_output - latin_output_start; 37| 246|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin17convertILNS_10endiannessE1EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 282| OutputPtr latin_output) { 18| 282| if (len == 0) { ------------------ | Branch (18:7): [True: 6, False: 276] ------------------ 19| 6| return 0; 20| 6| } 21| 276| size_t pos = 0; 22| 276| const auto latin_output_start = latin_output; 23| 276| uint16_t word = 0; 24| 276| uint16_t too_large = 0; 25| | 26| 2.36M| while (pos < len) { ------------------ | Branch (26:10): [True: 2.36M, False: 276] ------------------ 27| 2.36M| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (27:12): [True: 2.36M, Folded] ------------------ 28| 2.36M| too_large |= word; 29| 2.36M| *latin_output++ = char(word & 0xFF); 30| 2.36M| pos++; 31| 2.36M| } 32| 276| if ((too_large & 0xFF00) != 0) { ------------------ | Branch (32:7): [True: 118, False: 158] ------------------ 33| 118| return 0; 34| 118| } 35| | 36| 158| return latin_output - latin_output_start; 37| 276|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin119convert_with_errorsILNS_10endiannessE0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEENS_6resultES8_mS9_: 45| 396| OutputPtr latin_output) { 46| 396| if (len == 0) { ------------------ | Branch (46:7): [True: 4, False: 392] ------------------ 47| 4| return result(error_code::SUCCESS, 0); 48| 4| } 49| 392| size_t pos = 0; 50| 392| auto start = latin_output; 51| 392| uint16_t word; 52| | 53| 453k| while (pos < len) { ------------------ | Branch (53:10): [True: 453k, False: 168] ------------------ 54| |#if SIMDUTF_CPLUSPLUS23 55| | if !consteval 56| |#endif 57| 453k| { 58| 453k| if (pos + 16 <= len) { // if it is safe to read 32 more bytes, check that ------------------ | Branch (58:11): [True: 451k, False: 1.29k] ------------------ 59| | // they are Latin1 60| 451k| uint64_t v1, v2, v3, v4; 61| 451k| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 62| 451k| ::memcpy(&v2, data + pos + 4, sizeof(uint64_t)); 63| 451k| ::memcpy(&v3, data + pos + 8, sizeof(uint64_t)); 64| 451k| ::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| 451k| if (((v1 | v2 | v3 | v4) & 0xFF00FF00FF00FF00) == 0) { ------------------ | Branch (79:13): [True: 451k, False: 238] ------------------ 80| 451k| size_t final_pos = pos + 16; 81| 7.67M| while (pos < final_pos) { ------------------ | Branch (81:18): [True: 7.22M, False: 451k] ------------------ 82| 7.22M| *latin_output++ = !match_system(big_endian) ------------------ | Branch (82:31): [Folded, False: 7.22M] ------------------ 83| 7.22M| ? char(u16_swap_bytes(data[pos])) 84| 7.22M| : char(data[pos]); 85| 7.22M| pos++; 86| 7.22M| } 87| 451k| continue; 88| 451k| } 89| 451k| } 90| 453k| } 91| | 92| 1.53k| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (92:12): [Folded, False: 1.53k] ------------------ 93| 1.53k| if ((word & 0xFF00) == 0) { ------------------ | Branch (93:9): [True: 1.30k, False: 224] ------------------ 94| 1.30k| *latin_output++ = char(word & 0xFF); 95| 1.30k| pos++; 96| 1.30k| } else { 97| 224| return result(error_code::TOO_LARGE, pos); 98| 224| } 99| 1.53k| } 100| 168| return result(error_code::SUCCESS, latin_output - start); 101| 392|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin119convert_with_errorsILNS_10endiannessE1EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEENS_6resultES8_mS9_: 45| 450| OutputPtr latin_output) { 46| 450| if (len == 0) { ------------------ | Branch (46:7): [True: 4, False: 446] ------------------ 47| 4| return result(error_code::SUCCESS, 0); 48| 4| } 49| 446| size_t pos = 0; 50| 446| auto start = latin_output; 51| 446| uint16_t word; 52| | 53| 525k| while (pos < len) { ------------------ | Branch (53:10): [True: 525k, False: 208] ------------------ 54| |#if SIMDUTF_CPLUSPLUS23 55| | if !consteval 56| |#endif 57| 525k| { 58| 525k| if (pos + 16 <= len) { // if it is safe to read 32 more bytes, check that ------------------ | Branch (58:11): [True: 523k, False: 1.75k] ------------------ 59| | // they are Latin1 60| 523k| uint64_t v1, v2, v3, v4; 61| 523k| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 62| 523k| ::memcpy(&v2, data + pos + 4, sizeof(uint64_t)); 63| 523k| ::memcpy(&v3, data + pos + 8, sizeof(uint64_t)); 64| 523k| ::memcpy(&v4, data + pos + 12, sizeof(uint64_t)); 65| | 66| 523k| if constexpr (!match_system(big_endian)) { 67| 523k| v1 = (v1 >> 8) | (v1 << (64 - 8)); 68| 523k| } 69| 523k| if constexpr (!match_system(big_endian)) { 70| 523k| v2 = (v2 >> 8) | (v2 << (64 - 8)); 71| 523k| } 72| 523k| if constexpr (!match_system(big_endian)) { 73| 523k| v3 = (v3 >> 8) | (v3 << (64 - 8)); 74| 523k| } 75| 523k| if constexpr (!match_system(big_endian)) { 76| 523k| v4 = (v4 >> 8) | (v4 << (64 - 8)); 77| 523k| } 78| | 79| 523k| if (((v1 | v2 | v3 | v4) & 0xFF00FF00FF00FF00) == 0) { ------------------ | Branch (79:13): [True: 523k, False: 406] ------------------ 80| 523k| size_t final_pos = pos + 16; 81| 8.89M| while (pos < final_pos) { ------------------ | Branch (81:18): [True: 8.37M, False: 523k] ------------------ 82| 8.37M| *latin_output++ = !match_system(big_endian) ------------------ | Branch (82:31): [True: 8.37M, Folded] ------------------ 83| 8.37M| ? char(u16_swap_bytes(data[pos])) 84| 8.37M| : char(data[pos]); 85| 8.37M| pos++; 86| 8.37M| } 87| 523k| continue; 88| 523k| } 89| 523k| } 90| 525k| } 91| | 92| 2.15k| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (92:12): [True: 2.15k, Folded] ------------------ 93| 2.15k| if ((word & 0xFF00) == 0) { ------------------ | Branch (93:9): [True: 1.92k, False: 238] ------------------ 94| 1.92k| *latin_output++ = char(word & 0xFF); 95| 1.92k| pos++; 96| 1.92k| } else { 97| 238| return result(error_code::TOO_LARGE, pos); 98| 238| } 99| 2.15k| } 100| 208| return result(error_code::SUCCESS, latin_output - start); 101| 446|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf327convertILNS_10endiannessE0EEEmPKDsmPDi: 11| 712| char32_t *utf32_output) { 12| 712| size_t pos = 0; 13| 712| char32_t *start{utf32_output}; 14| 1.04M| while (pos < len) { ------------------ | Branch (14:10): [True: 1.04M, False: 410] ------------------ 15| 1.04M| uint16_t word = 16| 1.04M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [Folded, False: 1.04M] ------------------ 17| 1.04M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 1.04M, False: 2.76k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 1.04M| *utf32_output++ = char32_t(word); 20| 1.04M| pos++; 21| 1.04M| } else { 22| | // must be a surrogate pair 23| 2.76k| uint16_t diff = uint16_t(word - 0xD800); 24| 2.76k| if (diff > 0x3FF) { ------------------ | Branch (24:11): [True: 104, False: 2.65k] ------------------ 25| 104| return 0; 26| 104| } 27| 2.65k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 78, False: 2.58k] ------------------ 28| 78| return 0; 29| 78| } // minimal bound checking 30| 2.58k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (30:28): [Folded, False: 2.58k] ------------------ 31| 2.58k| ? u16_swap_bytes(data[pos + 1]) 32| 2.58k| : data[pos + 1]; 33| 2.58k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 34| 2.58k| if (diff2 > 0x3FF) { ------------------ | Branch (34:11): [True: 120, False: 2.46k] ------------------ 35| 120| return 0; 36| 120| } 37| 2.46k| uint32_t value = (diff << 10) + diff2 + 0x10000; 38| 2.46k| *utf32_output++ = char32_t(value); 39| 2.46k| pos += 2; 40| 2.46k| } 41| 1.04M| } 42| 410| return utf32_output - start; 43| 712|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf327convertILNS_10endiannessE1EEEmPKDsmPDi: 11| 708| char32_t *utf32_output) { 12| 708| size_t pos = 0; 13| 708| char32_t *start{utf32_output}; 14| 925k| while (pos < len) { ------------------ | Branch (14:10): [True: 924k, False: 402] ------------------ 15| 924k| uint16_t word = 16| 924k| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [True: 924k, Folded] ------------------ 17| 924k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 923k, False: 1.00k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 923k| *utf32_output++ = char32_t(word); 20| 923k| pos++; 21| 923k| } else { 22| | // must be a surrogate pair 23| 1.00k| uint16_t diff = uint16_t(word - 0xD800); 24| 1.00k| if (diff > 0x3FF) { ------------------ | Branch (24:11): [True: 88, False: 918] ------------------ 25| 88| return 0; 26| 88| } 27| 918| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 96, False: 822] ------------------ 28| 96| return 0; 29| 96| } // minimal bound checking 30| 822| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (30:28): [True: 822, Folded] ------------------ 31| 822| ? u16_swap_bytes(data[pos + 1]) 32| 822| : data[pos + 1]; 33| 822| uint16_t diff2 = uint16_t(next_word - 0xDC00); 34| 822| if (diff2 > 0x3FF) { ------------------ | Branch (34:11): [True: 122, False: 700] ------------------ 35| 122| return 0; 36| 122| } 37| 700| uint32_t value = (diff << 10) + diff2 + 0x10000; 38| 700| *utf32_output++ = char32_t(value); 39| 700| pos += 2; 40| 700| } 41| 924k| } 42| 402| return utf32_output - start; 43| 708|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3219convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKDsmPDi: 47| 892| char32_t *utf32_output) { 48| 892| size_t pos = 0; 49| 892| char32_t *start{utf32_output}; 50| 2.49M| while (pos < len) { ------------------ | Branch (50:10): [True: 2.49M, False: 492] ------------------ 51| 2.49M| uint16_t word = 52| 2.49M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (52:9): [Folded, False: 2.49M] ------------------ 53| 2.49M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (53:9): [True: 2.48M, False: 11.9k] ------------------ 54| | // No surrogate pair, extend 16-bit word to 32-bit word 55| 2.48M| *utf32_output++ = char32_t(word); 56| 2.48M| pos++; 57| 2.48M| } else { 58| | // must be a surrogate pair 59| 11.9k| uint16_t diff = uint16_t(word - 0xD800); 60| 11.9k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 118, False: 11.8k] ------------------ 61| 118| return result(error_code::SURROGATE, pos); 62| 118| } 63| 11.8k| if (pos + 1 >= len) { ------------------ | Branch (63:11): [True: 96, False: 11.7k] ------------------ 64| 96| return result(error_code::SURROGATE, pos); 65| 96| } // minimal bound checking 66| 11.7k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (66:28): [Folded, False: 11.7k] ------------------ 67| 11.7k| ? u16_swap_bytes(data[pos + 1]) 68| 11.7k| : data[pos + 1]; 69| 11.7k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 70| 11.7k| if (diff2 > 0x3FF) { ------------------ | Branch (70:11): [True: 186, False: 11.5k] ------------------ 71| 186| return result(error_code::SURROGATE, pos); 72| 186| } 73| 11.5k| uint32_t value = (diff << 10) + diff2 + 0x10000; 74| 11.5k| *utf32_output++ = char32_t(value); 75| 11.5k| pos += 2; 76| 11.5k| } 77| 2.49M| } 78| 492| return result(error_code::SUCCESS, utf32_output - start); 79| 892|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3219convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKDsmPDi: 47| 702| char32_t *utf32_output) { 48| 702| size_t pos = 0; 49| 702| char32_t *start{utf32_output}; 50| 3.39M| while (pos < len) { ------------------ | Branch (50:10): [True: 3.39M, False: 412] ------------------ 51| 3.39M| uint16_t word = 52| 3.39M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (52:9): [True: 3.39M, Folded] ------------------ 53| 3.39M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (53:9): [True: 3.38M, False: 10.4k] ------------------ 54| | // No surrogate pair, extend 16-bit word to 32-bit word 55| 3.38M| *utf32_output++ = char32_t(word); 56| 3.38M| pos++; 57| 3.38M| } else { 58| | // must be a surrogate pair 59| 10.4k| uint16_t diff = uint16_t(word - 0xD800); 60| 10.4k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 68, False: 10.4k] ------------------ 61| 68| return result(error_code::SURROGATE, pos); 62| 68| } 63| 10.4k| if (pos + 1 >= len) { ------------------ | Branch (63:11): [True: 102, False: 10.3k] ------------------ 64| 102| return result(error_code::SURROGATE, pos); 65| 102| } // minimal bound checking 66| 10.3k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (66:28): [True: 10.3k, Folded] ------------------ 67| 10.3k| ? u16_swap_bytes(data[pos + 1]) 68| 10.3k| : data[pos + 1]; 69| 10.3k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 70| 10.3k| if (diff2 > 0x3FF) { ------------------ | Branch (70:11): [True: 120, False: 10.1k] ------------------ 71| 120| return result(error_code::SURROGATE, pos); 72| 120| } 73| 10.1k| uint32_t value = (diff << 10) + diff2 + 0x10000; 74| 10.1k| *utf32_output++ = char32_t(value); 75| 10.1k| pos += 2; 76| 10.1k| } 77| 3.39M| } 78| 412| return result(error_code::SUCCESS, utf32_output - start); 79| 702|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3213convert_validILNS_10endiannessE0EEEmPKDsmPDi: 11| 112| char32_t *utf32_output) { 12| 112| size_t pos = 0; 13| 112| char32_t *start{utf32_output}; 14| 1.54M| while (pos < len) { ------------------ | Branch (14:10): [True: 1.54M, False: 112] ------------------ 15| 1.54M| uint16_t word = 16| 1.54M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [Folded, False: 1.54M] ------------------ 17| 1.54M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 1.53M, False: 2.60k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 1.53M| *utf32_output++ = char32_t(word); 20| 1.53M| pos++; 21| 1.53M| } else { 22| | // must be a surrogate pair 23| 2.60k| uint16_t diff = uint16_t(word - 0xD800); 24| 2.60k| if (pos + 1 >= len) { ------------------ | Branch (24:11): [True: 0, False: 2.60k] ------------------ 25| 0| return 0; 26| 0| } // minimal bound checking 27| 2.60k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (27:28): [Folded, False: 2.60k] ------------------ 28| 2.60k| ? u16_swap_bytes(data[pos + 1]) 29| 2.60k| : data[pos + 1]; 30| 2.60k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 31| 2.60k| uint32_t value = (diff << 10) + diff2 + 0x10000; 32| 2.60k| *utf32_output++ = char32_t(value); 33| 2.60k| pos += 2; 34| 2.60k| } 35| 1.54M| } 36| 112| return utf32_output - start; 37| 112|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3213convert_validILNS_10endiannessE1EEEmPKDsmPDi: 11| 140| char32_t *utf32_output) { 12| 140| size_t pos = 0; 13| 140| char32_t *start{utf32_output}; 14| 4.22M| while (pos < len) { ------------------ | Branch (14:10): [True: 4.22M, False: 140] ------------------ 15| 4.22M| uint16_t word = 16| 4.22M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [True: 4.22M, Folded] ------------------ 17| 4.22M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 4.22M, False: 1.26k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 4.22M| *utf32_output++ = char32_t(word); 20| 4.22M| pos++; 21| 4.22M| } else { 22| | // must be a surrogate pair 23| 1.26k| uint16_t diff = uint16_t(word - 0xD800); 24| 1.26k| if (pos + 1 >= len) { ------------------ | Branch (24:11): [True: 0, False: 1.26k] ------------------ 25| 0| return 0; 26| 0| } // minimal bound checking 27| 1.26k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (27:28): [True: 1.26k, Folded] ------------------ 28| 1.26k| ? u16_swap_bytes(data[pos + 1]) 29| 1.26k| : data[pos + 1]; 30| 1.26k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 31| 1.26k| uint32_t value = (diff << 10) + diff2 + 0x10000; 32| 1.26k| *utf32_output++ = char32_t(value); 33| 1.26k| pos += 2; 34| 1.26k| } 35| 4.22M| } 36| 140| return utf32_output - start; 37| 140|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf87convertILNS_10endiannessE0EPKDsPcQsr7simdutf6detailE18indexes_into_utf16IT0_EEEmS8_mT1_: 17| 1.79k| OutputPtr utf8_output) { 18| 1.79k| size_t pos = 0; 19| 1.79k| const auto start = utf8_output; 20| 3.00M| while (pos < len) { ------------------ | Branch (20:10): [True: 3.00M, False: 1.49k] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 3.00M| { 25| | // try to convert the next block of 8 bytes 26| 3.00M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 2.99M, False: 3.18k] ------------------ 27| | // they are ascii 28| 2.99M| uint64_t v; 29| 2.99M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| | if constexpr (!match_system(big_endian)) { 31| | v = (v >> 8) | (v << (64 - 8)); 32| | } 33| 2.99M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 2.42M, False: 571k] ------------------ 34| 2.42M| size_t final_pos = pos + 4; 35| 12.1M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 9.69M, False: 2.42M] ------------------ 36| 9.69M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [Folded, False: 9.69M] ------------------ 37| 9.69M| ? char(u16_swap_bytes(data[pos])) 38| 9.69M| : char(data[pos]); 39| 9.69M| pos++; 40| 9.69M| } 41| 2.42M| continue; 42| 2.42M| } 43| 2.99M| } 44| 3.00M| } 45| 575k| uint16_t word = 46| 575k| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (46:9): [Folded, False: 575k] ------------------ 47| 575k| if ((word & 0xFF80) == 0) { ------------------ | Branch (47:9): [True: 139k, False: 436k] ------------------ 48| | // will generate one UTF-8 bytes 49| 139k| *utf8_output++ = char(word); 50| 139k| pos++; 51| 436k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (51:16): [True: 91.6k, False: 344k] ------------------ 52| | // will generate two UTF-8 bytes 53| | // we have 0b110XXXXX 0b10XXXXXX 54| 91.6k| *utf8_output++ = char((word >> 6) | 0b11000000); 55| 91.6k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 56| 91.6k| pos++; 57| 344k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (57:16): [True: 339k, False: 4.42k] ------------------ 58| | // will generate three UTF-8 bytes 59| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 60| 339k| *utf8_output++ = char((word >> 12) | 0b11100000); 61| 339k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 62| 339k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 63| 339k| pos++; 64| 339k| } else { 65| | // must be a surrogate pair 66| 4.42k| if (pos + 1 >= len) { ------------------ | Branch (66:11): [True: 42, False: 4.37k] ------------------ 67| 42| return 0; 68| 42| } 69| 4.37k| uint16_t diff = uint16_t(word - 0xD800); 70| 4.37k| if (diff > 0x3FF) { ------------------ | Branch (70:11): [True: 102, False: 4.27k] ------------------ 71| 102| return 0; 72| 102| } 73| 4.27k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (73:28): [Folded, False: 4.27k] ------------------ 74| 4.27k| ? u16_swap_bytes(data[pos + 1]) 75| 4.27k| : data[pos + 1]; 76| 4.27k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 77| 4.27k| if (diff2 > 0x3FF) { ------------------ | Branch (77:11): [True: 152, False: 4.12k] ------------------ 78| 152| return 0; 79| 152| } 80| 4.12k| uint32_t value = (diff << 10) + diff2 + 0x10000; 81| | // will generate four UTF-8 bytes 82| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 83| 4.12k| *utf8_output++ = char((value >> 18) | 0b11110000); 84| 4.12k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 85| 4.12k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 86| 4.12k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 87| 4.12k| pos += 2; 88| 4.12k| } 89| 575k| } 90| 1.49k| return utf8_output - start; 91| 1.79k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf87convertILNS_10endiannessE1EPKDsPcQsr7simdutf6detailE18indexes_into_utf16IT0_EEEmS8_mT1_: 17| 1.68k| OutputPtr utf8_output) { 18| 1.68k| size_t pos = 0; 19| 1.68k| const auto start = utf8_output; 20| 3.79M| while (pos < len) { ------------------ | Branch (20:10): [True: 3.78M, False: 1.45k] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 3.78M| { 25| | // try to convert the next block of 8 bytes 26| 3.78M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 3.78M, False: 3.06k] ------------------ 27| | // they are ascii 28| 3.78M| uint64_t v; 29| 3.78M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| 3.78M| if constexpr (!match_system(big_endian)) { 31| 3.78M| v = (v >> 8) | (v << (64 - 8)); 32| 3.78M| } 33| 3.78M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.59M, False: 2.19M] ------------------ 34| 1.59M| size_t final_pos = pos + 4; 35| 7.96M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 6.36M, False: 1.59M] ------------------ 36| 6.36M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [True: 6.36M, Folded] ------------------ 37| 6.36M| ? char(u16_swap_bytes(data[pos])) 38| 6.36M| : char(data[pos]); 39| 6.36M| pos++; 40| 6.36M| } 41| 1.59M| continue; 42| 1.59M| } 43| 3.78M| } 44| 3.78M| } 45| 2.19M| uint16_t word = 46| 2.19M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (46:9): [True: 2.19M, Folded] ------------------ 47| 2.19M| if ((word & 0xFF80) == 0) { ------------------ | Branch (47:9): [True: 160k, False: 2.03M] ------------------ 48| | // will generate one UTF-8 bytes 49| 160k| *utf8_output++ = char(word); 50| 160k| pos++; 51| 2.03M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (51:16): [True: 57.5k, False: 1.97M] ------------------ 52| | // will generate two UTF-8 bytes 53| | // we have 0b110XXXXX 0b10XXXXXX 54| 57.5k| *utf8_output++ = char((word >> 6) | 0b11000000); 55| 57.5k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 56| 57.5k| pos++; 57| 1.97M| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (57:16): [True: 1.97M, False: 6.51k] ------------------ 58| | // will generate three UTF-8 bytes 59| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 60| 1.97M| *utf8_output++ = char((word >> 12) | 0b11100000); 61| 1.97M| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 62| 1.97M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 63| 1.97M| pos++; 64| 1.97M| } else { 65| | // must be a surrogate pair 66| 6.51k| if (pos + 1 >= len) { ------------------ | Branch (66:11): [True: 30, False: 6.48k] ------------------ 67| 30| return 0; 68| 30| } 69| 6.48k| uint16_t diff = uint16_t(word - 0xD800); 70| 6.48k| if (diff > 0x3FF) { ------------------ | Branch (70:11): [True: 80, False: 6.40k] ------------------ 71| 80| return 0; 72| 80| } 73| 6.40k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (73:28): [True: 6.40k, Folded] ------------------ 74| 6.40k| ? u16_swap_bytes(data[pos + 1]) 75| 6.40k| : data[pos + 1]; 76| 6.40k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 77| 6.40k| if (diff2 > 0x3FF) { ------------------ | Branch (77:11): [True: 122, False: 6.28k] ------------------ 78| 122| return 0; 79| 122| } 80| 6.28k| uint32_t value = (diff << 10) + diff2 + 0x10000; 81| | // will generate four UTF-8 bytes 82| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 83| 6.28k| *utf8_output++ = char((value >> 18) | 0b11110000); 84| 6.28k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 85| 6.28k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 86| 6.28k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 87| 6.28k| pos += 2; 88| 6.28k| } 89| 2.19M| } 90| 1.45k| return utf8_output - start; 91| 1.68k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf819convert_with_errorsILNS_10endiannessE0ELb0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT1_Esr7simdutf6detailE26index_assignable_from_charIT2_EEENS_11full_resultES8_mS9_m: 101| 1.64k| size_t utf8_len = 0) { 102| 1.64k| if (check_output && utf8_len == 0) { ------------------ | Branch (102:7): [Folded, False: 1.64k] | Branch (102:23): [True: 0, False: 0] ------------------ 103| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, 0, 0); 104| 0| } 105| | 106| 1.64k| size_t pos = 0; 107| 1.64k| auto start = utf8_output; 108| 1.64k| auto end = utf8_output + utf8_len; 109| | 110| 6.55M| while (pos < len) { ------------------ | Branch (110:10): [True: 6.55M, False: 1.23k] ------------------ 111| |#if SIMDUTF_CPLUSPLUS23 112| | if !consteval 113| |#endif 114| 6.55M| { 115| | // try to convert the next block of 8 bytes 116| 6.55M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (116:11): [True: 6.54M, False: 2.83k] ------------------ 117| | // they are ascii 118| 6.54M| uint64_t v; 119| 6.54M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 120| | if constexpr (!match_system(big_endian)) 121| | v = (v >> 8) | (v << (64 - 8)); 122| 6.54M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (122:13): [True: 1.95M, False: 4.59M] ------------------ 123| 1.95M| size_t final_pos = pos + 4; 124| 9.78M| while (pos < final_pos) { ------------------ | Branch (124:18): [True: 7.82M, False: 1.95M] ------------------ 125| 7.82M| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (125:17): [Folded, False: 7.82M] | 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| 7.82M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (129:30): [Folded, False: 7.82M] ------------------ 130| 7.82M| ? char(u16_swap_bytes(data[pos])) 131| 7.82M| : char(data[pos]); 132| 7.82M| pos++; 133| 7.82M| } 134| 1.95M| continue; 135| 1.95M| } 136| 6.54M| } 137| 6.55M| } 138| | 139| 4.59M| uint16_t word = 140| 4.59M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (140:9): [Folded, False: 4.59M] ------------------ 141| 4.59M| if ((word & 0xFF80) == 0) { ------------------ | Branch (141:9): [True: 240k, False: 4.35M] ------------------ 142| | // will generate one UTF-8 bytes 143| 240k| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (143:11): [Folded, False: 240k] | 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| 240k| *utf8_output++ = char(word); 148| 240k| pos++; 149| 4.35M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (149:16): [True: 1.48M, False: 2.86M] ------------------ 150| | // will generate two UTF-8 bytes 151| | // we have 0b110XXXXX 0b10XXXXXX 152| 1.48M| if (check_output && size_t(end - utf8_output) < 2) { ------------------ | Branch (152:11): [Folded, False: 1.48M] | 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| 1.48M| *utf8_output++ = char((word >> 6) | 0b11000000); 157| 1.48M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 158| 1.48M| pos++; 159| | 160| 2.86M| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (160:16): [True: 2.85M, False: 4.84k] ------------------ 161| | // will generate three UTF-8 bytes 162| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 163| 2.85M| if (check_output && size_t(end - utf8_output) < 3) { ------------------ | Branch (163:11): [Folded, False: 2.85M] | 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| 2.85M| *utf8_output++ = char((word >> 12) | 0b11100000); 168| 2.85M| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 169| 2.85M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 170| 2.85M| pos++; 171| 2.85M| } else { 172| | 173| 4.84k| if (check_output && size_t(end - utf8_output) < 4) { ------------------ | Branch (173:11): [Folded, False: 4.84k] | 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| 4.84k| if (pos + 1 >= len) { ------------------ | Branch (178:11): [True: 72, False: 4.77k] ------------------ 179| 72| return full_result(error_code::SURROGATE, pos, utf8_output - start); 180| 72| } 181| 4.77k| uint16_t diff = uint16_t(word - 0xD800); 182| 4.77k| if (diff > 0x3FF) { ------------------ | Branch (182:11): [True: 98, False: 4.67k] ------------------ 183| 98| return full_result(error_code::SURROGATE, pos, utf8_output - start); 184| 98| } 185| 4.67k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (185:28): [Folded, False: 4.67k] ------------------ 186| 4.67k| ? u16_swap_bytes(data[pos + 1]) 187| 4.67k| : data[pos + 1]; 188| 4.67k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 189| 4.67k| if (diff2 > 0x3FF) { ------------------ | Branch (189:11): [True: 238, False: 4.43k] ------------------ 190| 238| return full_result(error_code::SURROGATE, pos, utf8_output - start); 191| 238| } 192| 4.43k| uint32_t value = (diff << 10) + diff2 + 0x10000; 193| | // will generate four UTF-8 bytes 194| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 195| 4.43k| *utf8_output++ = char((value >> 18) | 0b11110000); 196| 4.43k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 197| 4.43k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 198| 4.43k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 199| 4.43k| pos += 2; 200| 4.43k| } 201| 4.59M| } 202| 1.23k| return full_result(error_code::SUCCESS, pos, utf8_output - start); 203| 1.64k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf819convert_with_errorsILNS_10endiannessE1ELb0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT1_Esr7simdutf6detailE26index_assignable_from_charIT2_EEENS_11full_resultES8_mS9_m: 101| 1.44k| size_t utf8_len = 0) { 102| 1.44k| if (check_output && utf8_len == 0) { ------------------ | Branch (102:7): [Folded, False: 1.44k] | Branch (102:23): [True: 0, False: 0] ------------------ 103| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, 0, 0); 104| 0| } 105| | 106| 1.44k| size_t pos = 0; 107| 1.44k| auto start = utf8_output; 108| 1.44k| auto end = utf8_output + utf8_len; 109| | 110| 2.49M| while (pos < len) { ------------------ | Branch (110:10): [True: 2.49M, False: 1.00k] ------------------ 111| |#if SIMDUTF_CPLUSPLUS23 112| | if !consteval 113| |#endif 114| 2.49M| { 115| | // try to convert the next block of 8 bytes 116| 2.49M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (116:11): [True: 2.49M, False: 2.25k] ------------------ 117| | // they are ascii 118| 2.49M| uint64_t v; 119| 2.49M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 120| | if constexpr (!match_system(big_endian)) 121| 2.49M| v = (v >> 8) | (v << (64 - 8)); 122| 2.49M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (122:13): [True: 1.66M, False: 829k] ------------------ 123| 1.66M| size_t final_pos = pos + 4; 124| 8.30M| while (pos < final_pos) { ------------------ | Branch (124:18): [True: 6.64M, False: 1.66M] ------------------ 125| 6.64M| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (125:17): [Folded, False: 6.64M] | Branch (125:33): [True: 0, False: 0] ------------------ 126| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, pos, 127| 0| utf8_output - start); 128| 0| } 129| 6.64M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (129:30): [True: 6.64M, Folded] ------------------ 130| 6.64M| ? char(u16_swap_bytes(data[pos])) 131| 6.64M| : char(data[pos]); 132| 6.64M| pos++; 133| 6.64M| } 134| 1.66M| continue; 135| 1.66M| } 136| 2.49M| } 137| 2.49M| } 138| | 139| 832k| uint16_t word = 140| 832k| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (140:9): [True: 832k, Folded] ------------------ 141| 832k| if ((word & 0xFF80) == 0) { ------------------ | Branch (141:9): [True: 255k, False: 576k] ------------------ 142| | // will generate one UTF-8 bytes 143| 255k| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (143:11): [Folded, False: 255k] | 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| 255k| *utf8_output++ = char(word); 148| 255k| pos++; 149| 576k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (149:16): [True: 110k, False: 465k] ------------------ 150| | // will generate two UTF-8 bytes 151| | // we have 0b110XXXXX 0b10XXXXXX 152| 110k| if (check_output && size_t(end - utf8_output) < 2) { ------------------ | Branch (152:11): [Folded, False: 110k] | 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| 110k| *utf8_output++ = char((word >> 6) | 0b11000000); 157| 110k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 158| 110k| pos++; 159| | 160| 465k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (160:16): [True: 457k, False: 8.40k] ------------------ 161| | // will generate three UTF-8 bytes 162| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 163| 457k| if (check_output && size_t(end - utf8_output) < 3) { ------------------ | Branch (163:11): [Folded, False: 457k] | 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| 457k| *utf8_output++ = char((word >> 12) | 0b11100000); 168| 457k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 169| 457k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 170| 457k| pos++; 171| 457k| } else { 172| | 173| 8.40k| if (check_output && size_t(end - utf8_output) < 4) { ------------------ | Branch (173:11): [Folded, False: 8.40k] | 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| 8.40k| if (pos + 1 >= len) { ------------------ | Branch (178:11): [True: 72, False: 8.33k] ------------------ 179| 72| return full_result(error_code::SURROGATE, pos, utf8_output - start); 180| 72| } 181| 8.33k| uint16_t diff = uint16_t(word - 0xD800); 182| 8.33k| if (diff > 0x3FF) { ------------------ | Branch (182:11): [True: 152, False: 8.18k] ------------------ 183| 152| return full_result(error_code::SURROGATE, pos, utf8_output - start); 184| 152| } 185| 8.18k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (185:28): [True: 8.18k, Folded] ------------------ 186| 8.18k| ? u16_swap_bytes(data[pos + 1]) 187| 8.18k| : data[pos + 1]; 188| 8.18k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 189| 8.18k| if (diff2 > 0x3FF) { ------------------ | Branch (189:11): [True: 220, False: 7.96k] ------------------ 190| 220| return full_result(error_code::SURROGATE, pos, utf8_output - start); 191| 220| } 192| 7.96k| uint32_t value = (diff << 10) + diff2 + 0x10000; 193| | // will generate four UTF-8 bytes 194| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 195| 7.96k| *utf8_output++ = char((value >> 18) | 0b11110000); 196| 7.96k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 197| 7.96k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 198| 7.96k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 199| 7.96k| pos += 2; 200| 7.96k| } 201| 832k| } 202| 1.00k| return full_result(error_code::SUCCESS, pos, utf8_output - start); 203| 1.44k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf813convert_validILNS_10endiannessE0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 416| OutputPtr utf8_output) { 18| 416| size_t pos = 0; 19| 416| auto start = utf8_output; 20| 2.15M| while (pos < len) { ------------------ | Branch (20:10): [True: 2.15M, False: 416] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 2.15M| { 25| | // try to convert the next block of 4 ASCII characters 26| 2.15M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 2.15M, False: 850] ------------------ 27| | // they are ascii 28| 2.15M| uint64_t v; 29| 2.15M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| | if constexpr (!match_system(big_endian)) { 31| | v = (v >> 8) | (v << (64 - 8)); 32| | } 33| 2.15M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.70M, False: 456k] ------------------ 34| 1.70M| size_t final_pos = pos + 4; 35| 8.50M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 6.80M, False: 1.70M] ------------------ 36| 6.80M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [Folded, False: 6.80M] ------------------ 37| 6.80M| ? char(u16_swap_bytes(data[pos])) 38| 6.80M| : char(data[pos]); 39| 6.80M| pos++; 40| 6.80M| } 41| 1.70M| continue; 42| 1.70M| } 43| 2.15M| } 44| 2.15M| } 45| | 46| 457k| uint16_t word = 47| 457k| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (47:9): [Folded, False: 457k] ------------------ 48| 457k| if ((word & 0xFF80) == 0) { ------------------ | Branch (48:9): [True: 177k, False: 279k] ------------------ 49| | // will generate one UTF-8 bytes 50| 177k| *utf8_output++ = char(word); 51| 177k| pos++; 52| 279k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (52:16): [True: 74.6k, False: 205k] ------------------ 53| | // will generate two UTF-8 bytes 54| | // we have 0b110XXXXX 0b10XXXXXX 55| 74.6k| *utf8_output++ = char((word >> 6) | 0b11000000); 56| 74.6k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 57| 74.6k| pos++; 58| 205k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (58:16): [True: 200k, False: 4.60k] ------------------ 59| | // will generate three UTF-8 bytes 60| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 61| 200k| *utf8_output++ = char((word >> 12) | 0b11100000); 62| 200k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 63| 200k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 64| 200k| pos++; 65| 200k| } else { 66| | // must be a surrogate pair 67| 4.60k| uint16_t diff = uint16_t(word - 0xD800); 68| 4.60k| if (pos + 1 >= len) { ------------------ | Branch (68:11): [True: 0, False: 4.60k] ------------------ 69| 0| return 0; 70| 0| } // minimal bound checking 71| 4.60k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (71:28): [Folded, False: 4.60k] ------------------ 72| 4.60k| ? u16_swap_bytes(data[pos + 1]) 73| 4.60k| : data[pos + 1]; 74| 4.60k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 75| 4.60k| uint32_t value = (diff << 10) + diff2 + 0x10000; 76| | // will generate four UTF-8 bytes 77| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 78| 4.60k| *utf8_output++ = char((value >> 18) | 0b11110000); 79| 4.60k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 80| 4.60k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 81| 4.60k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 82| 4.60k| pos += 2; 83| 4.60k| } 84| 457k| } 85| 416| return utf8_output - start; 86| 416|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf813convert_validILNS_10endiannessE1EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 442| OutputPtr utf8_output) { 18| 442| size_t pos = 0; 19| 442| auto start = utf8_output; 20| 3.33M| while (pos < len) { ------------------ | Branch (20:10): [True: 3.33M, False: 442] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 3.33M| { 25| | // try to convert the next block of 4 ASCII characters 26| 3.33M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 3.33M, False: 880] ------------------ 27| | // they are ascii 28| 3.33M| uint64_t v; 29| 3.33M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| 3.33M| if constexpr (!match_system(big_endian)) { 31| 3.33M| v = (v >> 8) | (v << (64 - 8)); 32| 3.33M| } 33| 3.33M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.91M, False: 1.41M] ------------------ 34| 1.91M| size_t final_pos = pos + 4; 35| 9.59M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 7.67M, False: 1.91M] ------------------ 36| 7.67M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [True: 7.67M, Folded] ------------------ 37| 7.67M| ? char(u16_swap_bytes(data[pos])) 38| 7.67M| : char(data[pos]); 39| 7.67M| pos++; 40| 7.67M| } 41| 1.91M| continue; 42| 1.91M| } 43| 3.33M| } 44| 3.33M| } 45| | 46| 1.41M| uint16_t word = 47| 1.41M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (47:9): [True: 1.41M, Folded] ------------------ 48| 1.41M| if ((word & 0xFF80) == 0) { ------------------ | Branch (48:9): [True: 361k, False: 1.05M] ------------------ 49| | // will generate one UTF-8 bytes 50| 361k| *utf8_output++ = char(word); 51| 361k| pos++; 52| 1.05M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (52:16): [True: 105k, False: 952k] ------------------ 53| | // will generate two UTF-8 bytes 54| | // we have 0b110XXXXX 0b10XXXXXX 55| 105k| *utf8_output++ = char((word >> 6) | 0b11000000); 56| 105k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 57| 105k| pos++; 58| 952k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (58:16): [True: 949k, False: 2.44k] ------------------ 59| | // will generate three UTF-8 bytes 60| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 61| 949k| *utf8_output++ = char((word >> 12) | 0b11100000); 62| 949k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 63| 949k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 64| 949k| pos++; 65| 949k| } else { 66| | // must be a surrogate pair 67| 2.44k| uint16_t diff = uint16_t(word - 0xD800); 68| 2.44k| if (pos + 1 >= len) { ------------------ | Branch (68:11): [True: 0, False: 2.44k] ------------------ 69| 0| return 0; 70| 0| } // minimal bound checking 71| 2.44k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (71:28): [True: 2.44k, Folded] ------------------ 72| 2.44k| ? u16_swap_bytes(data[pos + 1]) 73| 2.44k| : data[pos + 1]; 74| 2.44k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 75| 2.44k| uint32_t value = (diff << 10) + diff2 + 0x10000; 76| | // will generate four UTF-8 bytes 77| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 78| 2.44k| *utf8_output++ = char((value >> 18) | 0b11110000); 79| 2.44k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 80| 2.44k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 81| 2.44k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 82| 2.44k| pos += 2; 83| 2.44k| } 84| 1.41M| } 85| 442| return utf8_output - start; 86| 442|} _ZN7simdutf6scalar5utf328validateEPKDim: 25| 5.29k| size_t len) noexcept { 26| 5.29k| return validate(reinterpret_cast(buf), len); 27| 5.29k|} _ZN7simdutf6scalar5utf3220validate_with_errorsEPKDim: 49| 7.21k|validate_with_errors(const char32_t *buf, size_t len) noexcept { 50| 7.21k| return validate_with_errors(reinterpret_cast(buf), len); 51| 7.21k|} _ZN7simdutf6scalar5utf3222utf8_length_from_utf32EPKDim: 54| 2.45k| size_t len) { 55| | // We are not BOM aware. 56| 2.45k| size_t counter{0}; 57| 2.32M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (57:22): [True: 2.32M, False: 2.45k] ------------------ 58| | // credit: @ttsugriy for the vectorizable approach 59| 2.32M| counter++; // ASCII 60| 2.32M| counter += static_cast(p[i] > 0x7F); // two-byte 61| 2.32M| counter += static_cast(p[i] > 0x7FF); // three-byte 62| 2.32M| counter += static_cast(p[i] > 0xFFFF); // four-bytes 63| 2.32M| } 64| 2.45k| return counter; 65| 2.45k|} _ZN7simdutf6scalar5utf3223utf16_length_from_utf32EPKDim: 68| 3.88k|utf16_length_from_utf32(const char32_t *p, size_t len) { 69| | // We are not BOM aware. 70| 3.88k| size_t counter{0}; 71| 5.38M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (71:22): [True: 5.37M, False: 3.88k] ------------------ 72| 5.37M| counter++; // non-surrogate word 73| 5.37M| counter += static_cast(p[i] > 0xFFFF); // surrogate pair 74| 5.37M| } 75| 3.88k| return counter; 76| 3.88k|} _ZN7simdutf6scalar5utf328validateIPKjQsr7simdutf6detailE19indexes_into_uint32IT_EEEbS5_m: 13| 5.29k| size_t len) noexcept { 14| 5.29k| uint64_t pos = 0; 15| 1.02M| for (; pos < len; pos++) { ------------------ | Branch (15:10): [True: 1.02M, False: 2.39k] ------------------ 16| 1.02M| uint32_t word = data[pos]; 17| 1.02M| if (word > 0x10FFFF || (word >= 0xD800 && word <= 0xDFFF)) { ------------------ | Branch (17:9): [True: 2.41k, False: 1.01M] | Branch (17:29): [True: 31.7k, False: 986k] | Branch (17:47): [True: 487, False: 31.2k] ------------------ 18| 2.90k| return false; 19| 2.90k| } 20| 1.02M| } 21| 2.39k| return true; 22| 5.29k|} _ZN7simdutf6scalar5utf3220validate_with_errorsIPKjQsr7simdutf6detailE19indexes_into_uint32IT_EEENS_6resultES5_m: 34| 7.21k|validate_with_errors(InputPtr data, size_t len) noexcept { 35| 7.21k| size_t pos = 0; 36| 1.02M| for (; pos < len; pos++) { ------------------ | Branch (36:10): [True: 1.02M, False: 2.39k] ------------------ 37| 1.02M| uint32_t word = data[pos]; 38| 1.02M| if (word > 0x10FFFF) { ------------------ | Branch (38:9): [True: 3.92k, False: 1.02M] ------------------ 39| 3.92k| return result(error_code::TOO_LARGE, pos); 40| 3.92k| } 41| 1.02M| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (41:9): [True: 33.0k, False: 988k] | Branch (41:27): [True: 894, False: 32.1k] ------------------ 42| 894| return result(error_code::SURROGATE, pos); 43| 894| } 44| 1.02M| } 45| 2.39k| return result(error_code::SUCCESS, pos); 46| 7.21k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf32_to_latin17convertEPKDimPc: 12| 456| char *latin1_output) { 13| 456| char *start = latin1_output; 14| 456| uint32_t utf32_char; 15| 456| size_t pos = 0; 16| 456| uint32_t too_large = 0; 17| | 18| 1.22M| while (pos < len) { ------------------ | Branch (18:10): [True: 1.22M, False: 456] ------------------ 19| 1.22M| utf32_char = (uint32_t)data[pos]; 20| 1.22M| too_large |= utf32_char; 21| 1.22M| *latin1_output++ = (char)(utf32_char & 0xFF); 22| 1.22M| pos++; 23| 1.22M| } 24| 456| if ((too_large & 0xFFFFFF00) != 0) { ------------------ | Branch (24:7): [True: 308, False: 148] ------------------ 25| 308| return 0; 26| 308| } 27| 148| return latin1_output - start; 28| 456|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf32_to_latin119convert_with_errorsEPKDimPc: 32| 1.22k| char *latin1_output) { 33| 1.22k| char *start{latin1_output}; 34| 1.22k| size_t pos = 0; 35| 212k| while (pos < len) { ------------------ | Branch (35:10): [True: 212k, False: 110] ------------------ 36| |#if SIMDUTF_CPLUSPLUS23 37| | if !consteval 38| |#endif 39| 212k| { 40| 212k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (40:11): [True: 211k, False: 336] ------------------ 41| | // they are Latin1 42| 211k| uint64_t v; 43| 211k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 44| 211k| if ((v & 0xFFFFFF00FFFFFF00) == 0) { ------------------ | Branch (44:13): [True: 210k, False: 998] ------------------ 45| 210k| *latin1_output++ = char(data[pos]); 46| 210k| *latin1_output++ = char(data[pos + 1]); 47| 210k| pos += 2; 48| 210k| continue; 49| 210k| } 50| 211k| } 51| 212k| } 52| | 53| 1.33k| uint32_t utf32_char = data[pos]; 54| 1.33k| if ((utf32_char & 0xFFFFFF00) == ------------------ | Branch (54:9): [True: 224, False: 1.11k] ------------------ 55| 1.33k| 0) { // Check if the character can be represented in Latin-1 56| 224| *latin1_output++ = (char)(utf32_char & 0xFF); 57| 224| pos++; 58| 1.11k| } else { 59| 1.11k| return result(error_code::TOO_LARGE, pos); 60| 1.11k| }; 61| 224| } 62| 110| return result(error_code::SUCCESS, latin1_output - start); 63| 1.22k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf167convertILNS_10endiannessE0EEEmPKDimPDs: 11| 1.69k| char16_t *utf16_output) { 12| 1.69k| size_t pos = 0; 13| 1.69k| char16_t *start{utf16_output}; 14| 17.3k| while (pos < len) { ------------------ | Branch (14:10): [True: 16.3k, False: 1.01k] ------------------ 15| 16.3k| uint32_t word = data[pos]; 16| 16.3k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 12.9k, False: 3.40k] ------------------ 17| 12.9k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (17:11): [True: 1.71k, False: 11.2k] | Branch (17:29): [True: 176, False: 1.53k] ------------------ 18| 176| return 0; 19| 176| } 20| | // will not generate a surrogate pair 21| 12.7k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (21:25): [Folded, False: 12.7k] ------------------ 22| 12.7k| ? char16_t(u16_swap_bytes(uint16_t(word))) 23| 12.7k| : char16_t(word); 24| 12.7k| } else { 25| | // will generate a surrogate pair 26| 3.40k| if (word > 0x10FFFF) { ------------------ | Branch (26:11): [True: 506, False: 2.89k] ------------------ 27| 506| return 0; 28| 506| } 29| 2.89k| word -= 0x10000; 30| 2.89k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 31| 2.89k| 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| 2.89k| *utf16_output++ = char16_t(high_surrogate); 37| 2.89k| *utf16_output++ = char16_t(low_surrogate); 38| 2.89k| } 39| 15.6k| pos++; 40| 15.6k| } 41| 1.01k| return utf16_output - start; 42| 1.69k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf167convertILNS_10endiannessE1EEEmPKDimPDs: 11| 1.28k| char16_t *utf16_output) { 12| 1.28k| size_t pos = 0; 13| 1.28k| char16_t *start{utf16_output}; 14| 55.8k| while (pos < len) { ------------------ | Branch (14:10): [True: 55.3k, False: 502] ------------------ 15| 55.3k| uint32_t word = data[pos]; 16| 55.3k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 53.1k, False: 2.24k] ------------------ 17| 53.1k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (17:11): [True: 1.05k, False: 52.0k] | Branch (17:29): [True: 194, False: 860] ------------------ 18| 194| return 0; 19| 194| } 20| | // will not generate a surrogate pair 21| 52.9k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (21:25): [True: 52.9k, Folded] ------------------ 22| 52.9k| ? char16_t(u16_swap_bytes(uint16_t(word))) 23| 52.9k| : char16_t(word); 24| 52.9k| } else { 25| | // will generate a surrogate pair 26| 2.24k| if (word > 0x10FFFF) { ------------------ | Branch (26:11): [True: 586, False: 1.66k] ------------------ 27| 586| return 0; 28| 586| } 29| 1.66k| word -= 0x10000; 30| 1.66k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 31| 1.66k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 32| 1.66k| if constexpr (!match_system(big_endian)) { 33| 1.66k| high_surrogate = u16_swap_bytes(high_surrogate); 34| 1.66k| low_surrogate = u16_swap_bytes(low_surrogate); 35| 1.66k| } 36| 1.66k| *utf16_output++ = char16_t(high_surrogate); 37| 1.66k| *utf16_output++ = char16_t(low_surrogate); 38| 1.66k| } 39| 54.5k| pos++; 40| 54.5k| } 41| 502| return utf16_output - start; 42| 1.28k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1619convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKDimPDs: 46| 2.12k| char16_t *utf16_output) { 47| 2.12k| size_t pos = 0; 48| 2.12k| char16_t *start{utf16_output}; 49| 707k| while (pos < len) { ------------------ | Branch (49:10): [True: 707k, False: 530] ------------------ 50| 707k| uint32_t word = data[pos]; 51| 707k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (51:9): [True: 695k, False: 11.6k] ------------------ 52| 695k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (52:11): [True: 2.06k, False: 693k] | Branch (52:29): [True: 366, False: 1.69k] ------------------ 53| 366| return result(error_code::SURROGATE, pos); 54| 366| } 55| | // will not generate a surrogate pair 56| 695k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (56:25): [Folded, False: 695k] ------------------ 57| 695k| ? char16_t(u16_swap_bytes(uint16_t(word))) 58| 695k| : char16_t(word); 59| 695k| } else { 60| | // will generate a surrogate pair 61| 11.6k| if (word > 0x10FFFF) { ------------------ | Branch (61:11): [True: 1.22k, False: 10.4k] ------------------ 62| 1.22k| return result(error_code::TOO_LARGE, pos); 63| 1.22k| } 64| 10.4k| word -= 0x10000; 65| 10.4k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 66| 10.4k| 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| 10.4k| *utf16_output++ = char16_t(high_surrogate); 72| 10.4k| *utf16_output++ = char16_t(low_surrogate); 73| 10.4k| } 74| 705k| pos++; 75| 705k| } 76| 530| return result(error_code::SUCCESS, utf16_output - start); 77| 2.12k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1619convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKDimPDs: 46| 1.79k| char16_t *utf16_output) { 47| 1.79k| size_t pos = 0; 48| 1.79k| char16_t *start{utf16_output}; 49| 103k| while (pos < len) { ------------------ | Branch (49:10): [True: 102k, False: 490] ------------------ 50| 102k| uint32_t word = data[pos]; 51| 102k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (51:9): [True: 97.8k, False: 4.79k] ------------------ 52| 97.8k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (52:11): [True: 1.87k, False: 96.0k] | Branch (52:29): [True: 210, False: 1.66k] ------------------ 53| 210| return result(error_code::SURROGATE, pos); 54| 210| } 55| | // will not generate a surrogate pair 56| 97.6k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (56:25): [True: 97.6k, Folded] ------------------ 57| 97.6k| ? char16_t(u16_swap_bytes(uint16_t(word))) 58| 97.6k| : char16_t(word); 59| 97.6k| } else { 60| | // will generate a surrogate pair 61| 4.79k| if (word > 0x10FFFF) { ------------------ | Branch (61:11): [True: 1.09k, False: 3.69k] ------------------ 62| 1.09k| return result(error_code::TOO_LARGE, pos); 63| 1.09k| } 64| 3.69k| word -= 0x10000; 65| 3.69k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 66| 3.69k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 67| 3.69k| if constexpr (!match_system(big_endian)) { 68| 3.69k| high_surrogate = u16_swap_bytes(high_surrogate); 69| 3.69k| low_surrogate = u16_swap_bytes(low_surrogate); 70| 3.69k| } 71| 3.69k| *utf16_output++ = char16_t(high_surrogate); 72| 3.69k| *utf16_output++ = char16_t(low_surrogate); 73| 3.69k| } 74| 101k| pos++; 75| 101k| } 76| 490| return result(error_code::SUCCESS, utf16_output - start); 77| 1.79k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1613convert_validILNS_10endiannessE0EEEmPKDimPDs: 11| 290| char16_t *utf16_output) { 12| 290| size_t pos = 0; 13| 290| char16_t *start{utf16_output}; 14| 73.7k| while (pos < len) { ------------------ | Branch (14:10): [True: 73.4k, False: 290] ------------------ 15| 73.4k| uint32_t word = data[pos]; 16| 73.4k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 65.0k, False: 8.41k] ------------------ 17| | // will not generate a surrogate pair 18| 65.0k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (18:25): [Folded, False: 65.0k] ------------------ 19| 65.0k| ? char16_t(u16_swap_bytes(uint16_t(word))) 20| 65.0k| : char16_t(word); 21| 65.0k| pos++; 22| 65.0k| } else { 23| | // will generate a surrogate pair 24| 8.41k| word -= 0x10000; 25| 8.41k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 26| 8.41k| 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| 8.41k| *utf16_output++ = char16_t(high_surrogate); 32| 8.41k| *utf16_output++ = char16_t(low_surrogate); 33| 8.41k| pos++; 34| 8.41k| } 35| 73.4k| } 36| 290| return utf16_output - start; 37| 290|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1613convert_validILNS_10endiannessE1EEEmPKDimPDs: 11| 86| char16_t *utf16_output) { 12| 86| size_t pos = 0; 13| 86| char16_t *start{utf16_output}; 14| 6.61k| while (pos < len) { ------------------ | Branch (14:10): [True: 6.52k, False: 86] ------------------ 15| 6.52k| uint32_t word = data[pos]; 16| 6.52k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 2.92k, False: 3.60k] ------------------ 17| | // will not generate a surrogate pair 18| 2.92k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (18:25): [True: 2.92k, Folded] ------------------ 19| 2.92k| ? char16_t(u16_swap_bytes(uint16_t(word))) 20| 2.92k| : char16_t(word); 21| 2.92k| pos++; 22| 3.60k| } else { 23| | // will generate a surrogate pair 24| 3.60k| word -= 0x10000; 25| 3.60k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 26| 3.60k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 27| 3.60k| if constexpr (!match_system(big_endian)) { 28| 3.60k| high_surrogate = u16_swap_bytes(high_surrogate); 29| 3.60k| low_surrogate = u16_swap_bytes(low_surrogate); 30| 3.60k| } 31| 3.60k| *utf16_output++ = char16_t(high_surrogate); 32| 3.60k| *utf16_output++ = char16_t(low_surrogate); 33| 3.60k| pos++; 34| 3.60k| } 35| 6.52k| } 36| 86| return utf16_output - start; 37| 86|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf32_to_utf87convertIPKDiPcQaasr7simdutf6detailE18indexes_into_utf32IT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEEmS7_mS8_: 17| 1.67k| OutputPtr utf8_output) { 18| 1.67k| size_t pos = 0; 19| 1.67k| auto start = utf8_output; 20| 39.1k| while (pos < len) { ------------------ | Branch (20:10): [True: 38.4k, False: 684] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 38.4k| { // try to convert the next block of 2 ASCII characters 25| 38.4k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (25:11): [True: 37.4k, False: 948] ------------------ 26| | // they are ascii 27| 37.4k| uint64_t v; 28| 37.4k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 29| 37.4k| if ((v & 0xFFFFFF80FFFFFF80) == 0) { ------------------ | Branch (29:13): [True: 17.1k, False: 20.3k] ------------------ 30| 17.1k| *utf8_output++ = char(data[pos]); 31| 17.1k| *utf8_output++ = char(data[pos + 1]); 32| 17.1k| pos += 2; 33| 17.1k| continue; 34| 17.1k| } 35| 37.4k| } 36| 38.4k| } 37| | 38| 21.3k| uint32_t word = data[pos]; 39| 21.3k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (39:9): [True: 2.73k, False: 18.5k] ------------------ 40| | // will generate one UTF-8 bytes 41| 2.73k| *utf8_output++ = char(word); 42| 2.73k| pos++; 43| 18.5k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (43:16): [True: 6.79k, False: 11.7k] ------------------ 44| | // will generate two UTF-8 bytes 45| | // we have 0b110XXXXX 0b10XXXXXX 46| 6.79k| *utf8_output++ = char((word >> 6) | 0b11000000); 47| 6.79k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 48| 6.79k| pos++; 49| 11.7k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (49:16): [True: 7.97k, False: 3.80k] ------------------ 50| | // will generate three UTF-8 bytes 51| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 52| 7.97k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (52:11): [True: 5.98k, False: 1.98k] | Branch (52:29): [True: 192, False: 5.79k] ------------------ 53| 192| return 0; 54| 192| } 55| 7.78k| *utf8_output++ = char((word >> 12) | 0b11100000); 56| 7.78k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 57| 7.78k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 58| 7.78k| pos++; 59| 7.78k| } else { 60| | // will generate four UTF-8 bytes 61| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 62| 3.80k| if (word > 0x10FFFF) { ------------------ | Branch (62:11): [True: 794, False: 3.01k] ------------------ 63| 794| return 0; 64| 794| } 65| 3.01k| *utf8_output++ = char((word >> 18) | 0b11110000); 66| 3.01k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 67| 3.01k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 68| 3.01k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 69| 3.01k| pos++; 70| 3.01k| } 71| 21.3k| } 72| 684| return utf8_output - start; 73| 1.67k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf32_to_utf819convert_with_errorsIPKDiPcQaasr7simdutf6detailE18indexes_into_utf32IT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEENS_6resultES7_mS8_: 81| 2.48k| OutputPtr utf8_output) { 82| 2.48k| size_t pos = 0; 83| 2.48k| auto start = utf8_output; 84| 284k| while (pos < len) { ------------------ | Branch (84:10): [True: 283k, False: 586] ------------------ 85| |#if SIMDUTF_CPLUSPLUS23 86| | if !consteval 87| |#endif 88| 283k| { // try to convert the next block of 2 ASCII characters 89| 283k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (89:11): [True: 282k, False: 934] ------------------ 90| | // they are ascii 91| 282k| uint64_t v; 92| 282k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 93| 282k| if ((v & 0xFFFFFF80FFFFFF80) == 0) { ------------------ | Branch (93:13): [True: 228k, False: 54.8k] ------------------ 94| 228k| *utf8_output++ = char(data[pos]); 95| 228k| *utf8_output++ = char(data[pos + 1]); 96| 228k| pos += 2; 97| 228k| continue; 98| 228k| } 99| 282k| } 100| 283k| } 101| | 102| 55.7k| uint32_t word = data[pos]; 103| 55.7k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (103:9): [True: 5.47k, False: 50.2k] ------------------ 104| | // will generate one UTF-8 bytes 105| 5.47k| *utf8_output++ = char(word); 106| 5.47k| pos++; 107| 50.2k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (107:16): [True: 28.4k, False: 21.8k] ------------------ 108| | // will generate two UTF-8 bytes 109| | // we have 0b110XXXXX 0b10XXXXXX 110| 28.4k| *utf8_output++ = char((word >> 6) | 0b11000000); 111| 28.4k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 112| 28.4k| pos++; 113| 28.4k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (113:16): [True: 9.28k, False: 12.5k] ------------------ 114| | // will generate three UTF-8 bytes 115| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 116| 9.28k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (116:11): [True: 5.94k, False: 3.34k] | Branch (116:29): [True: 330, False: 5.61k] ------------------ 117| 330| return result(error_code::SURROGATE, pos); 118| 330| } 119| 8.95k| *utf8_output++ = char((word >> 12) | 0b11100000); 120| 8.95k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 121| 8.95k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 122| 8.95k| pos++; 123| 12.5k| } else { 124| | // will generate four UTF-8 bytes 125| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 126| 12.5k| if (word > 0x10FFFF) { ------------------ | Branch (126:11): [True: 1.56k, False: 10.9k] ------------------ 127| 1.56k| return result(error_code::TOO_LARGE, pos); 128| 1.56k| } 129| 10.9k| *utf8_output++ = char((word >> 18) | 0b11110000); 130| 10.9k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 131| 10.9k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 132| 10.9k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 133| 10.9k| pos++; 134| 10.9k| } 135| 55.7k| } 136| 586| return result(error_code::SUCCESS, utf8_output - start); 137| 2.48k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf32_to_utf813convert_validIPKDiPcQaasr7simdutf6detailE18indexes_into_utf32IT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEEmS7_mS8_: 17| 192| OutputPtr utf8_output) { 18| 192| size_t pos = 0; 19| 192| auto start = utf8_output; 20| 15.3k| while (pos < len) { ------------------ | Branch (20:10): [True: 15.1k, False: 192] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 15.1k| { // try to convert the next block of 2 ASCII characters 25| 15.1k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (25:11): [True: 14.9k, False: 174] ------------------ 26| | // they are ascii 27| 14.9k| uint64_t v; 28| 14.9k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 29| 14.9k| if ((v & 0xFFFFFF80FFFFFF80) == 0) { ------------------ | Branch (29:13): [True: 1.41k, False: 13.5k] ------------------ 30| 1.41k| *utf8_output++ = char(data[pos]); 31| 1.41k| *utf8_output++ = char(data[pos + 1]); 32| 1.41k| pos += 2; 33| 1.41k| continue; 34| 1.41k| } 35| 14.9k| } 36| 15.1k| } 37| | 38| 13.6k| uint32_t word = data[pos]; 39| 13.6k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (39:9): [True: 636, False: 13.0k] ------------------ 40| | // will generate one UTF-8 bytes 41| 636| *utf8_output++ = char(word); 42| 636| pos++; 43| 13.0k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (43:16): [True: 7.83k, False: 5.22k] ------------------ 44| | // will generate two UTF-8 bytes 45| | // we have 0b110XXXXX 0b10XXXXXX 46| 7.83k| *utf8_output++ = char((word >> 6) | 0b11000000); 47| 7.83k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 48| 7.83k| pos++; 49| 7.83k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (49:16): [True: 1.52k, False: 3.69k] ------------------ 50| | // will generate three UTF-8 bytes 51| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 52| 1.52k| *utf8_output++ = char((word >> 12) | 0b11100000); 53| 1.52k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 54| 1.52k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 55| 1.52k| pos++; 56| 3.69k| } else { 57| | // will generate four UTF-8 bytes 58| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 59| 3.69k| *utf8_output++ = char((word >> 18) | 0b11110000); 60| 3.69k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 61| 3.69k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 62| 3.69k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 63| 3.69k| pos++; 64| 3.69k| } 65| 13.6k| } 66| 192| return utf8_output - start; 67| 192|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf88validateEPKcm: 113| 3.22k| size_t len) noexcept { 114| 3.22k| return validate(reinterpret_cast(buf), len); 115| 3.22k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf88validateIPKhEEbT_m: 14| 3.22k| size_t len) noexcept { 15| 3.22k| static_assert( 16| 3.22k| std::is_same::type, uint8_t>::value, 17| 3.22k| "dereferencing the data pointer must result in a uint8_t"); 18| 3.22k| uint64_t pos = 0; 19| 3.22k| uint32_t code_point = 0; 20| 8.18M| while (pos < len) { ------------------ | Branch (20:10): [True: 8.18M, False: 616] ------------------ 21| 8.18M| uint64_t next_pos; 22| |#if SIMDUTF_CPLUSPLUS23 23| | if !consteval 24| |#endif 25| 8.18M| { // check if the next 16 bytes are ascii. 26| 8.18M| next_pos = pos + 16; 27| 8.18M| if (next_pos <= len) { // if it is safe to read 16 more bytes, check ------------------ | Branch (27:11): [True: 8.18M, False: 3.19k] ------------------ 28| | // that they are ascii 29| 8.18M| uint64_t v1{}; 30| 8.18M| std::memcpy(&v1, data + pos, sizeof(uint64_t)); 31| 8.18M| uint64_t v2{}; 32| 8.18M| std::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 33| 8.18M| uint64_t v{v1 | v2}; 34| 8.18M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (34:13): [True: 7.12M, False: 1.06M] ------------------ 35| 7.12M| pos = next_pos; 36| 7.12M| continue; 37| 7.12M| } 38| 8.18M| } 39| 8.18M| } 40| | 41| 1.06M| unsigned char byte = data[pos]; 42| | 43| 5.48M| while (byte < 0b10000000) { ------------------ | Branch (43:12): [True: 4.41M, False: 1.06M] ------------------ 44| 4.41M| if (++pos == len) { ------------------ | Branch (44:11): [True: 1.00k, False: 4.41M] ------------------ 45| 1.00k| return true; 46| 1.00k| } 47| 4.41M| byte = data[pos]; 48| 4.41M| } 49| | 50| 1.06M| if ((byte & 0b11100000) == 0b11000000) { ------------------ | Branch (50:9): [True: 706k, False: 357k] ------------------ 51| 706k| next_pos = pos + 2; 52| 706k| if (next_pos > len) { ------------------ | Branch (52:11): [True: 67, False: 706k] ------------------ 53| 67| return false; 54| 67| } 55| 706k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (55:11): [True: 252, False: 705k] ------------------ 56| 252| return false; 57| 252| } 58| | // range check 59| 705k| code_point = (byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 60| 705k| if (code_point < 0x80) { ------------------ | Branch (60:11): [True: 30, False: 705k] ------------------ 61| 30| return false; 62| 30| } 63| 705k| } else if ((byte & 0b11110000) == 0b11100000) { ------------------ | Branch (63:16): [True: 303k, False: 54.1k] ------------------ 64| 303k| next_pos = pos + 3; 65| 303k| if (next_pos > len) { ------------------ | Branch (65:11): [True: 39, False: 303k] ------------------ 66| 39| return false; 67| 39| } 68| 303k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (68:11): [True: 102, False: 303k] ------------------ 69| 102| return false; 70| 102| } 71| 303k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (71:11): [True: 48, False: 302k] ------------------ 72| 48| return false; 73| 48| } 74| | // range check 75| 302k| code_point = (byte & 0b00001111) << 12 | 76| 302k| (data[pos + 1] & 0b00111111) << 6 | 77| 302k| (data[pos + 2] & 0b00111111); 78| 302k| if ((code_point < 0x800) || ------------------ | Branch (78:11): [True: 39, False: 302k] ------------------ 79| 302k| (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (79:12): [True: 8.41k, False: 294k] | Branch (79:35): [True: 18, False: 8.39k] ------------------ 80| 57| return false; 81| 57| } 82| 302k| } else if ((byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (82:16): [True: 53.3k, False: 762] ------------------ 83| 53.3k| next_pos = pos + 4; 84| 53.3k| if (next_pos > len) { ------------------ | Branch (84:11): [True: 28, False: 53.3k] ------------------ 85| 28| return false; 86| 28| } 87| 53.3k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (87:11): [True: 91, False: 53.2k] ------------------ 88| 91| return false; 89| 91| } 90| 53.2k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (90:11): [True: 36, False: 53.2k] ------------------ 91| 36| return false; 92| 36| } 93| 53.2k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (93:11): [True: 32, False: 53.1k] ------------------ 94| 32| return false; 95| 32| } 96| | // range check 97| 53.1k| code_point = 98| 53.1k| (byte & 0b00000111) << 18 | (data[pos + 1] & 0b00111111) << 12 | 99| 53.1k| (data[pos + 2] & 0b00111111) << 6 | (data[pos + 3] & 0b00111111); 100| 53.1k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (100:11): [True: 27, False: 53.1k] | Branch (100:35): [True: 34, False: 53.1k] ------------------ 101| 61| return false; 102| 61| } 103| 53.1k| } else { 104| | // we may have a continuation 105| 762| return false; 106| 762| } 107| 1.06M| pos = next_pos; 108| 1.06M| } 109| 616| return true; 110| 3.22k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf820validate_with_errorsEPKcm: 221| 6.18k|validate_with_errors(const char *buf, size_t len) noexcept { 222| 6.18k| return validate_with_errors(reinterpret_cast(buf), len); 223| 6.18k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf820validate_with_errorsIPKhEENS_6resultET_m: 119| 6.18k|validate_with_errors(BytePtr data, size_t len) noexcept { 120| 6.18k| static_assert( 121| 6.18k| std::is_same::type, uint8_t>::value, 122| 6.18k| "dereferencing the data pointer must result in a uint8_t"); 123| 6.18k| size_t pos = 0; 124| 6.18k| uint32_t code_point = 0; 125| 8.19M| while (pos < len) { ------------------ | Branch (125:10): [True: 8.19M, False: 616] ------------------ 126| | // check of the next 16 bytes are ascii. 127| 8.19M| size_t next_pos = pos + 16; 128| 8.19M| if (next_pos <= ------------------ | Branch (128:9): [True: 8.18M, False: 4.86k] ------------------ 129| 8.19M| len) { // if it is safe to read 16 more bytes, check that they are ascii 130| 8.18M| uint64_t v1; 131| 8.18M| std::memcpy(&v1, data + pos, sizeof(uint64_t)); 132| 8.18M| uint64_t v2; 133| 8.18M| std::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 134| 8.18M| uint64_t v{v1 | v2}; 135| 8.18M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (135:11): [True: 7.12M, False: 1.06M] ------------------ 136| 7.12M| pos = next_pos; 137| 7.12M| continue; 138| 7.12M| } 139| 8.18M| } 140| 1.07M| unsigned char byte = data[pos]; 141| | 142| 5.50M| while (byte < 0b10000000) { ------------------ | Branch (142:12): [True: 4.43M, False: 1.06M] ------------------ 143| 4.43M| if (++pos == len) { ------------------ | Branch (143:11): [True: 1.00k, False: 4.43M] ------------------ 144| 1.00k| return result(error_code::SUCCESS, len); 145| 1.00k| } 146| 4.43M| byte = data[pos]; 147| 4.43M| } 148| | 149| 1.06M| if ((byte & 0b11100000) == 0b11000000) { ------------------ | Branch (149:9): [True: 709k, False: 360k] ------------------ 150| 709k| next_pos = pos + 2; 151| 709k| if (next_pos > len) { ------------------ | Branch (151:11): [True: 201, False: 708k] ------------------ 152| 201| return result(error_code::TOO_SHORT, pos); 153| 201| } 154| 708k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (154:11): [True: 756, False: 708k] ------------------ 155| 756| return result(error_code::TOO_SHORT, pos); 156| 756| } 157| | // range check 158| 708k| code_point = (byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 159| 708k| if (code_point < 0x80) { ------------------ | Branch (159:11): [True: 90, False: 708k] ------------------ 160| 90| return result(error_code::OVERLONG, pos); 161| 90| } 162| 708k| } else if ((byte & 0b11110000) == 0b11100000) { ------------------ | Branch (162:16): [True: 304k, False: 56.2k] ------------------ 163| 304k| next_pos = pos + 3; 164| 304k| if (next_pos > len) { ------------------ | Branch (164:11): [True: 117, False: 304k] ------------------ 165| 117| return result(error_code::TOO_SHORT, pos); 166| 117| } 167| 304k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (167:11): [True: 306, False: 304k] ------------------ 168| 306| return result(error_code::TOO_SHORT, pos); 169| 306| } 170| 304k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (170:11): [True: 144, False: 304k] ------------------ 171| 144| return result(error_code::TOO_SHORT, pos); 172| 144| } 173| | // range check 174| 304k| code_point = (byte & 0b00001111) << 12 | 175| 304k| (data[pos + 1] & 0b00111111) << 6 | 176| 304k| (data[pos + 2] & 0b00111111); 177| 304k| if (code_point < 0x800) { ------------------ | Branch (177:11): [True: 117, False: 303k] ------------------ 178| 117| return result(error_code::OVERLONG, pos); 179| 117| } 180| 303k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (180:11): [True: 9.03k, False: 294k] | Branch (180:34): [True: 54, False: 8.98k] ------------------ 181| 54| return result(error_code::SURROGATE, pos); 182| 54| } 183| 303k| } else if ((byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (183:16): [True: 54.1k, False: 2.04k] ------------------ 184| 54.1k| next_pos = pos + 4; 185| 54.1k| if (next_pos > len) { ------------------ | Branch (185:11): [True: 84, False: 54.0k] ------------------ 186| 84| return result(error_code::TOO_SHORT, pos); 187| 84| } 188| 54.0k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (188:11): [True: 273, False: 53.8k] ------------------ 189| 273| return result(error_code::TOO_SHORT, pos); 190| 273| } 191| 53.8k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (191:11): [True: 108, False: 53.7k] ------------------ 192| 108| return result(error_code::TOO_SHORT, pos); 193| 108| } 194| 53.7k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (194:11): [True: 96, False: 53.6k] ------------------ 195| 96| return result(error_code::TOO_SHORT, pos); 196| 96| } 197| | // range check 198| 53.6k| code_point = 199| 53.6k| (byte & 0b00000111) << 18 | (data[pos + 1] & 0b00111111) << 12 | 200| 53.6k| (data[pos + 2] & 0b00111111) << 6 | (data[pos + 3] & 0b00111111); 201| 53.6k| if (code_point <= 0xffff) { ------------------ | Branch (201:11): [True: 81, False: 53.5k] ------------------ 202| 81| return result(error_code::OVERLONG, pos); 203| 81| } 204| 53.5k| if (0x10ffff < code_point) { ------------------ | Branch (204:11): [True: 102, False: 53.4k] ------------------ 205| 102| return result(error_code::TOO_LARGE, pos); 206| 102| } 207| 53.5k| } else { 208| | // we either have too many continuation bytes or an invalid leading byte 209| 2.04k| if ((byte & 0b11000000) == 0b10000000) { ------------------ | Branch (209:11): [True: 995, False: 1.04k] ------------------ 210| 995| return result(error_code::TOO_LONG, pos); 211| 1.04k| } else { 212| 1.04k| return result(error_code::HEADER_BITS, pos); 213| 1.04k| } 214| 2.04k| } 215| 1.06M| pos = next_pos; 216| 1.06M| } 217| 616| return result(error_code::SUCCESS, len); 218| 6.18k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf831rewind_and_validate_with_errorsEPKcS4_m: 232| 3.21k| const char *start, const char *buf, size_t len) noexcept { 233| | // First check that we start with a leading byte 234| 3.21k| if ((*start & 0b11000000) == 0b10000000) { ------------------ | Branch (234:7): [True: 244, False: 2.96k] ------------------ 235| 244| return result(error_code::TOO_LONG, 0); 236| 244| } 237| 2.96k| size_t extra_len{0}; 238| | // A leading byte cannot be further than 4 bytes away 239| 3.16k| for (int i = 0; i < 5; i++) { ------------------ | Branch (239:19): [True: 3.16k, False: 0] ------------------ 240| 3.16k| unsigned char byte = *buf; 241| 3.16k| if ((byte & 0b11000000) != 0b10000000) { ------------------ | Branch (241:9): [True: 2.96k, False: 194] ------------------ 242| 2.96k| break; 243| 2.96k| } else { 244| 194| buf--; 245| 194| extra_len++; 246| 194| } 247| 3.16k| } 248| | 249| 2.96k| result res = validate_with_errors(buf, len + extra_len); 250| 2.96k| res.count -= extra_len; // Might underflow 251| 2.96k| return res; 252| 3.21k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf817count_code_pointsIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_m: 258| 14.0k|simdutf_constexpr23 size_t count_code_points(InputPtr data, size_t len) { 259| 14.0k| size_t counter{0}; 260| 256M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (260:22): [True: 256M, False: 14.0k] ------------------ 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: 2.84M] ------------------ 264| 253M| counter++; 265| 253M| } 266| 256M| } 267| 14.0k| return counter; 268| 14.0k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf822utf16_length_from_utf8IPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_m: 274| 5.24k|simdutf_constexpr23 size_t utf16_length_from_utf8(InputPtr data, size_t len) { 275| 5.24k| size_t counter{0}; 276| 95.6M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (276:22): [True: 95.6M, False: 5.24k] ------------------ 277| 95.6M| if (int8_t(data[i]) > -65) { ------------------ | Branch (277:9): [True: 94.1M, False: 1.47M] ------------------ 278| 94.1M| counter++; 279| 94.1M| } 280| 95.6M| if (uint8_t(data[i]) >= 240) { ------------------ | Branch (280:9): [True: 130k, False: 95.4M] ------------------ 281| 130k| counter++; 282| 130k| } 283| 95.6M| } 284| 5.24k| return counter; 285| 5.24k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf8_to_latin130rewind_and_convert_with_errorsEmPKcmPc: 176| 1.02k| char *latin1_output) { 177| 1.02k| size_t extra_len{0}; 178| | // We potentially need to go back in time and find a leading byte. 179| | // In theory '3' would be sufficient, but sometimes the error can go back 180| | // quite far. 181| 1.02k| size_t how_far_back = prior_bytes; 182| | // size_t how_far_back = 3; // 3 bytes in the past + current position 183| | // if(how_far_back >= prior_bytes) { how_far_back = prior_bytes; } 184| 1.02k| bool found_leading_bytes{false}; 185| | // important: it is i <= how_far_back and not 'i < how_far_back'. 186| 1.10k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (186:22): [True: 1.02k, False: 72] ------------------ 187| 1.02k| unsigned char byte = buf[-static_cast(i)]; 188| 1.02k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 189| 1.02k| if (found_leading_bytes) { ------------------ | Branch (189:9): [True: 952, False: 76] ------------------ 190| 952| if (i > 0 && byte < 128) { ------------------ | Branch (190:11): [True: 4, False: 948] | 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| 948| buf -= i; 196| 948| extra_len = i; 197| 948| break; 198| 952| } 199| 1.02k| } 200| | // 201| | // It is possible for this function to return a negative count in its result. 202| | // C++ Standard Section 18.1 defines size_t is in which is described 203| | // in C Standard as . C Standard Section 4.1.5 defines size_t as an 204| | // unsigned integral type of the result of the sizeof operator 205| | // 206| | // An unsigned type will simply wrap round arithmetically (well defined). 207| | // 208| 1.02k| if (!found_leading_bytes) { ------------------ | Branch (208:7): [True: 72, False: 948] ------------------ 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| 72| return result(error_code::TOO_LONG, 0 - how_far_back); 214| 72| } 215| 948| result res = convert_with_errors(buf, len + extra_len, latin1_output); 216| 948| if (res.error) { ------------------ | Branch (216:7): [True: 496, False: 452] ------------------ 217| 496| res.count -= extra_len; 218| 496| } 219| 948| return res; 220| 1.02k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf8_to_latin119convert_with_errorsIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEENS_6resultES6_mPc: 92| 1.46k| char *latin_output) { 93| 1.46k| size_t pos = 0; 94| 1.46k| char *start{latin_output}; 95| | 96| 1.78M| while (pos < len) { ------------------ | Branch (96:10): [True: 1.77M, False: 682] ------------------ 97| |#if SIMDUTF_CPLUSPLUS23 98| | if !consteval 99| |#endif 100| 1.77M| { 101| | // try to convert the next block of 16 ASCII bytes 102| 1.77M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (102:11): [True: 1.77M, False: 7.01k] ------------------ 103| | // they are ascii 104| 1.77M| uint64_t v1; 105| 1.77M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 106| 1.77M| uint64_t v2; 107| 1.77M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 108| 1.77M| uint64_t v{v1 | v2}; // We are only interested in these bits: 1000 1000 109| | // 1000 1000...etc 110| 1.77M| if ((v & 0x8080808080808080) == ------------------ | Branch (110:13): [True: 1.49M, False: 277k] ------------------ 111| 1.77M| 0) { // if NONE of these are set, e.g. all of them are zero, then 112| | // everything is ASCII 113| 1.49M| size_t final_pos = pos + 16; 114| 25.4M| while (pos < final_pos) { ------------------ | Branch (114:18): [True: 23.9M, False: 1.49M] ------------------ 115| 23.9M| *latin_output++ = char(data[pos]); 116| 23.9M| pos++; 117| 23.9M| } 118| 1.49M| continue; 119| 1.49M| } 120| 1.77M| } 121| 1.77M| } 122| | // suppose it is not an all ASCII byte sequence 123| 284k| uint8_t leading_byte = data[pos]; // leading byte 124| 284k| if (leading_byte < 0b10000000) { ------------------ | Branch (124:9): [True: 247k, False: 36.8k] ------------------ 125| | // converting one ASCII byte !!! 126| 247k| *latin_output++ = char(leading_byte); 127| 247k| pos++; 128| 247k| } else if ((leading_byte & 0b11100000) == ------------------ | Branch (128:16): [True: 36.4k, False: 434] ------------------ 129| 36.8k| 0b11000000) { // the first three bits indicate: 130| | // We have a two-byte UTF-8 131| 36.4k| if (pos + 1 >= len) { ------------------ | Branch (131:11): [True: 42, False: 36.4k] ------------------ 132| 42| return result(error_code::TOO_SHORT, pos); 133| 42| } // minimal bound checking 134| 36.4k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (134:11): [True: 222, False: 36.1k] ------------------ 135| 222| return result(error_code::TOO_SHORT, pos); 136| 222| } // 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| 36.1k| uint32_t code_point = 140| 36.1k| (leading_byte & 0b00011111) << 6 | 141| 36.1k| (data[pos + 1] & 142| 36.1k| 0b00111111); // assembles the Unicode code point from the two bytes. 143| | // It does this by discarding the leading 110 and 10 144| | // bits from the two bytes, shifting the remaining bits 145| | // of the first byte, and then combining the results 146| | // with a bitwise OR operation. 147| 36.1k| if (code_point < 0x80) { ------------------ | Branch (147:11): [True: 18, False: 36.1k] ------------------ 148| 18| return result(error_code::OVERLONG, pos); 149| 18| } 150| 36.1k| if (0xFF < code_point) { ------------------ | Branch (150:11): [True: 66, False: 36.0k] ------------------ 151| 66| return result(error_code::TOO_LARGE, pos); 152| 66| } // We only care about the range 129-255 which is Non-ASCII latin1 153| | // characters 154| 36.0k| *latin_output++ = char(code_point); 155| 36.0k| pos += 2; 156| 36.0k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (156:16): [True: 102, False: 332] ------------------ 157| | // We have a three-byte UTF-8 158| 102| return result(error_code::TOO_LARGE, pos); 159| 332| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (159:16): [True: 42, False: 290] ------------------ 160| | // we have a 4-byte UTF-8 word. 161| 42| return result(error_code::TOO_LARGE, pos); 162| 290| } else { 163| | // we either have too many continuation bytes or an invalid leading byte 164| 290| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (164:11): [True: 152, False: 138] ------------------ 165| 152| return result(error_code::TOO_LONG, pos); 166| 152| } 167| | 168| 138| return result(error_code::HEADER_BITS, pos); 169| 290| } 170| 284k| } 171| 682| return result(error_code::SUCCESS, latin_output - start); 172| 1.46k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf8_to_latin17convertIPKcPcQaasr7simdutf6detailE22indexes_into_byte_likeIT_Esr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mS8_: 17| 1.16k| OutputPtr latin_output) { 18| 1.16k| size_t pos = 0; 19| 1.16k| auto start = latin_output; 20| | 21| 1.58M| while (pos < len) { ------------------ | Branch (21:10): [True: 1.58M, False: 520] ------------------ 22| |#if SIMDUTF_CPLUSPLUS23 23| | if !consteval 24| |#endif 25| 1.58M| { 26| | // try to convert the next block of 16 ASCII bytes 27| 1.58M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (27:11): [True: 1.57M, False: 4.27k] ------------------ 28| | // they are ascii 29| 1.57M| uint64_t v1; 30| 1.57M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 31| 1.57M| uint64_t v2; 32| 1.57M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 33| 1.57M| uint64_t v{v1 | v2}; // We are only interested in these bits: 1000 1000 34| | // 1000 1000 .... etc 35| 1.57M| if ((v & 0x8080808080808080) == ------------------ | Branch (35:13): [True: 1.43M, False: 144k] ------------------ 36| 1.57M| 0) { // if NONE of these are set, e.g. all of them are zero, then 37| | // everything is ASCII 38| 1.43M| size_t final_pos = pos + 16; 39| 24.3M| while (pos < final_pos) { ------------------ | Branch (39:18): [True: 22.9M, False: 1.43M] ------------------ 40| 22.9M| *latin_output++ = char(data[pos]); 41| 22.9M| pos++; 42| 22.9M| } 43| 1.43M| continue; 44| 1.43M| } 45| 1.57M| } 46| 1.58M| } 47| | 48| | // suppose it is not an all ASCII byte sequence 49| 149k| uint8_t leading_byte = data[pos]; // leading byte 50| 149k| if (leading_byte < 0b10000000) { ------------------ | Branch (50:9): [True: 131k, False: 17.9k] ------------------ 51| | // converting one ASCII byte !!! 52| 131k| *latin_output++ = char(leading_byte); 53| 131k| pos++; 54| 131k| } else if ((leading_byte & 0b11100000) == ------------------ | Branch (54:16): [True: 17.4k, False: 450] ------------------ 55| 17.9k| 0b11000000) { // the first three bits indicate: 56| | // We have a two-byte UTF-8 57| 17.4k| if (pos + 1 >= len) { ------------------ | Branch (57:11): [True: 24, False: 17.4k] ------------------ 58| 24| return 0; 59| 24| } // minimal bound checking 60| 17.4k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (60:11): [True: 108, False: 17.3k] ------------------ 61| 108| return 0; 62| 108| } // 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| 17.3k| uint32_t code_point = 66| 17.3k| (leading_byte & 0b00011111) << 6 | 67| 17.3k| (data[pos + 1] & 68| 17.3k| 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| 17.3k| if (code_point < 0x80 || 0xFF < code_point) { ------------------ | Branch (73:11): [True: 16, False: 17.3k] | Branch (73:32): [True: 44, False: 17.2k] ------------------ 74| 60| 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| 60| } 78| 17.2k| *latin_output++ = char(code_point); 79| 17.2k| pos += 2; 80| 17.2k| } else { 81| 450| return 0; 82| 450| } 83| 149k| } 84| 520| return latin_output - start; 85| 1.16k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf167convertILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_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| 2.07M| while (pos < len) { ------------------ | Branch (19:10): [True: 2.07M, False: 670] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| | // try to convert the next block of 16 ASCII bytes 24| 2.07M| { 25| 2.07M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (25:11): [True: 2.06M, False: 4.16k] ------------------ 26| | // they are ascii 27| 2.06M| uint64_t v1; 28| 2.06M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 29| 2.06M| uint64_t v2; 30| 2.06M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 31| 2.06M| uint64_t v{v1 | v2}; 32| 2.06M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (32:13): [True: 1.16M, False: 899k] ------------------ 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| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (35:31): [Folded, False: 18.6M] ------------------ 36| 18.6M| ? char16_t(u16_swap_bytes(data[pos])) 37| 18.6M| : char16_t(data[pos]); 38| 18.6M| pos++; 39| 18.6M| } 40| 1.16M| continue; 41| 1.16M| } 42| 2.06M| } 43| 2.07M| } 44| | 45| 904k| uint8_t leading_byte = data[pos]; // leading byte 46| 904k| if (leading_byte < 0b10000000) { ------------------ | Branch (46:9): [True: 740k, False: 163k] ------------------ 47| | // converting one ASCII byte !!! 48| 740k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (48:25): [Folded, False: 740k] ------------------ 49| 740k| ? char16_t(u16_swap_bytes(leading_byte)) 50| 740k| : char16_t(leading_byte); 51| 740k| pos++; 52| 740k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (52:16): [True: 97.3k, False: 65.9k] ------------------ 53| | // We have a two-byte UTF-8, it should become 54| | // a single UTF-16 word. 55| 97.3k| if (pos + 1 >= len) { ------------------ | Branch (55:11): [True: 30, False: 97.3k] ------------------ 56| 30| return 0; 57| 30| } // minimal bound checking 58| 97.3k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (58:11): [True: 72, False: 97.2k] ------------------ 59| 72| return 0; 60| 72| } 61| | // range check 62| 97.2k| uint32_t code_point = 63| 97.2k| (leading_byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 64| 97.2k| if (code_point < 0x80) { ------------------ | Branch (64:11): [True: 18, False: 97.2k] ------------------ 65| 18| return 0; 66| 18| } 67| | if constexpr (!match_system(big_endian)) { 68| | code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 69| | } 70| 97.2k| *utf16_output++ = char16_t(code_point); 71| 97.2k| pos += 2; 72| 97.2k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (72:16): [True: 59.8k, False: 6.07k] ------------------ 73| | // We have a three-byte UTF-8, it should become 74| | // a single UTF-16 word. 75| 59.8k| if (pos + 2 >= len) { ------------------ | Branch (75:11): [True: 30, False: 59.8k] ------------------ 76| 30| return 0; 77| 30| } // minimal bound checking 78| | 79| 59.8k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (79:11): [True: 42, False: 59.8k] ------------------ 80| 42| return 0; 81| 42| } 82| 59.8k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (82:11): [True: 38, False: 59.7k] ------------------ 83| 38| return 0; 84| 38| } 85| | // range check 86| 59.7k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 87| 59.7k| (data[pos + 1] & 0b00111111) << 6 | 88| 59.7k| (data[pos + 2] & 0b00111111); 89| 59.7k| if (code_point < 0x800 || (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (89:11): [True: 44, False: 59.7k] | Branch (89:34): [True: 404, False: 59.3k] | Branch (89:57): [True: 18, False: 386] ------------------ 90| 62| return 0; 91| 62| } 92| | if constexpr (!match_system(big_endian)) { 93| | code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 94| | } 95| 59.7k| *utf16_output++ = char16_t(code_point); 96| 59.7k| pos += 3; 97| 59.7k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (97:16): [True: 5.74k, False: 326] ------------------ 98| | // we have a 4-byte UTF-8 word. 99| 5.74k| if (pos + 3 >= len) { ------------------ | Branch (99:11): [True: 12, False: 5.73k] ------------------ 100| 12| return 0; 101| 12| } // minimal bound checking 102| 5.73k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (102:11): [True: 40, False: 5.69k] ------------------ 103| 40| return 0; 104| 40| } 105| 5.69k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (105:11): [True: 22, False: 5.67k] ------------------ 106| 22| return 0; 107| 22| } 108| 5.67k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (108:11): [True: 28, False: 5.64k] ------------------ 109| 28| return 0; 110| 28| } 111| | 112| | // range check 113| 5.64k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 114| 5.64k| (data[pos + 1] & 0b00111111) << 12 | 115| 5.64k| (data[pos + 2] & 0b00111111) << 6 | 116| 5.64k| (data[pos + 3] & 0b00111111); 117| 5.64k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (117:11): [True: 10, False: 5.63k] | Branch (117:35): [True: 28, False: 5.60k] ------------------ 118| 38| return 0; 119| 38| } 120| 5.60k| code_point -= 0x10000; 121| 5.60k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 122| 5.60k| 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| 5.60k| *utf16_output++ = char16_t(high_surrogate); 128| 5.60k| *utf16_output++ = char16_t(low_surrogate); 129| 5.60k| pos += 4; 130| 5.60k| } else { 131| 326| return 0; 132| 326| } 133| 904k| } 134| 670| return utf16_output - start; 135| 1.42k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf167convertILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.46k| char16_t *utf16_output) { 17| 1.46k| size_t pos = 0; 18| 1.46k| char16_t *start{utf16_output}; 19| 1.59M| while (pos < len) { ------------------ | Branch (19:10): [True: 1.59M, False: 578] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| | // try to convert the next block of 16 ASCII bytes 24| 1.59M| { 25| 1.59M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (25:11): [True: 1.59M, False: 3.67k] ------------------ 26| | // they are ascii 27| 1.59M| uint64_t v1; 28| 1.59M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 29| 1.59M| uint64_t v2; 30| 1.59M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 31| 1.59M| uint64_t v{v1 | v2}; 32| 1.59M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (32:13): [True: 1.04M, False: 545k] ------------------ 33| 1.04M| size_t final_pos = pos + 16; 34| 17.8M| while (pos < final_pos) { ------------------ | Branch (34:18): [True: 16.7M, False: 1.04M] ------------------ 35| 16.7M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (35:31): [True: 16.7M, Folded] ------------------ 36| 16.7M| ? char16_t(u16_swap_bytes(data[pos])) 37| 16.7M| : char16_t(data[pos]); 38| 16.7M| pos++; 39| 16.7M| } 40| 1.04M| continue; 41| 1.04M| } 42| 1.59M| } 43| 1.59M| } 44| | 45| 549k| uint8_t leading_byte = data[pos]; // leading byte 46| 549k| if (leading_byte < 0b10000000) { ------------------ | Branch (46:9): [True: 441k, False: 107k] ------------------ 47| | // converting one ASCII byte !!! 48| 441k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (48:25): [True: 441k, Folded] ------------------ 49| 441k| ? char16_t(u16_swap_bytes(leading_byte)) 50| 441k| : char16_t(leading_byte); 51| 441k| pos++; 52| 441k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (52:16): [True: 70.3k, False: 37.5k] ------------------ 53| | // We have a two-byte UTF-8, it should become 54| | // a single UTF-16 word. 55| 70.3k| if (pos + 1 >= len) { ------------------ | Branch (55:11): [True: 60, False: 70.3k] ------------------ 56| 60| return 0; 57| 60| } // minimal bound checking 58| 70.3k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (58:11): [True: 110, False: 70.2k] ------------------ 59| 110| return 0; 60| 110| } 61| | // range check 62| 70.2k| uint32_t code_point = 63| 70.2k| (leading_byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 64| 70.2k| if (code_point < 0x80) { ------------------ | Branch (64:11): [True: 16, False: 70.1k] ------------------ 65| 16| return 0; 66| 16| } 67| 70.1k| if constexpr (!match_system(big_endian)) { 68| 70.1k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 69| 70.1k| } 70| 70.1k| *utf16_output++ = char16_t(code_point); 71| 70.1k| pos += 2; 72| 70.1k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (72:16): [True: 29.3k, False: 8.16k] ------------------ 73| | // We have a three-byte UTF-8, it should become 74| | // a single UTF-16 word. 75| 29.3k| if (pos + 2 >= len) { ------------------ | Branch (75:11): [True: 30, False: 29.3k] ------------------ 76| 30| return 0; 77| 30| } // minimal bound checking 78| | 79| 29.3k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (79:11): [True: 42, False: 29.3k] ------------------ 80| 42| return 0; 81| 42| } 82| 29.3k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (82:11): [True: 46, False: 29.2k] ------------------ 83| 46| return 0; 84| 46| } 85| | // range check 86| 29.2k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 87| 29.2k| (data[pos + 1] & 0b00111111) << 6 | 88| 29.2k| (data[pos + 2] & 0b00111111); 89| 29.2k| if (code_point < 0x800 || (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (89:11): [True: 38, False: 29.2k] | Branch (89:34): [True: 568, False: 28.6k] | Branch (89:57): [True: 16, False: 552] ------------------ 90| 54| return 0; 91| 54| } 92| 29.2k| if constexpr (!match_system(big_endian)) { 93| 29.2k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 94| 29.2k| } 95| 29.2k| *utf16_output++ = char16_t(code_point); 96| 29.2k| pos += 3; 97| 29.2k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (97:16): [True: 7.79k, False: 368] ------------------ 98| | // we have a 4-byte UTF-8 word. 99| 7.79k| if (pos + 3 >= len) { ------------------ | Branch (99:11): [True: 42, False: 7.75k] ------------------ 100| 42| return 0; 101| 42| } // minimal bound checking 102| 7.75k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (102:11): [True: 44, False: 7.70k] ------------------ 103| 44| return 0; 104| 44| } 105| 7.70k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (105:11): [True: 16, False: 7.69k] ------------------ 106| 16| return 0; 107| 16| } 108| 7.69k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (108:11): [True: 24, False: 7.66k] ------------------ 109| 24| return 0; 110| 24| } 111| | 112| | // range check 113| 7.66k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 114| 7.66k| (data[pos + 1] & 0b00111111) << 12 | 115| 7.66k| (data[pos + 2] & 0b00111111) << 6 | 116| 7.66k| (data[pos + 3] & 0b00111111); 117| 7.66k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (117:11): [True: 12, False: 7.65k] | Branch (117:35): [True: 22, False: 7.63k] ------------------ 118| 34| return 0; 119| 34| } 120| 7.63k| code_point -= 0x10000; 121| 7.63k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 122| 7.63k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 123| 7.63k| if constexpr (!match_system(big_endian)) { 124| 7.63k| high_surrogate = u16_swap_bytes(high_surrogate); 125| 7.63k| low_surrogate = u16_swap_bytes(low_surrogate); 126| 7.63k| } 127| 7.63k| *utf16_output++ = char16_t(high_surrogate); 128| 7.63k| *utf16_output++ = char16_t(low_surrogate); 129| 7.63k| pos += 4; 130| 7.63k| } else { 131| 368| return 0; 132| 368| } 133| 549k| } 134| 578| return utf16_output - start; 135| 1.46k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1619convert_with_errorsILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEENS_6resultES7_mPDs: 142| 1.64k| char16_t *utf16_output) { 143| 1.64k| size_t pos = 0; 144| 1.64k| char16_t *start{utf16_output}; 145| 3.10M| while (pos < len) { ------------------ | Branch (145:10): [True: 3.10M, False: 730] ------------------ 146| |#if SIMDUTF_CPLUSPLUS23 147| | if !consteval 148| |#endif 149| 3.10M| { 150| | // try to convert the next block of 16 ASCII bytes 151| 3.10M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (151:11): [True: 3.09M, False: 5.53k] ------------------ 152| | // they are ascii 153| 3.09M| uint64_t v1; 154| 3.09M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 155| 3.09M| uint64_t v2; 156| 3.09M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 157| 3.09M| uint64_t v{v1 | v2}; 158| 3.09M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (158:13): [True: 1.38M, False: 1.71M] ------------------ 159| 1.38M| size_t final_pos = pos + 16; 160| 23.5M| while (pos < final_pos) { ------------------ | Branch (160:18): [True: 22.1M, False: 1.38M] ------------------ 161| 22.1M| const char16_t byte = uint8_t(data[pos]); 162| 22.1M| *utf16_output++ = 163| 22.1M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (163:17): [Folded, False: 22.1M] ------------------ 164| 22.1M| pos++; 165| 22.1M| } 166| 1.38M| continue; 167| 1.38M| } 168| 3.09M| } 169| 3.10M| } 170| | 171| 1.71M| auto leading_byte = uint8_t(data[pos]); // leading byte 172| 1.71M| if (leading_byte < 0b10000000) { ------------------ | Branch (172:9): [True: 1.41M, False: 298k] ------------------ 173| | // converting one ASCII byte !!! 174| 1.41M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (174:25): [Folded, False: 1.41M] ------------------ 175| 1.41M| ? char16_t(u16_swap_bytes(leading_byte)) 176| 1.41M| : char16_t(leading_byte); 177| 1.41M| pos++; 178| 1.41M| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (178:16): [True: 193k, False: 104k] ------------------ 179| | // We have a two-byte UTF-8, it should become 180| | // a single UTF-16 word. 181| 193k| if (pos + 1 >= len) { ------------------ | Branch (181:11): [True: 54, False: 193k] ------------------ 182| 54| return result(error_code::TOO_SHORT, pos); 183| 54| } // minimal bound checking 184| 193k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (184:11): [True: 102, False: 193k] ------------------ 185| 102| return result(error_code::TOO_SHORT, pos); 186| 102| } 187| | // range check 188| 193k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 189| 193k| (uint8_t(data[pos + 1]) & 0b00111111); 190| 193k| if (code_point < 0x80) { ------------------ | Branch (190:11): [True: 30, False: 193k] ------------------ 191| 30| return result(error_code::OVERLONG, pos); 192| 30| } 193| | if constexpr (!match_system(big_endian)) { 194| | code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 195| | } 196| 193k| *utf16_output++ = char16_t(code_point); 197| 193k| pos += 2; 198| 193k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (198:16): [True: 87.0k, False: 17.5k] ------------------ 199| | // We have a three-byte UTF-8, it should become 200| | // a single UTF-16 word. 201| 87.0k| if (pos + 2 >= len) { ------------------ | Branch (201:11): [True: 36, False: 86.9k] ------------------ 202| 36| return result(error_code::TOO_SHORT, pos); 203| 36| } // minimal bound checking 204| | 205| 86.9k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (205:11): [True: 72, False: 86.9k] ------------------ 206| 72| return result(error_code::TOO_SHORT, pos); 207| 72| } 208| 86.9k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (208:11): [True: 30, False: 86.8k] ------------------ 209| 30| return result(error_code::TOO_SHORT, pos); 210| 30| } 211| | // range check 212| 86.8k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 213| 86.8k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 214| 86.8k| (uint8_t(data[pos + 2]) & 0b00111111); 215| 86.8k| if (code_point < 0x800) { ------------------ | Branch (215:11): [True: 18, False: 86.8k] ------------------ 216| 18| return result(error_code::OVERLONG, pos); 217| 18| } 218| 86.8k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (218:11): [True: 2.04k, False: 84.8k] | Branch (218:34): [True: 12, False: 2.03k] ------------------ 219| 12| return result(error_code::SURROGATE, pos); 220| 12| } 221| | if constexpr (!match_system(big_endian)) { 222| | code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 223| | } 224| 86.8k| *utf16_output++ = char16_t(code_point); 225| 86.8k| pos += 3; 226| 86.8k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (226:16): [True: 17.1k, False: 376] ------------------ 227| | // we have a 4-byte UTF-8 word. 228| 17.1k| if (pos + 3 >= len) { ------------------ | Branch (228:11): [True: 18, False: 17.1k] ------------------ 229| 18| return result(error_code::TOO_SHORT, pos); 230| 18| } // minimal bound checking 231| 17.1k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (231:11): [True: 72, False: 17.1k] ------------------ 232| 72| return result(error_code::TOO_SHORT, pos); 233| 72| } 234| 17.1k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (234:11): [True: 18, False: 17.0k] ------------------ 235| 18| return result(error_code::TOO_SHORT, pos); 236| 18| } 237| 17.0k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (237:11): [True: 24, False: 17.0k] ------------------ 238| 24| return result(error_code::TOO_SHORT, pos); 239| 24| } 240| | 241| | // range check 242| 17.0k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 243| 17.0k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 244| 17.0k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 245| 17.0k| (uint8_t(data[pos + 3]) & 0b00111111); 246| 17.0k| if (code_point <= 0xffff) { ------------------ | Branch (246:11): [True: 12, False: 17.0k] ------------------ 247| 12| return result(error_code::OVERLONG, pos); 248| 12| } 249| 17.0k| if (0x10ffff < code_point) { ------------------ | Branch (249:11): [True: 42, False: 17.0k] ------------------ 250| 42| return result(error_code::TOO_LARGE, pos); 251| 42| } 252| 17.0k| code_point -= 0x10000; 253| 17.0k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 254| 17.0k| 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| 17.0k| *utf16_output++ = char16_t(high_surrogate); 260| 17.0k| *utf16_output++ = char16_t(low_surrogate); 261| 17.0k| pos += 4; 262| 17.0k| } else { 263| | // we either have too many continuation bytes or an invalid leading byte 264| 376| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (264:11): [True: 154, False: 222] ------------------ 265| 154| return result(error_code::TOO_LONG, pos); 266| 222| } else { 267| 222| return result(error_code::HEADER_BITS, pos); 268| 222| } 269| 376| } 270| 1.71M| } 271| 730| return result(error_code::SUCCESS, utf16_output - start); 272| 1.64k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1619convert_with_errorsILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEENS_6resultES7_mPDs: 142| 2.01k| char16_t *utf16_output) { 143| 2.01k| size_t pos = 0; 144| 2.01k| char16_t *start{utf16_output}; 145| 3.62M| while (pos < len) { ------------------ | Branch (145:10): [True: 3.62M, False: 934] ------------------ 146| |#if SIMDUTF_CPLUSPLUS23 147| | if !consteval 148| |#endif 149| 3.62M| { 150| | // try to convert the next block of 16 ASCII bytes 151| 3.62M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (151:11): [True: 3.61M, False: 8.21k] ------------------ 152| | // they are ascii 153| 3.61M| uint64_t v1; 154| 3.61M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 155| 3.61M| uint64_t v2; 156| 3.61M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 157| 3.61M| uint64_t v{v1 | v2}; 158| 3.61M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (158:13): [True: 1.52M, False: 2.08M] ------------------ 159| 1.52M| size_t final_pos = pos + 16; 160| 25.9M| while (pos < final_pos) { ------------------ | Branch (160:18): [True: 24.4M, False: 1.52M] ------------------ 161| 24.4M| const char16_t byte = uint8_t(data[pos]); 162| 24.4M| *utf16_output++ = 163| 24.4M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (163:17): [True: 24.4M, Folded] ------------------ 164| 24.4M| pos++; 165| 24.4M| } 166| 1.52M| continue; 167| 1.52M| } 168| 3.61M| } 169| 3.62M| } 170| | 171| 2.09M| auto leading_byte = uint8_t(data[pos]); // leading byte 172| 2.09M| if (leading_byte < 0b10000000) { ------------------ | Branch (172:9): [True: 1.72M, False: 369k] ------------------ 173| | // converting one ASCII byte !!! 174| 1.72M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (174:25): [True: 1.72M, Folded] ------------------ 175| 1.72M| ? char16_t(u16_swap_bytes(leading_byte)) 176| 1.72M| : char16_t(leading_byte); 177| 1.72M| pos++; 178| 1.72M| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (178:16): [True: 227k, False: 141k] ------------------ 179| | // We have a two-byte UTF-8, it should become 180| | // a single UTF-16 word. 181| 227k| if (pos + 1 >= len) { ------------------ | Branch (181:11): [True: 60, False: 227k] ------------------ 182| 60| return result(error_code::TOO_SHORT, pos); 183| 60| } // minimal bound checking 184| 227k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (184:11): [True: 126, False: 227k] ------------------ 185| 126| return result(error_code::TOO_SHORT, pos); 186| 126| } 187| | // range check 188| 227k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 189| 227k| (uint8_t(data[pos + 1]) & 0b00111111); 190| 227k| if (code_point < 0x80) { ------------------ | Branch (190:11): [True: 12, False: 227k] ------------------ 191| 12| return result(error_code::OVERLONG, pos); 192| 12| } 193| 227k| if constexpr (!match_system(big_endian)) { 194| 227k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 195| 227k| } 196| 227k| *utf16_output++ = char16_t(code_point); 197| 227k| pos += 2; 198| 227k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (198:16): [True: 131k, False: 10.6k] ------------------ 199| | // We have a three-byte UTF-8, it should become 200| | // a single UTF-16 word. 201| 131k| if (pos + 2 >= len) { ------------------ | Branch (201:11): [True: 30, False: 131k] ------------------ 202| 30| return result(error_code::TOO_SHORT, pos); 203| 30| } // minimal bound checking 204| | 205| 131k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (205:11): [True: 90, False: 131k] ------------------ 206| 90| return result(error_code::TOO_SHORT, pos); 207| 90| } 208| 131k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (208:11): [True: 36, False: 131k] ------------------ 209| 36| return result(error_code::TOO_SHORT, pos); 210| 36| } 211| | // range check 212| 131k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 213| 131k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 214| 131k| (uint8_t(data[pos + 2]) & 0b00111111); 215| 131k| if (code_point < 0x800) { ------------------ | Branch (215:11): [True: 30, False: 131k] ------------------ 216| 30| return result(error_code::OVERLONG, pos); 217| 30| } 218| 131k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (218:11): [True: 3.53k, False: 127k] | Branch (218:34): [True: 18, False: 3.51k] ------------------ 219| 18| return result(error_code::SURROGATE, pos); 220| 18| } 221| 131k| if constexpr (!match_system(big_endian)) { 222| 131k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 223| 131k| } 224| 131k| *utf16_output++ = char16_t(code_point); 225| 131k| pos += 3; 226| 131k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (226:16): [True: 10.1k, False: 472] ------------------ 227| | // we have a 4-byte UTF-8 word. 228| 10.1k| if (pos + 3 >= len) { ------------------ | Branch (228:11): [True: 42, False: 10.1k] ------------------ 229| 42| return result(error_code::TOO_SHORT, pos); 230| 42| } // minimal bound checking 231| 10.1k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (231:11): [True: 24, False: 10.1k] ------------------ 232| 24| return result(error_code::TOO_SHORT, pos); 233| 24| } 234| 10.1k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (234:11): [True: 42, False: 10.0k] ------------------ 235| 42| return result(error_code::TOO_SHORT, pos); 236| 42| } 237| 10.0k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (237:11): [True: 36, False: 10.0k] ------------------ 238| 36| return result(error_code::TOO_SHORT, pos); 239| 36| } 240| | 241| | // range check 242| 10.0k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 243| 10.0k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 244| 10.0k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 245| 10.0k| (uint8_t(data[pos + 3]) & 0b00111111); 246| 10.0k| if (code_point <= 0xffff) { ------------------ | Branch (246:11): [True: 36, False: 9.98k] ------------------ 247| 36| return result(error_code::OVERLONG, pos); 248| 36| } 249| 9.98k| if (0x10ffff < code_point) { ------------------ | Branch (249:11): [True: 30, False: 9.95k] ------------------ 250| 30| return result(error_code::TOO_LARGE, pos); 251| 30| } 252| 9.95k| code_point -= 0x10000; 253| 9.95k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 254| 9.95k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 255| 9.95k| if constexpr (!match_system(big_endian)) { 256| 9.95k| high_surrogate = u16_swap_bytes(high_surrogate); 257| 9.95k| low_surrogate = u16_swap_bytes(low_surrogate); 258| 9.95k| } 259| 9.95k| *utf16_output++ = char16_t(high_surrogate); 260| 9.95k| *utf16_output++ = char16_t(low_surrogate); 261| 9.95k| pos += 4; 262| 9.95k| } else { 263| | // we either have too many continuation bytes or an invalid leading byte 264| 472| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (264:11): [True: 208, False: 264] ------------------ 265| 208| return result(error_code::TOO_LONG, pos); 266| 264| } else { 267| 264| return result(error_code::HEADER_BITS, pos); 268| 264| } 269| 472| } 270| 2.09M| } 271| 934| return result(error_code::SUCCESS, utf16_output - start); 272| 2.01k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1630rewind_and_convert_with_errorsILNS_10endiannessE0EEENS_6resultEmPKcmPDs: 292| 1.14k| char16_t *utf16_output) { 293| 1.14k| 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| 1.14k| 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| 1.14k| bool found_leading_bytes{false}; 301| | // important: it is i <= how_far_back and not 'i < how_far_back'. 302| 1.20k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (302:22): [True: 1.15k, False: 56] ------------------ 303| 1.15k| unsigned char byte = buf[-static_cast(i)]; 304| 1.15k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 305| 1.15k| if (found_leading_bytes) { ------------------ | Branch (305:9): [True: 1.08k, False: 68] ------------------ 306| 1.08k| if (i > 0 && byte < 128) { ------------------ | Branch (306:11): [True: 12, False: 1.07k] | 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| 1.07k| buf -= i; 312| 1.07k| extra_len = i; 313| 1.07k| break; 314| 1.08k| } 315| 1.15k| } 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| 1.12k| if (!found_leading_bytes) { ------------------ | Branch (324:7): [True: 56, False: 1.07k] ------------------ 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| 56| return result(error_code::TOO_LONG, 0 - how_far_back); 330| 56| } 331| 1.07k| result res = convert_with_errors(buf, len + extra_len, utf16_output); 332| 1.07k| if (res.error) { ------------------ | Branch (332:7): [True: 588, False: 484] ------------------ 333| 588| res.count -= extra_len; 334| 588| } 335| 1.07k| return res; 336| 1.12k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1630rewind_and_convert_with_errorsILNS_10endiannessE1EEENS_6resultEmPKcmPDs: 292| 1.39k| char16_t *utf16_output) { 293| 1.39k| 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| 1.39k| 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| 1.39k| bool found_leading_bytes{false}; 301| | // important: it is i <= how_far_back and not 'i < how_far_back'. 302| 1.47k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (302:22): [True: 1.40k, False: 72] ------------------ 303| 1.40k| unsigned char byte = buf[-static_cast(i)]; 304| 1.40k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 305| 1.40k| if (found_leading_bytes) { ------------------ | Branch (305:9): [True: 1.32k, False: 80] ------------------ 306| 1.32k| if (i > 0 && byte < 128) { ------------------ | Branch (306:11): [True: 8, False: 1.31k] | Branch (306:20): [True: 8, False: 0] ------------------ 307| | // If we had to go back and the leading byte is ascii 308| | // then we can stop right away. 309| 8| return result(error_code::TOO_LONG, 0 - i + 1); 310| 8| } 311| 1.31k| buf -= i; 312| 1.31k| extra_len = i; 313| 1.31k| break; 314| 1.32k| } 315| 1.40k| } 316| | // 317| | // It is possible for this function to return a negative count in its result. 318| | // C++ Standard Section 18.1 defines size_t is in which is described 319| | // in C Standard as . C Standard Section 4.1.5 defines size_t as an 320| | // unsigned integral type of the result of the sizeof operator 321| | // 322| | // An unsigned type will simply wrap round arithmetically (well defined). 323| | // 324| 1.38k| if (!found_leading_bytes) { ------------------ | Branch (324:7): [True: 72, False: 1.31k] ------------------ 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| 72| return result(error_code::TOO_LONG, 0 - how_far_back); 330| 72| } 331| 1.31k| result res = convert_with_errors(buf, len + extra_len, utf16_output); 332| 1.31k| if (res.error) { ------------------ | Branch (332:7): [True: 696, False: 620] ------------------ 333| 696| res.count -= extra_len; 334| 696| } 335| 1.31k| return res; 336| 1.38k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.10k| char16_t *utf16_output) { 17| 1.10k| size_t pos = 0; 18| 1.10k| char16_t *start{utf16_output}; 19| 4.10M| while (pos < len) { ------------------ | Branch (19:10): [True: 4.10M, False: 1.10k] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 4.10M| { // try to convert the next block of 8 ASCII bytes 24| 4.10M| if (pos + 8 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (24:11): [True: 4.10M, False: 3.36k] ------------------ 25| | // they are ascii 26| 4.10M| uint64_t v; 27| 4.10M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 28| 4.10M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (28:13): [True: 2.85M, False: 1.24M] ------------------ 29| 2.85M| size_t final_pos = pos + 8; 30| 25.6M| while (pos < final_pos) { ------------------ | Branch (30:18): [True: 22.8M, False: 2.85M] ------------------ 31| 22.8M| const char16_t byte = uint8_t(data[pos]); 32| 22.8M| *utf16_output++ = 33| 22.8M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (33:17): [Folded, False: 22.8M] ------------------ 34| 22.8M| pos++; 35| 22.8M| } 36| 2.85M| continue; 37| 2.85M| } 38| 4.10M| } 39| 4.10M| } 40| | 41| 1.25M| auto leading_byte = uint8_t(data[pos]); // leading byte 42| 1.25M| if (leading_byte < 0b10000000) { ------------------ | Branch (42:9): [True: 845k, False: 406k] ------------------ 43| | // converting one ASCII byte !!! 44| 845k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (44:25): [Folded, False: 845k] ------------------ 45| 845k| ? char16_t(u16_swap_bytes(leading_byte)) 46| 845k| : char16_t(leading_byte); 47| 845k| pos++; 48| 845k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (48:16): [True: 237k, False: 169k] ------------------ 49| | // We have a two-byte UTF-8, it should become 50| | // a single UTF-16 word. 51| 237k| if (pos + 1 >= len) { ------------------ | Branch (51:11): [True: 0, False: 237k] ------------------ 52| 0| break; 53| 0| } // minimal bound checking 54| 237k| uint16_t code_point = uint16_t(((leading_byte & 0b00011111) << 6) | 55| 237k| (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| 237k| *utf16_output++ = char16_t(code_point); 60| 237k| pos += 2; 61| 237k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (61:16): [True: 154k, False: 14.7k] ------------------ 62| | // We have a three-byte UTF-8, it should become 63| | // a single UTF-16 word. 64| 154k| if (pos + 2 >= len) { ------------------ | Branch (64:11): [True: 0, False: 154k] ------------------ 65| 0| break; 66| 0| } // minimal bound checking 67| 154k| uint16_t code_point = 68| 154k| uint16_t(((leading_byte & 0b00001111) << 12) | 69| 154k| ((uint8_t(data[pos + 1]) & 0b00111111) << 6) | 70| 154k| (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| 154k| *utf16_output++ = char16_t(code_point); 75| 154k| pos += 3; 76| 154k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (76:16): [True: 14.7k, False: 0] ------------------ 77| | // we have a 4-byte UTF-8 word. 78| 14.7k| if (pos + 3 >= len) { ------------------ | Branch (78:11): [True: 0, False: 14.7k] ------------------ 79| 0| break; 80| 0| } // minimal bound checking 81| 14.7k| uint32_t code_point = ((leading_byte & 0b00000111) << 18) | 82| 14.7k| ((uint8_t(data[pos + 1]) & 0b00111111) << 12) | 83| 14.7k| ((uint8_t(data[pos + 2]) & 0b00111111) << 6) | 84| 14.7k| (uint8_t(data[pos + 3]) & 0b00111111); 85| 14.7k| code_point -= 0x10000; 86| 14.7k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 87| 14.7k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 88| | if constexpr (!match_system(big_endian)) { 89| | high_surrogate = u16_swap_bytes(high_surrogate); 90| | low_surrogate = u16_swap_bytes(low_surrogate); 91| | } 92| 14.7k| *utf16_output++ = char16_t(high_surrogate); 93| 14.7k| *utf16_output++ = char16_t(low_surrogate); 94| 14.7k| pos += 4; 95| 14.7k| } else { 96| | // we may have a continuation but we do not do error checking 97| 0| return 0; 98| 0| } 99| 1.25M| } 100| 1.10k| return utf16_output - start; 101| 1.10k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.31k| char16_t *utf16_output) { 17| 1.31k| size_t pos = 0; 18| 1.31k| char16_t *start{utf16_output}; 19| 2.99M| while (pos < len) { ------------------ | Branch (19:10): [True: 2.98M, False: 1.31k] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 2.98M| { // try to convert the next block of 8 ASCII bytes 24| 2.98M| if (pos + 8 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (24:11): [True: 2.98M, False: 4.19k] ------------------ 25| | // they are ascii 26| 2.98M| uint64_t v; 27| 2.98M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 28| 2.98M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (28:13): [True: 2.44M, False: 541k] ------------------ 29| 2.44M| size_t final_pos = pos + 8; 30| 21.9M| while (pos < final_pos) { ------------------ | Branch (30:18): [True: 19.5M, False: 2.44M] ------------------ 31| 19.5M| const char16_t byte = uint8_t(data[pos]); 32| 19.5M| *utf16_output++ = 33| 19.5M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (33:17): [True: 19.5M, Folded] ------------------ 34| 19.5M| pos++; 35| 19.5M| } 36| 2.44M| continue; 37| 2.44M| } 38| 2.98M| } 39| 2.98M| } 40| | 41| 545k| auto leading_byte = uint8_t(data[pos]); // leading byte 42| 545k| if (leading_byte < 0b10000000) { ------------------ | Branch (42:9): [True: 294k, False: 250k] ------------------ 43| | // converting one ASCII byte !!! 44| 294k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (44:25): [True: 294k, Folded] ------------------ 45| 294k| ? char16_t(u16_swap_bytes(leading_byte)) 46| 294k| : char16_t(leading_byte); 47| 294k| pos++; 48| 294k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (48:16): [True: 200k, False: 49.8k] ------------------ 49| | // We have a two-byte UTF-8, it should become 50| | // a single UTF-16 word. 51| 200k| if (pos + 1 >= len) { ------------------ | Branch (51:11): [True: 0, False: 200k] ------------------ 52| 0| break; 53| 0| } // minimal bound checking 54| 200k| uint16_t code_point = uint16_t(((leading_byte & 0b00011111) << 6) | 55| 200k| (uint8_t(data[pos + 1]) & 0b00111111)); 56| 200k| if constexpr (!match_system(big_endian)) { 57| 200k| code_point = u16_swap_bytes(uint16_t(code_point)); 58| 200k| } 59| 200k| *utf16_output++ = char16_t(code_point); 60| 200k| pos += 2; 61| 200k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (61:16): [True: 28.2k, False: 21.6k] ------------------ 62| | // We have a three-byte UTF-8, it should become 63| | // a single UTF-16 word. 64| 28.2k| if (pos + 2 >= len) { ------------------ | Branch (64:11): [True: 0, False: 28.2k] ------------------ 65| 0| break; 66| 0| } // minimal bound checking 67| 28.2k| uint16_t code_point = 68| 28.2k| uint16_t(((leading_byte & 0b00001111) << 12) | 69| 28.2k| ((uint8_t(data[pos + 1]) & 0b00111111) << 6) | 70| 28.2k| (uint8_t(data[pos + 2]) & 0b00111111)); 71| 28.2k| if constexpr (!match_system(big_endian)) { 72| 28.2k| code_point = u16_swap_bytes(uint16_t(code_point)); 73| 28.2k| } 74| 28.2k| *utf16_output++ = char16_t(code_point); 75| 28.2k| pos += 3; 76| 28.2k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (76:16): [True: 21.6k, False: 0] ------------------ 77| | // we have a 4-byte UTF-8 word. 78| 21.6k| if (pos + 3 >= len) { ------------------ | Branch (78:11): [True: 0, False: 21.6k] ------------------ 79| 0| break; 80| 0| } // minimal bound checking 81| 21.6k| uint32_t code_point = ((leading_byte & 0b00000111) << 18) | 82| 21.6k| ((uint8_t(data[pos + 1]) & 0b00111111) << 12) | 83| 21.6k| ((uint8_t(data[pos + 2]) & 0b00111111) << 6) | 84| 21.6k| (uint8_t(data[pos + 3]) & 0b00111111); 85| 21.6k| code_point -= 0x10000; 86| 21.6k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 87| 21.6k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 88| 21.6k| if constexpr (!match_system(big_endian)) { 89| 21.6k| high_surrogate = u16_swap_bytes(high_surrogate); 90| 21.6k| low_surrogate = u16_swap_bytes(low_surrogate); 91| 21.6k| } 92| 21.6k| *utf16_output++ = char16_t(high_surrogate); 93| 21.6k| *utf16_output++ = char16_t(low_surrogate); 94| 21.6k| pos += 4; 95| 21.6k| } else { 96| | // we may have a continuation but we do not do error checking 97| 0| return 0; 98| 0| } 99| 545k| } 100| 1.31k| return utf16_output - start; 101| 1.31k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf3230rewind_and_convert_with_errorsEmPKcmPDi: 249| 1.46k| char32_t *utf32_output) { 250| 1.46k| size_t extra_len{0}; 251| | // We potentially need to go back in time and find a leading byte. 252| 1.46k| size_t how_far_back = 3; // 3 bytes in the past + current position 253| 1.46k| if (how_far_back > prior_bytes) { ------------------ | Branch (253:7): [True: 760, False: 708] ------------------ 254| 760| how_far_back = prior_bytes; 255| 760| } 256| 1.46k| bool found_leading_bytes{false}; 257| | // important: it is i <= how_far_back and not 'i < how_far_back'. 258| 1.54k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (258:22): [True: 1.47k, False: 68] ------------------ 259| 1.47k| unsigned char byte = buf[-static_cast(i)]; 260| 1.47k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 261| 1.47k| if (found_leading_bytes) { ------------------ | Branch (261:9): [True: 1.40k, False: 72] ------------------ 262| 1.40k| if (i > 0 && byte < 128) { ------------------ | Branch (262:11): [True: 4, False: 1.39k] | Branch (262:20): [True: 4, False: 0] ------------------ 263| | // If we had to go back and the leading byte is ascii 264| | // then we can stop right away. 265| 4| return result(error_code::TOO_LONG, 0 - i + 1); 266| 4| } 267| 1.39k| buf -= i; 268| 1.39k| extra_len = i; 269| 1.39k| break; 270| 1.40k| } 271| 1.47k| } 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| 1.46k| if (!found_leading_bytes) { ------------------ | Branch (280:7): [True: 68, False: 1.39k] ------------------ 281| | // If how_far_back == 3, we may have four consecutive continuation bytes!!! 282| | // [....] [continuation] [continuation] [continuation] | [buf is 283| | // continuation] Or we possibly have a stream that does not start with a 284| | // leading byte. 285| 68| return result(error_code::TOO_LONG, 0 - how_far_back); 286| 68| } 287| | 288| 1.39k| result res = convert_with_errors(buf, len + extra_len, utf32_output); 289| 1.39k| if (res.error) { ------------------ | Branch (289:7): [True: 724, False: 672] ------------------ 290| 724| res.count -= extra_len; 291| 724| } 292| 1.39k| return res; 293| 1.46k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf3219convert_with_errorsIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEENS_6resultES6_mPDi: 121| 2.13k| char32_t *utf32_output) { 122| 2.13k| size_t pos = 0; 123| 2.13k| char32_t *start{utf32_output}; 124| 2.35M| while (pos < len) { ------------------ | Branch (124:10): [True: 2.35M, False: 1.01k] ------------------ 125| |#if SIMDUTF_CPLUSPLUS23 126| | if !consteval 127| |#endif 128| 2.35M| { 129| | // try to convert the next block of 16 ASCII bytes 130| 2.35M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (130:11): [True: 2.34M, False: 9.21k] ------------------ 131| | // they are ascii 132| 2.34M| uint64_t v1; 133| 2.34M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 134| 2.34M| uint64_t v2; 135| 2.34M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 136| 2.34M| uint64_t v{v1 | v2}; 137| 2.34M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (137:13): [True: 1.69M, False: 654k] ------------------ 138| 1.69M| size_t final_pos = pos + 16; 139| 28.7M| while (pos < final_pos) { ------------------ | Branch (139:18): [True: 27.0M, False: 1.69M] ------------------ 140| 27.0M| *utf32_output++ = uint8_t(data[pos]); 141| 27.0M| pos++; 142| 27.0M| } 143| 1.69M| continue; 144| 1.69M| } 145| 2.34M| } 146| 2.35M| } 147| 663k| auto leading_byte = uint8_t(data[pos]); // leading byte 148| 663k| if (leading_byte < 0b10000000) { ------------------ | Branch (148:9): [True: 505k, False: 158k] ------------------ 149| | // converting one ASCII byte !!! 150| 505k| *utf32_output++ = char32_t(leading_byte); 151| 505k| pos++; 152| 505k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (152:16): [True: 112k, False: 46.2k] ------------------ 153| | // We have a two-byte UTF-8 154| 112k| if (pos + 1 >= len) { ------------------ | Branch (154:11): [True: 72, False: 112k] ------------------ 155| 72| return result(error_code::TOO_SHORT, pos); 156| 72| } // minimal bound checking 157| 112k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (157:11): [True: 186, False: 111k] ------------------ 158| 186| return result(error_code::TOO_SHORT, pos); 159| 186| } 160| | // range check 161| 111k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 162| 111k| (uint8_t(data[pos + 1]) & 0b00111111); 163| 111k| if (code_point < 0x80) { ------------------ | Branch (163:11): [True: 12, False: 111k] ------------------ 164| 12| return result(error_code::OVERLONG, pos); 165| 12| } 166| 111k| *utf32_output++ = char32_t(code_point); 167| 111k| pos += 2; 168| 111k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (168:16): [True: 37.5k, False: 8.63k] ------------------ 169| | // We have a three-byte UTF-8 170| 37.5k| if (pos + 2 >= len) { ------------------ | Branch (170:11): [True: 42, False: 37.5k] ------------------ 171| 42| return result(error_code::TOO_SHORT, pos); 172| 42| } // minimal bound checking 173| | 174| 37.5k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (174:11): [True: 66, False: 37.4k] ------------------ 175| 66| return result(error_code::TOO_SHORT, pos); 176| 66| } 177| 37.4k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (177:11): [True: 48, False: 37.4k] ------------------ 178| 48| return result(error_code::TOO_SHORT, pos); 179| 48| } 180| | // range check 181| 37.4k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 182| 37.4k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 183| 37.4k| (uint8_t(data[pos + 2]) & 0b00111111); 184| 37.4k| if (code_point < 0x800) { ------------------ | Branch (184:11): [True: 36, False: 37.3k] ------------------ 185| 36| return result(error_code::OVERLONG, pos); 186| 36| } 187| 37.3k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (187:11): [True: 3.81k, False: 33.5k] | Branch (187:34): [True: 6, False: 3.80k] ------------------ 188| 6| return result(error_code::SURROGATE, pos); 189| 6| } 190| 37.3k| *utf32_output++ = char32_t(code_point); 191| 37.3k| pos += 3; 192| 37.3k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (192:16): [True: 8.17k, False: 462] ------------------ 193| | // we have a 4-byte UTF-8 word. 194| 8.17k| if (pos + 3 >= len) { ------------------ | Branch (194:11): [True: 30, False: 8.14k] ------------------ 195| 30| return result(error_code::TOO_SHORT, pos); 196| 30| } // minimal bound checking 197| 8.14k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (197:11): [True: 72, False: 8.06k] ------------------ 198| 72| return result(error_code::TOO_SHORT, pos); 199| 72| } 200| 8.06k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (200:11): [True: 36, False: 8.03k] ------------------ 201| 36| return result(error_code::TOO_SHORT, pos); 202| 36| } 203| 8.03k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (203:11): [True: 18, False: 8.01k] ------------------ 204| 18| return result(error_code::TOO_SHORT, pos); 205| 18| } 206| | 207| | // range check 208| 8.01k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 209| 8.01k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 210| 8.01k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 211| 8.01k| (uint8_t(data[pos + 3]) & 0b00111111); 212| 8.01k| if (code_point <= 0xffff) { ------------------ | Branch (212:11): [True: 12, False: 8.00k] ------------------ 213| 12| return result(error_code::OVERLONG, pos); 214| 12| } 215| 8.00k| if (0x10ffff < code_point) { ------------------ | Branch (215:11): [True: 24, False: 7.97k] ------------------ 216| 24| return result(error_code::TOO_LARGE, pos); 217| 24| } 218| 7.97k| *utf32_output++ = char32_t(code_point); 219| 7.97k| pos += 4; 220| 7.97k| } else { 221| | // we either have too many continuation bytes or an invalid leading byte 222| 462| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (222:11): [True: 228, False: 234] ------------------ 223| 228| return result(error_code::TOO_LONG, pos); 224| 234| } else { 225| 234| return result(error_code::HEADER_BITS, pos); 226| 234| } 227| 462| } 228| 663k| } 229| 1.01k| return result(error_code::SUCCESS, utf32_output - start); 230| 2.13k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf327convertIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mPDi: 16| 1.60k| char32_t *utf32_output) { 17| 1.60k| size_t pos = 0; 18| 1.60k| char32_t *start{utf32_output}; 19| 2.10M| while (pos < len) { ------------------ | Branch (19:10): [True: 2.10M, False: 824] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 2.10M| { 24| | // try to convert the next block of 16 ASCII bytes 25| 2.10M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (25:11): [True: 2.09M, False: 5.39k] ------------------ 26| | // they are ascii 27| 2.09M| uint64_t v1; 28| 2.09M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 29| 2.09M| uint64_t v2; 30| 2.09M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 31| 2.09M| uint64_t v{v1 | v2}; 32| 2.09M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (32:13): [True: 1.24M, False: 850k] ------------------ 33| 1.24M| size_t final_pos = pos + 16; 34| 21.1M| while (pos < final_pos) { ------------------ | Branch (34:18): [True: 19.9M, False: 1.24M] ------------------ 35| 19.9M| *utf32_output++ = uint8_t(data[pos]); 36| 19.9M| pos++; 37| 19.9M| } 38| 1.24M| continue; 39| 1.24M| } 40| 2.09M| } 41| 2.10M| } 42| 855k| auto leading_byte = uint8_t(data[pos]); // leading byte 43| 855k| if (leading_byte < 0b10000000) { ------------------ | Branch (43:9): [True: 697k, False: 158k] ------------------ 44| | // converting one ASCII byte !!! 45| 697k| *utf32_output++ = char32_t(leading_byte); 46| 697k| pos++; 47| 697k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (47:16): [True: 97.4k, False: 60.7k] ------------------ 48| | // We have a two-byte UTF-8 49| 97.4k| if (pos + 1 >= len) { ------------------ | Branch (49:11): [True: 54, False: 97.4k] ------------------ 50| 54| return 0; 51| 54| } // minimal bound checking 52| 97.4k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (52:11): [True: 106, False: 97.3k] ------------------ 53| 106| return 0; 54| 106| } 55| | // range check 56| 97.3k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 57| 97.3k| (uint8_t(data[pos + 1]) & 0b00111111); 58| 97.3k| if (code_point < 0x80) { ------------------ | Branch (58:11): [True: 12, False: 97.3k] ------------------ 59| 12| return 0; 60| 12| } 61| 97.3k| *utf32_output++ = char32_t(code_point); 62| 97.3k| pos += 2; 63| 97.3k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (63:16): [True: 53.7k, False: 6.99k] ------------------ 64| | // We have a three-byte UTF-8 65| 53.7k| if (pos + 2 >= len) { ------------------ | Branch (65:11): [True: 18, False: 53.7k] ------------------ 66| 18| return 0; 67| 18| } // minimal bound checking 68| | 69| 53.7k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (69:11): [True: 52, False: 53.6k] ------------------ 70| 52| return 0; 71| 52| } 72| 53.6k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (72:11): [True: 16, False: 53.6k] ------------------ 73| 16| return 0; 74| 16| } 75| | // range check 76| 53.6k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 77| 53.6k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 78| 53.6k| (uint8_t(data[pos + 2]) & 0b00111111); 79| 53.6k| if (code_point < 0x800 || (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (79:11): [True: 40, False: 53.5k] | Branch (79:34): [True: 828, False: 52.7k] | Branch (79:57): [True: 10, False: 818] ------------------ 80| 50| return 0; 81| 50| } 82| 53.5k| *utf32_output++ = char32_t(code_point); 83| 53.5k| pos += 3; 84| 53.5k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (84:16): [True: 6.70k, False: 290] ------------------ 85| | // we have a 4-byte UTF-8 word. 86| 6.70k| if (pos + 3 >= len) { ------------------ | Branch (86:11): [True: 6, False: 6.69k] ------------------ 87| 6| return 0; 88| 6| } // minimal bound checking 89| 6.69k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (89:11): [True: 48, False: 6.65k] ------------------ 90| 48| return 0; 91| 48| } 92| 6.65k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (92:11): [True: 34, False: 6.61k] ------------------ 93| 34| return 0; 94| 34| } 95| 6.61k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (95:11): [True: 10, False: 6.60k] ------------------ 96| 10| return 0; 97| 10| } 98| | 99| | // range check 100| 6.60k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 101| 6.60k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 102| 6.60k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 103| 6.60k| (uint8_t(data[pos + 3]) & 0b00111111); 104| 6.60k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (104:11): [True: 48, False: 6.55k] | Branch (104:35): [True: 32, False: 6.52k] ------------------ 105| 80| return 0; 106| 80| } 107| 6.52k| *utf32_output++ = char32_t(code_point); 108| 6.52k| pos += 4; 109| 6.52k| } else { 110| 290| return 0; 111| 290| } 112| 855k| } 113| 824| return utf32_output - start; 114| 1.60k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf3213convert_validIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mPDi: 16| 1.18k| char32_t *utf32_output) { 17| 1.18k| size_t pos = 0; 18| 1.18k| char32_t *start{utf32_output}; 19| 1.70M| while (pos < len) { ------------------ | Branch (19:10): [True: 1.69M, False: 1.18k] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 1.69M| { 24| | // try to convert the next block of 8 ASCII bytes 25| 1.69M| if (pos + 8 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (25:11): [True: 1.69M, False: 3.51k] ------------------ 26| | // they are ascii 27| 1.69M| uint64_t v; 28| 1.69M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 29| 1.69M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (29:13): [True: 1.28M, False: 408k] ------------------ 30| 1.28M| size_t final_pos = pos + 8; 31| 11.5M| while (pos < final_pos) { ------------------ | Branch (31:18): [True: 10.2M, False: 1.28M] ------------------ 32| 10.2M| *utf32_output++ = uint8_t(data[pos]); 33| 10.2M| pos++; 34| 10.2M| } 35| 1.28M| continue; 36| 1.28M| } 37| 1.69M| } 38| 1.69M| } 39| 412k| auto leading_byte = uint8_t(data[pos]); // leading byte 40| 412k| if (leading_byte < 0b10000000) { ------------------ | Branch (40:9): [True: 236k, False: 176k] ------------------ 41| | // converting one ASCII byte !!! 42| 236k| *utf32_output++ = char32_t(leading_byte); 43| 236k| pos++; 44| 236k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (44:16): [True: 129k, False: 46.9k] ------------------ 45| | // We have a two-byte UTF-8 46| 129k| if (pos + 1 >= len) { ------------------ | Branch (46:11): [True: 0, False: 129k] ------------------ 47| 0| break; 48| 0| } // minimal bound checking 49| 129k| *utf32_output++ = char32_t(((leading_byte & 0b00011111) << 6) | 50| 129k| (uint8_t(data[pos + 1]) & 0b00111111)); 51| 129k| pos += 2; 52| 129k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (52:16): [True: 29.8k, False: 17.0k] ------------------ 53| | // We have a three-byte UTF-8 54| 29.8k| if (pos + 2 >= len) { ------------------ | Branch (54:11): [True: 0, False: 29.8k] ------------------ 55| 0| break; 56| 0| } // minimal bound checking 57| 29.8k| *utf32_output++ = char32_t(((leading_byte & 0b00001111) << 12) | 58| 29.8k| ((uint8_t(data[pos + 1]) & 0b00111111) << 6) | 59| 29.8k| (uint8_t(data[pos + 2]) & 0b00111111)); 60| 29.8k| pos += 3; 61| 29.8k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (61:16): [True: 17.0k, False: 0] ------------------ 62| | // we have a 4-byte UTF-8 word. 63| 17.0k| if (pos + 3 >= len) { ------------------ | Branch (63:11): [True: 0, False: 17.0k] ------------------ 64| 0| break; 65| 0| } // minimal bound checking 66| 17.0k| uint32_t code_word = ((leading_byte & 0b00000111) << 18) | 67| 17.0k| ((uint8_t(data[pos + 1]) & 0b00111111) << 12) | 68| 17.0k| ((uint8_t(data[pos + 2]) & 0b00111111) << 6) | 69| 17.0k| (uint8_t(data[pos + 3]) & 0b00111111); 70| 17.0k| *utf32_output++ = char32_t(code_word); 71| 17.0k| pos += 4; 72| 17.0k| } else { 73| | // we may have a continuation but we do not do error checking 74| 0| return 0; 75| 0| } 76| 412k| } 77| 1.18k| return utf32_output - start; 78| 1.18k|} _ZNK7simdutf8fallback14implementation13validate_utf8EPKcm: 37| 3.22k|implementation::validate_utf8(const char *buf, size_t len) const noexcept { 38| 3.22k| return scalar::utf8::validate(buf, len); 39| 3.22k|} _ZNK7simdutf8fallback14implementation25validate_utf8_with_errorsEPKcm: 44| 3.22k| const char *buf, size_t len) const noexcept { 45| 3.22k| return scalar::utf8::validate_with_errors(buf, len); 46| 3.22k|} _ZNK7simdutf8fallback14implementation16validate_utf16leEPKDsm: 77| 1.30k| size_t len) const noexcept { 78| 1.30k| return scalar::utf16::validate(buf, len); 79| 1.30k|} _ZNK7simdutf8fallback14implementation16validate_utf16beEPKDsm: 85| 1.23k| size_t len) const noexcept { 86| 1.23k| return scalar::utf16::validate(buf, len); 87| 1.23k|} _ZNK7simdutf8fallback14implementation28validate_utf16le_with_errorsEPKDsm: 90| 1.30k| const char16_t *buf, size_t len) const noexcept { 91| 1.30k| return scalar::utf16::validate_with_errors(buf, len); 92| 1.30k|} _ZNK7simdutf8fallback14implementation28validate_utf16be_with_errorsEPKDsm: 95| 1.23k| const char16_t *buf, size_t len) const noexcept { 96| 1.23k| return scalar::utf16::validate_with_errors(buf, len); 97| 1.23k|} _ZNK7simdutf8fallback14implementation14validate_utf32EPKDim: 114| 2.41k|implementation::validate_utf32(const char32_t *buf, size_t len) const noexcept { 115| 2.41k| return scalar::utf32::validate(buf, len); 116| 2.41k|} _ZNK7simdutf8fallback14implementation26validate_utf32_with_errorsEPKDim: 121| 2.41k| const char32_t *buf, size_t len) const noexcept { 122| 2.41k| return scalar::utf32::validate_with_errors(buf, len); 123| 2.41k|} _ZNK7simdutf8fallback14implementation22convert_latin1_to_utf8EPKcmPc: 128| 538| const char *buf, size_t len, char *utf8_output) const noexcept { 129| 538| return scalar::latin1_to_utf8::convert(buf, len, utf8_output); 130| 538|} _ZNK7simdutf8fallback14implementation25convert_latin1_to_utf16leEPKcmPDs: 135| 80| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 136| 80| return scalar::latin1_to_utf16::convert(buf, len, 137| 80| utf16_output); 138| 80|} _ZNK7simdutf8fallback14implementation25convert_latin1_to_utf16beEPKcmPDs: 141| 80| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 142| 80| return scalar::latin1_to_utf16::convert(buf, len, 143| 80| utf16_output); 144| 80|} _ZNK7simdutf8fallback14implementation23convert_latin1_to_utf32EPKcmPDi: 149| 90| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 150| 90| return scalar::latin1_to_utf32::convert(buf, len, utf32_output); 151| 90|} _ZNK7simdutf8fallback14implementation22convert_utf8_to_latin1EPKcmPc: 156| 562| const char *buf, size_t len, char *latin1_output) const noexcept { 157| 562| return scalar::utf8_to_latin1::convert(buf, len, latin1_output); 158| 562|} _ZNK7simdutf8fallback14implementation34convert_utf8_to_latin1_with_errorsEPKcmPc: 161| 516| const char *buf, size_t len, char *latin1_output) const noexcept { 162| 516| return scalar::utf8_to_latin1::convert_with_errors(buf, len, latin1_output); 163| 516|} _ZNK7simdutf8fallback14implementation23convert_utf8_to_utf16leEPKcmPDs: 173| 660| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 174| 660| return scalar::utf8_to_utf16::convert(buf, len, 175| 660| utf16_output); 176| 660|} _ZNK7simdutf8fallback14implementation23convert_utf8_to_utf16beEPKcmPDs: 179| 712| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 180| 712| return scalar::utf8_to_utf16::convert(buf, len, 181| 712| utf16_output); 182| 712|} _ZNK7simdutf8fallback14implementation35convert_utf8_to_utf16le_with_errorsEPKcmPDs: 185| 574| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 186| 574| return scalar::utf8_to_utf16::convert_with_errors( 187| 574| buf, len, utf16_output); 188| 574|} _ZNK7simdutf8fallback14implementation35convert_utf8_to_utf16be_with_errorsEPKcmPDs: 191| 702| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 192| 702| return scalar::utf8_to_utf16::convert_with_errors( 193| 702| buf, len, utf16_output); 194| 702|} _ZNK7simdutf8fallback14implementation29convert_valid_utf8_to_utf16leEPKcmPDs: 197| 368| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 198| 368| return scalar::utf8_to_utf16::convert_valid(buf, len, 199| 368| utf16_output); 200| 368|} _ZNK7simdutf8fallback14implementation29convert_valid_utf8_to_utf16beEPKcmPDs: 203| 438| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 204| 438| return scalar::utf8_to_utf16::convert_valid(buf, len, 205| 438| utf16_output); 206| 438|} _ZNK7simdutf8fallback14implementation21convert_utf8_to_utf32EPKcmPDi: 211| 712| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 212| 712| return scalar::utf8_to_utf32::convert(buf, len, utf32_output); 213| 712|} _ZNK7simdutf8fallback14implementation33convert_utf8_to_utf32_with_errorsEPKcmPDi: 216| 738| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 217| 738| return scalar::utf8_to_utf32::convert_with_errors(buf, len, utf32_output); 218| 738|} _ZNK7simdutf8fallback14implementation27convert_valid_utf8_to_utf32EPKcmPDi: 221| 394| const char *input, size_t size, char32_t *utf32_output) const noexcept { 222| 394| return scalar::utf8_to_utf32::convert_valid(input, size, utf32_output); 223| 394|} _ZNK7simdutf8fallback14implementation25convert_utf16le_to_latin1EPKDsmPc: 228| 124| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 229| 124| return scalar::utf16_to_latin1::convert(buf, len, 230| 124| latin1_output); 231| 124|} _ZNK7simdutf8fallback14implementation25convert_utf16be_to_latin1EPKDsmPc: 234| 126| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 235| 126| return scalar::utf16_to_latin1::convert(buf, len, 236| 126| latin1_output); 237| 126|} _ZNK7simdutf8fallback14implementation37convert_utf16le_to_latin1_with_errorsEPKDsmPc: 241| 230| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 242| 230| return scalar::utf16_to_latin1::convert_with_errors( 243| 230| buf, len, latin1_output); 244| 230|} _ZNK7simdutf8fallback14implementation37convert_utf16be_to_latin1_with_errorsEPKDsmPc: 248| 252| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 249| 252| return scalar::utf16_to_latin1::convert_with_errors( 250| 252| buf, len, latin1_output); 251| 252|} _ZNK7simdutf8fallback14implementation23convert_utf16le_to_utf8EPKDsmPc: 268| 386| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 269| 386| return scalar::utf16_to_utf8::convert(buf, len, 270| 386| utf8_output); 271| 386|} _ZNK7simdutf8fallback14implementation23convert_utf16be_to_utf8EPKDsmPc: 274| 326| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 275| 326| return scalar::utf16_to_utf8::convert(buf, len, utf8_output); 276| 326|} _ZNK7simdutf8fallback14implementation35convert_utf16le_to_utf8_with_errorsEPKDsmPc: 279| 630| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 280| 630| return scalar::utf16_to_utf8::convert_with_errors( 281| 630| buf, len, utf8_output); 282| 630|} _ZNK7simdutf8fallback14implementation35convert_utf16be_to_utf8_with_errorsEPKDsmPc: 285| 586| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 286| 586| return scalar::utf16_to_utf8::convert_with_errors( 287| 586| buf, len, utf8_output); 288| 586|} _ZNK7simdutf8fallback14implementation29convert_valid_utf16le_to_utf8EPKDsmPc: 291| 416| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 292| 416| return scalar::utf16_to_utf8::convert_valid(buf, len, 293| 416| utf8_output); 294| 416|} _ZNK7simdutf8fallback14implementation29convert_valid_utf16be_to_utf8EPKDsmPc: 297| 442| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 298| 442| return scalar::utf16_to_utf8::convert_valid(buf, len, 299| 442| utf8_output); 300| 442|} _ZNK7simdutf8fallback14implementation23convert_utf32_to_latin1EPKDimPc: 305| 188| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 306| 188| return scalar::utf32_to_latin1::convert(buf, len, latin1_output); 307| 188|} _ZNK7simdutf8fallback14implementation35convert_utf32_to_latin1_with_errorsEPKDimPc: 310| 420| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 311| 420| return scalar::utf32_to_latin1::convert_with_errors(buf, len, latin1_output); 312| 420|} _ZNK7simdutf8fallback14implementation21convert_utf32_to_utf8EPKDimPc: 322| 604| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 323| 604| return scalar::utf32_to_utf8::convert(buf, len, utf8_output); 324| 604|} _ZNK7simdutf8fallback14implementation33convert_utf32_to_utf8_with_errorsEPKDimPc: 327| 830| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 328| 830| return scalar::utf32_to_utf8::convert_with_errors(buf, len, utf8_output); 329| 830|} _ZNK7simdutf8fallback14implementation27convert_valid_utf32_to_utf8EPKDimPc: 332| 192| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 333| 192| return scalar::utf32_to_utf8::convert_valid(buf, len, utf8_output); 334| 192|} _ZNK7simdutf8fallback14implementation24convert_utf32_to_utf16leEPKDimPDs: 339| 432| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 340| 432| return scalar::utf32_to_utf16::convert(buf, len, 341| 432| utf16_output); 342| 432|} _ZNK7simdutf8fallback14implementation24convert_utf32_to_utf16beEPKDimPDs: 345| 442| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 346| 442| return scalar::utf32_to_utf16::convert(buf, len, 347| 442| utf16_output); 348| 442|} _ZNK7simdutf8fallback14implementation36convert_utf32_to_utf16le_with_errorsEPKDimPDs: 351| 716| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 352| 716| return scalar::utf32_to_utf16::convert_with_errors( 353| 716| buf, len, utf16_output); 354| 716|} _ZNK7simdutf8fallback14implementation36convert_utf32_to_utf16be_with_errorsEPKDimPDs: 357| 606| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 358| 606| return scalar::utf32_to_utf16::convert_with_errors( 359| 606| buf, len, utf16_output); 360| 606|} _ZNK7simdutf8fallback14implementation30convert_valid_utf32_to_utf16leEPKDimPDs: 363| 290| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 364| 290| return scalar::utf32_to_utf16::convert_valid( 365| 290| buf, len, utf16_output); 366| 290|} _ZNK7simdutf8fallback14implementation30convert_valid_utf32_to_utf16beEPKDimPDs: 369| 86| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 370| 86| return scalar::utf32_to_utf16::convert_valid(buf, len, 371| 86| utf16_output); 372| 86|} _ZNK7simdutf8fallback14implementation24convert_utf16le_to_utf32EPKDsmPDi: 375| 272| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 376| 272| return scalar::utf16_to_utf32::convert(buf, len, 377| 272| utf32_output); 378| 272|} _ZNK7simdutf8fallback14implementation24convert_utf16be_to_utf32EPKDsmPDi: 381| 226| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 382| 226| return scalar::utf16_to_utf32::convert(buf, len, 383| 226| utf32_output); 384| 226|} _ZNK7simdutf8fallback14implementation36convert_utf16le_to_utf32_with_errorsEPKDsmPDi: 387| 414| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 388| 414| return scalar::utf16_to_utf32::convert_with_errors( 389| 414| buf, len, utf32_output); 390| 414|} _ZNK7simdutf8fallback14implementation36convert_utf16be_to_utf32_with_errorsEPKDsmPDi: 393| 300| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 394| 300| return scalar::utf16_to_utf32::convert_with_errors( 395| 300| buf, len, utf32_output); 396| 300|} _ZNK7simdutf8fallback14implementation30convert_valid_utf16le_to_utf32EPKDsmPDi: 399| 112| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 400| 112| return scalar::utf16_to_utf32::convert_valid( 401| 112| buf, len, utf32_output); 402| 112|} _ZNK7simdutf8fallback14implementation30convert_valid_utf16be_to_utf32EPKDsmPDi: 405| 140| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 406| 140| return scalar::utf16_to_utf32::convert_valid(buf, len, 407| 140| utf32_output); 408| 140|} _ZNK7simdutf8fallback14implementation13count_utf16leEPKDsm: 419| 1.30k| const char16_t *input, size_t length) const noexcept { 420| 1.30k| return scalar::utf16::count_code_points(input, length); 421| 1.30k|} _ZNK7simdutf8fallback14implementation13count_utf16beEPKDsm: 424| 1.23k| const char16_t *input, size_t length) const noexcept { 425| 1.23k| return scalar::utf16::count_code_points(input, length); 426| 1.23k|} _ZNK7simdutf8fallback14implementation10count_utf8EPKcm: 431| 3.22k|implementation::count_utf8(const char *input, size_t length) const noexcept { 432| 3.22k| return scalar::utf8::count_code_points(input, length); 433| 3.22k|} _ZNK7simdutf8fallback14implementation23latin1_length_from_utf8EPKcm: 438| 539| const char *buf, size_t len) const noexcept { 439| 539| return scalar::utf8::count_code_points(buf, len); 440| 539|} _ZNK7simdutf8fallback14implementation23utf8_length_from_latin1EPKcm: 445| 269| const char *input, size_t length) const noexcept { 446| 269| return scalar::latin1_to_utf8::utf8_length_from_latin1(input, length); 447| 269|} _ZNK7simdutf8fallback14implementation24utf8_length_from_utf16leEPKDsm: 452| 727| const char16_t *input, size_t length) const noexcept { 453| 727| return scalar::utf16::utf8_length_from_utf16(input, 454| 727| length); 455| 727|} _ZNK7simdutf8fallback14implementation24utf8_length_from_utf16beEPKDsm: 458| 696| const char16_t *input, size_t length) const noexcept { 459| 696| return scalar::utf16::utf8_length_from_utf16(input, length); 460| 696|} _ZNK7simdutf8fallback14implementation25utf32_length_from_utf16leEPKDsm: 465| 404| const char16_t *input, size_t length) const noexcept { 466| 404| return scalar::utf16::utf32_length_from_utf16(input, 467| 404| length); 468| 404|} _ZNK7simdutf8fallback14implementation25utf32_length_from_utf16beEPKDsm: 471| 351| const char16_t *input, size_t length) const noexcept { 472| 351| return scalar::utf16::utf32_length_from_utf16(input, length); 473| 351|} _ZNK7simdutf8fallback14implementation22utf16_length_from_utf8EPKcm: 478| 1.74k| const char *input, size_t length) const noexcept { 479| 1.74k| return scalar::utf8::utf16_length_from_utf8(input, length); 480| 1.74k|} _ZNK7simdutf8fallback14implementation22utf8_length_from_utf32EPKDim: 513| 819| const char32_t *input, size_t length) const noexcept { 514| 819| return scalar::utf32::utf8_length_from_utf32(input, length); 515| 819|} _ZNK7simdutf8fallback14implementation23utf16_length_from_utf32EPKDim: 520| 1.29k| const char32_t *input, size_t length) const noexcept { 521| 1.29k| return scalar::utf32::utf16_length_from_utf32(input, length); 522| 1.29k|} _ZNK7simdutf8fallback14implementation22utf32_length_from_utf8EPKcm: 527| 933| const char *input, size_t length) const noexcept { 528| 933| return scalar::utf8::count_code_points(input, length); 529| 933|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE13get_remainderEPh: 57| 5.63k|buf_block_reader::get_remainder(uint8_t *dst) const { 58| 5.63k| if (len == idx) { ------------------ | Branch (58:7): [True: 36, False: 5.59k] ------------------ 59| 36| return 0; 60| 36| } // memcpy(dst, null, 0) will trigger an error with some sanitizers 61| 5.59k| std::memset(dst, 0x20, 62| 5.59k| STEP_SIZE); // std::memset STEP_SIZE because it is more efficient 63| | // to write out 8 or 16 bytes at once. 64| 5.59k| std::memcpy(dst, buf + idx, len - idx); 65| 5.59k| return len - idx; 66| 5.63k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE13get_remainderEPh: 57| 5.63k|buf_block_reader::get_remainder(uint8_t *dst) const { 58| 5.63k| if (len == idx) { ------------------ | Branch (58:7): [True: 36, False: 5.59k] ------------------ 59| 36| return 0; 60| 36| } // memcpy(dst, null, 0) will trigger an error with some sanitizers 61| 5.59k| std::memset(dst, 0x20, 62| 5.59k| STEP_SIZE); // std::memset STEP_SIZE because it is more efficient 63| | // to write out 8 or 16 bytes at once. 64| 5.59k| std::memcpy(dst, buf + idx, len - idx); 65| 5.59k| return len - idx; 66| 5.63k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EEC2EPKhm: 36| 6.44k| : buf{_buf}, len{_len}, lenminusstep{len < STEP_SIZE ? 0 : len - STEP_SIZE}, ------------------ | Branch (36:42): [True: 2.43k, False: 4.00k] ------------------ 37| 6.44k| idx{0} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE14has_full_blockEv: 45| 4.64M|simdutf_really_inline bool buf_block_reader::has_full_block() const { 46| 4.64M| return idx < lenminusstep; 47| 4.64M|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE10full_blockEv: 51| 4.63M|buf_block_reader::full_block() const { 52| 4.63M| return &buf[idx]; 53| 4.63M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE7advanceEv: 69| 4.64M|simdutf_really_inline void buf_block_reader::advance() { 70| 4.64M| idx += STEP_SIZE; 71| 4.64M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EEC2EPKhm: 36| 6.44k| : buf{_buf}, len{_len}, lenminusstep{len < STEP_SIZE ? 0 : len - STEP_SIZE}, ------------------ | Branch (36:42): [True: 2.43k, False: 4.00k] ------------------ 37| 6.44k| idx{0} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE14has_full_blockEv: 45| 4.64M|simdutf_really_inline bool buf_block_reader::has_full_block() const { 46| 4.64M| return idx < lenminusstep; 47| 4.64M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE10full_blockEv: 51| 4.63M|buf_block_reader::full_block() const { 52| 4.63M| return &buf[idx]; 53| 4.63M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE7advanceEv: 69| 4.64M|simdutf_really_inline void buf_block_reader::advance() { 70| 4.64M| idx += STEP_SIZE; 71| 4.64M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE0EEEmPKDsm: 8| 1.71k| size_t size) { 9| 1.71k| size_t pos = 0; 10| 1.71k| size_t count = 0; 11| 1.11M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 1.10M, False: 1.71k] ------------------ 12| 1.10M| simd16x32 input(reinterpret_cast(in + pos)); 13| | if constexpr (!match_system(big_endian)) { 14| | input.swap_bytes(); 15| | } 16| 1.10M| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 1.10M| count += count_ones(not_pair) / 2; 18| 1.10M| } 19| 1.71k| return count + 20| 1.71k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.71k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE1EEEmPKDsm: 8| 1.58k| size_t size) { 9| 1.58k| size_t pos = 0; 10| 1.58k| size_t count = 0; 11| 1.20M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 1.20M, False: 1.58k] ------------------ 12| 1.20M| simd16x32 input(reinterpret_cast(in + pos)); 13| 1.20M| if constexpr (!match_system(big_endian)) { 14| 1.20M| input.swap_bytes(); 15| 1.20M| } 16| 1.20M| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 1.20M| count += count_ones(not_pair) / 2; 18| 1.20M| } 19| 1.58k| return count + 20| 1.58k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.58k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 51| 404| size_t size) { 52| 404| return count_code_points(in, size); 53| 404|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 51| 351| size_t size) { 52| 351| return count_code_points(in, size); 53| 351|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE0EEEmPKDsm: 8| 1.71k| size_t size) { 9| 1.71k| size_t pos = 0; 10| 1.71k| size_t count = 0; 11| 1.11M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 1.10M, False: 1.71k] ------------------ 12| 1.10M| simd16x32 input(reinterpret_cast(in + pos)); 13| | if constexpr (!match_system(big_endian)) { 14| | input.swap_bytes(); 15| | } 16| 1.10M| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 1.10M| count += count_ones(not_pair) / 2; 18| 1.10M| } 19| 1.71k| return count + 20| 1.71k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.71k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE1EEEmPKDsm: 8| 1.58k| size_t size) { 9| 1.58k| size_t pos = 0; 10| 1.58k| size_t count = 0; 11| 1.20M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 1.20M, False: 1.58k] ------------------ 12| 1.20M| simd16x32 input(reinterpret_cast(in + pos)); 13| 1.20M| if constexpr (!match_system(big_endian)) { 14| 1.20M| input.swap_bytes(); 15| 1.20M| } 16| 1.20M| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 1.20M| count += count_ones(not_pair) / 2; 18| 1.20M| } 19| 1.58k| return count + 20| 1.58k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.58k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 51| 404| size_t size) { 52| 404| return count_code_points(in, size); 53| 404|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 51| 351| size_t size) { 52| 351| return count_code_points(in, size); 53| 351|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE0EEEmPKDsm: 10| 727| size_t size) { 11| 727| size_t pos = 0; 12| | 13| 727| using vector_u16 = simd16; 14| 727| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 727| const auto one = vector_u16::splat(1); 17| | 18| 727| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 727| 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| 727| constexpr size_t max_iterations = 65535 / 2; 26| 727| size_t iteration = max_iterations; 27| | 28| 1.19M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 1.19M, False: 727] ------------------ 29| 1.19M| 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| 1.19M| 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| 1.19M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 1.19M| 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| 1.19M| v_count += c0; 68| 1.19M| v_count += c1; 69| 1.19M| v_count += vector_u16(is_surrogate); 70| | 71| 1.19M| iteration -= 1; 72| 1.19M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 17, False: 1.19M] ------------------ 73| 17| count += v_count.sum(); 74| 17| v_count = vector_u16::zero(); 75| 17| iteration = max_iterations; 76| 17| } 77| 1.19M| } 78| | 79| 727| if (iteration > 0) { ------------------ | Branch (79:7): [True: 727, False: 0] ------------------ 80| 727| count += v_count.sum(); 81| 727| } 82| | 83| 727| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 727| size - pos); 85| 727|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE1EEEmPKDsm: 10| 696| size_t size) { 11| 696| size_t pos = 0; 12| | 13| 696| using vector_u16 = simd16; 14| 696| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 696| const auto one = vector_u16::splat(1); 17| | 18| 696| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 696| 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| 696| constexpr size_t max_iterations = 65535 / 2; 26| 696| size_t iteration = max_iterations; 27| | 28| 1.11M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 1.11M, False: 696] ------------------ 29| 1.11M| auto input = vector_u16::load(reinterpret_cast(in + pos)); 30| 1.11M| if constexpr (!match_system(big_endian)) { 31| 1.11M| input = input.swap_bytes(); 32| 1.11M| } 33| | // 0xd800 .. 0xdbff - low surrogate 34| | // 0xdc00 .. 0xdfff - high surrogate 35| 1.11M| const auto is_surrogate = ((input & uint16_t(0xf800)) == uint16_t(0xd800)); 36| | 37| | // c0 - chars that yield 2- or 3-byte UTF-8 codes 38| 1.11M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 1.11M| const auto c1 = min(input & uint16_t(0xf800), one); 42| | 43| | /* 44| | Explanation how the counting works. 45| | 46| | In the case of a non-surrogate character we count: 47| | * always 1 -- see how `count` is initialized above; 48| | * c0 = 1 if the current char yields 2 or 3 bytes; 49| | * c1 = 1 if the current char yields 3 bytes. 50| | 51| | Thus, we always have correct count for the current char: 52| | from 1, 2 or 3 bytes. 53| | 54| | A trickier part is how we count surrogate pairs. Whether 55| | we encounter a surrogate (low or high), we count it as 56| | 3 chars and then minus 1 (`is_surrogate` is -1 or 0). 57| | Each surrogate char yields 2. A surrogate pair, that 58| | is a low surrogate followed by a high one, yields 59| | the expected 4 bytes. 60| | 61| | It also correctly handles cases when low surrogate is 62| | processed by the this loop, but high surrogate is counted 63| | by the scalar procedure. The scalar procedure uses exactly 64| | the described approach, thanks to that for valid UTF-16 65| | strings it always count correctly. 66| | */ 67| 1.11M| v_count += c0; 68| 1.11M| v_count += c1; 69| 1.11M| v_count += vector_u16(is_surrogate); 70| | 71| 1.11M| iteration -= 1; 72| 1.11M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 19, False: 1.11M] ------------------ 73| 19| count += v_count.sum(); 74| 19| v_count = vector_u16::zero(); 75| 19| iteration = max_iterations; 76| 19| } 77| 1.11M| } 78| | 79| 696| if (iteration > 0) { ------------------ | Branch (79:7): [True: 696, False: 0] ------------------ 80| 696| count += v_count.sum(); 81| 696| } 82| | 83| 696| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 696| size - pos); 85| 696|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE0EEEmPKDsm: 10| 727| size_t size) { 11| 727| size_t pos = 0; 12| | 13| 727| using vector_u16 = simd16; 14| 727| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 727| const auto one = vector_u16::splat(1); 17| | 18| 727| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 727| 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| 727| constexpr size_t max_iterations = 65535 / 2; 26| 727| size_t iteration = max_iterations; 27| | 28| 2.38M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 2.38M, False: 727] ------------------ 29| 2.38M| 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| 2.38M| 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| 2.38M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 2.38M| 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| 2.38M| v_count += c0; 68| 2.38M| v_count += c1; 69| 2.38M| v_count += vector_u16(is_surrogate); 70| | 71| 2.38M| iteration -= 1; 72| 2.38M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 51, False: 2.38M] ------------------ 73| 51| count += v_count.sum(); 74| 51| v_count = vector_u16::zero(); 75| 51| iteration = max_iterations; 76| 51| } 77| 2.38M| } 78| | 79| 727| if (iteration > 0) { ------------------ | Branch (79:7): [True: 727, False: 0] ------------------ 80| 727| count += v_count.sum(); 81| 727| } 82| | 83| 727| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 727| size - pos); 85| 727|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE1EEEmPKDsm: 10| 696| size_t size) { 11| 696| size_t pos = 0; 12| | 13| 696| using vector_u16 = simd16; 14| 696| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 696| const auto one = vector_u16::splat(1); 17| | 18| 696| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 696| 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| 696| constexpr size_t max_iterations = 65535 / 2; 26| 696| size_t iteration = max_iterations; 27| | 28| 2.23M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 2.23M, False: 696] ------------------ 29| 2.23M| auto input = vector_u16::load(reinterpret_cast(in + pos)); 30| 2.23M| if constexpr (!match_system(big_endian)) { 31| 2.23M| input = input.swap_bytes(); 32| 2.23M| } 33| | // 0xd800 .. 0xdbff - low surrogate 34| | // 0xdc00 .. 0xdfff - high surrogate 35| 2.23M| 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| 2.23M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 2.23M| 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| 2.23M| v_count += c0; 68| 2.23M| v_count += c1; 69| 2.23M| v_count += vector_u16(is_surrogate); 70| | 71| 2.23M| iteration -= 1; 72| 2.23M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 48, False: 2.23M] ------------------ 73| 48| count += v_count.sum(); 74| 48| v_count = vector_u16::zero(); 75| 48| iteration = max_iterations; 76| 48| } 77| 2.23M| } 78| | 79| 696| if (iteration > 0) { ------------------ | Branch (79:7): [True: 696, False: 0] ------------------ 80| 696| count += v_count.sum(); 81| 696| } 82| | 83| 696| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 696| size - pos); 85| 696|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf3222utf8_length_from_utf32EPKDim: 11| 819| size_t length) { 12| 819| using vector_u32 = simd32; 13| | 14| 819| 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| 819| const size_t max_increment = 3; 19| | 20| 819| 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| 819| const auto v_ffffff80 = vector_u32::splat(0xffffff80); 28| 819| const auto v_fffff800 = vector_u32::splat(0xfffff800); 29| 819| const auto v_ffff0000 = vector_u32::splat(0xffff0000); 30| 819| const auto one = vector_u32::splat(1); 31| 819|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 32| | 33| 819| size_t counter = 0; 34| | 35| | // 1. vectorized loop unrolled 4 times 36| 819| { 37| | // we use vector of uint32 counters, this is why this limit is used 38| 819| const size_t max_iterations = 39| 819| std::numeric_limits::max() / (max_increment * 4); 40| 819| size_t blocks = length / (N * 4); 41| 819| length -= blocks * (N * 4); 42| 1.01k| while (blocks != 0) { ------------------ | Branch (42:12): [True: 191, False: 819] ------------------ 43| 191| const size_t iterations = min(blocks, max_iterations); 44| 191| blocks -= iterations; 45| | 46| 191| simd32 acc = vector_u32::zero(); 47| 72.3k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (47:26): [True: 72.1k, False: 191] ------------------ 48| 72.1k| const auto in0 = vector_u32(input + 0 * N); 49| 72.1k| const auto in1 = vector_u32(input + 1 * N); 50| 72.1k| const auto in2 = vector_u32(input + 2 * N); 51| 72.1k| 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| 72.1k| acc += min(one, in0 & v_ffffff80); 70| 72.1k| acc += min(one, in1 & v_ffffff80); 71| 72.1k| acc += min(one, in2 & v_ffffff80); 72| 72.1k| acc += min(one, in3 & v_ffffff80); 73| | 74| 72.1k| acc += min(one, in0 & v_fffff800); 75| 72.1k| acc += min(one, in1 & v_fffff800); 76| 72.1k| acc += min(one, in2 & v_fffff800); 77| 72.1k| acc += min(one, in3 & v_fffff800); 78| | 79| 72.1k| acc += min(one, in0 & v_ffff0000); 80| 72.1k| acc += min(one, in1 & v_ffff0000); 81| 72.1k| acc += min(one, in2 & v_ffff0000); 82| 72.1k| acc += min(one, in3 & v_ffff0000); 83| 72.1k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 84| | 85| 72.1k| input += 4 * N; 86| 72.1k| } 87| | 88| 191| counter += acc.sum(); 89| 191| } 90| 819| } 91| | 92| | // 2. vectorized loop for tail 93| 819| { 94| 819| const size_t max_iterations = 95| 819| std::numeric_limits::max() / max_increment; 96| 819| size_t blocks = length / N; 97| 819| length -= blocks * N; 98| 1.22k| while (blocks != 0) { ------------------ | Branch (98:12): [True: 408, False: 819] ------------------ 99| 408| const size_t iterations = min(blocks, max_iterations); 100| 408| blocks -= iterations; 101| | 102| 408| auto acc = vector_u32::zero(); 103| 1.41k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (103:26): [True: 1.00k, False: 408] ------------------ 104| 1.00k| 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| 1.00k| acc += min(one, in & v_ffffff80); 112| 1.00k| acc += min(one, in & v_fffff800); 113| 1.00k| acc += min(one, in & v_ffff0000); 114| 1.00k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 115| | 116| 1.00k| input += N; 117| 1.00k| } 118| | 119| 408| counter += acc.sum(); 120| 408| } 121| 819| } 122| | 123| 819| const size_t consumed = input - start; 124| 819| if (consumed != 0) { ------------------ | Branch (124:7): [True: 448, False: 371] ------------------ 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| 448| counter += consumed; 128| 448| } 129| | 130| 819| return counter + scalar::utf32::utf8_length_from_utf32(input, length); 131| 819|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf323minImEET_S4_S4_: 8| 599|template T min(T a, T b) { return a <= b ? a : b; } ------------------ | Branch (8:48): [True: 599, False: 0] ------------------ simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf3222utf8_length_from_utf32EPKDim: 11| 819| size_t length) { 12| 819| using vector_u32 = simd32; 13| | 14| 819| 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| 819| const size_t max_increment = 3; 19| | 20| 819| 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| 819| const auto v_ffffff80 = vector_u32::splat(0xffffff80); 28| 819| const auto v_fffff800 = vector_u32::splat(0xfffff800); 29| 819| const auto v_ffff0000 = vector_u32::splat(0xffff0000); 30| 819| const auto one = vector_u32::splat(1); 31| 819|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 32| | 33| 819| size_t counter = 0; 34| | 35| | // 1. vectorized loop unrolled 4 times 36| 819| { 37| | // we use vector of uint32 counters, this is why this limit is used 38| 819| const size_t max_iterations = 39| 819| std::numeric_limits::max() / (max_increment * 4); 40| 819| size_t blocks = length / (N * 4); 41| 819| length -= blocks * (N * 4); 42| 1.24k| while (blocks != 0) { ------------------ | Branch (42:12): [True: 430, False: 819] ------------------ 43| 430| const size_t iterations = min(blocks, max_iterations); 44| 430| blocks -= iterations; 45| | 46| 430| simd32 acc = vector_u32::zero(); 47| 145k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (47:26): [True: 144k, False: 430] ------------------ 48| 144k| const auto in0 = vector_u32(input + 0 * N); 49| 144k| const auto in1 = vector_u32(input + 1 * N); 50| 144k| const auto in2 = vector_u32(input + 2 * N); 51| 144k| 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| 144k| acc += min(one, in0 & v_ffffff80); 70| 144k| acc += min(one, in1 & v_ffffff80); 71| 144k| acc += min(one, in2 & v_ffffff80); 72| 144k| acc += min(one, in3 & v_ffffff80); 73| | 74| 144k| acc += min(one, in0 & v_fffff800); 75| 144k| acc += min(one, in1 & v_fffff800); 76| 144k| acc += min(one, in2 & v_fffff800); 77| 144k| acc += min(one, in3 & v_fffff800); 78| | 79| 144k| acc += min(one, in0 & v_ffff0000); 80| 144k| acc += min(one, in1 & v_ffff0000); 81| 144k| acc += min(one, in2 & v_ffff0000); 82| 144k| acc += min(one, in3 & v_ffff0000); 83| 144k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 84| | 85| 144k| input += 4 * N; 86| 144k| } 87| | 88| 430| counter += acc.sum(); 89| 430| } 90| 819| } 91| | 92| | // 2. vectorized loop for tail 93| 819| { 94| 819| const size_t max_iterations = 95| 819| std::numeric_limits::max() / max_increment; 96| 819| size_t blocks = length / N; 97| 819| length -= blocks * N; 98| 1.26k| while (blocks != 0) { ------------------ | Branch (98:12): [True: 447, False: 819] ------------------ 99| 447| const size_t iterations = min(blocks, max_iterations); 100| 447| blocks -= iterations; 101| | 102| 447| auto acc = vector_u32::zero(); 103| 1.53k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (103:26): [True: 1.09k, False: 447] ------------------ 104| 1.09k| 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| 1.09k| acc += min(one, in & v_ffffff80); 112| 1.09k| acc += min(one, in & v_fffff800); 113| 1.09k| acc += min(one, in & v_ffff0000); 114| 1.09k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 115| | 116| 1.09k| input += N; 117| 1.09k| } 118| | 119| 447| counter += acc.sum(); 120| 447| } 121| 819| } 122| | 123| 819| const size_t consumed = input - start; 124| 819| if (consumed != 0) { ------------------ | Branch (124:7): [True: 490, False: 329] ------------------ 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| 490| counter += consumed; 128| 490| } 129| | 130| 819| return counter + scalar::utf32::utf8_length_from_utf32(input, length); 131| 819|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf323minImEET_S4_S4_: 8| 877|template T min(T a, T b) { return a <= b ? a : b; } ------------------ | Branch (8:48): [True: 877, False: 0] ------------------ simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14utf826count_code_points_bytemaskEPKcm: 21| 3.76k|count_code_points_bytemask(const char *in, size_t size) { 22| 3.76k| using vector_i8 = simd8; 23| 3.76k| using vector_u8 = simd8; 24| 3.76k| using vector_u64 = simd64; 25| | 26| 3.76k| constexpr size_t N = vector_i8::SIZE; 27| 3.76k| constexpr size_t max_iterations = 255 / 4; 28| | 29| 3.76k| size_t pos = 0; 30| 3.76k| size_t count = 0; 31| | 32| 3.76k| auto counters = vector_u64::zero(); 33| 3.76k| auto local = vector_u8::zero(); 34| 3.76k| size_t iterations = 0; 35| 1.67M| for (; pos + 4 * N <= size; pos += 4 * N) { ------------------ | Branch (35:10): [True: 1.66M, False: 3.76k] ------------------ 36| 1.66M| const auto input0 = 37| 1.66M| simd8::load(reinterpret_cast(in + pos + 0 * N)); 38| 1.66M| const auto input1 = 39| 1.66M| simd8::load(reinterpret_cast(in + pos + 1 * N)); 40| 1.66M| const auto input2 = 41| 1.66M| simd8::load(reinterpret_cast(in + pos + 2 * N)); 42| 1.66M| const auto input3 = 43| 1.66M| simd8::load(reinterpret_cast(in + pos + 3 * N)); 44| 1.66M| const auto mask0 = input0 > int8_t(-65); 45| 1.66M| const auto mask1 = input1 > int8_t(-65); 46| 1.66M| const auto mask2 = input2 > int8_t(-65); 47| 1.66M| const auto mask3 = input3 > int8_t(-65); 48| | 49| 1.66M| local -= vector_u8(mask0); 50| 1.66M| local -= vector_u8(mask1); 51| 1.66M| local -= vector_u8(mask2); 52| 1.66M| local -= vector_u8(mask3); 53| | 54| 1.66M| iterations += 1; 55| 1.66M| if (iterations == max_iterations) { ------------------ | Branch (55:9): [True: 26.2k, False: 1.64M] ------------------ 56| 26.2k| counters += sum_8bytes(local); 57| 26.2k| local = vector_u8::zero(); 58| 26.2k| iterations = 0; 59| 26.2k| } 60| 1.66M| } 61| | 62| 3.76k| if (iterations > 0) { ------------------ | Branch (62:7): [True: 1.43k, False: 2.32k] ------------------ 63| 1.43k| count += local.sum_bytes(); 64| 1.43k| } 65| | 66| 3.76k| count += counters.sum(); 67| | 68| 3.76k| return count + scalar::utf8::count_code_points(in + pos, size - pos); 69| 3.76k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14utf817count_code_pointsEPKcm: 8| 933|simdutf_really_inline size_t count_code_points(const char *in, size_t size) { 9| 933| size_t pos = 0; 10| 933| size_t count = 0; 11| 666k| for (; pos + 64 <= size; pos += 64) { ------------------ | Branch (11:10): [True: 665k, False: 933] ------------------ 12| 665k| simd8x64 input(reinterpret_cast(in + pos)); 13| 665k| uint64_t utf8_continuation_mask = input.gt(-65); 14| 665k| count += count_ones(utf8_continuation_mask); 15| 665k| } 16| 933| return count + scalar::utf8::count_code_points(in + pos, size - pos); 17| 933|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14utf826count_code_points_bytemaskEPKcm: 21| 3.76k|count_code_points_bytemask(const char *in, size_t size) { 22| 3.76k| using vector_i8 = simd8; 23| 3.76k| using vector_u8 = simd8; 24| 3.76k| using vector_u64 = simd64; 25| | 26| 3.76k| constexpr size_t N = vector_i8::SIZE; 27| 3.76k| constexpr size_t max_iterations = 255 / 4; 28| | 29| 3.76k| size_t pos = 0; 30| 3.76k| size_t count = 0; 31| | 32| 3.76k| auto counters = vector_u64::zero(); 33| 3.76k| auto local = vector_u8::zero(); 34| 3.76k| size_t iterations = 0; 35| 3.34M| for (; pos + 4 * N <= size; pos += 4 * N) { ------------------ | Branch (35:10): [True: 3.33M, False: 3.76k] ------------------ 36| 3.33M| const auto input0 = 37| 3.33M| simd8::load(reinterpret_cast(in + pos + 0 * N)); 38| 3.33M| const auto input1 = 39| 3.33M| simd8::load(reinterpret_cast(in + pos + 1 * N)); 40| 3.33M| const auto input2 = 41| 3.33M| simd8::load(reinterpret_cast(in + pos + 2 * N)); 42| 3.33M| const auto input3 = 43| 3.33M| simd8::load(reinterpret_cast(in + pos + 3 * N)); 44| 3.33M| const auto mask0 = input0 > int8_t(-65); 45| 3.33M| const auto mask1 = input1 > int8_t(-65); 46| 3.33M| const auto mask2 = input2 > int8_t(-65); 47| 3.33M| const auto mask3 = input3 > int8_t(-65); 48| | 49| 3.33M| local -= vector_u8(mask0); 50| 3.33M| local -= vector_u8(mask1); 51| 3.33M| local -= vector_u8(mask2); 52| 3.33M| local -= vector_u8(mask3); 53| | 54| 3.33M| iterations += 1; 55| 3.33M| if (iterations == max_iterations) { ------------------ | Branch (55:9): [True: 52.6k, False: 3.28M] ------------------ 56| 52.6k| counters += sum_8bytes(local); 57| 52.6k| local = vector_u8::zero(); 58| 52.6k| iterations = 0; 59| 52.6k| } 60| 3.33M| } 61| | 62| 3.76k| if (iterations > 0) { ------------------ | Branch (62:7): [True: 2.31k, False: 1.44k] ------------------ 63| 2.31k| count += local.sum_bytes(); 64| 2.31k| } 65| | 66| 3.76k| count += counters.sum(); 67| | 68| 3.76k| return count + scalar::utf8::count_code_points(in + pos, size - pos); 69| 3.76k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14utf817count_code_pointsEPKcm: 8| 933|simdutf_really_inline size_t count_code_points(const char *in, size_t size) { 9| 933| size_t pos = 0; 10| 933| size_t count = 0; 11| 666k| for (; pos + 64 <= size; pos += 64) { ------------------ | Branch (11:10): [True: 665k, False: 933] ------------------ 12| 665k| simd8x64 input(reinterpret_cast(in + pos)); 13| 665k| uint64_t utf8_continuation_mask = input.gt(-65); 14| 665k| count += count_ones(utf8_continuation_mask); 15| 665k| } 16| 933| return count + scalar::utf8::count_code_points(in + pos, size - pos); 17| 933|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14utf831utf16_length_from_utf8_bytemaskEPKcm: 9| 1.74k| size_t size) { 10| 1.74k| using vector_i8 = simd8; 11| 1.74k| using vector_u8 = simd8; 12| 1.74k| using vector_u64 = simd64; 13| | 14| 1.74k| constexpr size_t N = vector_i8::SIZE; 15| 1.74k| constexpr size_t max_iterations = 255 / 2; 16| | 17| 1.74k| auto counters = vector_u64::zero(); 18| 1.74k| auto local = vector_u8::zero(); 19| | 20| 1.74k| size_t iterations = 0; 21| 1.74k| size_t pos = 0; 22| 1.74k| size_t count = 0; 23| 2.98M| for (; pos + N <= size; pos += N) { ------------------ | Branch (23:10): [True: 2.98M, False: 1.74k] ------------------ 24| 2.98M| const auto input = 25| 2.98M| vector_i8::load(reinterpret_cast(in + pos)); 26| | 27| 2.98M| const auto continuation = input > int8_t(-65); 28| 2.98M| const auto utf_4bytes = vector_u8(input.value) >= uint8_t(240); 29| | 30| 2.98M| local -= vector_u8(continuation); 31| 2.98M| local -= vector_u8(utf_4bytes); 32| | 33| 2.98M| iterations += 1; 34| 2.98M| if (iterations == max_iterations) { ------------------ | Branch (34:9): [True: 23.3k, False: 2.96M] ------------------ 35| 23.3k| counters += sum_8bytes(local); 36| 23.3k| local = vector_u8::zero(); 37| 23.3k| iterations = 0; 38| 23.3k| } 39| 2.98M| } 40| | 41| 1.74k| if (iterations > 0) { ------------------ | Branch (41:7): [True: 1.11k, False: 636] ------------------ 42| 1.11k| count += local.sum_bytes(); 43| 1.11k| } 44| | 45| 1.74k| count += counters.sum(); 46| | 47| 1.74k| return count + scalar::utf8::utf16_length_from_utf8(in + pos, size - pos); 48| 1.74k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14utf831utf16_length_from_utf8_bytemaskEPKcm: 9| 1.74k| size_t size) { 10| 1.74k| using vector_i8 = simd8; 11| 1.74k| using vector_u8 = simd8; 12| 1.74k| using vector_u64 = simd64; 13| | 14| 1.74k| constexpr size_t N = vector_i8::SIZE; 15| 1.74k| constexpr size_t max_iterations = 255 / 2; 16| | 17| 1.74k| auto counters = vector_u64::zero(); 18| 1.74k| auto local = vector_u8::zero(); 19| | 20| 1.74k| size_t iterations = 0; 21| 1.74k| size_t pos = 0; 22| 1.74k| size_t count = 0; 23| 5.97M| for (; pos + N <= size; pos += N) { ------------------ | Branch (23:10): [True: 5.97M, False: 1.74k] ------------------ 24| 5.97M| const auto input = 25| 5.97M| vector_i8::load(reinterpret_cast(in + pos)); 26| | 27| 5.97M| const auto continuation = input > int8_t(-65); 28| 5.97M| const auto utf_4bytes = vector_u8(input.value) >= uint8_t(240); 29| | 30| 5.97M| local -= vector_u8(continuation); 31| 5.97M| local -= vector_u8(utf_4bytes); 32| | 33| 5.97M| iterations += 1; 34| 5.97M| if (iterations == max_iterations) { ------------------ | Branch (34:9): [True: 46.8k, False: 5.92M] ------------------ 35| 46.8k| counters += sum_8bytes(local); 36| 46.8k| local = vector_u8::zero(); 37| 46.8k| iterations = 0; 38| 46.8k| } 39| 5.97M| } 40| | 41| 1.74k| if (iterations > 0) { ------------------ | Branch (41:7): [True: 1.21k, False: 535] ------------------ 42| 1.21k| count += local.sum_bytes(); 43| 1.21k| } 44| | 45| 1.74k| count += counters.sum(); 46| | 47| 1.74k| return count + scalar::utf8::utf16_length_from_utf8(in + pos, size - pos); 48| 1.74k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoderC2Ev: 101| 1.07k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder7convertEPKcmPc: 115| 562| char *latin1_output) { 116| 562| size_t pos = 0; 117| 562| 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| 562| size_t leading_byte = 0; 125| 562| size_t margin = size; 126| 9.10k| for (; margin > 0 && leading_byte < 16; margin--) { ------------------ | Branch (126:12): [True: 8.98k, False: 114] | Branch (126:26): [True: 8.53k, False: 448] ------------------ 127| 8.53k| leading_byte += (int8_t(in[margin - 1]) > 128| 8.53k| -65); // twos complement of -65 is 1011 1111 ... 129| 8.53k| } 130| | // If the input is long enough, then we have that margin-1 is the eight last 131| | // leading byte. 132| 562| const size_t safety_margin = size - margin + 1; // to avoid overruns! 133| 684k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (133:12): [True: 683k, False: 538] ------------------ 134| 683k| simd8x64 input(reinterpret_cast(in + pos)); 135| 683k| if (input.is_ascii()) { ------------------ | Branch (135:11): [True: 644k, False: 38.9k] ------------------ 136| 644k| input.store((int8_t *)latin1_output); 137| 644k| latin1_output += 64; 138| 644k| pos += 64; 139| 644k| } else { 140| | // you might think that a for-loop would work, but under Visual Studio, 141| | // it is not good enough. 142| 38.9k| static_assert( 143| 38.9k| (simd8x64::NUM_CHUNKS == 2) || 144| 38.9k| (simd8x64::NUM_CHUNKS == 4), 145| 38.9k| "We support either two or four chunks per 64-byte block."); 146| 38.9k| auto zero = simd8{uint8_t(0)}; 147| 38.9k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 148| 38.9k| this->check_utf8_bytes(input.chunks[0], zero); 149| 38.9k| 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| 38.9k| uint64_t utf8_continuation_mask = 157| 38.9k| input.lt(-65 + 1); // -64 is 1100 0000 in twos complement. Note: in 158| | // this case, we also have ASCII to account for. 159| 38.9k| if (utf8_continuation_mask & 1) { ------------------ | Branch (159:13): [True: 24, False: 38.9k] ------------------ 160| 24| return 0; // error 161| 24| } 162| 38.9k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 163| 38.9k| 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| 38.9k| size_t max_starting_point = (pos + 64) - 12; 168| | // Next loop is going to run at least five times. 169| 386k| while (pos < max_starting_point) { ------------------ | Branch (169:16): [True: 347k, False: 38.9k] ------------------ 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| 347k| size_t consumed = convert_masked_utf8_to_latin1( 180| 347k| in + pos, utf8_end_of_code_point_mask, latin1_output); 181| 347k| pos += consumed; 182| 347k| utf8_end_of_code_point_mask >>= consumed; 183| 347k| } 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| 38.9k| } 189| 683k| } 190| 538| if (errors()) { ------------------ | Branch (190:9): [True: 234, False: 304] ------------------ 191| 234| return 0; 192| 234| } 193| 304| if (pos < size) { ------------------ | Branch (193:9): [True: 300, False: 4] ------------------ 194| 300| size_t howmany = 195| 300| scalar::utf8_to_latin1::convert(in + pos, size - pos, latin1_output); 196| 300| if (howmany == 0) { ------------------ | Branch (196:11): [True: 128, False: 172] ------------------ 197| 128| return 0; 198| 128| } 199| 172| latin1_output += howmany; 200| 172| } 201| 176| return latin1_output - start; 202| 304| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 106| 121k| 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| 121k| simd8 prev1 = input.prev<1>(prev_input); 111| 121k| this->error |= check_special_cases(input, prev1); 112| 121k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin119check_special_casesENS1_4simd5simd8IhEES5_: 8| 121k|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| 121k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 20| | // 11______ 11______ 21| 121k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 22| 121k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 23| 121k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 24| 121k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 25| 121k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 26| 121k| 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| 121k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 35| | // 11110101 1000____ 36| | // 1111011_ 1000____ 37| | // 11111___ 1000____ 38| 121k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 39| 121k| constexpr const uint8_t FORBIDDEN = 0xff; 40| | 41| 121k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 42| | // 0_______ ________ 43| 121k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 44| 121k| TOO_LONG, 45| | // 10______ ________ 46| 121k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 47| | // 1100____ ________ 48| 121k| TOO_SHORT | OVERLONG_2, 49| | // 1101____ ________ 50| 121k| FORBIDDEN, 51| | // 1110____ ________ 52| 121k| FORBIDDEN, 53| | // 1111____ ________ 54| 121k| FORBIDDEN); 55| 121k| constexpr const uint8_t CARRY = 56| 121k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 57| 121k| const simd8 byte_1_low = 58| 121k| (prev1 & 0x0F) 59| 121k| .lookup_16( 60| | // ____0000 ________ 61| 121k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 62| | // ____0001 ________ 63| 121k| CARRY | OVERLONG_2, 64| | // ____001_ ________ 65| 121k| CARRY, CARRY, 66| | 67| | // ____0100 ________ 68| 121k| FORBIDDEN, 69| | // ____0101 ________ 70| 121k| FORBIDDEN, 71| | // ____011_ ________ 72| 121k| FORBIDDEN, FORBIDDEN, 73| | 74| | // ____1___ ________ 75| 121k| FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, 76| | // ____1101 ________ 77| 121k| FORBIDDEN, FORBIDDEN, FORBIDDEN); 78| 121k| const simd8 byte_2_high = input.shr<4>().lookup_16( 79| | // ________ 0_______ 80| 121k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 81| 121k| TOO_SHORT, TOO_SHORT, 82| | 83| | // ________ 1000____ 84| 121k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 85| 121k| OVERLONG_4, 86| | // ________ 1001____ 87| 121k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 88| | // ________ 101_____ 89| 121k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 90| 121k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 91| | 92| | // ________ 11______ 93| 121k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 94| 121k| return (byte_1_high & byte_1_low & byte_2_high); 95| 121k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder6errorsEv: 308| 22.8k| simdutf_really_inline bool errors() const { 309| 22.8k| return this->error.any_bits_set_anywhere(); 310| 22.8k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder19convert_with_errorsEPKcmPc: 205| 516| char *latin1_output) { 206| 516| size_t pos = 0; 207| 516| 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| 516| size_t leading_byte = 0; 215| 516| size_t margin = size; 216| 4.72k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (216:12): [True: 4.60k, False: 122] | Branch (216:26): [True: 4.20k, False: 394] ------------------ 217| 4.20k| leading_byte += (int8_t(in[margin - 1]) > -65); 218| 4.20k| } 219| | // If the input is long enough, then we have that margin-1 is the eight last 220| | // leading byte. 221| 516| const size_t safety_margin = size - margin + 1; // to avoid overruns! 222| 381k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (222:12): [True: 381k, False: 432] ------------------ 223| 381k| simd8x64 input(reinterpret_cast(in + pos)); 224| 381k| if (input.is_ascii()) { ------------------ | Branch (224:11): [True: 359k, False: 21.8k] ------------------ 225| 359k| input.store((int8_t *)latin1_output); 226| 359k| latin1_output += 64; 227| 359k| pos += 64; 228| 359k| } else { 229| | // you might think that a for-loop would work, but under Visual Studio, 230| | // it is not good enough. 231| 21.8k| static_assert( 232| 21.8k| (simd8x64::NUM_CHUNKS == 2) || 233| 21.8k| (simd8x64::NUM_CHUNKS == 4), 234| 21.8k| "We support either two or four chunks per 64-byte block."); 235| 21.8k| auto zero = simd8{uint8_t(0)}; 236| 21.8k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 237| 21.8k| this->check_utf8_bytes(input.chunks[0], zero); 238| 21.8k| 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| 21.8k| if (errors()) { ------------------ | Branch (245:13): [True: 84, False: 21.7k] ------------------ 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| 84| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 250| 84| pos, in + pos, size - pos, latin1_output); 251| 84| res.count += pos; 252| 84| return res; 253| 84| } 254| 21.7k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 255| 21.7k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 256| 21.7k| 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| 21.7k| size_t max_starting_point = (pos + 64) - 12; 261| | // Next loop is going to run at least five times. 262| 217k| while (pos < max_starting_point) { ------------------ | Branch (262:16): [True: 195k, False: 21.7k] ------------------ 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| 195k| size_t consumed = convert_masked_utf8_to_latin1( 273| 195k| in + pos, utf8_end_of_code_point_mask, latin1_output); 274| 195k| pos += consumed; 275| 195k| utf8_end_of_code_point_mask >>= consumed; 276| 195k| } 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| 21.7k| } 282| 381k| } 283| 432| if (errors()) { ------------------ | Branch (283:9): [True: 0, False: 432] ------------------ 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| 432| if (pos < size) { ------------------ | Branch (292:9): [True: 428, 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| 428| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 297| 428| pos, in + pos, size - pos, latin1_output); 298| 428| if (res.error) { // In case of error, we want the error position ------------------ | Branch (298:11): [True: 202, False: 226] ------------------ 299| 202| res.count += pos; 300| 202| return res; 301| 226| } else { // In case of success, we want the number of word written 302| 226| latin1_output += res.count; 303| 226| } 304| 428| } 305| 230| return result(error_code::SUCCESS, latin1_output - start); 306| 432| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoderC2Ev: 101| 1.07k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder7convertEPKcmPc: 115| 562| char *latin1_output) { 116| 562| size_t pos = 0; 117| 562| 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| 562| size_t leading_byte = 0; 125| 562| size_t margin = size; 126| 9.10k| for (; margin > 0 && leading_byte < 16; margin--) { ------------------ | Branch (126:12): [True: 8.98k, False: 114] | Branch (126:26): [True: 8.53k, False: 448] ------------------ 127| 8.53k| leading_byte += (int8_t(in[margin - 1]) > 128| 8.53k| -65); // twos complement of -65 is 1011 1111 ... 129| 8.53k| } 130| | // If the input is long enough, then we have that margin-1 is the eight last 131| | // leading byte. 132| 562| const size_t safety_margin = size - margin + 1; // to avoid overruns! 133| 684k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (133:12): [True: 683k, False: 538] ------------------ 134| 683k| simd8x64 input(reinterpret_cast(in + pos)); 135| 683k| if (input.is_ascii()) { ------------------ | Branch (135:11): [True: 644k, False: 38.9k] ------------------ 136| 644k| input.store((int8_t *)latin1_output); 137| 644k| latin1_output += 64; 138| 644k| pos += 64; 139| 644k| } else { 140| | // you might think that a for-loop would work, but under Visual Studio, 141| | // it is not good enough. 142| 38.9k| static_assert( 143| 38.9k| (simd8x64::NUM_CHUNKS == 2) || 144| 38.9k| (simd8x64::NUM_CHUNKS == 4), 145| 38.9k| "We support either two or four chunks per 64-byte block."); 146| 38.9k| 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| 38.9k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 151| 38.9k| this->check_utf8_bytes(input.chunks[0], zero); 152| 38.9k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 153| 38.9k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 154| 38.9k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 155| 38.9k| } 156| 38.9k| uint64_t utf8_continuation_mask = 157| 38.9k| input.lt(-65 + 1); // -64 is 1100 0000 in twos complement. Note: in 158| | // this case, we also have ASCII to account for. 159| 38.9k| if (utf8_continuation_mask & 1) { ------------------ | Branch (159:13): [True: 24, False: 38.9k] ------------------ 160| 24| return 0; // error 161| 24| } 162| 38.9k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 163| 38.9k| 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| 38.9k| size_t max_starting_point = (pos + 64) - 12; 168| | // Next loop is going to run at least five times. 169| 386k| while (pos < max_starting_point) { ------------------ | Branch (169:16): [True: 347k, False: 38.9k] ------------------ 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| 347k| size_t consumed = convert_masked_utf8_to_latin1( 180| 347k| in + pos, utf8_end_of_code_point_mask, latin1_output); 181| 347k| pos += consumed; 182| 347k| utf8_end_of_code_point_mask >>= consumed; 183| 347k| } 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| 38.9k| } 189| 683k| } 190| 538| if (errors()) { ------------------ | Branch (190:9): [True: 234, False: 304] ------------------ 191| 234| return 0; 192| 234| } 193| 304| if (pos < size) { ------------------ | Branch (193:9): [True: 300, False: 4] ------------------ 194| 300| size_t howmany = 195| 300| scalar::utf8_to_latin1::convert(in + pos, size - pos, latin1_output); 196| 300| if (howmany == 0) { ------------------ | Branch (196:11): [True: 128, False: 172] ------------------ 197| 128| return 0; 198| 128| } 199| 172| latin1_output += howmany; 200| 172| } 201| 176| return latin1_output - start; 202| 304| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 106| 243k| 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| 243k| simd8 prev1 = input.prev<1>(prev_input); 111| 243k| this->error |= check_special_cases(input, prev1); 112| 243k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin119check_special_casesENS1_4simd5simd8IhEES5_: 8| 243k|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| 243k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 20| | // 11______ 11______ 21| 243k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 22| 243k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 23| 243k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 24| 243k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 25| 243k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 26| 243k| 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| 243k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 35| | // 11110101 1000____ 36| | // 1111011_ 1000____ 37| | // 11111___ 1000____ 38| 243k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 39| 243k| constexpr const uint8_t FORBIDDEN = 0xff; 40| | 41| 243k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 42| | // 0_______ ________ 43| 243k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 44| 243k| TOO_LONG, 45| | // 10______ ________ 46| 243k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 47| | // 1100____ ________ 48| 243k| TOO_SHORT | OVERLONG_2, 49| | // 1101____ ________ 50| 243k| FORBIDDEN, 51| | // 1110____ ________ 52| 243k| FORBIDDEN, 53| | // 1111____ ________ 54| 243k| FORBIDDEN); 55| 243k| constexpr const uint8_t CARRY = 56| 243k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 57| 243k| const simd8 byte_1_low = 58| 243k| (prev1 & 0x0F) 59| 243k| .lookup_16( 60| | // ____0000 ________ 61| 243k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 62| | // ____0001 ________ 63| 243k| CARRY | OVERLONG_2, 64| | // ____001_ ________ 65| 243k| CARRY, CARRY, 66| | 67| | // ____0100 ________ 68| 243k| FORBIDDEN, 69| | // ____0101 ________ 70| 243k| FORBIDDEN, 71| | // ____011_ ________ 72| 243k| FORBIDDEN, FORBIDDEN, 73| | 74| | // ____1___ ________ 75| 243k| FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, 76| | // ____1101 ________ 77| 243k| FORBIDDEN, FORBIDDEN, FORBIDDEN); 78| 243k| const simd8 byte_2_high = input.shr<4>().lookup_16( 79| | // ________ 0_______ 80| 243k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 81| 243k| TOO_SHORT, TOO_SHORT, 82| | 83| | // ________ 1000____ 84| 243k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 85| 243k| OVERLONG_4, 86| | // ________ 1001____ 87| 243k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 88| | // ________ 101_____ 89| 243k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 90| 243k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 91| | 92| | // ________ 11______ 93| 243k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 94| 243k| return (byte_1_high & byte_1_low & byte_2_high); 95| 243k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder6errorsEv: 308| 22.8k| simdutf_really_inline bool errors() const { 309| 22.8k| return this->error.any_bits_set_anywhere(); 310| 22.8k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder19convert_with_errorsEPKcmPc: 205| 516| char *latin1_output) { 206| 516| size_t pos = 0; 207| 516| 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| 516| size_t leading_byte = 0; 215| 516| size_t margin = size; 216| 4.72k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (216:12): [True: 4.60k, False: 122] | Branch (216:26): [True: 4.20k, False: 394] ------------------ 217| 4.20k| leading_byte += (int8_t(in[margin - 1]) > -65); 218| 4.20k| } 219| | // If the input is long enough, then we have that margin-1 is the eight last 220| | // leading byte. 221| 516| const size_t safety_margin = size - margin + 1; // to avoid overruns! 222| 381k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (222:12): [True: 381k, False: 432] ------------------ 223| 381k| simd8x64 input(reinterpret_cast(in + pos)); 224| 381k| if (input.is_ascii()) { ------------------ | Branch (224:11): [True: 359k, False: 21.8k] ------------------ 225| 359k| input.store((int8_t *)latin1_output); 226| 359k| latin1_output += 64; 227| 359k| pos += 64; 228| 359k| } else { 229| | // you might think that a for-loop would work, but under Visual Studio, 230| | // it is not good enough. 231| 21.8k| static_assert( 232| 21.8k| (simd8x64::NUM_CHUNKS == 2) || 233| 21.8k| (simd8x64::NUM_CHUNKS == 4), 234| 21.8k| "We support either two or four chunks per 64-byte block."); 235| 21.8k| 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| 21.8k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 240| 21.8k| this->check_utf8_bytes(input.chunks[0], zero); 241| 21.8k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 242| 21.8k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 243| 21.8k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 244| 21.8k| } 245| 21.8k| if (errors()) { ------------------ | Branch (245:13): [True: 84, False: 21.7k] ------------------ 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| 84| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 250| 84| pos, in + pos, size - pos, latin1_output); 251| 84| res.count += pos; 252| 84| return res; 253| 84| } 254| 21.7k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 255| 21.7k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 256| 21.7k| 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| 21.7k| size_t max_starting_point = (pos + 64) - 12; 261| | // Next loop is going to run at least five times. 262| 217k| while (pos < max_starting_point) { ------------------ | Branch (262:16): [True: 195k, False: 21.7k] ------------------ 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| 195k| size_t consumed = convert_masked_utf8_to_latin1( 273| 195k| in + pos, utf8_end_of_code_point_mask, latin1_output); 274| 195k| pos += consumed; 275| 195k| utf8_end_of_code_point_mask >>= consumed; 276| 195k| } 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| 21.7k| } 282| 381k| } 283| 432| if (errors()) { ------------------ | Branch (283:9): [True: 0, False: 432] ------------------ 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| 432| if (pos < size) { ------------------ | Branch (292:9): [True: 428, 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| 428| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 297| 428| pos, in + pos, size - pos, latin1_output); 298| 428| if (res.error) { // In case of error, we want the error position ------------------ | Branch (298:11): [True: 202, False: 226] ------------------ 299| 202| res.count += pos; 300| 202| return res; 301| 226| } else { // In case of success, we want the number of word written 302| 226| latin1_output += res.count; 303| 226| } 304| 428| } 305| 230| return result(error_code::SUCCESS, latin1_output - start); 306| 432| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoderC2Ev: 114| 2.64k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoderC2Ev: 114| 2.64k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE0EEEmPKcmPDs: 130| 660| char16_t *utf16_output) { 131| 660| size_t pos = 0; 132| 660| 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| 660| size_t leading_byte = 0; 140| 660| size_t margin = size; 141| 6.34k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 6.13k, False: 218] | Branch (141:26): [True: 5.68k, False: 442] ------------------ 142| 5.68k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 5.68k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 660| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 586k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 585k, False: 638] ------------------ 148| 585k| simd8x64 input(reinterpret_cast(in + pos)); 149| 585k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 506k, False: 79.1k] ------------------ 150| 506k| input.store_ascii_as_utf16(utf16_output); 151| 506k| utf16_output += 64; 152| 506k| pos += 64; 153| 506k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 79.1k| static_assert( 157| 79.1k| (simd8x64::NUM_CHUNKS == 2) || 158| 79.1k| (simd8x64::NUM_CHUNKS == 4), 159| 79.1k| "We support either two or four chunks per 64-byte block."); 160| 79.1k| auto zero = simd8{uint8_t(0)}; 161| 79.1k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 162| 79.1k| this->check_utf8_bytes(input.chunks[0], zero); 163| 79.1k| 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| 79.1k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 79.1k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 22, False: 79.1k] ------------------ 172| 22| return 0; // error 173| 22| } 174| 79.1k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 79.1k| 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| 79.1k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 779k| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 700k, False: 79.1k] ------------------ 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| 700k| size_t consumed = convert_masked_utf8_to_utf16( 192| 700k| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 700k| pos += consumed; 194| 700k| utf8_end_of_code_point_mask >>= consumed; 195| 700k| } 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| 79.1k| } 201| 585k| } 202| 638| if (errors()) { ------------------ | Branch (202:9): [True: 250, False: 388] ------------------ 203| 250| return 0; 204| 250| } 205| 388| if (pos < size) { ------------------ | Branch (205:9): [True: 384, False: 4] ------------------ 206| 384| size_t howmany = scalar::utf8_to_utf16::convert( 207| 384| in + pos, size - pos, utf16_output); 208| 384| if (howmany == 0) { ------------------ | Branch (208:11): [True: 162, False: 222] ------------------ 209| 162| return 0; 210| 162| } 211| 222| utf16_output += howmany; 212| 222| } 213| 226| return utf16_output - start; 214| 388| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 960k| 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| 960k| simd8 prev1 = input.prev<1>(prev_input); 124| 960k| simd8 sc = check_special_cases(input, prev1); 125| 960k| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 960k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1619check_special_casesENS1_4simd5simd8IhEES5_: 8| 960k|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| 960k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 960k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 960k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 960k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 960k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 960k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 960k| 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| 960k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 960k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 960k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 960k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 960k| TOO_LONG, 40| | // 10______ ________ 41| 960k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 960k| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 960k| TOO_SHORT, 46| | // 1110____ ________ 47| 960k| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 960k| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 960k| constexpr const uint8_t CARRY = 51| 960k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 960k| const simd8 byte_1_low = 53| 960k| (prev1 & 0x0F) 54| 960k| .lookup_16( 55| | // ____0000 ________ 56| 960k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 960k| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 960k| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 960k| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 960k| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 960k| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 960k| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 960k| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 960k| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 960k| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 960k| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 960k| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 960k| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 960k| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 960k| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 960k| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 960k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 960k| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 960k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 960k| OVERLONG_4, 88| | // ________ 1001____ 89| 960k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 960k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 960k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 960k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 960k| return (byte_1_high & byte_1_low & byte_2_high); 97| 960k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1623check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 960k| const simd8 sc) { 102| 960k| simd8 prev2 = input.prev<2>(prev_input); 103| 960k| simd8 prev3 = input.prev<3>(prev_input); 104| 960k| simd8 must23 = 105| 960k| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 960k| simd8 must23_80 = must23 & uint8_t(0x80); 107| 960k| return must23_80 ^ sc; 108| 960k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder6errorsEv: 324| 233k| simdutf_really_inline bool errors() const { 325| 233k| return this->error.any_bits_set_anywhere(); 326| 233k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE1EEEmPKcmPDs: 130| 712| char16_t *utf16_output) { 131| 712| size_t pos = 0; 132| 712| 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| 712| size_t leading_byte = 0; 140| 712| size_t margin = size; 141| 6.53k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 6.32k, False: 214] | Branch (141:26): [True: 5.82k, False: 498] ------------------ 142| 5.82k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 5.82k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 712| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 700k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 699k, False: 688] ------------------ 148| 699k| simd8x64 input(reinterpret_cast(in + pos)); 149| 699k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 529k, False: 170k] ------------------ 150| 529k| input.store_ascii_as_utf16(utf16_output); 151| 529k| utf16_output += 64; 152| 529k| pos += 64; 153| 529k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 170k| static_assert( 157| 170k| (simd8x64::NUM_CHUNKS == 2) || 158| 170k| (simd8x64::NUM_CHUNKS == 4), 159| 170k| "We support either two or four chunks per 64-byte block."); 160| 170k| auto zero = simd8{uint8_t(0)}; 161| 170k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 162| 170k| this->check_utf8_bytes(input.chunks[0], zero); 163| 170k| 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| 170k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 170k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 24, False: 170k] ------------------ 172| 24| return 0; // error 173| 24| } 174| 170k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 170k| 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| 170k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 1.68M| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 1.51M, False: 170k] ------------------ 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| 1.51M| size_t consumed = convert_masked_utf8_to_utf16( 192| 1.51M| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 1.51M| pos += consumed; 194| 1.51M| utf8_end_of_code_point_mask >>= consumed; 195| 1.51M| } 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| 170k| } 201| 699k| } 202| 688| if (errors()) { ------------------ | Branch (202:9): [True: 310, False: 378] ------------------ 203| 310| return 0; 204| 310| } 205| 378| if (pos < size) { ------------------ | Branch (205:9): [True: 376, False: 2] ------------------ 206| 376| size_t howmany = scalar::utf8_to_utf16::convert( 207| 376| in + pos, size - pos, utf16_output); 208| 376| if (howmany == 0) { ------------------ | Branch (208:11): [True: 184, False: 192] ------------------ 209| 184| return 0; 210| 184| } 211| 192| utf16_output += howmany; 212| 192| } 213| 194| return utf16_output - start; 214| 378| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKcmPDs: 218| 574| char16_t *utf16_output) { 219| 574| size_t pos = 0; 220| 574| 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| 574| size_t leading_byte = 0; 228| 574| size_t margin = size; 229| 5.00k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 4.78k, False: 220] | Branch (229:26): [True: 4.42k, False: 354] ------------------ 230| 4.42k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 4.42k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 574| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 393k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 393k, False: 488] ------------------ 236| 393k| simd8x64 input(reinterpret_cast(in + pos)); 237| 393k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 293k, False: 100k] ------------------ 238| 293k| input.store_ascii_as_utf16(utf16_output); 239| 293k| utf16_output += 64; 240| 293k| pos += 64; 241| 293k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 100k| static_assert( 245| 100k| (simd8x64::NUM_CHUNKS == 2) || 246| 100k| (simd8x64::NUM_CHUNKS == 4), 247| 100k| "We support either two or four chunks per 64-byte block."); 248| 100k| auto zero = simd8{uint8_t(0)}; 249| 100k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 250| 100k| this->check_utf8_bytes(input.chunks[0], zero); 251| 100k| 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| 100k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 100k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 86, False: 99.9k] | Branch (259:25): [True: 0, False: 99.9k] ------------------ 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| 86| result res = 264| 86| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 86| pos, in + pos, size - pos, utf16_output); 266| 86| res.count += pos; 267| 86| return res; 268| 86| } 269| 99.9k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 99.9k| 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| 99.9k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 985k| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 885k, False: 99.9k] ------------------ 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| 885k| size_t consumed = convert_masked_utf8_to_utf16( 287| 885k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 885k| pos += consumed; 289| 885k| utf8_end_of_code_point_mask >>= consumed; 290| 885k| } 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| 99.9k| } 296| 393k| } 297| 488| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 488] ------------------ 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| 488| if (pos < size) { ------------------ | Branch (307:9): [True: 484, 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| 484| result res = 312| 484| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 484| pos, in + pos, size - pos, utf16_output); 314| 484| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 242, False: 242] ------------------ 315| 242| res.count += pos; 316| 242| return res; 317| 242| } else { // In case of success, we want the number of word written 318| 242| utf16_output += res.count; 319| 242| } 320| 484| } 321| 246| return result(error_code::SUCCESS, utf16_output - start); 322| 488| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKcmPDs: 218| 702| char16_t *utf16_output) { 219| 702| size_t pos = 0; 220| 702| 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| 702| size_t leading_byte = 0; 228| 702| size_t margin = size; 229| 6.13k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 5.88k, False: 250] | Branch (229:26): [True: 5.42k, False: 452] ------------------ 230| 5.42k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 5.42k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 702| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 441k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 440k, False: 634] ------------------ 236| 440k| simd8x64 input(reinterpret_cast(in + pos)); 237| 440k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 309k, False: 131k] ------------------ 238| 309k| input.store_ascii_as_utf16(utf16_output); 239| 309k| utf16_output += 64; 240| 309k| pos += 64; 241| 309k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 131k| static_assert( 245| 131k| (simd8x64::NUM_CHUNKS == 2) || 246| 131k| (simd8x64::NUM_CHUNKS == 4), 247| 131k| "We support either two or four chunks per 64-byte block."); 248| 131k| auto zero = simd8{uint8_t(0)}; 249| 131k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 250| 131k| this->check_utf8_bytes(input.chunks[0], zero); 251| 131k| 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| 131k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 131k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 68, False: 130k] | Branch (259:25): [True: 0, False: 130k] ------------------ 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| 68| result res = 264| 68| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 68| pos, in + pos, size - pos, utf16_output); 266| 68| res.count += pos; 267| 68| return res; 268| 68| } 269| 130k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 130k| 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| 130k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 1.29M| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 1.16M, False: 130k] ------------------ 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| 1.16M| size_t consumed = convert_masked_utf8_to_utf16( 287| 1.16M| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 1.16M| pos += consumed; 289| 1.16M| utf8_end_of_code_point_mask >>= consumed; 290| 1.16M| } 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| 130k| } 296| 440k| } 297| 634| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 634] ------------------ 298| | // rewind_and_convert_with_errors will seek a potential error from in+pos 299| | // onward, with the ability to go back up to pos bytes, and read size-pos 300| | // bytes forward. 301| 0| result res = 302| 0| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 303| 0| pos, in + pos, size - pos, utf16_output); 304| 0| res.count += pos; 305| 0| return res; 306| 0| } 307| 634| if (pos < size) { ------------------ | Branch (307:9): [True: 630, False: 4] ------------------ 308| | // rewind_and_convert_with_errors will seek a potential error from in+pos 309| | // onward, with the ability to go back up to pos bytes, and read size-pos 310| | // bytes forward. 311| 630| result res = 312| 630| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 630| pos, in + pos, size - pos, utf16_output); 314| 630| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 320, False: 310] ------------------ 315| 320| res.count += pos; 316| 320| return res; 317| 320| } else { // In case of success, we want the number of word written 318| 310| utf16_output += res.count; 319| 310| } 320| 630| } 321| 314| return result(error_code::SUCCESS, utf16_output - start); 322| 634| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE0EEEmPKcmPDs: 130| 660| char16_t *utf16_output) { 131| 660| size_t pos = 0; 132| 660| 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| 660| size_t leading_byte = 0; 140| 660| size_t margin = size; 141| 6.34k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 6.13k, False: 218] | Branch (141:26): [True: 5.68k, False: 442] ------------------ 142| 5.68k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 5.68k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 660| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 586k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 585k, False: 638] ------------------ 148| 585k| simd8x64 input(reinterpret_cast(in + pos)); 149| 585k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 506k, False: 79.1k] ------------------ 150| 506k| input.store_ascii_as_utf16(utf16_output); 151| 506k| utf16_output += 64; 152| 506k| pos += 64; 153| 506k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 79.1k| static_assert( 157| 79.1k| (simd8x64::NUM_CHUNKS == 2) || 158| 79.1k| (simd8x64::NUM_CHUNKS == 4), 159| 79.1k| "We support either two or four chunks per 64-byte block."); 160| 79.1k| 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| 79.1k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 165| 79.1k| this->check_utf8_bytes(input.chunks[0], zero); 166| 79.1k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 167| 79.1k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 168| 79.1k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 169| 79.1k| } 170| 79.1k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 79.1k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 22, False: 79.1k] ------------------ 172| 22| return 0; // error 173| 22| } 174| 79.1k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 79.1k| 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| 79.1k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 779k| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 700k, False: 79.1k] ------------------ 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| 700k| size_t consumed = convert_masked_utf8_to_utf16( 192| 700k| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 700k| pos += consumed; 194| 700k| utf8_end_of_code_point_mask >>= consumed; 195| 700k| } 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| 79.1k| } 201| 585k| } 202| 638| if (errors()) { ------------------ | Branch (202:9): [True: 250, False: 388] ------------------ 203| 250| return 0; 204| 250| } 205| 388| if (pos < size) { ------------------ | Branch (205:9): [True: 384, False: 4] ------------------ 206| 384| size_t howmany = scalar::utf8_to_utf16::convert( 207| 384| in + pos, size - pos, utf16_output); 208| 384| if (howmany == 0) { ------------------ | Branch (208:11): [True: 162, False: 222] ------------------ 209| 162| return 0; 210| 162| } 211| 222| utf16_output += howmany; 212| 222| } 213| 226| return utf16_output - start; 214| 388| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 1.92M| 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| 1.92M| simd8 prev1 = input.prev<1>(prev_input); 124| 1.92M| simd8 sc = check_special_cases(input, prev1); 125| 1.92M| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 1.92M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1619check_special_casesENS1_4simd5simd8IhEES5_: 8| 1.92M|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| 1.92M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 1.92M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 1.92M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 1.92M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 1.92M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 1.92M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 1.92M| 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| 1.92M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 1.92M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 1.92M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 1.92M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 1.92M| TOO_LONG, 40| | // 10______ ________ 41| 1.92M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 1.92M| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 1.92M| TOO_SHORT, 46| | // 1110____ ________ 47| 1.92M| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 1.92M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 1.92M| constexpr const uint8_t CARRY = 51| 1.92M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 1.92M| const simd8 byte_1_low = 53| 1.92M| (prev1 & 0x0F) 54| 1.92M| .lookup_16( 55| | // ____0000 ________ 56| 1.92M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 1.92M| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 1.92M| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 1.92M| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 1.92M| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 1.92M| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 1.92M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 1.92M| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 1.92M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 1.92M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 1.92M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 1.92M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 1.92M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 1.92M| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 1.92M| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 1.92M| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 1.92M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 1.92M| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 1.92M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 1.92M| OVERLONG_4, 88| | // ________ 1001____ 89| 1.92M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 1.92M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 1.92M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 1.92M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 1.92M| return (byte_1_high & byte_1_low & byte_2_high); 97| 1.92M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1623check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 1.92M| const simd8 sc) { 102| 1.92M| simd8 prev2 = input.prev<2>(prev_input); 103| 1.92M| simd8 prev3 = input.prev<3>(prev_input); 104| 1.92M| simd8 must23 = 105| 1.92M| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 1.92M| simd8 must23_80 = must23 & uint8_t(0x80); 107| 1.92M| return must23_80 ^ sc; 108| 1.92M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder6errorsEv: 324| 233k| simdutf_really_inline bool errors() const { 325| 233k| return this->error.any_bits_set_anywhere(); 326| 233k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE1EEEmPKcmPDs: 130| 712| char16_t *utf16_output) { 131| 712| size_t pos = 0; 132| 712| 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| 712| size_t leading_byte = 0; 140| 712| size_t margin = size; 141| 6.53k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 6.32k, False: 214] | Branch (141:26): [True: 5.82k, False: 498] ------------------ 142| 5.82k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 5.82k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 712| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 700k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 699k, False: 688] ------------------ 148| 699k| simd8x64 input(reinterpret_cast(in + pos)); 149| 699k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 529k, False: 170k] ------------------ 150| 529k| input.store_ascii_as_utf16(utf16_output); 151| 529k| utf16_output += 64; 152| 529k| pos += 64; 153| 529k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 170k| static_assert( 157| 170k| (simd8x64::NUM_CHUNKS == 2) || 158| 170k| (simd8x64::NUM_CHUNKS == 4), 159| 170k| "We support either two or four chunks per 64-byte block."); 160| 170k| 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| 170k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 165| 170k| this->check_utf8_bytes(input.chunks[0], zero); 166| 170k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 167| 170k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 168| 170k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 169| 170k| } 170| 170k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 170k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 24, False: 170k] ------------------ 172| 24| return 0; // error 173| 24| } 174| 170k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 170k| 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| 170k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 1.68M| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 1.51M, False: 170k] ------------------ 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| 1.51M| size_t consumed = convert_masked_utf8_to_utf16( 192| 1.51M| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 1.51M| pos += consumed; 194| 1.51M| utf8_end_of_code_point_mask >>= consumed; 195| 1.51M| } 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| 170k| } 201| 699k| } 202| 688| if (errors()) { ------------------ | Branch (202:9): [True: 310, False: 378] ------------------ 203| 310| return 0; 204| 310| } 205| 378| if (pos < size) { ------------------ | Branch (205:9): [True: 376, False: 2] ------------------ 206| 376| size_t howmany = scalar::utf8_to_utf16::convert( 207| 376| in + pos, size - pos, utf16_output); 208| 376| if (howmany == 0) { ------------------ | Branch (208:11): [True: 184, False: 192] ------------------ 209| 184| return 0; 210| 184| } 211| 192| utf16_output += howmany; 212| 192| } 213| 194| return utf16_output - start; 214| 378| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKcmPDs: 218| 574| char16_t *utf16_output) { 219| 574| size_t pos = 0; 220| 574| 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| 574| size_t leading_byte = 0; 228| 574| size_t margin = size; 229| 5.00k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 4.78k, False: 220] | Branch (229:26): [True: 4.42k, False: 354] ------------------ 230| 4.42k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 4.42k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 574| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 393k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 393k, False: 488] ------------------ 236| 393k| simd8x64 input(reinterpret_cast(in + pos)); 237| 393k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 293k, False: 100k] ------------------ 238| 293k| input.store_ascii_as_utf16(utf16_output); 239| 293k| utf16_output += 64; 240| 293k| pos += 64; 241| 293k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 100k| static_assert( 245| 100k| (simd8x64::NUM_CHUNKS == 2) || 246| 100k| (simd8x64::NUM_CHUNKS == 4), 247| 100k| "We support either two or four chunks per 64-byte block."); 248| 100k| 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| 100k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 253| 100k| this->check_utf8_bytes(input.chunks[0], zero); 254| 100k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 255| 100k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 256| 100k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 257| 100k| } 258| 100k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 100k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 86, False: 99.9k] | Branch (259:25): [True: 0, False: 99.9k] ------------------ 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| 86| result res = 264| 86| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 86| pos, in + pos, size - pos, utf16_output); 266| 86| res.count += pos; 267| 86| return res; 268| 86| } 269| 99.9k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 99.9k| 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| 99.9k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 985k| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 885k, False: 99.9k] ------------------ 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| 885k| size_t consumed = convert_masked_utf8_to_utf16( 287| 885k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 885k| pos += consumed; 289| 885k| utf8_end_of_code_point_mask >>= consumed; 290| 885k| } 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| 99.9k| } 296| 393k| } 297| 488| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 488] ------------------ 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| 488| if (pos < size) { ------------------ | Branch (307:9): [True: 484, 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| 484| result res = 312| 484| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 484| pos, in + pos, size - pos, utf16_output); 314| 484| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 242, False: 242] ------------------ 315| 242| res.count += pos; 316| 242| return res; 317| 242| } else { // In case of success, we want the number of word written 318| 242| utf16_output += res.count; 319| 242| } 320| 484| } 321| 246| return result(error_code::SUCCESS, utf16_output - start); 322| 488| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKcmPDs: 218| 702| char16_t *utf16_output) { 219| 702| size_t pos = 0; 220| 702| 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| 702| size_t leading_byte = 0; 228| 702| size_t margin = size; 229| 6.13k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 5.88k, False: 250] | Branch (229:26): [True: 5.42k, False: 452] ------------------ 230| 5.42k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 5.42k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 702| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 441k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 440k, False: 634] ------------------ 236| 440k| simd8x64 input(reinterpret_cast(in + pos)); 237| 440k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 309k, False: 131k] ------------------ 238| 309k| input.store_ascii_as_utf16(utf16_output); 239| 309k| utf16_output += 64; 240| 309k| pos += 64; 241| 309k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 131k| static_assert( 245| 131k| (simd8x64::NUM_CHUNKS == 2) || 246| 131k| (simd8x64::NUM_CHUNKS == 4), 247| 131k| "We support either two or four chunks per 64-byte block."); 248| 131k| 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| 131k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 253| 131k| this->check_utf8_bytes(input.chunks[0], zero); 254| 131k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 255| 131k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 256| 131k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 257| 131k| } 258| 131k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 131k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 68, False: 130k] | Branch (259:25): [True: 0, False: 130k] ------------------ 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| 68| result res = 264| 68| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 68| pos, in + pos, size - pos, utf16_output); 266| 68| res.count += pos; 267| 68| return res; 268| 68| } 269| 130k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 130k| 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| 130k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 1.29M| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 1.16M, False: 130k] ------------------ 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| 1.16M| size_t consumed = convert_masked_utf8_to_utf16( 287| 1.16M| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 1.16M| pos += consumed; 289| 1.16M| utf8_end_of_code_point_mask >>= consumed; 290| 1.16M| } 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| 130k| } 296| 440k| } 297| 634| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 634] ------------------ 298| | // rewind_and_convert_with_errors will seek a potential error from in+pos 299| | // onward, with the ability to go back up to pos bytes, and read size-pos 300| | // bytes forward. 301| 0| result res = 302| 0| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 303| 0| pos, in + pos, size - pos, utf16_output); 304| 0| res.count += pos; 305| 0| return res; 306| 0| } 307| 634| if (pos < size) { ------------------ | Branch (307:9): [True: 630, False: 4] ------------------ 308| | // rewind_and_convert_with_errors will seek a potential error from in+pos 309| | // onward, with the ability to go back up to pos bytes, and read size-pos 310| | // bytes forward. 311| 630| result res = 312| 630| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 630| pos, in + pos, size - pos, utf16_output); 314| 630| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 320, False: 310] ------------------ 315| 320| res.count += pos; 316| 320| return res; 317| 320| } else { // In case of success, we want the number of word written 318| 310| utf16_output += res.count; 319| 310| } 320| 630| } 321| 314| return result(error_code::SUCCESS, utf16_output - start); 322| 634| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE0EEEmPKcmPDs: 10| 368| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 368| size_t pos = 0; 14| 368| char16_t *start{utf16_output}; 15| 368| const size_t safety_margin = 16; // to avoid overruns! 16| 406k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 405k, False: 368] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 405k| simd8x64 in(reinterpret_cast(input + pos)); 20| 405k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 261k, False: 144k] ------------------ 21| 261k| in.store_ascii_as_utf16(utf16_output); 22| 261k| utf16_output += 64; 23| 261k| pos += 64; 24| 261k| } else { 25| | // Slow path. We hope that the compiler will recognize that this is a slow 26| | // path. Anything that is not a continuation mask is a 'leading byte', 27| | // that is, the start of a new code point. 28| 144k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 144k| 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| 144k| 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| 144k| 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| 1.42M| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 1.28M, False: 144k] ------------------ 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| 1.28M| size_t consumed = convert_masked_utf8_to_utf16( 56| 1.28M| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 1.28M| pos += consumed; 58| 1.28M| utf8_end_of_code_point_mask >>= consumed; 59| 1.28M| } 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| 144k| } 65| 405k| } 66| 368| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 368| input + pos, size - pos, utf16_output); 68| 368| return utf16_output - start; 69| 368|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE1EEEmPKcmPDs: 10| 438| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 438| size_t pos = 0; 14| 438| char16_t *start{utf16_output}; 15| 438| const size_t safety_margin = 16; // to avoid overruns! 16| 325k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 325k, False: 438] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 325k| simd8x64 in(reinterpret_cast(input + pos)); 20| 325k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 276k, False: 49.2k] ------------------ 21| 276k| in.store_ascii_as_utf16(utf16_output); 22| 276k| utf16_output += 64; 23| 276k| pos += 64; 24| 276k| } 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| 49.2k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 49.2k| 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| 49.2k| 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| 49.2k| 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| 464k| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 415k, False: 49.2k] ------------------ 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| 415k| size_t consumed = convert_masked_utf8_to_utf16( 56| 415k| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 415k| pos += consumed; 58| 415k| utf8_end_of_code_point_mask >>= consumed; 59| 415k| } 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| 49.2k| } 65| 325k| } 66| 438| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 438| input + pos, size - pos, utf16_output); 68| 438| return utf16_output - start; 69| 438|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE0EEEmPKcmPDs: 10| 368| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 368| size_t pos = 0; 14| 368| char16_t *start{utf16_output}; 15| 368| const size_t safety_margin = 16; // to avoid overruns! 16| 406k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 405k, False: 368] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 405k| simd8x64 in(reinterpret_cast(input + pos)); 20| 405k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 261k, False: 144k] ------------------ 21| 261k| in.store_ascii_as_utf16(utf16_output); 22| 261k| utf16_output += 64; 23| 261k| pos += 64; 24| 261k| } else { 25| | // Slow path. We hope that the compiler will recognize that this is a slow 26| | // path. Anything that is not a continuation mask is a 'leading byte', 27| | // that is, the start of a new code point. 28| 144k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 144k| 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| 144k| 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| 144k| 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| 1.42M| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 1.28M, False: 144k] ------------------ 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| 1.28M| size_t consumed = convert_masked_utf8_to_utf16( 56| 1.28M| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 1.28M| pos += consumed; 58| 1.28M| utf8_end_of_code_point_mask >>= consumed; 59| 1.28M| } 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| 144k| } 65| 405k| } 66| 368| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 368| input + pos, size - pos, utf16_output); 68| 368| return utf16_output - start; 69| 368|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE1EEEmPKcmPDs: 10| 438| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 438| size_t pos = 0; 14| 438| char16_t *start{utf16_output}; 15| 438| const size_t safety_margin = 16; // to avoid overruns! 16| 325k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 325k, False: 438] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 325k| simd8x64 in(reinterpret_cast(input + pos)); 20| 325k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 276k, False: 49.2k] ------------------ 21| 276k| in.store_ascii_as_utf16(utf16_output); 22| 276k| utf16_output += 64; 23| 276k| pos += 64; 24| 276k| } 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| 49.2k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 49.2k| 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| 49.2k| 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| 49.2k| 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| 464k| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 415k, False: 49.2k] ------------------ 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| 415k| size_t consumed = convert_masked_utf8_to_utf16( 56| 415k| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 415k| pos += consumed; 58| 415k| utf8_end_of_code_point_mask >>= consumed; 59| 415k| } 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| 49.2k| } 65| 325k| } 66| 438| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 438| input + pos, size - pos, utf16_output); 68| 438| return utf16_output - start; 69| 438|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoderC2Ev: 114| 1.45k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder7convertEPKcmPDi: 129| 712| char32_t *utf32_output) { 130| 712| size_t pos = 0; 131| 712| 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| 712| size_t leading_byte = 0; 139| 712| size_t margin = size; 140| 6.55k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (140:12): [True: 6.32k, False: 232] | Branch (140:26): [True: 5.84k, False: 480] ------------------ 141| 5.84k| leading_byte += (int8_t(in[margin - 1]) > -65); 142| 5.84k| } 143| | // If the input is long enough, then we have that margin-1 is the fourth 144| | // last leading byte. 145| 712| const size_t safety_margin = size - margin + 1; // to avoid overruns! 146| 724k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (146:12): [True: 724k, False: 690] ------------------ 147| 724k| simd8x64 input(reinterpret_cast(in + pos)); 148| 724k| if (input.is_ascii()) { ------------------ | Branch (148:11): [True: 473k, False: 250k] ------------------ 149| 473k| input.store_ascii_as_utf32(utf32_output); 150| 473k| utf32_output += 64; 151| 473k| pos += 64; 152| 473k| } 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: 22, False: 250k] ------------------ 171| 22| return 0; // we have an error 172| 22| } 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: 2.23M, 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| 2.23M| size_t consumed = convert_masked_utf8_to_utf32( 191| 2.23M| in + pos, utf8_end_of_code_point_mask, utf32_output); 192| 2.23M| pos += consumed; 193| 2.23M| utf8_end_of_code_point_mask >>= consumed; 194| 2.23M| } 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| 724k| } 201| 690| if (errors()) { ------------------ | Branch (201:9): [True: 244, False: 446] ------------------ 202| 244| return 0; 203| 244| } 204| 446| if (pos < size) { ------------------ | Branch (204:9): [True: 444, False: 2] ------------------ 205| 444| size_t howmany = 206| 444| scalar::utf8_to_utf32::convert(in + pos, size - pos, utf32_output); 207| 444| if (howmany == 0) { ------------------ | Branch (207:11): [True: 170, False: 274] ------------------ 208| 170| return 0; 209| 170| } 210| 274| utf32_output += howmany; 211| 274| } 212| 276| return utf32_output - start; 213| 446| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 582k| 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| 582k| simd8 prev1 = input.prev<1>(prev_input); 124| 582k| simd8 sc = check_special_cases(input, prev1); 125| 582k| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 582k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3219check_special_casesENS1_4simd5simd8IhEES5_: 8| 582k|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| 582k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 582k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 582k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 582k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 582k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 582k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 582k| 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| 582k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 582k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 582k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 582k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 582k| TOO_LONG, 40| | // 10______ ________ 41| 582k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 582k| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 582k| TOO_SHORT, 46| | // 1110____ ________ 47| 582k| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 582k| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 582k| constexpr const uint8_t CARRY = 51| 582k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 582k| const simd8 byte_1_low = 53| 582k| (prev1 & 0x0F) 54| 582k| .lookup_16( 55| | // ____0000 ________ 56| 582k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 582k| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 582k| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 582k| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 582k| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 582k| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 582k| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 582k| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 582k| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 582k| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 582k| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 582k| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 582k| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 582k| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 582k| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 582k| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 582k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 582k| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 582k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 582k| OVERLONG_4, 88| | // ________ 1001____ 89| 582k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 582k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 582k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 582k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 582k| return (byte_1_high & byte_1_low & byte_2_high); 97| 582k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3223check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 582k| const simd8 sc) { 102| 582k| simd8 prev2 = input.prev<2>(prev_input); 103| 582k| simd8 prev3 = input.prev<3>(prev_input); 104| 582k| simd8 must23 = 105| 582k| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 582k| simd8 must23_80 = must23 & uint8_t(0x80); 107| 582k| return must23_80 ^ sc; 108| 582k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder6errorsEv: 310| 42.1k| simdutf_really_inline bool errors() const { 311| 42.1k| return this->error.any_bits_set_anywhere(); 312| 42.1k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder19convert_with_errorsEPKcmPDi: 216| 738| char32_t *utf32_output) { 217| 738| size_t pos = 0; 218| 738| 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| 738| size_t leading_byte = 0; 226| 738| size_t margin = size; 227| 6.50k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (227:12): [True: 6.27k, False: 224] | Branch (227:26): [True: 5.76k, False: 514] ------------------ 228| 5.76k| leading_byte += (int8_t(in[margin - 1]) > -65); 229| 5.76k| } 230| | // If the input is long enough, then we have that margin-1 is the fourth 231| | // last leading byte. 232| 738| const size_t safety_margin = size - margin + 1; // to avoid overruns! 233| 442k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (233:12): [True: 441k, False: 652] ------------------ 234| 441k| simd8x64 input(reinterpret_cast(in + pos)); 235| 441k| if (input.is_ascii()) { ------------------ | Branch (235:11): [True: 400k, False: 40.8k] ------------------ 236| 400k| input.store_ascii_as_utf32(utf32_output); 237| 400k| utf32_output += 64; 238| 400k| pos += 64; 239| 400k| } else { 240| | // you might think that a for-loop would work, but under Visual Studio, 241| | // it is not good enough. 242| 40.8k| static_assert( 243| 40.8k| (simd8x64::NUM_CHUNKS == 2) || 244| 40.8k| (simd8x64::NUM_CHUNKS == 4), 245| 40.8k| "We support either two or four chunks per 64-byte block."); 246| 40.8k| auto zero = simd8{uint8_t(0)}; 247| 40.8k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 248| 40.8k| this->check_utf8_bytes(input.chunks[0], zero); 249| 40.8k| 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| 40.8k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 257| 40.8k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (257:13): [True: 86, False: 40.7k] | Branch (257:25): [True: 0, False: 40.7k] ------------------ 258| 86| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 259| 86| pos, in + pos, size - pos, utf32_output); 260| 86| res.count += pos; 261| 86| return res; 262| 86| } 263| 40.7k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 264| 40.7k| 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| 40.7k| size_t max_starting_point = (pos + 64) - 12; 269| | // Next loop is going to run at least five times. 270| 395k| while (pos < max_starting_point) { ------------------ | Branch (270:16): [True: 354k, False: 40.7k] ------------------ 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| 354k| size_t consumed = convert_masked_utf8_to_utf32( 281| 354k| in + pos, utf8_end_of_code_point_mask, utf32_output); 282| 354k| pos += consumed; 283| 354k| utf8_end_of_code_point_mask >>= consumed; 284| 354k| } 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| 40.7k| } 290| 441k| } 291| 652| if (errors()) { ------------------ | Branch (291:9): [True: 0, False: 652] ------------------ 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| 652| if (pos < size) { ------------------ | Branch (297:9): [True: 648, False: 4] ------------------ 298| 648| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 299| 648| pos, in + pos, size - pos, utf32_output); 300| 648| if (res.error) { // In case of error, we want the error position ------------------ | Branch (300:11): [True: 312, False: 336] ------------------ 301| 312| res.count += pos; 302| 312| return res; 303| 336| } else { // In case of success, we want the number of word written 304| 336| utf32_output += res.count; 305| 336| } 306| 648| } 307| 340| return result(error_code::SUCCESS, utf32_output - start); 308| 652| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoderC2Ev: 114| 1.45k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder7convertEPKcmPDi: 129| 712| char32_t *utf32_output) { 130| 712| size_t pos = 0; 131| 712| 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| 712| size_t leading_byte = 0; 139| 712| size_t margin = size; 140| 6.55k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (140:12): [True: 6.32k, False: 232] | Branch (140:26): [True: 5.84k, False: 480] ------------------ 141| 5.84k| leading_byte += (int8_t(in[margin - 1]) > -65); 142| 5.84k| } 143| | // If the input is long enough, then we have that margin-1 is the fourth 144| | // last leading byte. 145| 712| const size_t safety_margin = size - margin + 1; // to avoid overruns! 146| 724k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (146:12): [True: 724k, False: 690] ------------------ 147| 724k| simd8x64 input(reinterpret_cast(in + pos)); 148| 724k| if (input.is_ascii()) { ------------------ | Branch (148:11): [True: 473k, False: 250k] ------------------ 149| 473k| input.store_ascii_as_utf32(utf32_output); 150| 473k| utf32_output += 64; 151| 473k| pos += 64; 152| 473k| } 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: 22, False: 250k] ------------------ 171| 22| return 0; // we have an error 172| 22| } 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: 2.23M, 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| 2.23M| size_t consumed = convert_masked_utf8_to_utf32( 191| 2.23M| in + pos, utf8_end_of_code_point_mask, utf32_output); 192| 2.23M| pos += consumed; 193| 2.23M| utf8_end_of_code_point_mask >>= consumed; 194| 2.23M| } 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| 724k| } 201| 690| if (errors()) { ------------------ | Branch (201:9): [True: 244, False: 446] ------------------ 202| 244| return 0; 203| 244| } 204| 446| if (pos < size) { ------------------ | Branch (204:9): [True: 444, False: 2] ------------------ 205| 444| size_t howmany = 206| 444| scalar::utf8_to_utf32::convert(in + pos, size - pos, utf32_output); 207| 444| if (howmany == 0) { ------------------ | Branch (207:11): [True: 170, False: 274] ------------------ 208| 170| return 0; 209| 170| } 210| 274| utf32_output += howmany; 211| 274| } 212| 276| return utf32_output - start; 213| 446| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 1.16M| 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| 1.16M| simd8 prev1 = input.prev<1>(prev_input); 124| 1.16M| simd8 sc = check_special_cases(input, prev1); 125| 1.16M| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 1.16M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3219check_special_casesENS1_4simd5simd8IhEES5_: 8| 1.16M|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| 1.16M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 1.16M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 1.16M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 1.16M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 1.16M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 1.16M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 1.16M| 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| 1.16M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 1.16M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 1.16M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 1.16M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 1.16M| TOO_LONG, 40| | // 10______ ________ 41| 1.16M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 1.16M| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 1.16M| TOO_SHORT, 46| | // 1110____ ________ 47| 1.16M| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 1.16M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 1.16M| constexpr const uint8_t CARRY = 51| 1.16M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 1.16M| const simd8 byte_1_low = 53| 1.16M| (prev1 & 0x0F) 54| 1.16M| .lookup_16( 55| | // ____0000 ________ 56| 1.16M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 1.16M| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 1.16M| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 1.16M| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 1.16M| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 1.16M| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 1.16M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 1.16M| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 1.16M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 1.16M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 1.16M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 1.16M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 1.16M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 1.16M| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 1.16M| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 1.16M| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 1.16M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 1.16M| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 1.16M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 1.16M| OVERLONG_4, 88| | // ________ 1001____ 89| 1.16M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 1.16M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 1.16M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 1.16M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 1.16M| return (byte_1_high & byte_1_low & byte_2_high); 97| 1.16M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3223check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 1.16M| const simd8 sc) { 102| 1.16M| simd8 prev2 = input.prev<2>(prev_input); 103| 1.16M| simd8 prev3 = input.prev<3>(prev_input); 104| 1.16M| simd8 must23 = 105| 1.16M| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 1.16M| simd8 must23_80 = must23 & uint8_t(0x80); 107| 1.16M| return must23_80 ^ sc; 108| 1.16M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder6errorsEv: 310| 42.1k| simdutf_really_inline bool errors() const { 311| 42.1k| return this->error.any_bits_set_anywhere(); 312| 42.1k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder19convert_with_errorsEPKcmPDi: 216| 738| char32_t *utf32_output) { 217| 738| size_t pos = 0; 218| 738| 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| 738| size_t leading_byte = 0; 226| 738| size_t margin = size; 227| 6.50k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (227:12): [True: 6.27k, False: 224] | Branch (227:26): [True: 5.76k, False: 514] ------------------ 228| 5.76k| leading_byte += (int8_t(in[margin - 1]) > -65); 229| 5.76k| } 230| | // If the input is long enough, then we have that margin-1 is the fourth 231| | // last leading byte. 232| 738| const size_t safety_margin = size - margin + 1; // to avoid overruns! 233| 442k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (233:12): [True: 441k, False: 652] ------------------ 234| 441k| simd8x64 input(reinterpret_cast(in + pos)); 235| 441k| if (input.is_ascii()) { ------------------ | Branch (235:11): [True: 400k, False: 40.8k] ------------------ 236| 400k| input.store_ascii_as_utf32(utf32_output); 237| 400k| utf32_output += 64; 238| 400k| pos += 64; 239| 400k| } else { 240| | // you might think that a for-loop would work, but under Visual Studio, 241| | // it is not good enough. 242| 40.8k| static_assert( 243| 40.8k| (simd8x64::NUM_CHUNKS == 2) || 244| 40.8k| (simd8x64::NUM_CHUNKS == 4), 245| 40.8k| "We support either two or four chunks per 64-byte block."); 246| 40.8k| 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| 40.8k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 251| 40.8k| this->check_utf8_bytes(input.chunks[0], zero); 252| 40.8k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 253| 40.8k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 254| 40.8k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 255| 40.8k| } 256| 40.8k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 257| 40.8k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (257:13): [True: 86, False: 40.7k] | Branch (257:25): [True: 0, False: 40.7k] ------------------ 258| 86| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 259| 86| pos, in + pos, size - pos, utf32_output); 260| 86| res.count += pos; 261| 86| return res; 262| 86| } 263| 40.7k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 264| 40.7k| 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| 40.7k| size_t max_starting_point = (pos + 64) - 12; 269| | // Next loop is going to run at least five times. 270| 395k| while (pos < max_starting_point) { ------------------ | Branch (270:16): [True: 354k, False: 40.7k] ------------------ 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| 354k| size_t consumed = convert_masked_utf8_to_utf32( 281| 354k| in + pos, utf8_end_of_code_point_mask, utf32_output); 282| 354k| pos += consumed; 283| 354k| utf8_end_of_code_point_mask >>= consumed; 284| 354k| } 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| 40.7k| } 290| 441k| } 291| 652| if (errors()) { ------------------ | Branch (291:9): [True: 0, False: 652] ------------------ 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| 652| if (pos < size) { ------------------ | Branch (297:9): [True: 648, False: 4] ------------------ 298| 648| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 299| 648| pos, in + pos, size - pos, utf32_output); 300| 648| if (res.error) { // In case of error, we want the error position ------------------ | Branch (300:11): [True: 312, False: 336] ------------------ 301| 312| res.count += pos; 302| 312| return res; 303| 336| } else { // In case of success, we want the number of word written 304| 336| utf32_output += res.count; 305| 336| } 306| 648| } 307| 340| return result(error_code::SUCCESS, utf32_output - start); 308| 652| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3213convert_validEPKcmPDi: 9| 394| char32_t *utf32_output) noexcept { 10| 394| size_t pos = 0; 11| 394| char32_t *start{utf32_output}; 12| 394| const size_t safety_margin = 16; // to avoid overruns! 13| 177k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (13:10): [True: 176k, False: 394] ------------------ 14| 176k| simd8x64 in(reinterpret_cast(input + pos)); 15| 176k| if (in.is_ascii()) { ------------------ | Branch (15:9): [True: 131k, False: 45.3k] ------------------ 16| 131k| in.store_ascii_as_utf32(utf32_output); 17| 131k| utf32_output += 64; 18| 131k| pos += 64; 19| 131k| } else { 20| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 21| | // byte 22| 45.3k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 23| 45.3k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 24| 45.3k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 25| 45.3k| size_t max_starting_point = (pos + 64) - 12; 26| 439k| while (pos < max_starting_point) { ------------------ | Branch (26:14): [True: 394k, False: 45.3k] ------------------ 27| 394k| size_t consumed = convert_masked_utf8_to_utf32( 28| 394k| input + pos, utf8_end_of_code_point_mask, utf32_output); 29| 394k| pos += consumed; 30| 394k| utf8_end_of_code_point_mask >>= consumed; 31| 394k| } 32| 45.3k| } 33| 176k| } 34| 394| utf32_output += scalar::utf8_to_utf32::convert_valid(input + pos, size - pos, 35| 394| utf32_output); 36| 394| return utf32_output - start; 37| 394|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3213convert_validEPKcmPDi: 9| 394| char32_t *utf32_output) noexcept { 10| 394| size_t pos = 0; 11| 394| char32_t *start{utf32_output}; 12| 394| const size_t safety_margin = 16; // to avoid overruns! 13| 177k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (13:10): [True: 176k, False: 394] ------------------ 14| 176k| simd8x64 in(reinterpret_cast(input + pos)); 15| 176k| if (in.is_ascii()) { ------------------ | Branch (15:9): [True: 131k, False: 45.3k] ------------------ 16| 131k| in.store_ascii_as_utf32(utf32_output); 17| 131k| utf32_output += 64; 18| 131k| pos += 64; 19| 131k| } else { 20| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 21| | // byte 22| 45.3k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 23| 45.3k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 24| 45.3k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 25| 45.3k| size_t max_starting_point = (pos + 64) - 12; 26| 439k| while (pos < max_starting_point) { ------------------ | Branch (26:14): [True: 394k, False: 45.3k] ------------------ 27| 394k| size_t consumed = convert_masked_utf8_to_utf32( 28| 394k| input + pos, utf8_end_of_code_point_mask, utf32_output); 29| 394k| pos += consumed; 30| 394k| utf8_end_of_code_point_mask >>= consumed; 31| 394k| } 32| 45.3k| } 33| 176k| } 34| 394| utf32_output += scalar::utf8_to_utf32::convert_valid(input + pos, size - pos, 35| 394| utf32_output); 36| 394| return utf32_output - start; 37| 394|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker16check_next_inputERKNS1_4simd8simd8x64IhEE: 187| 4.64M| simdutf_really_inline void check_next_input(const simd8x64 &input) { 188| 4.64M| if (simdutf_likely(is_ascii(input))) { ------------------ | | 90| 4.64M| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.81M, False: 828k] | | ------------------ ------------------ 189| 3.81M| this->error |= this->prev_incomplete; 190| 3.81M| } else { 191| | // you might think that a for-loop would work, but under Visual Studio, it 192| | // is not good enough. 193| 828k| static_assert((simd8x64::NUM_CHUNKS == 2) || 194| 828k| (simd8x64::NUM_CHUNKS == 4), 195| 828k| "We support either two or four chunks per 64-byte block."); 196| 828k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 197| 828k| this->check_utf8_bytes(input.chunks[0], this->prev_input_block); 198| 828k| 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| 828k| this->prev_incomplete = 206| 828k| is_incomplete(input.chunks[simd8x64::NUM_CHUNKS - 1]); 207| 828k| this->prev_input_block = input.chunks[simd8x64::NUM_CHUNKS - 1]; 208| 828k| } 209| 4.64M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker16check_utf8_bytesENS1_4simd5simd8IhEES6_: 169| 1.65M| 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| 1.65M| simd8 prev1 = input.prev<1>(prev_input); 174| 1.65M| simd8 sc = check_special_cases(input, prev1); 175| 1.65M| this->error |= check_multibyte_lengths(input, prev_input, sc); 176| 1.65M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation19check_special_casesENS1_4simd5simd8IhEES5_: 9| 1.65M|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| 1.65M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 17| | // 11______ 11______ 18| 1.65M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 19| 1.65M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 20| 1.65M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 21| 1.65M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 22| 1.65M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 23| 1.65M| 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| 1.65M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 32| | // 11110101 1000____ 33| | // 1111011_ 1000____ 34| | // 11111___ 1000____ 35| 1.65M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 36| | 37| 1.65M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 38| | // 0_______ ________ 39| 1.65M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 40| 1.65M| TOO_LONG, 41| | // 10______ ________ 42| 1.65M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 43| | // 1100____ ________ 44| 1.65M| TOO_SHORT | OVERLONG_2, 45| | // 1101____ ________ 46| 1.65M| TOO_SHORT, 47| | // 1110____ ________ 48| 1.65M| TOO_SHORT | OVERLONG_3 | SURROGATE, 49| | // 1111____ ________ 50| 1.65M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 51| 1.65M| constexpr const uint8_t CARRY = 52| 1.65M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 53| 1.65M| const simd8 byte_1_low = 54| 1.65M| (prev1 & 0x0F) 55| 1.65M| .lookup_16( 56| | // ____0000 ________ 57| 1.65M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 58| | // ____0001 ________ 59| 1.65M| CARRY | OVERLONG_2, 60| | // ____001_ ________ 61| 1.65M| CARRY, CARRY, 62| | 63| | // ____0100 ________ 64| 1.65M| CARRY | TOO_LARGE, 65| | // ____0101 ________ 66| 1.65M| CARRY | TOO_LARGE | TOO_LARGE_1000, 67| | // ____011_ ________ 68| 1.65M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| 1.65M| CARRY | TOO_LARGE | TOO_LARGE_1000, 70| | 71| | // ____1___ ________ 72| 1.65M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 1.65M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 1.65M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 1.65M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| 1.65M| CARRY | TOO_LARGE | TOO_LARGE_1000, 77| | // ____1101 ________ 78| 1.65M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 79| 1.65M| CARRY | TOO_LARGE | TOO_LARGE_1000, 80| 1.65M| CARRY | TOO_LARGE | TOO_LARGE_1000); 81| 1.65M| const simd8 byte_2_high = input.shr<4>().lookup_16( 82| | // ________ 0_______ 83| 1.65M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 84| 1.65M| TOO_SHORT, TOO_SHORT, 85| | 86| | // ________ 1000____ 87| 1.65M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 88| 1.65M| OVERLONG_4, 89| | // ________ 1001____ 90| 1.65M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 91| | // ________ 101_____ 92| 1.65M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| 1.65M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 94| | 95| | // ________ 11______ 96| 1.65M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 97| 1.65M| return (byte_1_high & byte_1_low & byte_2_high); 98| 1.65M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation23check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 102| 1.65M| const simd8 sc) { 103| 1.65M| simd8 prev2 = input.prev<2>(prev_input); 104| 1.65M| simd8 prev3 = input.prev<3>(prev_input); 105| 1.65M| simd8 must23 = 106| 1.65M| simd8(must_be_2_3_continuation(prev2, prev3)); 107| 1.65M| simd8 must23_80 = must23 & uint8_t(0x80); 108| 1.65M| return must23_80 ^ sc; 109| 1.65M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation13is_incompleteENS1_4simd5simd8IhEE: 115| 828k|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| 828k| static const uint8_t max_array[32] = {255, 120| 828k| 255, 121| 828k| 255, 122| 828k| 255, 123| 828k| 255, 124| 828k| 255, 125| 828k| 255, 126| 828k| 255, 127| 828k| 255, 128| 828k| 255, 129| 828k| 255, 130| 828k| 255, 131| 828k| 255, 132| 828k| 255, 133| 828k| 255, 134| 828k| 255, 135| 828k| 255, 136| 828k| 255, 137| 828k| 255, 138| 828k| 255, 139| 828k| 255, 140| 828k| 255, 141| 828k| 255, 142| 828k| 255, 143| 828k| 255, 144| 828k| 255, 145| 828k| 255, 146| 828k| 255, 147| 828k| 255, 148| 828k| 0b11110000u - 1, 149| 828k| 0b11100000u - 1, 150| 828k| 0b11000000u - 1}; 151| 828k| const simd8 max_value( 152| 828k| &max_array[sizeof(max_array) - sizeof(simd8)]); 153| 828k| return input.gt_bits(max_value); 154| 828k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker9check_eofEv: 181| 5.63k| 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| 5.63k| this->error |= this->prev_incomplete; 185| 5.63k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker6errorsEv: 212| 1.89M| simdutf_really_inline bool errors() const { 213| 1.89M| return this->error.any_bits_set_anywhere(); 214| 1.89M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker16check_next_inputERKNS1_4simd8simd8x64IhEE: 187| 4.64M| simdutf_really_inline void check_next_input(const simd8x64 &input) { 188| 4.64M| if (simdutf_likely(is_ascii(input))) { ------------------ | | 90| 4.64M| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.81M, False: 828k] | | ------------------ ------------------ 189| 3.81M| this->error |= this->prev_incomplete; 190| 3.81M| } else { 191| | // you might think that a for-loop would work, but under Visual Studio, it 192| | // is not good enough. 193| 828k| static_assert((simd8x64::NUM_CHUNKS == 2) || 194| 828k| (simd8x64::NUM_CHUNKS == 4), 195| 828k| "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| 828k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 200| 828k| this->check_utf8_bytes(input.chunks[0], this->prev_input_block); 201| 828k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 202| 828k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 203| 828k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 204| 828k| } 205| 828k| this->prev_incomplete = 206| 828k| is_incomplete(input.chunks[simd8x64::NUM_CHUNKS - 1]); 207| 828k| this->prev_input_block = input.chunks[simd8x64::NUM_CHUNKS - 1]; 208| 828k| } 209| 4.64M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker16check_utf8_bytesENS1_4simd5simd8IhEES6_: 169| 3.31M| 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| 3.31M| simd8 prev1 = input.prev<1>(prev_input); 174| 3.31M| simd8 sc = check_special_cases(input, prev1); 175| 3.31M| this->error |= check_multibyte_lengths(input, prev_input, sc); 176| 3.31M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation19check_special_casesENS1_4simd5simd8IhEES5_: 9| 3.31M|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| 3.31M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 17| | // 11______ 11______ 18| 3.31M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 19| 3.31M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 20| 3.31M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 21| 3.31M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 22| 3.31M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 23| 3.31M| 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| 3.31M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 32| | // 11110101 1000____ 33| | // 1111011_ 1000____ 34| | // 11111___ 1000____ 35| 3.31M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 36| | 37| 3.31M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 38| | // 0_______ ________ 39| 3.31M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 40| 3.31M| TOO_LONG, 41| | // 10______ ________ 42| 3.31M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 43| | // 1100____ ________ 44| 3.31M| TOO_SHORT | OVERLONG_2, 45| | // 1101____ ________ 46| 3.31M| TOO_SHORT, 47| | // 1110____ ________ 48| 3.31M| TOO_SHORT | OVERLONG_3 | SURROGATE, 49| | // 1111____ ________ 50| 3.31M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 51| 3.31M| constexpr const uint8_t CARRY = 52| 3.31M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 53| 3.31M| const simd8 byte_1_low = 54| 3.31M| (prev1 & 0x0F) 55| 3.31M| .lookup_16( 56| | // ____0000 ________ 57| 3.31M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 58| | // ____0001 ________ 59| 3.31M| CARRY | OVERLONG_2, 60| | // ____001_ ________ 61| 3.31M| CARRY, CARRY, 62| | 63| | // ____0100 ________ 64| 3.31M| CARRY | TOO_LARGE, 65| | // ____0101 ________ 66| 3.31M| CARRY | TOO_LARGE | TOO_LARGE_1000, 67| | // ____011_ ________ 68| 3.31M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| 3.31M| CARRY | TOO_LARGE | TOO_LARGE_1000, 70| | 71| | // ____1___ ________ 72| 3.31M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 3.31M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 3.31M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 3.31M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| 3.31M| CARRY | TOO_LARGE | TOO_LARGE_1000, 77| | // ____1101 ________ 78| 3.31M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 79| 3.31M| CARRY | TOO_LARGE | TOO_LARGE_1000, 80| 3.31M| CARRY | TOO_LARGE | TOO_LARGE_1000); 81| 3.31M| const simd8 byte_2_high = input.shr<4>().lookup_16( 82| | // ________ 0_______ 83| 3.31M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 84| 3.31M| TOO_SHORT, TOO_SHORT, 85| | 86| | // ________ 1000____ 87| 3.31M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 88| 3.31M| OVERLONG_4, 89| | // ________ 1001____ 90| 3.31M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 91| | // ________ 101_____ 92| 3.31M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| 3.31M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 94| | 95| | // ________ 11______ 96| 3.31M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 97| 3.31M| return (byte_1_high & byte_1_low & byte_2_high); 98| 3.31M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation23check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 102| 3.31M| const simd8 sc) { 103| 3.31M| simd8 prev2 = input.prev<2>(prev_input); 104| 3.31M| simd8 prev3 = input.prev<3>(prev_input); 105| 3.31M| simd8 must23 = 106| 3.31M| simd8(must_be_2_3_continuation(prev2, prev3)); 107| 3.31M| simd8 must23_80 = must23 & uint8_t(0x80); 108| 3.31M| return must23_80 ^ sc; 109| 3.31M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation13is_incompleteENS1_4simd5simd8IhEE: 115| 828k|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| 828k| static const uint8_t max_array[32] = {255, 120| 828k| 255, 121| 828k| 255, 122| 828k| 255, 123| 828k| 255, 124| 828k| 255, 125| 828k| 255, 126| 828k| 255, 127| 828k| 255, 128| 828k| 255, 129| 828k| 255, 130| 828k| 255, 131| 828k| 255, 132| 828k| 255, 133| 828k| 255, 134| 828k| 255, 135| 828k| 255, 136| 828k| 255, 137| 828k| 255, 138| 828k| 255, 139| 828k| 255, 140| 828k| 255, 141| 828k| 255, 142| 828k| 255, 143| 828k| 255, 144| 828k| 255, 145| 828k| 255, 146| 828k| 255, 147| 828k| 255, 148| 828k| 0b11110000u - 1, 149| 828k| 0b11100000u - 1, 150| 828k| 0b11000000u - 1}; 151| 828k| const simd8 max_value( 152| 828k| &max_array[sizeof(max_array) - sizeof(simd8)]); 153| 828k| return input.gt_bits(max_value); 154| 828k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker9check_eofEv: 181| 5.63k| 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| 5.63k| this->error |= this->prev_incomplete; 185| 5.63k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker6errorsEv: 212| 1.89M| simdutf_really_inline bool errors() const { 213| 1.89M| return this->error.any_bits_set_anywhere(); 214| 1.89M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation21generic_validate_utf8EPKcm: 27| 3.22k|bool generic_validate_utf8(const char *input, size_t length) { 28| 3.22k| return generic_validate_utf8( 29| 3.22k| reinterpret_cast(input), length); 30| 3.22k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation21generic_validate_utf8INS2_12utf8_checkerEEEbPKhm: 10| 3.22k|bool generic_validate_utf8(const uint8_t *input, size_t length) { 11| 3.22k| checker c{}; 12| 3.22k| buf_block_reader<64> reader(input, length); 13| 2.75M| while (reader.has_full_block()) { ------------------ | Branch (13:10): [True: 2.74M, False: 3.22k] ------------------ 14| 2.74M| simd::simd8x64 in(reader.full_block()); 15| 2.74M| c.check_next_input(in); 16| 2.74M| reader.advance(); 17| 2.74M| } 18| 3.22k| uint8_t block[64]{}; 19| 3.22k| reader.get_remainder(block); 20| 3.22k| simd::simd8x64 in(block); 21| 3.22k| c.check_next_input(in); 22| 3.22k| reader.advance(); 23| 3.22k| c.check_eof(); 24| 3.22k| return !c.errors(); 25| 3.22k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsEPKcm: 76| 3.22k|result generic_validate_utf8_with_errors(const char *input, size_t length) { 77| 3.22k| return generic_validate_utf8_with_errors( 78| 3.22k| reinterpret_cast(input), length); 79| 3.22k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsINS2_12utf8_checkerEEENS_6resultEPKhm: 36| 3.22k|result generic_validate_utf8_with_errors(const uint8_t *input, size_t length) { 37| 3.22k| checker c{}; 38| 3.22k| buf_block_reader<64> reader(input, length); 39| 3.22k| size_t count{0}; 40| 1.89M| while (reader.has_full_block()) { ------------------ | Branch (40:10): [True: 1.88M, False: 2.41k] ------------------ 41| 1.88M| simd::simd8x64 in(reader.full_block()); 42| 1.88M| c.check_next_input(in); 43| 1.88M| if (c.errors()) { ------------------ | Branch (43:9): [True: 808, False: 1.88M] ------------------ 44| 808| if (count != 0) { ------------------ | Branch (44:11): [True: 208, False: 600] ------------------ 45| 208| count--; 46| 208| } // Sometimes the error is only detected in the next chunk 47| 808| result res = scalar::utf8::rewind_and_validate_with_errors( 48| 808| reinterpret_cast(input), 49| 808| reinterpret_cast(input + count), length - count); 50| 808| res.count += count; 51| 808| return res; 52| 808| } 53| 1.88M| reader.advance(); 54| 1.88M| count += 64; 55| 1.88M| } 56| 2.41k| uint8_t block[64]{}; 57| 2.41k| reader.get_remainder(block); 58| 2.41k| simd::simd8x64 in(block); 59| 2.41k| c.check_next_input(in); 60| 2.41k| reader.advance(); 61| 2.41k| c.check_eof(); 62| 2.41k| if (c.errors()) { ------------------ | Branch (62:7): [True: 797, False: 1.61k] ------------------ 63| 797| if (count != 0) { ------------------ | Branch (63:9): [True: 231, False: 566] ------------------ 64| 231| count--; 65| 231| } // Sometimes the error is only detected in the next chunk 66| 797| result res = scalar::utf8::rewind_and_validate_with_errors( 67| 797| reinterpret_cast(input), 68| 797| reinterpret_cast(input) + count, length - count); 69| 797| res.count += count; 70| 797| return res; 71| 1.61k| } else { 72| 1.61k| return result(error_code::SUCCESS, length); 73| 1.61k| } 74| 2.41k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation21generic_validate_utf8EPKcm: 27| 3.22k|bool generic_validate_utf8(const char *input, size_t length) { 28| 3.22k| return generic_validate_utf8( 29| 3.22k| reinterpret_cast(input), length); 30| 3.22k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation21generic_validate_utf8INS2_12utf8_checkerEEEbPKhm: 10| 3.22k|bool generic_validate_utf8(const uint8_t *input, size_t length) { 11| 3.22k| checker c{}; 12| 3.22k| buf_block_reader<64> reader(input, length); 13| 2.75M| while (reader.has_full_block()) { ------------------ | Branch (13:10): [True: 2.74M, False: 3.22k] ------------------ 14| 2.74M| simd::simd8x64 in(reader.full_block()); 15| 2.74M| c.check_next_input(in); 16| 2.74M| reader.advance(); 17| 2.74M| } 18| 3.22k| uint8_t block[64]{}; 19| 3.22k| reader.get_remainder(block); 20| 3.22k| simd::simd8x64 in(block); 21| 3.22k| c.check_next_input(in); 22| 3.22k| reader.advance(); 23| 3.22k| c.check_eof(); 24| 3.22k| return !c.errors(); 25| 3.22k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsEPKcm: 76| 3.22k|result generic_validate_utf8_with_errors(const char *input, size_t length) { 77| 3.22k| return generic_validate_utf8_with_errors( 78| 3.22k| reinterpret_cast(input), length); 79| 3.22k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsINS2_12utf8_checkerEEENS_6resultEPKhm: 36| 3.22k|result generic_validate_utf8_with_errors(const uint8_t *input, size_t length) { 37| 3.22k| checker c{}; 38| 3.22k| buf_block_reader<64> reader(input, length); 39| 3.22k| size_t count{0}; 40| 1.89M| while (reader.has_full_block()) { ------------------ | Branch (40:10): [True: 1.88M, False: 2.41k] ------------------ 41| 1.88M| simd::simd8x64 in(reader.full_block()); 42| 1.88M| c.check_next_input(in); 43| 1.88M| if (c.errors()) { ------------------ | Branch (43:9): [True: 808, False: 1.88M] ------------------ 44| 808| if (count != 0) { ------------------ | Branch (44:11): [True: 208, False: 600] ------------------ 45| 208| count--; 46| 208| } // Sometimes the error is only detected in the next chunk 47| 808| result res = scalar::utf8::rewind_and_validate_with_errors( 48| 808| reinterpret_cast(input), 49| 808| reinterpret_cast(input + count), length - count); 50| 808| res.count += count; 51| 808| return res; 52| 808| } 53| 1.88M| reader.advance(); 54| 1.88M| count += 64; 55| 1.88M| } 56| 2.41k| uint8_t block[64]{}; 57| 2.41k| reader.get_remainder(block); 58| 2.41k| simd::simd8x64 in(block); 59| 2.41k| c.check_next_input(in); 60| 2.41k| reader.advance(); 61| 2.41k| c.check_eof(); 62| 2.41k| if (c.errors()) { ------------------ | Branch (62:7): [True: 797, False: 1.61k] ------------------ 63| 797| if (count != 0) { ------------------ | Branch (63:9): [True: 231, False: 566] ------------------ 64| 231| count--; 65| 231| } // Sometimes the error is only detected in the next chunk 66| 797| result res = scalar::utf8::rewind_and_validate_with_errors( 67| 797| reinterpret_cast(input), 68| 797| reinterpret_cast(input) + count, length - count); 69| 797| res.count += count; 70| 797| return res; 71| 1.61k| } else { 72| 1.61k| return result(error_code::SUCCESS, length); 73| 1.61k| } 74| 2.41k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE0EEEKNS_6resultEPKDsm: 49| 2.59k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.59k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.59k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 17, False: 2.58k] | | ------------------ ------------------ 51| 17| return result(error_code::SUCCESS, 0); 52| 17| } 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| 2.70M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.70M, False: 2.42k] ------------------ 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| 2.70M| auto in0 = simd16(input); 67| 2.70M| auto in1 = 68| 2.70M| 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| 2.70M| 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| 2.70M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.70M| const uint16_t surrogates_bitmask = 77| 2.70M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.70M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.68M, False: 20.5k] ------------------ 79| 2.68M| input += 16; 80| 2.68M| } else { 81| | // 2. We have some surrogates that have to be distinguished: 82| | // - low surrogates: 0b1101'10xx'yyyy'yyyy (0xD800..0xDBFF) 83| | // - high surrogates: 0b1101'11xx'yyyy'yyyy (0xDC00..0xDFFF) 84| | // 85| | // Fact: high surrogate has 11th bit set (3rd bit in the higher byte) 86| | 87| | // V - non-surrogate code units 88| | // V = not surrogates_wordmask 89| 20.5k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 20.5k| const auto vH = (in & v_fc) == v_dc; 93| 20.5k| const uint16_t H = static_cast(vH.to_bitmask()); 94| | 95| | // L - word mask for low surrogates 96| | // L = not H and surrogates_wordmask 97| 20.5k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 20.5k| const uint16_t a = static_cast( 100| 20.5k| L & (H >> 1)); // A low surrogate must be followed by high one. 101| | // (A low surrogate placed in the 7th register's word 102| | // is an exception we handle.) 103| 20.5k| const uint16_t b = static_cast( 104| 20.5k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 20.5k| const uint16_t c = static_cast( 107| 20.5k| V | a | b); // Combine all the masks into the final one. 108| | 109| 20.5k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 15.4k, False: 5.16k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 15.4k| input += 16; 113| 15.4k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 5.00k, False: 158] ------------------ 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| 5.00k| input += 15; 119| 5.00k| } else { 120| 158| return result(error_code::SURROGATE, input - start); 121| 158| } 122| 20.5k| } 123| 2.70M| } 124| | 125| 2.42k| return result(error_code::SUCCESS, input - start); 126| 2.58k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE1EEEKNS_6resultEPKDsm: 49| 2.45k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.45k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.45k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 15, False: 2.44k] | | ------------------ ------------------ 51| 15| return result(error_code::SUCCESS, 0); 52| 15| } 53| | 54| 2.44k| const char16_t *start = input; 55| 2.44k| const char16_t *end = input + size; 56| | 57| 2.44k| const auto v_d8 = simd8::splat(0xd8); 58| 2.44k| const auto v_f8 = simd8::splat(0xf8); 59| 2.44k| const auto v_fc = simd8::splat(0xfc); 60| 2.44k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.80M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.80M, False: 2.24k] ------------------ 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| 2.80M| auto in0 = simd16(input); 67| 2.80M| auto in1 = 68| 2.80M| 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| 2.80M| 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| 2.80M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.80M| const uint16_t surrogates_bitmask = 77| 2.80M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.80M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.78M, False: 19.3k] ------------------ 79| 2.78M| input += 16; 80| 2.78M| } else { 81| | // 2. We have some surrogates that have to be distinguished: 82| | // - low surrogates: 0b1101'10xx'yyyy'yyyy (0xD800..0xDBFF) 83| | // - high surrogates: 0b1101'11xx'yyyy'yyyy (0xDC00..0xDFFF) 84| | // 85| | // Fact: high surrogate has 11th bit set (3rd bit in the higher byte) 86| | 87| | // V - non-surrogate code units 88| | // V = not surrogates_wordmask 89| 19.3k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 19.3k| const auto vH = (in & v_fc) == v_dc; 93| 19.3k| const uint16_t H = static_cast(vH.to_bitmask()); 94| | 95| | // L - word mask for low surrogates 96| | // L = not H and surrogates_wordmask 97| 19.3k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 19.3k| const uint16_t a = static_cast( 100| 19.3k| L & (H >> 1)); // A low surrogate must be followed by high one. 101| | // (A low surrogate placed in the 7th register's word 102| | // is an exception we handle.) 103| 19.3k| const uint16_t b = static_cast( 104| 19.3k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 19.3k| const uint16_t c = static_cast( 107| 19.3k| V | a | b); // Combine all the masks into the final one. 108| | 109| 19.3k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 15.0k, False: 4.30k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 15.0k| input += 16; 113| 15.0k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 4.10k, False: 196] ------------------ 114| | // The 15 lower code units of the input register contains valid UTF-16. 115| | // The 15th word may be either a low or high surrogate. It the next 116| | // iteration we 1) check if the low surrogate is followed by a high 117| | // one, 2) reject sole high surrogate. 118| 4.10k| input += 15; 119| 4.10k| } else { 120| 196| return result(error_code::SURROGATE, input - start); 121| 196| } 122| 19.3k| } 123| 2.80M| } 124| | 125| 2.24k| return result(error_code::SUCCESS, input - start); 126| 2.44k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE0EEEKNS_6resultEPKDsm: 49| 2.59k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.59k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.59k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 17, False: 2.58k] | | ------------------ ------------------ 51| 17| return result(error_code::SUCCESS, 0); 52| 17| } 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| 2.70M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.70M, False: 2.33k] ------------------ 63| | // 0. Load data: since the validation takes into account only higher 64| | // byte of each word, we compress the two vectors into one which 65| | // consists only the higher bytes. 66| 2.70M| auto in0 = simd16(input); 67| 2.70M| auto in1 = 68| 2.70M| 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| 2.70M| 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| 2.70M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.70M| const uint16_t surrogates_bitmask = 77| 2.70M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.70M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.68M, False: 21.1k] ------------------ 79| 2.68M| input += 16; 80| 2.68M| } 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| 21.1k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 21.1k| const auto vH = (in & v_fc) == v_dc; 93| 21.1k| const uint16_t H = static_cast(vH.to_bitmask()); 94| | 95| | // L - word mask for low surrogates 96| | // L = not H and surrogates_wordmask 97| 21.1k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 21.1k| const uint16_t a = static_cast( 100| 21.1k| L & (H >> 1)); // A low surrogate must be followed by high one. 101| | // (A low surrogate placed in the 7th register's word 102| | // is an exception we handle.) 103| 21.1k| const uint16_t b = static_cast( 104| 21.1k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 21.1k| const uint16_t c = static_cast( 107| 21.1k| V | a | b); // Combine all the masks into the final one. 108| | 109| 21.1k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 15.7k, False: 5.34k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 15.7k| input += 16; 113| 15.7k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 5.09k, False: 250] ------------------ 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| 5.09k| input += 15; 119| 5.09k| } else { 120| 250| return result(error_code::SURROGATE, input - start); 121| 250| } 122| 21.1k| } 123| 2.70M| } 124| | 125| 2.33k| return result(error_code::SUCCESS, input - start); 126| 2.58k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE1EEEKNS_6resultEPKDsm: 49| 2.45k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.45k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.45k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 15, False: 2.44k] | | ------------------ ------------------ 51| 15| return result(error_code::SUCCESS, 0); 52| 15| } 53| | 54| 2.44k| const char16_t *start = input; 55| 2.44k| const char16_t *end = input + size; 56| | 57| 2.44k| const auto v_d8 = simd8::splat(0xd8); 58| 2.44k| const auto v_f8 = simd8::splat(0xf8); 59| 2.44k| const auto v_fc = simd8::splat(0xfc); 60| 2.44k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.80M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.80M, False: 2.18k] ------------------ 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| 2.80M| auto in0 = simd16(input); 67| 2.80M| auto in1 = 68| 2.80M| 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| 2.80M| 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| 2.80M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.80M| const uint16_t surrogates_bitmask = 77| 2.80M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.80M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.78M, False: 19.7k] ------------------ 79| 2.78M| input += 16; 80| 2.78M| } else { 81| | // 2. We have some surrogates that have to be distinguished: 82| | // - low surrogates: 0b1101'10xx'yyyy'yyyy (0xD800..0xDBFF) 83| | // - high surrogates: 0b1101'11xx'yyyy'yyyy (0xDC00..0xDFFF) 84| | // 85| | // Fact: high surrogate has 11th bit set (3rd bit in the higher byte) 86| | 87| | // V - non-surrogate code units 88| | // V = not surrogates_wordmask 89| 19.7k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 19.7k| const auto vH = (in & v_fc) == v_dc; 93| 19.7k| const uint16_t H = static_cast(vH.to_bitmask()); 94| | 95| | // L - word mask for low surrogates 96| | // L = not H and surrogates_wordmask 97| 19.7k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 19.7k| const uint16_t a = static_cast( 100| 19.7k| L & (H >> 1)); // A low surrogate must be followed by high one. 101| | // (A low surrogate placed in the 7th register's word 102| | // is an exception we handle.) 103| 19.7k| const uint16_t b = static_cast( 104| 19.7k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 19.7k| const uint16_t c = static_cast( 107| 19.7k| V | a | b); // Combine all the masks into the final one. 108| | 109| 19.7k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 15.2k, False: 4.47k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 15.2k| input += 16; 113| 15.2k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 4.21k, False: 262] ------------------ 114| | // The 15 lower code units of the input register contains valid UTF-16. 115| | // The 15th word may be either a low or high surrogate. It the next 116| | // iteration we 1) check if the low surrogate is followed by a high 117| | // one, 2) reject sole high surrogate. 118| 4.21k| input += 15; 119| 4.21k| } else { 120| 262| return result(error_code::SURROGATE, input - start); 121| 262| } 122| 19.7k| } 123| 2.80M| } 124| | 125| 2.18k| return result(error_code::SUCCESS, input - start); 126| 2.44k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf328validateEPKDim: 6| 2.41k|simdutf_really_inline bool validate(const char32_t *input, size_t size) { 7| 2.41k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.41k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 21, False: 2.39k] | | ------------------ ------------------ 8| | // empty input is valid UTF-32. protect the implementation from 9| | // handling nullptr 10| 21| return true; 11| 21| } 12| | 13| 2.39k| const char32_t *end = input + size; 14| | 15| 2.39k| using vector_u32 = simd32; 16| | 17| 2.39k| const auto standardmax = vector_u32::splat(0x10ffff); 18| 2.39k| const auto offset = vector_u32::splat(0xffff2000); 19| 2.39k| const auto standardoffsetmax = vector_u32::splat(0xfffff7ff); 20| 2.39k| auto currentmax = vector_u32::zero(); 21| 2.39k| auto currentoffsetmax = vector_u32::zero(); 22| | 23| 2.39k| constexpr size_t N = vector_u32::ELEMENTS; 24| | 25| 1.13M| while (input + N < end) { ------------------ | Branch (25:10): [True: 1.13M, False: 2.39k] ------------------ 26| 1.13M| auto in = vector_u32(input); 27| 1.13M| if constexpr (!match_system(endianness::BIG)) { 28| 1.13M| in.swap_bytes(); 29| 1.13M| } 30| | 31| 1.13M| currentmax = max(currentmax, in); 32| 1.13M| currentoffsetmax = max(currentoffsetmax, in + offset); 33| 1.13M| input += N; 34| 1.13M| } 35| | 36| 2.39k| const auto too_large = currentmax > standardmax; 37| 2.39k| if (too_large.any()) { ------------------ | Branch (37:7): [True: 845, False: 1.55k] ------------------ 38| 845| return false; 39| 845| } 40| | 41| 1.55k| const auto surrogate = currentoffsetmax > standardoffsetmax; 42| 1.55k| if (surrogate.any()) { ------------------ | Branch (42:7): [True: 34, False: 1.51k] ------------------ 43| 34| return false; 44| 34| } 45| | 46| 1.51k| return scalar::utf32::validate(input, end - input); 47| 1.55k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf3220validate_with_errorsEPKDim: 50| 2.41k| size_t size) { 51| 2.41k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.41k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 21, False: 2.39k] | | ------------------ ------------------ 52| | // empty input is valid UTF-32. protect the implementation from 53| | // handling nullptr 54| 21| return result(error_code::SUCCESS, 0); 55| 21| } 56| | 57| 2.39k| const char32_t *start = input; 58| 2.39k| const char32_t *end = input + size; 59| | 60| 2.39k| using vector_u32 = simd32; 61| | 62| 2.39k| const auto standardmax = vector_u32::splat(0x10ffff + 1); 63| 2.39k| const auto surrogate_mask = vector_u32::splat(0xfffff800); 64| 2.39k| const auto surrogate_byte = vector_u32::splat(0x0000d800); 65| | 66| 2.39k| constexpr size_t N = vector_u32::ELEMENTS; 67| | 68| 128k| while (input + N < end) { ------------------ | Branch (68:10): [True: 126k, False: 1.51k] ------------------ 69| 126k| auto in = vector_u32(input); 70| 126k| if constexpr (!match_system(endianness::BIG)) { 71| 126k| in.swap_bytes(); 72| 126k| } 73| | 74| 126k| const auto too_large = in >= standardmax; 75| 126k| const auto surrogate = (in & surrogate_mask) == surrogate_byte; 76| | 77| 126k| const auto combined = too_large | surrogate; 78| 126k| if (simdutf_unlikely(combined.any())) { ------------------ | | 93| 126k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 879, False: 125k] | | ------------------ ------------------ 79| 879| const size_t consumed = input - start; 80| 879| auto sr = scalar::utf32::validate_with_errors(input, end - input); 81| 879| sr.count += consumed; 82| | 83| 879| return sr; 84| 879| } 85| | 86| 125k| input += N; 87| 125k| } 88| | 89| 1.51k| const size_t consumed = input - start; 90| 1.51k| auto sr = scalar::utf32::validate_with_errors(input, end - input); 91| 1.51k| sr.count += consumed; 92| | 93| 1.51k| return sr; 94| 2.39k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf328validateEPKDim: 6| 2.41k|simdutf_really_inline bool validate(const char32_t *input, size_t size) { 7| 2.41k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.41k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 21, False: 2.39k] | | ------------------ ------------------ 8| | // empty input is valid UTF-32. protect the implementation from 9| | // handling nullptr 10| 21| return true; 11| 21| } 12| | 13| 2.39k| const char32_t *end = input + size; 14| | 15| 2.39k| using vector_u32 = simd32; 16| | 17| 2.39k| const auto standardmax = vector_u32::splat(0x10ffff); 18| 2.39k| const auto offset = vector_u32::splat(0xffff2000); 19| 2.39k| const auto standardoffsetmax = vector_u32::splat(0xfffff7ff); 20| 2.39k| auto currentmax = vector_u32::zero(); 21| 2.39k| auto currentoffsetmax = vector_u32::zero(); 22| | 23| 2.39k| constexpr size_t N = vector_u32::ELEMENTS; 24| | 25| 2.26M| while (input + N < end) { ------------------ | Branch (25:10): [True: 2.26M, False: 2.39k] ------------------ 26| 2.26M| auto in = vector_u32(input); 27| 2.26M| if constexpr (!match_system(endianness::BIG)) { 28| 2.26M| in.swap_bytes(); 29| 2.26M| } 30| | 31| 2.26M| currentmax = max(currentmax, in); 32| 2.26M| currentoffsetmax = max(currentoffsetmax, in + offset); 33| 2.26M| input += N; 34| 2.26M| } 35| | 36| 2.39k| const auto too_large = currentmax > standardmax; 37| 2.39k| if (too_large.any()) { ------------------ | Branch (37:7): [True: 998, False: 1.39k] ------------------ 38| 998| return false; 39| 998| } 40| | 41| 1.39k| const auto surrogate = currentoffsetmax > standardoffsetmax; 42| 1.39k| if (surrogate.any()) { ------------------ | Branch (42:7): [True: 40, False: 1.35k] ------------------ 43| 40| return false; 44| 40| } 45| | 46| 1.35k| return scalar::utf32::validate(input, end - input); 47| 1.39k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf3220validate_with_errorsEPKDim: 50| 2.41k| size_t size) { 51| 2.41k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.41k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 21, False: 2.39k] | | ------------------ ------------------ 52| | // empty input is valid UTF-32. protect the implementation from 53| | // handling nullptr 54| 21| return result(error_code::SUCCESS, 0); 55| 21| } 56| | 57| 2.39k| const char32_t *start = input; 58| 2.39k| const char32_t *end = input + size; 59| | 60| 2.39k| using vector_u32 = simd32; 61| | 62| 2.39k| const auto standardmax = vector_u32::splat(0x10ffff + 1); 63| 2.39k| const auto surrogate_mask = vector_u32::splat(0xfffff800); 64| 2.39k| const auto surrogate_byte = vector_u32::splat(0x0000d800); 65| | 66| 2.39k| constexpr size_t N = vector_u32::ELEMENTS; 67| | 68| 254k| while (input + N < end) { ------------------ | Branch (68:10): [True: 253k, False: 1.35k] ------------------ 69| 253k| auto in = vector_u32(input); 70| 253k| if constexpr (!match_system(endianness::BIG)) { 71| 253k| in.swap_bytes(); 72| 253k| } 73| | 74| 253k| const auto too_large = in >= standardmax; 75| 253k| const auto surrogate = (in & surrogate_mask) == surrogate_byte; 76| | 77| 253k| const auto combined = too_large | surrogate; 78| 253k| if (simdutf_unlikely(combined.any())) { ------------------ | | 93| 253k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 1.03k, False: 252k] | | ------------------ ------------------ 79| 1.03k| const size_t consumed = input - start; 80| 1.03k| auto sr = scalar::utf32::validate_with_errors(input, end - input); 81| 1.03k| sr.count += consumed; 82| | 83| 1.03k| return sr; 84| 1.03k| } 85| | 86| 252k| input += N; 87| 252k| } 88| | 89| 1.35k| const size_t consumed = input - start; 90| 1.35k| auto sr = scalar::utf32::validate_with_errors(input, end - input); 91| 1.35k| sr.count += consumed; 92| | 93| 1.35k| return sr; 94| 2.39k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_latin1_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 80| char16_t *utf16_output) { 5| 80| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| | 7| 80| size_t i = 0; 8| 159k| for (; i < rounded_len; i += 16) { ------------------ | Branch (8:10): [True: 159k, False: 80] ------------------ 9| | // Load 16 bytes from the address (input + i) into a xmm register 10| 159k| const __m128i latin1 = 11| 159k| _mm_loadu_si128(reinterpret_cast(latin1_input + i)); 12| | 13| | // Zero extend each byte in `in` to word 14| 159k| __m256i utf16 = _mm256_cvtepu8_epi16(latin1); 15| | 16| 159k| if (big_endian) { ------------------ | Branch (16:9): [Folded, False: 159k] ------------------ 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| 159k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output + i), utf16); 25| 159k| } 26| | 27| 80| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 28| 80|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_latin1_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 80| char16_t *utf16_output) { 5| 80| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| | 7| 80| size_t i = 0; 8| 147k| for (; i < rounded_len; i += 16) { ------------------ | Branch (8:10): [True: 147k, False: 80] ------------------ 9| | // Load 16 bytes from the address (input + i) into a xmm register 10| 147k| const __m128i latin1 = 11| 147k| _mm_loadu_si128(reinterpret_cast(latin1_input + i)); 12| | 13| | // Zero extend each byte in `in` to word 14| 147k| __m256i utf16 = _mm256_cvtepu8_epi16(latin1); 15| | 16| 147k| if (big_endian) { ------------------ | Branch (16:9): [True: 147k, Folded] ------------------ 17| 147k| const __m128i swap128 = 18| 147k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 19| 147k| const __m256i swap = _mm256_set_m128i(swap128, swap128); ------------------ | | 4| 147k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 147k| _mm256_castsi128_si256(xmm2), 2) ------------------ 20| 147k| utf16 = _mm256_shuffle_epi8(utf16, swap); 21| 147k| } 22| | 23| | // Store the contents of xmm1 into the address pointed by (output + i) 24| 147k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output + i), utf16); 25| 147k| } 26| | 27| 80| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 28| 80|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_latin1_to_utf32EPKcmPDi: 3| 90| char32_t *utf32_output) { 4| 90| size_t rounded_len = ((len | 7) ^ 7); // Round down to nearest multiple of 8 5| | 6| 336k| for (size_t i = 0; i < rounded_len; i += 8) { ------------------ | Branch (6:22): [True: 336k, False: 90] ------------------ 7| | // Load 8 Latin1 characters into a 64-bit register 8| 336k| __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| 336k| __m256i out = _mm256_cvtepu8_epi32(in); 13| | 14| | // Store the results back to memory 15| 336k| _mm256_storeu_si256((__m256i *)&utf32_output[i], out); 16| 336k| } 17| | 18| | // return pointers pointing to where we left off 19| 90| return std::make_pair(buf + rounded_len, utf32_output + rounded_len); 20| 90|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_latin1_to_utf8EPKcmPc: 3| 538| char *utf8_output) { 4| 538| const char *end = latin1_input + len; 5| 538| const __m256i v_0000 = _mm256_setzero_si256(); 6| 538| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 7| 538| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 8| 538| const size_t safety_margin = 12; 9| | 10| 2.47M| while (end - latin1_input >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (10:10): [True: 2.47M, False: 538] ------------------ 11| 2.47M| __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| 2.47M| const __m128i v_80 = _mm_set1_epi8((char)0x80); 14| 2.47M| if (_mm_testz_si128(in8, v_80)) { // ASCII fast path!!!! ------------------ | Branch (14:9): [True: 2.26M, False: 205k] ------------------ 15| | // 1. store (16 bytes) 16| 2.26M| _mm_storeu_si128((__m128i *)utf8_output, in8); 17| | // 2. adjust pointers 18| 2.26M| latin1_input += 16; 19| 2.26M| utf8_output += 16; 20| 2.26M| continue; // we are done for this round! 21| 2.26M| } 22| | // We proceed only with the first 16 bytes. 23| 205k| 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| 205k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 29| 205k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 30| | 31| | // t0 = [0000|00aa|bbbb|bb00] 32| 205k| const __m256i t0 = _mm256_slli_epi16(in, 2); 33| | // t1 = [0000|00aa|0000|0000] 34| 205k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 35| | // t2 = [0000|0000|00bb|bbbb] 36| 205k| const __m256i t2 = _mm256_and_si256(in, v_003f); 37| | // t3 = [000a|aaaa|00bb|bbbb] 38| 205k| const __m256i t3 = _mm256_or_si256(t1, t2); 39| | // t4 = [1100|00aa|10bb|bbbb] 40| 205k| 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| 205k| const __m256i one_byte_bytemask = 46| 205k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 47| 205k| const uint32_t one_byte_bitmask = 48| 205k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 49| | 50| 205k| const __m256i utf8_unpacked = _mm256_blendv_epi8(t4, in, one_byte_bytemask); 51| | 52| | // 3. prepare bitmask for 8-bit lookup 53| 205k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 54| 205k| const uint32_t M1 = M0 >> 7; 55| 205k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 56| | // 4. pack the bytes 57| | 58| 205k| const uint8_t *row = 59| 205k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 60| 205k| const uint8_t *row_2 = 61| 205k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 16)] 62| 205k| [0]; 63| | 64| 205k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 65| 205k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 66| | 67| 205k| const __m256i utf8_packed = _mm256_shuffle_epi8( 68| 205k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 205k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 205k| _mm256_castsi128_si256(xmm2), 2) ------------------ 69| | // 5. store bytes 70| 205k| _mm_storeu_si128((__m128i *)utf8_output, 71| 205k| _mm256_castsi256_si128(utf8_packed)); 72| 205k| utf8_output += row[0]; 73| 205k| _mm_storeu_si128((__m128i *)utf8_output, 74| 205k| _mm256_extractf128_si256(utf8_packed, 1)); 75| 205k| utf8_output += row_2[0]; 76| | 77| | // 6. adjust pointers 78| 205k| latin1_input += 16; 79| 205k| continue; 80| | 81| 2.47M| } // while 82| 538| return std::make_pair(latin1_input, utf8_output); 83| 538|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_utf16_to_latin1ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 124| char *latin1_output) { 5| 124| const char16_t *end = buf + len; 6| 9.88k| while (end - buf >= 32) { ------------------ | Branch (6:10): [True: 9.78k, False: 94] ------------------ 7| | // Load 16 UTF-16 characters into 256-bit AVX2 register 8| 9.78k| __m256i in0 = _mm256_loadu_si256(reinterpret_cast(buf)); 9| 9.78k| __m256i in1 = 10| 9.78k| _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| 9.78k| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 21| 9.78k| if (_mm256_testz_si256(_mm256_or_si256(in0, in1), high_byte_mask)) { ------------------ | Branch (21:9): [True: 9.75k, False: 30] ------------------ 22| | // Pack 16-bit characters into 8-bit and store in latin1_output 23| 9.75k| const __m256i packed = _mm256_packus_epi16(in0, in1); 24| | 25| 9.75k| const __m256i result = _mm256_permute4x64_epi64(packed, 0b11011000); 26| | 27| 9.75k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(latin1_output), result); 28| | // Adjust pointers for the next iteration 29| 9.75k| buf += 32; 30| 9.75k| latin1_output += 32; 31| 9.75k| } else { 32| 30| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 33| 30| } 34| 9.78k| } // while 35| 94| return std::make_pair(buf, latin1_output); 36| 124|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_utf16_to_latin1ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 126| char *latin1_output) { 5| 126| const char16_t *end = buf + len; 6| 17.4k| while (end - buf >= 32) { ------------------ | Branch (6:10): [True: 17.3k, False: 104] ------------------ 7| | // Load 16 UTF-16 characters into 256-bit AVX2 register 8| 17.3k| __m256i in0 = _mm256_loadu_si256(reinterpret_cast(buf)); 9| 17.3k| __m256i in1 = 10| 17.3k| _mm256_loadu_si256(reinterpret_cast(buf + 16)); 11| | 12| 17.3k| if constexpr (!match_system(big_endian)) { 13| 17.3k| const __m256i swap = _mm256_setr_epi8( 14| 17.3k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 15| 17.3k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 16| 17.3k| in0 = _mm256_shuffle_epi8(in0, swap); 17| 17.3k| in1 = _mm256_shuffle_epi8(in1, swap); 18| 17.3k| } 19| | 20| 17.3k| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 21| 17.3k| if (_mm256_testz_si256(_mm256_or_si256(in0, in1), high_byte_mask)) { ------------------ | Branch (21:9): [True: 17.2k, False: 22] ------------------ 22| | // Pack 16-bit characters into 8-bit and store in latin1_output 23| 17.2k| const __m256i packed = _mm256_packus_epi16(in0, in1); 24| | 25| 17.2k| const __m256i result = _mm256_permute4x64_epi64(packed, 0b11011000); 26| | 27| 17.2k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(latin1_output), result); 28| | // Adjust pointers for the next iteration 29| 17.2k| buf += 32; 30| 17.2k| latin1_output += 32; 31| 17.2k| } else { 32| 22| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 33| 22| } 34| 17.3k| } // while 35| 104| return std::make_pair(buf, latin1_output); 36| 126|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_140avx2_convert_utf16_to_latin1_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 41| 230| char *latin1_output) { 42| 230| const char16_t *start = buf; 43| 230| const char16_t *end = buf + len; 44| 451k| while (end - buf >= 16) { ------------------ | Branch (44:10): [True: 451k, False: 138] ------------------ 45| 451k| __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| 451k| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 55| 451k| if (_mm256_testz_si256(in, high_byte_mask)) { ------------------ | Branch (55:9): [True: 451k, False: 92] ------------------ 56| 451k| __m128i lo = _mm256_extractf128_si256(in, 0); 57| 451k| __m128i hi = _mm256_extractf128_si256(in, 1); 58| 451k| __m128i latin1_packed_lo = _mm_packus_epi16(lo, lo); 59| 451k| __m128i latin1_packed_hi = _mm_packus_epi16(hi, hi); 60| 451k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 61| 451k| latin1_packed_lo); 62| 451k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output + 8), 63| 451k| latin1_packed_hi); 64| 451k| buf += 16; 65| 451k| latin1_output += 16; 66| 451k| } else { 67| | // Fallback to scalar code for handling errors 68| 490| for (int k = 0; k < 16; k++) { ------------------ | Branch (68:23): [True: 490, False: 0] ------------------ 69| 490| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 70| 490| if (word <= 0xff) { ------------------ | Branch (70:13): [True: 398, False: 92] ------------------ 71| 398| *latin1_output++ = char(word); 72| 398| } else { 73| 92| return std::make_pair( 74| 92| result{error_code::TOO_LARGE, (size_t)(buf - start + k)}, 75| 92| latin1_output); 76| 92| } 77| 490| } 78| 0| buf += 16; 79| 0| } 80| 451k| } // while 81| 138| return std::make_pair(result{error_code::SUCCESS, (size_t)(buf - start)}, 82| 138| latin1_output); 83| 230|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_140avx2_convert_utf16_to_latin1_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 41| 252| char *latin1_output) { 42| 252| const char16_t *start = buf; 43| 252| const char16_t *end = buf + len; 44| 523k| while (end - buf >= 16) { ------------------ | Branch (44:10): [True: 523k, False: 150] ------------------ 45| 523k| __m256i in = _mm256_loadu_si256(reinterpret_cast(buf)); 46| | 47| 523k| if constexpr (!match_system(big_endian)) { 48| 523k| const __m256i swap = _mm256_setr_epi8( 49| 523k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 50| 523k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 51| 523k| in = _mm256_shuffle_epi8(in, swap); 52| 523k| } 53| | 54| 523k| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 55| 523k| if (_mm256_testz_si256(in, high_byte_mask)) { ------------------ | Branch (55:9): [True: 523k, False: 102] ------------------ 56| 523k| __m128i lo = _mm256_extractf128_si256(in, 0); 57| 523k| __m128i hi = _mm256_extractf128_si256(in, 1); 58| 523k| __m128i latin1_packed_lo = _mm_packus_epi16(lo, lo); 59| 523k| __m128i latin1_packed_hi = _mm_packus_epi16(hi, hi); 60| 523k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 61| 523k| latin1_packed_lo); 62| 523k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output + 8), 63| 523k| latin1_packed_hi); 64| 523k| buf += 16; 65| 523k| latin1_output += 16; 66| 523k| } else { 67| | // Fallback to scalar code for handling errors 68| 574| for (int k = 0; k < 16; k++) { ------------------ | Branch (68:23): [True: 574, False: 0] ------------------ 69| 574| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 70| 574| if (word <= 0xff) { ------------------ | Branch (70:13): [True: 472, False: 102] ------------------ 71| 472| *latin1_output++ = char(word); 72| 472| } else { 73| 102| return std::make_pair( 74| 102| result{error_code::TOO_LARGE, (size_t)(buf - start + k)}, 75| 102| latin1_output); 76| 102| } 77| 574| } 78| 0| buf += 16; 79| 0| } 80| 523k| } // while 81| 150| return std::make_pair(result{error_code::SUCCESS, (size_t)(buf - start)}, 82| 150| latin1_output); 83| 252|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf16_to_utf32ILNS_10endiannessE0EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 384| char32_t *utf32_output) { 58| 384| const char16_t *end = buf + len; 59| 384| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 384| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| | 62| 162k| while (end - buf >= 16) { ------------------ | Branch (62:10): [True: 162k, False: 288] ------------------ 63| 162k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 64| 162k| if (big_endian) { ------------------ | Branch (64:9): [Folded, False: 162k] ------------------ 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| 162k| const __m256i surrogates_bytemask = 75| 162k| _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| 162k| const uint32_t surrogates_bitmask = 80| 162k| 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| 162k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (83:9): [True: 158k, False: 3.89k] ------------------ 84| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 85| | // units 86| 158k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 87| 158k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 88| 158k| _mm256_storeu_si256( 89| 158k| reinterpret_cast<__m256i *>(utf32_output + 8), 90| 158k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 91| 158k| utf32_output += 16; 92| 158k| buf += 16; 93| | // surrogate pair(s) in a register 94| 158k| } 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| 3.89k| size_t forward = 15; 99| 3.89k| size_t k = 0; 100| 3.89k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (100:11): [True: 0, False: 3.89k] ------------------ 101| 0| forward = size_t(end - buf - 1); 102| 0| } 103| 56.8k| for (; k < forward; k++) { ------------------ | Branch (103:14): [True: 53.0k, False: 3.80k] ------------------ 104| 53.0k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 105| 53.0k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (105:13): [True: 48.1k, False: 4.87k] ------------------ 106| | // No surrogate pair 107| 48.1k| *utf32_output++ = char32_t(word); 108| 48.1k| } else { 109| | // must be a surrogate pair 110| 4.87k| uint16_t diff = uint16_t(word - 0xD800); 111| 4.87k| uint16_t next_word = 112| 4.87k| scalar::utf16::swap_if_needed(buf[k + 1]); 113| 4.87k| k++; 114| 4.87k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 115| 4.87k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (115:15): [True: 96, False: 4.77k] ------------------ 116| 96| return std::make_pair(nullptr, utf32_output); 117| 96| } 118| 4.77k| uint32_t value = (diff << 10) + diff2 + 0x10000; 119| 4.77k| *utf32_output++ = char32_t(value); 120| 4.77k| } 121| 53.0k| } 122| 3.80k| buf += k; 123| 3.80k| } 124| 162k| } // while 125| 288| return std::make_pair(buf, utf32_output); 126| 384|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf16_to_utf32ILNS_10endiannessE1EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 366| char32_t *utf32_output) { 58| 366| const char16_t *end = buf + len; 59| 366| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 366| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| | 62| 322k| while (end - buf >= 16) { ------------------ | Branch (62:10): [True: 321k, False: 310] ------------------ 63| 321k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 64| 321k| if (big_endian) { ------------------ | Branch (64:9): [True: 321k, Folded] ------------------ 65| 321k| const __m256i swap = _mm256_setr_epi8( 66| 321k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 67| 321k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 68| 321k| in = _mm256_shuffle_epi8(in, swap); 69| 321k| } 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| 321k| const __m256i surrogates_bytemask = 75| 321k| _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| 321k| const uint32_t surrogates_bitmask = 80| 321k| 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| 321k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (83:9): [True: 320k, False: 1.07k] ------------------ 84| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 85| | // units 86| 320k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 87| 320k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 88| 320k| _mm256_storeu_si256( 89| 320k| reinterpret_cast<__m256i *>(utf32_output + 8), 90| 320k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 91| 320k| utf32_output += 16; 92| 320k| buf += 16; 93| | // surrogate pair(s) in a register 94| 320k| } 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| 1.07k| size_t forward = 15; 99| 1.07k| size_t k = 0; 100| 1.07k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (100:11): [True: 0, False: 1.07k] ------------------ 101| 0| forward = size_t(end - buf - 1); 102| 0| } 103| 15.0k| for (; k < forward; k++) { ------------------ | Branch (103:14): [True: 14.0k, False: 1.02k] ------------------ 104| 14.0k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 105| 14.0k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (105:13): [True: 12.3k, False: 1.77k] ------------------ 106| | // No surrogate pair 107| 12.3k| *utf32_output++ = char32_t(word); 108| 12.3k| } else { 109| | // must be a surrogate pair 110| 1.77k| uint16_t diff = uint16_t(word - 0xD800); 111| 1.77k| uint16_t next_word = 112| 1.77k| scalar::utf16::swap_if_needed(buf[k + 1]); 113| 1.77k| k++; 114| 1.77k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 115| 1.77k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (115:15): [True: 56, False: 1.71k] ------------------ 116| 56| return std::make_pair(nullptr, utf32_output); 117| 56| } 118| 1.71k| uint32_t value = (diff << 10) + diff2 + 0x10000; 119| 1.71k| *utf32_output++ = char32_t(value); 120| 1.71k| } 121| 14.0k| } 122| 1.02k| buf += k; 123| 1.02k| } 124| 321k| } // while 125| 310| return std::make_pair(buf, utf32_output); 126| 366|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf16_to_utf32_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 138| 414| char32_t *utf32_output) { 139| 414| const char16_t *start = buf; 140| 414| const char16_t *end = buf + len; 141| 414| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 142| 414| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 143| | 144| 157k| while (end - buf >= 16) { ------------------ | Branch (144:10): [True: 157k, False: 290] ------------------ 145| 157k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 146| 157k| if (big_endian) { ------------------ | Branch (146:9): [Folded, False: 157k] ------------------ 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| 157k| const __m256i surrogates_bytemask = 157| 157k| _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| 157k| const uint32_t surrogates_bitmask = 162| 157k| 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| 157k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (165:9): [True: 150k, False: 6.77k] ------------------ 166| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 167| | // units 168| 150k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 169| 150k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 170| 150k| _mm256_storeu_si256( 171| 150k| reinterpret_cast<__m256i *>(utf32_output + 8), 172| 150k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 173| 150k| utf32_output += 16; 174| 150k| buf += 16; 175| | // surrogate pair(s) in a register 176| 150k| } 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| 6.77k| size_t forward = 15; 181| 6.77k| size_t k = 0; 182| 6.77k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (182:11): [True: 0, False: 6.77k] ------------------ 183| 0| forward = size_t(end - buf - 1); 184| 0| } 185| 96.5k| for (; k < forward; k++) { ------------------ | Branch (185:14): [True: 89.9k, False: 6.64k] ------------------ 186| 89.9k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 187| 89.9k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (187:13): [True: 78.5k, False: 11.4k] ------------------ 188| | // No surrogate pair 189| 78.5k| *utf32_output++ = char32_t(word); 190| 78.5k| } else { 191| | // must be a surrogate pair 192| 11.4k| uint16_t diff = uint16_t(word - 0xD800); 193| 11.4k| uint16_t next_word = 194| 11.4k| scalar::utf16::swap_if_needed(buf[k + 1]); 195| 11.4k| k++; 196| 11.4k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 197| 11.4k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (197:15): [True: 124, False: 11.2k] ------------------ 198| 124| return std::make_pair( 199| 124| result(error_code::SURROGATE, buf - start + k - 1), 200| 124| utf32_output); 201| 124| } 202| 11.2k| uint32_t value = (diff << 10) + diff2 + 0x10000; 203| 11.2k| *utf32_output++ = char32_t(value); 204| 11.2k| } 205| 89.9k| } 206| 6.64k| buf += k; 207| 6.64k| } 208| 157k| } // while 209| 290| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 210| 414|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf16_to_utf32_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 138| 300| char32_t *utf32_output) { 139| 300| const char16_t *start = buf; 140| 300| const char16_t *end = buf + len; 141| 300| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 142| 300| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 143| | 144| 213k| while (end - buf >= 16) { ------------------ | Branch (144:10): [True: 213k, False: 246] ------------------ 145| 213k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 146| 213k| if (big_endian) { ------------------ | Branch (146:9): [True: 213k, Folded] ------------------ 147| 213k| const __m256i swap = _mm256_setr_epi8( 148| 213k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 149| 213k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 150| 213k| in = _mm256_shuffle_epi8(in, swap); 151| 213k| } 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| 213k| const __m256i surrogates_bytemask = 157| 213k| _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| 213k| const uint32_t surrogates_bitmask = 162| 213k| 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| 213k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (165:9): [True: 209k, False: 3.95k] ------------------ 166| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 167| | // units 168| 209k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 169| 209k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 170| 209k| _mm256_storeu_si256( 171| 209k| reinterpret_cast<__m256i *>(utf32_output + 8), 172| 209k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 173| 209k| utf32_output += 16; 174| 209k| buf += 16; 175| | // surrogate pair(s) in a register 176| 209k| } 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| 3.95k| size_t forward = 15; 181| 3.95k| size_t k = 0; 182| 3.95k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (182:11): [True: 0, False: 3.95k] ------------------ 183| 0| forward = size_t(end - buf - 1); 184| 0| } 185| 53.6k| for (; k < forward; k++) { ------------------ | Branch (185:14): [True: 49.7k, False: 3.89k] ------------------ 186| 49.7k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 187| 49.7k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (187:13): [True: 39.8k, False: 9.99k] ------------------ 188| | // No surrogate pair 189| 39.8k| *utf32_output++ = char32_t(word); 190| 39.8k| } else { 191| | // must be a surrogate pair 192| 9.99k| uint16_t diff = uint16_t(word - 0xD800); 193| 9.99k| uint16_t next_word = 194| 9.99k| scalar::utf16::swap_if_needed(buf[k + 1]); 195| 9.99k| k++; 196| 9.99k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 197| 9.99k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (197:15): [True: 54, False: 9.93k] ------------------ 198| 54| return std::make_pair( 199| 54| result(error_code::SURROGATE, buf - start + k - 1), 200| 54| utf32_output); 201| 54| } 202| 9.93k| uint32_t value = (diff << 10) + diff2 + 0x10000; 203| 9.93k| *utf32_output++ = char32_t(value); 204| 9.93k| } 205| 49.7k| } 206| 3.89k| buf += k; 207| 3.89k| } 208| 213k| } // while 209| 246| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 210| 300|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_126avx2_convert_utf16_to_utf8ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 802|avx2_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| 802| const char16_t *end = buf + len; 58| 802| const __m256i v_0000 = _mm256_setzero_si256(); 59| 802| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 802| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| 802| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 62| 802| const size_t safety_margin = 63| 802| 12; // to avoid overruns, see issue 64| | // https://github.com/simdutf/simdutf/issues/92 65| | 66| 1.09M| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (66:10): [True: 1.09M, False: 702] ------------------ 67| 1.09M| __m256i in = _mm256_loadu_si256((__m256i *)buf); 68| 1.09M| if (big_endian) { ------------------ | Branch (68:9): [Folded, False: 1.09M] ------------------ 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| 1.09M| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 76| 1.09M| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 987k, False: 108k] ------------------ 77| | // 1. pack the bytes 78| 987k| const __m128i utf8_packed = _mm_packus_epi16( 79| 987k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 80| | // 2. store (16 bytes) 81| 987k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 82| | // 3. adjust pointers 83| 987k| buf += 16; 84| 987k| utf8_output += 16; 85| 987k| continue; // we are done for this round! 86| 987k| } 87| | // no bits set above 7th bit 88| 108k| const __m256i one_byte_bytemask = 89| 108k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 90| 108k| const uint32_t one_byte_bitmask = 91| 108k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 92| | 93| | // no bits set above 11th bit 94| 108k| const __m256i one_or_two_bytes_bytemask = 95| 108k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 96| 108k| const uint32_t one_or_two_bytes_bitmask = 97| 108k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 98| 108k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (98:9): [True: 31.2k, False: 77.0k] ------------------ 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| 31.2k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 104| 31.2k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 105| | 106| | // t0 = [000a|aaaa|bbbb|bb00] 107| 31.2k| const __m256i t0 = _mm256_slli_epi16(in, 2); 108| | // t1 = [000a|aaaa|0000|0000] 109| 31.2k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 110| | // t2 = [0000|0000|00bb|bbbb] 111| 31.2k| const __m256i t2 = _mm256_and_si256(in, v_003f); 112| | // t3 = [000a|aaaa|00bb|bbbb] 113| 31.2k| const __m256i t3 = _mm256_or_si256(t1, t2); 114| | // t4 = [110a|aaaa|10bb|bbbb] 115| 31.2k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 116| | 117| | // 2. merge ASCII and 2-byte codewords 118| 31.2k| const __m256i utf8_unpacked = 119| 31.2k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 120| | 121| | // 3. prepare bitmask for 8-bit lookup 122| 31.2k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 123| 31.2k| const uint32_t M1 = M0 >> 7; 124| 31.2k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 125| | // 4. pack the bytes 126| | 127| 31.2k| const uint8_t *row = 128| 31.2k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 129| 31.2k| const uint8_t *row_2 = 130| 31.2k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 131| 31.2k| 16)][0]; 132| | 133| 31.2k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 134| 31.2k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 135| | 136| 31.2k| const __m256i utf8_packed = _mm256_shuffle_epi8( 137| 31.2k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 31.2k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 31.2k| _mm256_castsi128_si256(xmm2), 2) ------------------ 138| | // 5. store bytes 139| 31.2k| _mm_storeu_si128((__m128i *)utf8_output, 140| 31.2k| _mm256_castsi256_si128(utf8_packed)); 141| 31.2k| utf8_output += row[0]; 142| 31.2k| _mm_storeu_si128((__m128i *)utf8_output, 143| 31.2k| _mm256_extractf128_si256(utf8_packed, 1)); 144| 31.2k| utf8_output += row_2[0]; 145| | 146| | // 6. adjust pointers 147| 31.2k| buf += 16; 148| 31.2k| continue; 149| 31.2k| } 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| 77.0k| const __m256i surrogates_bytemask = 154| 77.0k| _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| 77.0k| const uint32_t surrogates_bitmask = 159| 77.0k| 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| 77.0k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (162:9): [True: 69.7k, False: 7.28k] ------------------ 163| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 164| 69.7k| const __m256i dup_even = _mm256_setr_epi16( 165| 69.7k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 166| 69.7k| 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| 69.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 196| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 197| 69.7k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 198| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 199| 69.7k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 195| 69.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 200| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 201| 69.7k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 195| 69.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 202| | 203| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 204| 69.7k| const __m256i s0 = _mm256_srli_epi16(in, 4); 205| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 206| 69.7k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 195| 69.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 207| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 208| 69.7k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 195| 69.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 209| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 210| 69.7k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 195| 69.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 211| 69.7k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 212| 69.7k| simdutf_vec(0b0100000000000000)); ------------------ | | 195| 69.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 213| 69.7k| const __m256i s4 = _mm256_xor_si256(s3, m0); 214| 69.7k|#undef simdutf_vec 215| | 216| | // 4. expand code units 16-bit => 32-bit 217| 69.7k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 218| 69.7k| 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| 69.7k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 222| 69.7k| (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| 69.7k| const uint8_t mask0 = uint8_t(mask); 243| 69.7k| const uint8_t *row0 = 244| 69.7k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 245| 69.7k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 246| 69.7k| const __m128i utf8_0 = 247| 69.7k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 248| | 249| 69.7k| const uint8_t mask1 = static_cast(mask >> 8); 250| 69.7k| const uint8_t *row1 = 251| 69.7k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 252| 69.7k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 253| 69.7k| const __m128i utf8_1 = 254| 69.7k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 255| | 256| 69.7k| const uint8_t mask2 = static_cast(mask >> 16); 257| 69.7k| const uint8_t *row2 = 258| 69.7k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 259| 69.7k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 260| 69.7k| const __m128i utf8_2 = 261| 69.7k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 262| | 263| 69.7k| const uint8_t mask3 = static_cast(mask >> 24); 264| 69.7k| const uint8_t *row3 = 265| 69.7k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 266| 69.7k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 267| 69.7k| const __m128i utf8_3 = 268| 69.7k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 269| | 270| 69.7k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 271| 69.7k| utf8_output += row0[0]; 272| 69.7k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 273| 69.7k| utf8_output += row1[0]; 274| 69.7k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 275| 69.7k| utf8_output += row2[0]; 276| 69.7k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 277| 69.7k| utf8_output += row3[0]; 278| 69.7k| buf += 16; 279| | // surrogate pair(s) in a register 280| 69.7k| } else { 281| | // Let us do a scalar fallback. 282| | // It may seem wasteful to use scalar code, but being efficient with SIMD 283| | // in the presence of surrogate pairs may require non-trivial tables. 284| 7.28k| size_t forward = 15; 285| 7.28k| size_t k = 0; 286| 7.28k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (286:11): [True: 0, False: 7.28k] ------------------ 287| 0| forward = size_t(end - buf - 1); 288| 0| } 289| 107k| for (; k < forward; k++) { ------------------ | Branch (289:14): [True: 100k, False: 7.18k] ------------------ 290| 100k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 291| 100k| if ((word & 0xFF80) == 0) { ------------------ | Branch (291:13): [True: 14.1k, False: 86.5k] ------------------ 292| 14.1k| *utf8_output++ = char(word); 293| 86.5k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (293:20): [True: 4.95k, False: 81.5k] ------------------ 294| 4.95k| *utf8_output++ = char((word >> 6) | 0b11000000); 295| 4.95k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 296| 81.5k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (296:20): [True: 73.4k, False: 8.17k] ------------------ 297| 73.4k| *utf8_output++ = char((word >> 12) | 0b11100000); 298| 73.4k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 299| 73.4k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 300| 73.4k| } else { 301| | // must be a surrogate pair 302| 8.17k| uint16_t diff = uint16_t(word - 0xD800); 303| 8.17k| uint16_t next_word = 304| 8.17k| scalar::utf16::swap_if_needed(buf[k + 1]); 305| 8.17k| k++; 306| 8.17k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 307| 8.17k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (307:15): [True: 100, False: 8.07k] ------------------ 308| 100| return std::make_pair(nullptr, utf8_output); 309| 100| } 310| 8.07k| uint32_t value = (diff << 10) + diff2 + 0x10000; 311| 8.07k| *utf8_output++ = char((value >> 18) | 0b11110000); 312| 8.07k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 313| 8.07k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 314| 8.07k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 315| 8.07k| } 316| 100k| } 317| 7.18k| buf += k; 318| 7.18k| } 319| 77.0k| } // while 320| 702| return std::make_pair(buf, utf8_output); 321| 802|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_126avx2_convert_utf16_to_utf8ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 768|avx2_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| 768| const char16_t *end = buf + len; 58| 768| const __m256i v_0000 = _mm256_setzero_si256(); 59| 768| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 768| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| 768| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 62| 768| const size_t safety_margin = 63| 768| 12; // to avoid overruns, see issue 64| | // https://github.com/simdutf/simdutf/issues/92 65| | 66| 1.10M| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (66:10): [True: 1.10M, False: 682] ------------------ 67| 1.10M| __m256i in = _mm256_loadu_si256((__m256i *)buf); 68| 1.10M| if (big_endian) { ------------------ | Branch (68:9): [True: 1.10M, Folded] ------------------ 69| 1.10M| const __m256i swap = _mm256_setr_epi8( 70| 1.10M| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 71| 1.10M| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 72| 1.10M| in = _mm256_shuffle_epi8(in, swap); 73| 1.10M| } 74| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 75| 1.10M| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 76| 1.10M| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 799k, False: 303k] ------------------ 77| | // 1. pack the bytes 78| 799k| const __m128i utf8_packed = _mm_packus_epi16( 79| 799k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 80| | // 2. store (16 bytes) 81| 799k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 82| | // 3. adjust pointers 83| 799k| buf += 16; 84| 799k| utf8_output += 16; 85| 799k| continue; // we are done for this round! 86| 799k| } 87| | // no bits set above 7th bit 88| 303k| const __m256i one_byte_bytemask = 89| 303k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 90| 303k| const uint32_t one_byte_bitmask = 91| 303k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 92| | 93| | // no bits set above 11th bit 94| 303k| const __m256i one_or_two_bytes_bytemask = 95| 303k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 96| 303k| const uint32_t one_or_two_bytes_bitmask = 97| 303k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 98| 303k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (98:9): [True: 14.8k, False: 288k] ------------------ 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| 14.8k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 104| 14.8k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 105| | 106| | // t0 = [000a|aaaa|bbbb|bb00] 107| 14.8k| const __m256i t0 = _mm256_slli_epi16(in, 2); 108| | // t1 = [000a|aaaa|0000|0000] 109| 14.8k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 110| | // t2 = [0000|0000|00bb|bbbb] 111| 14.8k| const __m256i t2 = _mm256_and_si256(in, v_003f); 112| | // t3 = [000a|aaaa|00bb|bbbb] 113| 14.8k| const __m256i t3 = _mm256_or_si256(t1, t2); 114| | // t4 = [110a|aaaa|10bb|bbbb] 115| 14.8k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 116| | 117| | // 2. merge ASCII and 2-byte codewords 118| 14.8k| const __m256i utf8_unpacked = 119| 14.8k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 120| | 121| | // 3. prepare bitmask for 8-bit lookup 122| 14.8k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 123| 14.8k| const uint32_t M1 = M0 >> 7; 124| 14.8k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 125| | // 4. pack the bytes 126| | 127| 14.8k| const uint8_t *row = 128| 14.8k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 129| 14.8k| const uint8_t *row_2 = 130| 14.8k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 131| 14.8k| 16)][0]; 132| | 133| 14.8k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 134| 14.8k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 135| | 136| 14.8k| const __m256i utf8_packed = _mm256_shuffle_epi8( 137| 14.8k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 14.8k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 14.8k| _mm256_castsi128_si256(xmm2), 2) ------------------ 138| | // 5. store bytes 139| 14.8k| _mm_storeu_si128((__m128i *)utf8_output, 140| 14.8k| _mm256_castsi256_si128(utf8_packed)); 141| 14.8k| utf8_output += row[0]; 142| 14.8k| _mm_storeu_si128((__m128i *)utf8_output, 143| 14.8k| _mm256_extractf128_si256(utf8_packed, 1)); 144| 14.8k| utf8_output += row_2[0]; 145| | 146| | // 6. adjust pointers 147| 14.8k| buf += 16; 148| 14.8k| continue; 149| 14.8k| } 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| 288k| const __m256i surrogates_bytemask = 154| 288k| _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| 288k| const uint32_t surrogates_bitmask = 159| 288k| 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| 288k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (162:9): [True: 281k, False: 6.99k] ------------------ 163| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 164| 281k| const __m256i dup_even = _mm256_setr_epi16( 165| 281k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 166| 281k| 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| 281k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 196| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 197| 281k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 198| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 199| 281k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 195| 281k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 200| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 201| 281k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 195| 281k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 202| | 203| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 204| 281k| const __m256i s0 = _mm256_srli_epi16(in, 4); 205| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 206| 281k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 195| 281k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 207| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 208| 281k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 195| 281k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 209| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 210| 281k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 195| 281k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 211| 281k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 212| 281k| simdutf_vec(0b0100000000000000)); ------------------ | | 195| 281k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 213| 281k| const __m256i s4 = _mm256_xor_si256(s3, m0); 214| 281k|#undef simdutf_vec 215| | 216| | // 4. expand code units 16-bit => 32-bit 217| 281k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 218| 281k| 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| 281k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 222| 281k| (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| 281k| const uint8_t mask0 = uint8_t(mask); 243| 281k| const uint8_t *row0 = 244| 281k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 245| 281k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 246| 281k| const __m128i utf8_0 = 247| 281k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 248| | 249| 281k| const uint8_t mask1 = static_cast(mask >> 8); 250| 281k| const uint8_t *row1 = 251| 281k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 252| 281k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 253| 281k| const __m128i utf8_1 = 254| 281k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 255| | 256| 281k| const uint8_t mask2 = static_cast(mask >> 16); 257| 281k| const uint8_t *row2 = 258| 281k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 259| 281k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 260| 281k| const __m128i utf8_2 = 261| 281k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 262| | 263| 281k| const uint8_t mask3 = static_cast(mask >> 24); 264| 281k| const uint8_t *row3 = 265| 281k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 266| 281k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 267| 281k| const __m128i utf8_3 = 268| 281k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 269| | 270| 281k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 271| 281k| utf8_output += row0[0]; 272| 281k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 273| 281k| utf8_output += row1[0]; 274| 281k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 275| 281k| utf8_output += row2[0]; 276| 281k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 277| 281k| utf8_output += row3[0]; 278| 281k| buf += 16; 279| | // surrogate pair(s) in a register 280| 281k| } 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| 6.99k| size_t forward = 15; 285| 6.99k| size_t k = 0; 286| 6.99k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (286:11): [True: 0, False: 6.99k] ------------------ 287| 0| forward = size_t(end - buf - 1); 288| 0| } 289| 103k| for (; k < forward; k++) { ------------------ | Branch (289:14): [True: 96.3k, False: 6.91k] ------------------ 290| 96.3k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 291| 96.3k| if ((word & 0xFF80) == 0) { ------------------ | Branch (291:13): [True: 11.8k, False: 84.5k] ------------------ 292| 11.8k| *utf8_output++ = char(word); 293| 84.5k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (293:20): [True: 2.75k, False: 81.7k] ------------------ 294| 2.75k| *utf8_output++ = char((word >> 6) | 0b11000000); 295| 2.75k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 296| 81.7k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (296:20): [True: 73.5k, False: 8.17k] ------------------ 297| 73.5k| *utf8_output++ = char((word >> 12) | 0b11100000); 298| 73.5k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 299| 73.5k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 300| 73.5k| } else { 301| | // must be a surrogate pair 302| 8.17k| uint16_t diff = uint16_t(word - 0xD800); 303| 8.17k| uint16_t next_word = 304| 8.17k| scalar::utf16::swap_if_needed(buf[k + 1]); 305| 8.17k| k++; 306| 8.17k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 307| 8.17k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (307:15): [True: 86, False: 8.08k] ------------------ 308| 86| return std::make_pair(nullptr, utf8_output); 309| 86| } 310| 8.08k| uint32_t value = (diff << 10) + diff2 + 0x10000; 311| 8.08k| *utf8_output++ = char((value >> 18) | 0b11110000); 312| 8.08k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 313| 8.08k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 314| 8.08k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 315| 8.08k| } 316| 96.3k| } 317| 6.91k| buf += k; 318| 6.91k| } 319| 288k| } // while 320| 682| return std::make_pair(buf, utf8_output); 321| 768|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_138avx2_convert_utf16_to_utf8_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 333| 630| char *utf8_output) { 334| 630| const char16_t *start = buf; 335| 630| const char16_t *end = buf + len; 336| | 337| 630| const __m256i v_0000 = _mm256_setzero_si256(); 338| 630| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 339| 630| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 340| 630| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 341| 630| const size_t safety_margin = 342| 630| 12; // to avoid overruns, see issue 343| | // https://github.com/simdutf/simdutf/issues/92 344| | 345| 776k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (345:10): [True: 776k, False: 502] ------------------ 346| 776k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 347| 776k| if (big_endian) { ------------------ | Branch (347:9): [Folded, False: 776k] ------------------ 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| 776k| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 355| 776k| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (355:9): [True: 459k, False: 316k] ------------------ 356| | // 1. pack the bytes 357| 459k| const __m128i utf8_packed = _mm_packus_epi16( 358| 459k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 359| | // 2. store (16 bytes) 360| 459k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 361| | // 3. adjust pointers 362| 459k| buf += 16; 363| 459k| utf8_output += 16; 364| 459k| continue; // we are done for this round! 365| 459k| } 366| | // no bits set above 7th bit 367| 316k| const __m256i one_byte_bytemask = 368| 316k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 369| 316k| const uint32_t one_byte_bitmask = 370| 316k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 371| | 372| | // no bits set above 11th bit 373| 316k| const __m256i one_or_two_bytes_bytemask = 374| 316k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 375| 316k| const uint32_t one_or_two_bytes_bitmask = 376| 316k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 377| 316k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (377:9): [True: 95.0k, False: 221k] ------------------ 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| 95.0k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 383| 95.0k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 384| | 385| | // t0 = [000a|aaaa|bbbb|bb00] 386| 95.0k| const __m256i t0 = _mm256_slli_epi16(in, 2); 387| | // t1 = [000a|aaaa|0000|0000] 388| 95.0k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 389| | // t2 = [0000|0000|00bb|bbbb] 390| 95.0k| const __m256i t2 = _mm256_and_si256(in, v_003f); 391| | // t3 = [000a|aaaa|00bb|bbbb] 392| 95.0k| const __m256i t3 = _mm256_or_si256(t1, t2); 393| | // t4 = [110a|aaaa|10bb|bbbb] 394| 95.0k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 395| | 396| | // 2. merge ASCII and 2-byte codewords 397| 95.0k| const __m256i utf8_unpacked = 398| 95.0k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 399| | 400| | // 3. prepare bitmask for 8-bit lookup 401| 95.0k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 402| 95.0k| const uint32_t M1 = M0 >> 7; 403| 95.0k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 404| | // 4. pack the bytes 405| | 406| 95.0k| const uint8_t *row = 407| 95.0k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 408| 95.0k| const uint8_t *row_2 = 409| 95.0k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 410| 95.0k| 16)][0]; 411| | 412| 95.0k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 413| 95.0k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 414| | 415| 95.0k| const __m256i utf8_packed = _mm256_shuffle_epi8( 416| 95.0k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 95.0k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 95.0k| _mm256_castsi128_si256(xmm2), 2) ------------------ 417| | // 5. store bytes 418| 95.0k| _mm_storeu_si128((__m128i *)utf8_output, 419| 95.0k| _mm256_castsi256_si128(utf8_packed)); 420| 95.0k| utf8_output += row[0]; 421| 95.0k| _mm_storeu_si128((__m128i *)utf8_output, 422| 95.0k| _mm256_extractf128_si256(utf8_packed, 1)); 423| 95.0k| utf8_output += row_2[0]; 424| | 425| | // 6. adjust pointers 426| 95.0k| buf += 16; 427| 95.0k| continue; 428| 95.0k| } 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| 221k| const __m256i surrogates_bytemask = 433| 221k| _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| 221k| const uint32_t surrogates_bitmask = 438| 221k| 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| 221k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (441:9): [True: 217k, False: 3.69k] ------------------ 442| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 443| 217k| const __m256i dup_even = _mm256_setr_epi16( 444| 217k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 445| 217k| 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| 217k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 475| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 476| 217k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 477| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 478| 217k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 474| 217k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 479| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 480| 217k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 474| 217k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 481| | 482| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 483| 217k| const __m256i s0 = _mm256_srli_epi16(in, 4); 484| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 485| 217k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 474| 217k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 486| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 487| 217k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 474| 217k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 488| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 489| 217k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 474| 217k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 490| 217k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 491| 217k| simdutf_vec(0b0100000000000000)); ------------------ | | 474| 217k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 492| 217k| const __m256i s4 = _mm256_xor_si256(s3, m0); 493| 217k|#undef simdutf_vec 494| | 495| | // 4. expand code units 16-bit => 32-bit 496| 217k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 497| 217k| 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| 217k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 501| 217k| (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| 217k| const uint8_t mask0 = uint8_t(mask); 522| 217k| const uint8_t *row0 = 523| 217k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 524| 217k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 525| 217k| const __m128i utf8_0 = 526| 217k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 527| | 528| 217k| const uint8_t mask1 = static_cast(mask >> 8); 529| 217k| const uint8_t *row1 = 530| 217k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 531| 217k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 532| 217k| const __m128i utf8_1 = 533| 217k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 534| | 535| 217k| const uint8_t mask2 = static_cast(mask >> 16); 536| 217k| const uint8_t *row2 = 537| 217k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 538| 217k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 539| 217k| const __m128i utf8_2 = 540| 217k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 541| | 542| 217k| const uint8_t mask3 = static_cast(mask >> 24); 543| 217k| const uint8_t *row3 = 544| 217k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 545| 217k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 546| 217k| const __m128i utf8_3 = 547| 217k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 548| | 549| 217k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 550| 217k| utf8_output += row0[0]; 551| 217k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 552| 217k| utf8_output += row1[0]; 553| 217k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 554| 217k| utf8_output += row2[0]; 555| 217k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 556| 217k| utf8_output += row3[0]; 557| 217k| buf += 16; 558| | // surrogate pair(s) in a register 559| 217k| } 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| 3.69k| size_t forward = 15; 564| 3.69k| size_t k = 0; 565| 3.69k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (565:11): [True: 0, False: 3.69k] ------------------ 566| 0| forward = size_t(end - buf - 1); 567| 0| } 568| 53.8k| for (; k < forward; k++) { ------------------ | Branch (568:14): [True: 50.3k, False: 3.56k] ------------------ 569| 50.3k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 570| 50.3k| if ((word & 0xFF80) == 0) { ------------------ | Branch (570:13): [True: 9.40k, False: 40.9k] ------------------ 571| 9.40k| *utf8_output++ = char(word); 572| 40.9k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (572:20): [True: 2.12k, False: 38.7k] ------------------ 573| 2.12k| *utf8_output++ = char((word >> 6) | 0b11000000); 574| 2.12k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 575| 38.7k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (575:20): [True: 34.5k, False: 4.21k] ------------------ 576| 34.5k| *utf8_output++ = char((word >> 12) | 0b11100000); 577| 34.5k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 578| 34.5k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 579| 34.5k| } else { 580| | // must be a surrogate pair 581| 4.21k| uint16_t diff = uint16_t(word - 0xD800); 582| 4.21k| uint16_t next_word = 583| 4.21k| scalar::utf16::swap_if_needed(buf[k + 1]); 584| 4.21k| k++; 585| 4.21k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 586| 4.21k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (586:15): [True: 128, False: 4.09k] ------------------ 587| 128| return std::make_pair( 588| 128| result(error_code::SURROGATE, buf - start + k - 1), 589| 128| utf8_output); 590| 128| } 591| 4.09k| uint32_t value = (diff << 10) + diff2 + 0x10000; 592| 4.09k| *utf8_output++ = char((value >> 18) | 0b11110000); 593| 4.09k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 594| 4.09k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 595| 4.09k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 596| 4.09k| } 597| 50.3k| } 598| 3.56k| buf += k; 599| 3.56k| } 600| 221k| } // while 601| 502| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 602| 630|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_138avx2_convert_utf16_to_utf8_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 333| 586| char *utf8_output) { 334| 586| const char16_t *start = buf; 335| 586| const char16_t *end = buf + len; 336| | 337| 586| const __m256i v_0000 = _mm256_setzero_si256(); 338| 586| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 339| 586| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 340| 586| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 341| 586| const size_t safety_margin = 342| 586| 12; // to avoid overruns, see issue 343| | // https://github.com/simdutf/simdutf/issues/92 344| | 345| 467k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (345:10): [True: 467k, False: 422] ------------------ 346| 467k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 347| 467k| if (big_endian) { ------------------ | Branch (347:9): [True: 467k, Folded] ------------------ 348| 467k| const __m256i swap = _mm256_setr_epi8( 349| 467k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 350| 467k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 351| 467k| in = _mm256_shuffle_epi8(in, swap); 352| 467k| } 353| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 354| 467k| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 355| 467k| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (355:9): [True: 369k, False: 97.4k] ------------------ 356| | // 1. pack the bytes 357| 369k| const __m128i utf8_packed = _mm_packus_epi16( 358| 369k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 359| | // 2. store (16 bytes) 360| 369k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 361| | // 3. adjust pointers 362| 369k| buf += 16; 363| 369k| utf8_output += 16; 364| 369k| continue; // we are done for this round! 365| 369k| } 366| | // no bits set above 7th bit 367| 97.4k| const __m256i one_byte_bytemask = 368| 97.4k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 369| 97.4k| const uint32_t one_byte_bitmask = 370| 97.4k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 371| | 372| | // no bits set above 11th bit 373| 97.4k| const __m256i one_or_two_bytes_bytemask = 374| 97.4k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 375| 97.4k| const uint32_t one_or_two_bytes_bitmask = 376| 97.4k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 377| 97.4k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (377:9): [True: 24.3k, False: 73.0k] ------------------ 378| | 379| | // 1. prepare 2-byte values 380| | // input 16-bit word : [0000|0aaa|aabb|bbbb] x 8 381| | // expected output : [110a|aaaa|10bb|bbbb] x 8 382| 24.3k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 383| 24.3k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 384| | 385| | // t0 = [000a|aaaa|bbbb|bb00] 386| 24.3k| const __m256i t0 = _mm256_slli_epi16(in, 2); 387| | // t1 = [000a|aaaa|0000|0000] 388| 24.3k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 389| | // t2 = [0000|0000|00bb|bbbb] 390| 24.3k| const __m256i t2 = _mm256_and_si256(in, v_003f); 391| | // t3 = [000a|aaaa|00bb|bbbb] 392| 24.3k| const __m256i t3 = _mm256_or_si256(t1, t2); 393| | // t4 = [110a|aaaa|10bb|bbbb] 394| 24.3k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 395| | 396| | // 2. merge ASCII and 2-byte codewords 397| 24.3k| const __m256i utf8_unpacked = 398| 24.3k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 399| | 400| | // 3. prepare bitmask for 8-bit lookup 401| 24.3k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 402| 24.3k| const uint32_t M1 = M0 >> 7; 403| 24.3k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 404| | // 4. pack the bytes 405| | 406| 24.3k| const uint8_t *row = 407| 24.3k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 408| 24.3k| const uint8_t *row_2 = 409| 24.3k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 410| 24.3k| 16)][0]; 411| | 412| 24.3k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 413| 24.3k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 414| | 415| 24.3k| const __m256i utf8_packed = _mm256_shuffle_epi8( 416| 24.3k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 24.3k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 24.3k| _mm256_castsi128_si256(xmm2), 2) ------------------ 417| | // 5. store bytes 418| 24.3k| _mm_storeu_si128((__m128i *)utf8_output, 419| 24.3k| _mm256_castsi256_si128(utf8_packed)); 420| 24.3k| utf8_output += row[0]; 421| 24.3k| _mm_storeu_si128((__m128i *)utf8_output, 422| 24.3k| _mm256_extractf128_si256(utf8_packed, 1)); 423| 24.3k| utf8_output += row_2[0]; 424| | 425| | // 6. adjust pointers 426| 24.3k| buf += 16; 427| 24.3k| continue; 428| 24.3k| } 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| 73.0k| const __m256i surrogates_bytemask = 433| 73.0k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_d800); 434| | 435| | // bitmask = 0x0000 if there are no surrogates 436| | // = 0xc000 if the last word is a surrogate 437| 73.0k| const uint32_t surrogates_bitmask = 438| 73.0k| static_cast(_mm256_movemask_epi8(surrogates_bytemask)); 439| | // It might seem like checking for surrogates_bitmask == 0xc000 could help. 440| | // However, it is likely an uncommon occurrence. 441| 73.0k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (441:9): [True: 65.9k, False: 7.10k] ------------------ 442| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 443| 65.9k| const __m256i dup_even = _mm256_setr_epi16( 444| 65.9k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 445| 65.9k| 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| 65.9k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 475| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 476| 65.9k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 477| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 478| 65.9k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 474| 65.9k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 479| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 480| 65.9k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 474| 65.9k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 481| | 482| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 483| 65.9k| const __m256i s0 = _mm256_srli_epi16(in, 4); 484| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 485| 65.9k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 474| 65.9k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 486| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 487| 65.9k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 474| 65.9k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 488| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 489| 65.9k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 474| 65.9k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 490| 65.9k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 491| 65.9k| simdutf_vec(0b0100000000000000)); ------------------ | | 474| 65.9k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 492| 65.9k| const __m256i s4 = _mm256_xor_si256(s3, m0); 493| 65.9k|#undef simdutf_vec 494| | 495| | // 4. expand code units 16-bit => 32-bit 496| 65.9k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 497| 65.9k| 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| 65.9k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 501| 65.9k| (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| 65.9k| const uint8_t mask0 = uint8_t(mask); 522| 65.9k| const uint8_t *row0 = 523| 65.9k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 524| 65.9k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 525| 65.9k| const __m128i utf8_0 = 526| 65.9k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 527| | 528| 65.9k| const uint8_t mask1 = static_cast(mask >> 8); 529| 65.9k| const uint8_t *row1 = 530| 65.9k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 531| 65.9k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 532| 65.9k| const __m128i utf8_1 = 533| 65.9k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 534| | 535| 65.9k| const uint8_t mask2 = static_cast(mask >> 16); 536| 65.9k| const uint8_t *row2 = 537| 65.9k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 538| 65.9k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 539| 65.9k| const __m128i utf8_2 = 540| 65.9k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 541| | 542| 65.9k| const uint8_t mask3 = static_cast(mask >> 24); 543| 65.9k| const uint8_t *row3 = 544| 65.9k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 545| 65.9k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 546| 65.9k| const __m128i utf8_3 = 547| 65.9k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 548| | 549| 65.9k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 550| 65.9k| utf8_output += row0[0]; 551| 65.9k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 552| 65.9k| utf8_output += row1[0]; 553| 65.9k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 554| 65.9k| utf8_output += row2[0]; 555| 65.9k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 556| 65.9k| utf8_output += row3[0]; 557| 65.9k| buf += 16; 558| | // surrogate pair(s) in a register 559| 65.9k| } else { 560| | // Let us do a scalar fallback. 561| | // It may seem wasteful to use scalar code, but being efficient with SIMD 562| | // in the presence of surrogate pairs may require non-trivial tables. 563| 7.10k| size_t forward = 15; 564| 7.10k| size_t k = 0; 565| 7.10k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (565:11): [True: 0, False: 7.10k] ------------------ 566| 0| forward = size_t(end - buf - 1); 567| 0| } 568| 104k| for (; k < forward; k++) { ------------------ | Branch (568:14): [True: 97.8k, False: 6.94k] ------------------ 569| 97.8k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 570| 97.8k| if ((word & 0xFF80) == 0) { ------------------ | Branch (570:13): [True: 33.5k, False: 64.2k] ------------------ 571| 33.5k| *utf8_output++ = char(word); 572| 64.2k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (572:20): [True: 5.21k, False: 59.0k] ------------------ 573| 5.21k| *utf8_output++ = char((word >> 6) | 0b11000000); 574| 5.21k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 575| 59.0k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (575:20): [True: 51.1k, False: 7.86k] ------------------ 576| 51.1k| *utf8_output++ = char((word >> 12) | 0b11100000); 577| 51.1k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 578| 51.1k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 579| 51.1k| } else { 580| | // must be a surrogate pair 581| 7.86k| uint16_t diff = uint16_t(word - 0xD800); 582| 7.86k| uint16_t next_word = 583| 7.86k| scalar::utf16::swap_if_needed(buf[k + 1]); 584| 7.86k| k++; 585| 7.86k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 586| 7.86k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (586:15): [True: 164, False: 7.70k] ------------------ 587| 164| return std::make_pair( 588| 164| result(error_code::SURROGATE, buf - start + k - 1), 589| 164| utf8_output); 590| 164| } 591| 7.70k| uint32_t value = (diff << 10) + diff2 + 0x10000; 592| 7.70k| *utf8_output++ = char((value >> 18) | 0b11110000); 593| 7.70k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 594| 7.70k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 595| 7.70k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 596| 7.70k| } 597| 97.8k| } 598| 6.94k| buf += k; 599| 6.94k| } 600| 73.0k| } // while 601| 422| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 602| 586|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_utf32_to_latin1EPKDimPc: 3| 188| char *latin1_output) { 4| 188| const size_t rounded_len = 5| 188| len & ~0x1F; // Round down to nearest multiple of 32 6| | 7| 188| const __m256i high_bytes_mask = _mm256_set1_epi32(0xFFFFFF00); 8| | 9| 2.93k| for (size_t i = 0; i < rounded_len; i += 4 * 8) { ------------------ | Branch (9:22): [True: 2.77k, False: 160] ------------------ 10| 2.77k| __m256i a = _mm256_loadu_si256((__m256i *)(buf + 0 * 8)); 11| 2.77k| __m256i b = _mm256_loadu_si256((__m256i *)(buf + 1 * 8)); 12| 2.77k| __m256i c = _mm256_loadu_si256((__m256i *)(buf + 2 * 8)); 13| 2.77k| __m256i d = _mm256_loadu_si256((__m256i *)(buf + 3 * 8)); 14| | 15| 2.77k| const __m256i check_combined = 16| 2.77k| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 17| | 18| 2.77k| if (!_mm256_testz_si256(check_combined, high_bytes_mask)) { ------------------ | Branch (18:9): [True: 28, False: 2.75k] ------------------ 19| 28| return std::make_pair(nullptr, latin1_output); 20| 28| } 21| | 22| 2.75k| b = _mm256_slli_epi32(b, 1 * 8); 23| 2.75k| c = _mm256_slli_epi32(c, 2 * 8); 24| 2.75k| 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| 2.75k| const __m256i t0 = 35| 2.75k| _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| 2.75k| const __m256i shuffle_bytes = 40| 2.75k| _mm256_setr_epi8(0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15, 41| 2.75k| 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15); 42| | 43| 2.75k| 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| 2.75k| const __m256i shuffle_dwords = _mm256_setr_epi32(0, 4, 1, 5, 2, 6, 3, 7); 48| 2.75k| const __m256i t2 = _mm256_permutevar8x32_epi32(t1, shuffle_dwords); 49| |// clang format on 50| | 51| 2.75k| _mm256_storeu_si256((__m256i *)latin1_output, t2); 52| | 53| 2.75k| latin1_output += 32; 54| 2.75k| buf += 32; 55| 2.75k| } 56| | 57| 160| return std::make_pair(buf, latin1_output); 58| 188|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_140avx2_convert_utf32_to_latin1_with_errorsEPKDimPc: 62| 420| char *latin1_output) { 63| 420| const size_t rounded_len = 64| 420| len & ~0x1F; // Round down to nearest multiple of 32 65| | 66| 420| const char32_t *start = buf; 67| | 68| 420| const __m256i high_bytes_mask = _mm256_set1_epi32(0xFFFFFF00); 69| | 70| 13.5k| for (size_t i = 0; i < rounded_len; i += 4 * 8) { ------------------ | Branch (70:22): [True: 13.2k, False: 320] ------------------ 71| 13.2k| __m256i a = _mm256_loadu_si256((__m256i *)(buf + 0 * 8)); 72| 13.2k| __m256i b = _mm256_loadu_si256((__m256i *)(buf + 1 * 8)); 73| 13.2k| __m256i c = _mm256_loadu_si256((__m256i *)(buf + 2 * 8)); 74| 13.2k| __m256i d = _mm256_loadu_si256((__m256i *)(buf + 3 * 8)); 75| | 76| 13.2k| const __m256i check_combined = 77| 13.2k| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 78| | 79| 13.2k| if (!_mm256_testz_si256(check_combined, high_bytes_mask)) { ------------------ | Branch (79:9): [True: 100, False: 13.1k] ------------------ 80| | // Fallback to scalar code for handling errors 81| 754| for (int k = 0; k < 4 * 8; k++) { ------------------ | Branch (81:23): [True: 754, False: 0] ------------------ 82| 754| char32_t codepoint = buf[k]; 83| 754| if (codepoint <= 0xFF) { ------------------ | Branch (83:13): [True: 654, False: 100] ------------------ 84| 654| *latin1_output++ = static_cast(codepoint); 85| 654| } else { 86| 100| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 87| 100| latin1_output); 88| 100| } 89| 754| } 90| 100| } 91| | 92| 13.1k| b = _mm256_slli_epi32(b, 1 * 8); 93| 13.1k| c = _mm256_slli_epi32(c, 2 * 8); 94| 13.1k| d = _mm256_slli_epi32(d, 3 * 8); 95| | 96| 13.1k| const __m256i t0 = 97| 13.1k| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 98| | 99| 13.1k| const __m256i shuffle_bytes = 100| 13.1k| _mm256_setr_epi8(0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15, 101| 13.1k| 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15); 102| | 103| 13.1k| const __m256i t1 = _mm256_shuffle_epi8(t0, shuffle_bytes); 104| | 105| 13.1k| const __m256i shuffle_dwords = _mm256_setr_epi32(0, 4, 1, 5, 2, 6, 3, 7); 106| 13.1k| const __m256i t2 = _mm256_permutevar8x32_epi32(t1, shuffle_dwords); 107| | 108| 13.1k| _mm256_storeu_si256((__m256i *)latin1_output, t2); 109| | 110| 13.1k| latin1_output += 32; 111| 13.1k| buf += 32; 112| 13.1k| } 113| | 114| 320| return std::make_pair(result(error_code::SUCCESS, buf - start), 115| 320| latin1_output); 116| 420|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf32_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKDiPDsEES7_mS8_: 4| 722| char16_t *utf16_output) { 5| 722| const char32_t *end = buf + len; 6| | 7| 722| const size_t safety_margin = 8| 722| 12; // to avoid overruns, see issue 9| | // https://github.com/simdutf/simdutf/issues/92 10| 722| __m256i forbidden_bytemask = _mm256_setzero_si256(); 11| | 12| 722| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 13| 722| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 14| 722| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 15| | 16| 10.9k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (16:10): [True: 10.3k, False: 650] ------------------ 17| 10.3k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 18| | 19| 10.3k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 10.3k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 7.99k, False: 2.31k] | | ------------------ ------------------ 20| | // no bits set above 16th bit <=> can pack to UTF16 21| | // without surrogate pairs 22| 7.99k| forbidden_bytemask = _mm256_or_si256( 23| 7.99k| forbidden_bytemask, 24| 7.99k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800)); 25| | 26| 7.99k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 27| 7.99k| _mm256_extractf128_si256(in, 1)); 28| 7.99k| if (big_endian) { ------------------ | Branch (28:11): [Folded, False: 7.99k] ------------------ 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| 7.99k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 34| 7.99k| utf16_output += 8; 35| 7.99k| buf += 8; 36| 7.99k| } else { 37| 2.31k| size_t forward = 7; 38| 2.31k| size_t k = 0; 39| 2.31k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (39:11): [True: 0, False: 2.31k] ------------------ 40| 0| forward = size_t(end - buf - 1); 41| 0| } 42| 18.2k| for (; k < forward; k++) { ------------------ | Branch (42:14): [True: 15.9k, False: 2.24k] ------------------ 43| 15.9k| uint32_t word = buf[k]; 44| 15.9k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (44:13): [True: 7.23k, False: 8.73k] ------------------ 45| | // will not generate a surrogate pair 46| 7.23k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (46:15): [True: 844, False: 6.38k] | Branch (46:33): [True: 2, False: 842] ------------------ 47| 2| return std::make_pair(nullptr, utf16_output); 48| 2| } 49| 7.22k| *utf16_output++ = 50| 7.22k| big_endian ------------------ | Branch (50:15): [Folded, False: 7.22k] ------------------ 51| 7.22k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 52| 7.22k| : char16_t(word); 53| 8.73k| } else { 54| | // will generate a surrogate pair 55| 8.73k| if (word > 0x10FFFF) { ------------------ | Branch (55:15): [True: 70, False: 8.66k] ------------------ 56| 70| return std::make_pair(nullptr, utf16_output); 57| 70| } 58| 8.66k| word -= 0x10000; 59| 8.66k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 60| 8.66k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 61| 8.66k| if (big_endian) { ------------------ | Branch (61:15): [Folded, False: 8.66k] ------------------ 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| 8.66k| *utf16_output++ = char16_t(high_surrogate); 68| 8.66k| *utf16_output++ = char16_t(low_surrogate); 69| 8.66k| } 70| 15.9k| } 71| 2.24k| buf += k; 72| 2.24k| } 73| 10.3k| } 74| | 75| | // check for invalid input 76| 650| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (76:7): [True: 44, False: 606] ------------------ 77| 44| return std::make_pair(nullptr, utf16_output); 78| 44| } 79| | 80| 606| return std::make_pair(buf, utf16_output); 81| 650|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf32_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKDiPDsEES7_mS8_: 4| 528| char16_t *utf16_output) { 5| 528| const char32_t *end = buf + len; 6| | 7| 528| const size_t safety_margin = 8| 528| 12; // to avoid overruns, see issue 9| | // https://github.com/simdutf/simdutf/issues/92 10| 528| __m256i forbidden_bytemask = _mm256_setzero_si256(); 11| | 12| 528| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 13| 528| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 14| 528| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 15| | 16| 7.73k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (16:10): [True: 7.33k, False: 400] ------------------ 17| 7.33k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 18| | 19| 7.33k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 7.33k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 6.50k, False: 828] | | ------------------ ------------------ 20| | // no bits set above 16th bit <=> can pack to UTF16 21| | // without surrogate pairs 22| 6.50k| forbidden_bytemask = _mm256_or_si256( 23| 6.50k| forbidden_bytemask, 24| 6.50k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800)); 25| | 26| 6.50k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 27| 6.50k| _mm256_extractf128_si256(in, 1)); 28| 6.50k| if (big_endian) { ------------------ | Branch (28:11): [True: 6.50k, Folded] ------------------ 29| 6.50k| const __m128i swap = 30| 6.50k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 31| 6.50k| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 32| 6.50k| } 33| 6.50k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 34| 6.50k| utf16_output += 8; 35| 6.50k| buf += 8; 36| 6.50k| } else { 37| 828| size_t forward = 7; 38| 828| size_t k = 0; 39| 828| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (39:11): [True: 0, False: 828] ------------------ 40| 0| forward = size_t(end - buf - 1); 41| 0| } 42| 6.00k| for (; k < forward; k++) { ------------------ | Branch (42:14): [True: 5.30k, False: 700] ------------------ 43| 5.30k| uint32_t word = buf[k]; 44| 5.30k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (44:13): [True: 1.33k, False: 3.96k] ------------------ 45| | // will not generate a surrogate pair 46| 1.33k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (46:15): [True: 262, False: 1.07k] | Branch (46:33): [True: 20, False: 242] ------------------ 47| 20| return std::make_pair(nullptr, utf16_output); 48| 20| } 49| 1.31k| *utf16_output++ = 50| 1.31k| big_endian ------------------ | Branch (50:15): [True: 1.31k, Folded] ------------------ 51| 1.31k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 52| 1.31k| : char16_t(word); 53| 3.96k| } else { 54| | // will generate a surrogate pair 55| 3.96k| if (word > 0x10FFFF) { ------------------ | Branch (55:15): [True: 108, False: 3.85k] ------------------ 56| 108| return std::make_pair(nullptr, utf16_output); 57| 108| } 58| 3.85k| word -= 0x10000; 59| 3.85k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 60| 3.85k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 61| 3.85k| if (big_endian) { ------------------ | Branch (61:15): [True: 3.85k, Folded] ------------------ 62| 3.85k| high_surrogate = 63| 3.85k| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 64| 3.85k| low_surrogate = 65| 3.85k| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 66| 3.85k| } 67| 3.85k| *utf16_output++ = char16_t(high_surrogate); 68| 3.85k| *utf16_output++ = char16_t(low_surrogate); 69| 3.85k| } 70| 5.30k| } 71| 700| buf += k; 72| 700| } 73| 7.33k| } 74| | 75| | // check for invalid input 76| 400| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (76:7): [True: 36, False: 364] ------------------ 77| 36| return std::make_pair(nullptr, utf16_output); 78| 36| } 79| | 80| 364| return std::make_pair(buf, utf16_output); 81| 400|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf32_to_utf16_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 86| 716| char16_t *utf16_output) { 87| 716| const char32_t *start = buf; 88| 716| const char32_t *end = buf + len; 89| | 90| 716| const size_t safety_margin = 91| 716| 12; // to avoid overruns, see issue 92| | // https://github.com/simdutf/simdutf/issues/92 93| | 94| 716| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 95| 716| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 96| 716| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 97| | 98| 88.6k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (98:10): [True: 88.1k, False: 498] ------------------ 99| 88.1k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 100| | 101| 88.1k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 88.1k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 85.3k, False: 2.81k] | | ------------------ ------------------ 102| | // no bits set above 16th bit <=> can pack to UTF16 without surrogate 103| | // pairs 104| 85.3k| const __m256i forbidden_bytemask = 105| 85.3k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800); 106| 85.3k| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != ------------------ | Branch (106:11): [True: 32, False: 85.3k] ------------------ 107| 85.3k| 0x0) { 108| 32| return std::make_pair(result(error_code::SURROGATE, buf - start), 109| 32| utf16_output); 110| 32| } 111| | 112| 85.3k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 113| 85.3k| _mm256_extractf128_si256(in, 1)); 114| 85.3k| if (big_endian) { ------------------ | Branch (114:11): [Folded, False: 85.3k] ------------------ 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| 85.3k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 120| 85.3k| utf16_output += 8; 121| 85.3k| buf += 8; 122| 85.3k| } else { 123| 2.81k| size_t forward = 7; 124| 2.81k| size_t k = 0; 125| 2.81k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (125:11): [True: 0, False: 2.81k] ------------------ 126| 0| forward = size_t(end - buf - 1); 127| 0| } 128| 21.6k| for (; k < forward; k++) { ------------------ | Branch (128:14): [True: 19.0k, False: 2.62k] ------------------ 129| 19.0k| uint32_t word = buf[k]; 130| 19.0k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (130:13): [True: 9.45k, False: 9.55k] ------------------ 131| | // will not generate a surrogate pair 132| 9.45k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (132:15): [True: 820, False: 8.63k] | Branch (132:33): [True: 16, False: 804] ------------------ 133| 16| return std::make_pair( 134| 16| result(error_code::SURROGATE, buf - start + k), utf16_output); 135| 16| } 136| 9.43k| *utf16_output++ = 137| 9.43k| big_endian ------------------ | Branch (137:15): [Folded, False: 9.43k] ------------------ 138| 9.43k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 139| 9.43k| : char16_t(word); 140| 9.55k| } else { 141| | // will generate a surrogate pair 142| 9.55k| if (word > 0x10FFFF) { ------------------ | Branch (142:15): [True: 170, False: 9.38k] ------------------ 143| 170| return std::make_pair( 144| 170| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 145| 170| } 146| 9.38k| word -= 0x10000; 147| 9.38k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 148| 9.38k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 149| 9.38k| if (big_endian) { ------------------ | Branch (149:15): [Folded, False: 9.38k] ------------------ 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| 9.38k| *utf16_output++ = char16_t(high_surrogate); 156| 9.38k| *utf16_output++ = char16_t(low_surrogate); 157| 9.38k| } 158| 19.0k| } 159| 2.62k| buf += k; 160| 2.62k| } 161| 88.1k| } 162| | 163| 498| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 164| 716|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf32_to_utf16_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 86| 606| char16_t *utf16_output) { 87| 606| const char32_t *start = buf; 88| 606| const char32_t *end = buf + len; 89| | 90| 606| const size_t safety_margin = 91| 606| 12; // to avoid overruns, see issue 92| | // https://github.com/simdutf/simdutf/issues/92 93| | 94| 606| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 95| 606| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 96| 606| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 97| | 98| 12.9k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (98:10): [True: 12.4k, False: 458] ------------------ 99| 12.4k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 100| | 101| 12.4k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 12.4k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 11.4k, False: 1.04k] | | ------------------ ------------------ 102| | // no bits set above 16th bit <=> can pack to UTF16 without surrogate 103| | // pairs 104| 11.4k| const __m256i forbidden_bytemask = 105| 11.4k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800); 106| 11.4k| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != ------------------ | Branch (106:11): [True: 8, False: 11.4k] ------------------ 107| 11.4k| 0x0) { 108| 8| return std::make_pair(result(error_code::SURROGATE, buf - start), 109| 8| utf16_output); 110| 8| } 111| | 112| 11.4k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 113| 11.4k| _mm256_extractf128_si256(in, 1)); 114| 11.4k| if (big_endian) { ------------------ | Branch (114:11): [True: 11.4k, Folded] ------------------ 115| 11.4k| const __m128i swap = 116| 11.4k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 117| 11.4k| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 118| 11.4k| } 119| 11.4k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 120| 11.4k| utf16_output += 8; 121| 11.4k| buf += 8; 122| 11.4k| } else { 123| 1.04k| size_t forward = 7; 124| 1.04k| size_t k = 0; 125| 1.04k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (125:11): [True: 0, False: 1.04k] ------------------ 126| 0| forward = size_t(end - buf - 1); 127| 0| } 128| 7.66k| for (; k < forward; k++) { ------------------ | Branch (128:14): [True: 6.76k, False: 900] ------------------ 129| 6.76k| uint32_t word = buf[k]; 130| 6.76k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (130:13): [True: 3.82k, False: 2.94k] ------------------ 131| | // will not generate a surrogate pair 132| 3.82k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (132:15): [True: 786, False: 3.03k] | Branch (132:33): [True: 6, False: 780] ------------------ 133| 6| return std::make_pair( 134| 6| result(error_code::SURROGATE, buf - start + k), utf16_output); 135| 6| } 136| 3.81k| *utf16_output++ = 137| 3.81k| big_endian ------------------ | Branch (137:15): [True: 3.81k, Folded] ------------------ 138| 3.81k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 139| 3.81k| : char16_t(word); 140| 3.81k| } else { 141| | // will generate a surrogate pair 142| 2.94k| if (word > 0x10FFFF) { ------------------ | Branch (142:15): [True: 134, False: 2.80k] ------------------ 143| 134| return std::make_pair( 144| 134| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 145| 134| } 146| 2.80k| word -= 0x10000; 147| 2.80k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 148| 2.80k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 149| 2.80k| if (big_endian) { ------------------ | Branch (149:15): [True: 2.80k, Folded] ------------------ 150| 2.80k| high_surrogate = 151| 2.80k| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 152| 2.80k| low_surrogate = 153| 2.80k| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 154| 2.80k| } 155| 2.80k| *utf16_output++ = char16_t(high_surrogate); 156| 2.80k| *utf16_output++ = char16_t(low_surrogate); 157| 2.80k| } 158| 6.76k| } 159| 900| buf += k; 160| 900| } 161| 12.4k| } 162| | 163| 458| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 164| 606|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_126avx2_convert_utf32_to_utf8EPKDimPc: 2| 796|avx2_convert_utf32_to_utf8(const char32_t *buf, size_t len, char *utf8_output) { 3| 796| const char32_t *end = buf + len; 4| 796| const __m256i v_0000 = _mm256_setzero_si256(); 5| 796| const __m256i v_ffff0000 = _mm256_set1_epi32((uint32_t)0xffff0000); 6| 796| const __m256i v_ff80 = _mm256_set1_epi16((uint16_t)0xff80); 7| 796| const __m256i v_f800 = _mm256_set1_epi16((uint16_t)0xf800); 8| 796| const __m256i v_c080 = _mm256_set1_epi16((uint16_t)0xc080); 9| 796| const __m256i v_7fffffff = _mm256_set1_epi32((uint32_t)0x7fffffff); 10| 796| __m256i running_max = _mm256_setzero_si256(); 11| 796| __m256i forbidden_bytemask = _mm256_setzero_si256(); 12| | 13| 796| const size_t safety_margin = 14| 796| 12; // to avoid overruns, see issue 15| | // https://github.com/simdutf/simdutf/issues/92 16| | 17| 4.96k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (17:10): [True: 4.33k, False: 630] ------------------ 18| 4.33k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 19| 4.33k| __m256i nextin = _mm256_loadu_si256((__m256i *)buf + 1); 20| 4.33k| 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| 4.33k| __m256i in_16 = _mm256_packus_epi32(_mm256_and_si256(in, v_7fffffff), 25| 4.33k| _mm256_and_si256(nextin, v_7fffffff)); 26| 4.33k| 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| 4.33k| if (_mm256_testz_si256(in_16, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (31:9): [True: 822, False: 3.51k] ------------------ 32| | // 1. pack the bytes 33| 822| const __m128i utf8_packed = _mm_packus_epi16( 34| 822| _mm256_castsi256_si128(in_16), _mm256_extractf128_si256(in_16, 1)); 35| | // 2. store (16 bytes) 36| 822| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 37| | // 3. adjust pointers 38| 822| buf += 16; 39| 822| utf8_output += 16; 40| 822| continue; // we are done for this round! 41| 822| } 42| | // no bits set above 7th bit 43| 3.51k| const __m256i one_byte_bytemask = 44| 3.51k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_ff80), v_0000); 45| 3.51k| const uint32_t one_byte_bitmask = 46| 3.51k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 47| | 48| | // no bits set above 11th bit 49| 3.51k| const __m256i one_or_two_bytes_bytemask = 50| 3.51k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_0000); 51| 3.51k| const uint32_t one_or_two_bytes_bitmask = 52| 3.51k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 53| 3.51k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (53:9): [True: 1.46k, False: 2.05k] ------------------ 54| | // 1. prepare 2-byte values 55| | // input 16-bit word : [0000|0aaa|aabb|bbbb] x 8 56| | // expected output : [110a|aaaa|10bb|bbbb] x 8 57| 1.46k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 58| 1.46k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 59| | 60| | // t0 = [000a|aaaa|bbbb|bb00] 61| 1.46k| const __m256i t0 = _mm256_slli_epi16(in_16, 2); 62| | // t1 = [000a|aaaa|0000|0000] 63| 1.46k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 64| | // t2 = [0000|0000|00bb|bbbb] 65| 1.46k| const __m256i t2 = _mm256_and_si256(in_16, v_003f); 66| | // t3 = [000a|aaaa|00bb|bbbb] 67| 1.46k| const __m256i t3 = _mm256_or_si256(t1, t2); 68| | // t4 = [110a|aaaa|10bb|bbbb] 69| 1.46k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 70| | 71| | // 2. merge ASCII and 2-byte codewords 72| 1.46k| const __m256i utf8_unpacked = 73| 1.46k| _mm256_blendv_epi8(t4, in_16, one_byte_bytemask); 74| | 75| | // 3. prepare bitmask for 8-bit lookup 76| 1.46k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 77| 1.46k| const uint32_t M1 = M0 >> 7; 78| 1.46k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 79| | // 4. pack the bytes 80| | 81| 1.46k| const uint8_t *row = 82| 1.46k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 83| 1.46k| const uint8_t *row_2 = 84| 1.46k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 85| 1.46k| 16)][0]; 86| | 87| 1.46k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 88| 1.46k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 89| | 90| 1.46k| const __m256i utf8_packed = _mm256_shuffle_epi8( 91| 1.46k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 1.46k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 1.46k| _mm256_castsi128_si256(xmm2), 2) ------------------ 92| | // 5. store bytes 93| 1.46k| _mm_storeu_si128((__m128i *)utf8_output, 94| 1.46k| _mm256_castsi256_si128(utf8_packed)); 95| 1.46k| utf8_output += row[0]; 96| 1.46k| _mm_storeu_si128((__m128i *)utf8_output, 97| 1.46k| _mm256_extractf128_si256(utf8_packed, 1)); 98| 1.46k| utf8_output += row_2[0]; 99| | 100| | // 6. adjust pointers 101| 1.46k| buf += 16; 102| 1.46k| continue; 103| 1.46k| } 104| | // Must check for overflow in packing 105| 2.05k| const __m256i saturation_bytemask = _mm256_cmpeq_epi32( 106| 2.05k| _mm256_and_si256(_mm256_or_si256(in, nextin), v_ffff0000), v_0000); 107| 2.05k| const uint32_t saturation_bitmask = 108| 2.05k| static_cast(_mm256_movemask_epi8(saturation_bytemask)); 109| 2.05k| if (saturation_bitmask == 0xffffffff) { ------------------ | Branch (109:9): [True: 786, False: 1.26k] ------------------ 110| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 111| 786| const __m256i v_d800 = _mm256_set1_epi16((uint16_t)0xd800); 112| 786| forbidden_bytemask = _mm256_or_si256( 113| 786| forbidden_bytemask, 114| 786| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_d800)); 115| | 116| 786| const __m256i dup_even = _mm256_setr_epi16( 117| 786| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 118| 786| 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| 786|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 148| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 149| 786| const __m256i t0 = _mm256_shuffle_epi8(in_16, dup_even); 150| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 151| 786| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 147| 786|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 152| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 153| 786| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 147| 786|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 154| | 155| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 156| 786| const __m256i s0 = _mm256_srli_epi16(in_16, 4); 157| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 158| 786| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 147| 786|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 159| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 160| 786| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 147| 786|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 161| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 162| 786| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 147| 786|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 163| 786| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 164| 786| simdutf_vec(0b0100000000000000)); ------------------ | | 147| 786|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 165| 786| const __m256i s4 = _mm256_xor_si256(s3, m0); 166| 786|#undef simdutf_vec 167| | 168| | // 4. expand code units 16-bit => 32-bit 169| 786| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 170| 786| 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| 786| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 174| 786| (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| 786| const uint8_t mask0 = uint8_t(mask); 195| 786| const uint8_t *row0 = 196| 786| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 197| 786| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 198| 786| const __m128i utf8_0 = 199| 786| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 200| | 201| 786| const uint8_t mask1 = static_cast(mask >> 8); 202| 786| const uint8_t *row1 = 203| 786| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 204| 786| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 205| 786| const __m128i utf8_1 = 206| 786| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 207| | 208| 786| const uint8_t mask2 = static_cast(mask >> 16); 209| 786| const uint8_t *row2 = 210| 786| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 211| 786| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 212| 786| const __m128i utf8_2 = 213| 786| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 214| | 215| 786| const uint8_t mask3 = static_cast(mask >> 24); 216| 786| const uint8_t *row3 = 217| 786| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 218| 786| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 219| 786| const __m128i utf8_3 = 220| 786| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 221| | 222| 786| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 223| 786| utf8_output += row0[0]; 224| 786| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 225| 786| utf8_output += row1[0]; 226| 786| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 227| 786| utf8_output += row2[0]; 228| 786| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 229| 786| utf8_output += row3[0]; 230| 786| buf += 16; 231| 1.26k| } else { 232| | // case: at least one 32-bit word is larger than 0xFFFF <=> it will 233| | // produce four UTF-8 bytes. Let us do a scalar fallback. It may seem 234| | // wasteful to use scalar code, but being efficient with SIMD may require 235| | // large, non-trivial tables? 236| 1.26k| size_t forward = 15; 237| 1.26k| size_t k = 0; 238| 1.26k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (238:11): [True: 0, False: 1.26k] ------------------ 239| 0| forward = size_t(end - buf - 1); 240| 0| } 241| 18.5k| for (; k < forward; k++) { ------------------ | Branch (241:14): [True: 17.4k, False: 1.09k] ------------------ 242| 17.4k| uint32_t word = buf[k]; 243| 17.4k| if ((word & 0xFFFFFF80) == 0) { // 1-byte (ASCII) ------------------ | Branch (243:13): [True: 8.54k, False: 8.87k] ------------------ 244| 8.54k| *utf8_output++ = char(word); 245| 8.87k| } else if ((word & 0xFFFFF800) == 0) { // 2-byte ------------------ | Branch (245:20): [True: 1.06k, False: 7.81k] ------------------ 246| 1.06k| *utf8_output++ = char((word >> 6) | 0b11000000); 247| 1.06k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 248| 7.81k| } else if ((word & 0xFFFF0000) == 0) { // 3-byte ------------------ | Branch (248:20): [True: 2.53k, False: 5.27k] ------------------ 249| 2.53k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (249:15): [True: 1.09k, False: 1.44k] | Branch (249:33): [True: 12, False: 1.08k] ------------------ 250| 12| return std::make_pair(nullptr, utf8_output); 251| 12| } 252| 2.52k| *utf8_output++ = char((word >> 12) | 0b11100000); 253| 2.52k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 254| 2.52k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 255| 5.27k| } else { // 4-byte 256| 5.27k| if (word > 0x10FFFF) { ------------------ | Branch (256:15): [True: 154, False: 5.12k] ------------------ 257| 154| return std::make_pair(nullptr, utf8_output); 258| 154| } 259| 5.12k| *utf8_output++ = char((word >> 18) | 0b11110000); 260| 5.12k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 261| 5.12k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 262| 5.12k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 263| 5.12k| } 264| 17.4k| } 265| 1.09k| buf += k; 266| 1.09k| } 267| 2.05k| } // while 268| | 269| | // check for invalid input 270| 630| const __m256i v_10ffff = _mm256_set1_epi32((uint32_t)0x10ffff); 271| 630| if (static_cast(_mm256_movemask_epi8(_mm256_cmpeq_epi32( ------------------ | Branch (271:7): [True: 40, False: 590] ------------------ 272| 630| _mm256_max_epu32(running_max, v_10ffff), v_10ffff))) != 0xffffffff) { 273| 40| return std::make_pair(nullptr, utf8_output); 274| 40| } 275| | 276| 590| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (276:7): [True: 68, False: 522] ------------------ 277| 68| return std::make_pair(nullptr, utf8_output); 278| 68| } 279| | 280| 522| return std::make_pair(buf, utf8_output); 281| 590|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_138avx2_convert_utf32_to_utf8_with_errorsEPKDimPc: 285| 830| char *utf8_output) { 286| 830| const char32_t *end = buf + len; 287| 830| const char32_t *start = buf; 288| | 289| 830| const __m256i v_0000 = _mm256_setzero_si256(); 290| 830| const __m256i v_ffff0000 = _mm256_set1_epi32((uint32_t)0xffff0000); 291| 830| const __m256i v_ff80 = _mm256_set1_epi16((uint16_t)0xff80); 292| 830| const __m256i v_f800 = _mm256_set1_epi16((uint16_t)0xf800); 293| 830| const __m256i v_c080 = _mm256_set1_epi16((uint16_t)0xc080); 294| 830| const __m256i v_7fffffff = _mm256_set1_epi32((uint32_t)0x7fffffff); 295| 830| const __m256i v_10ffff = _mm256_set1_epi32((uint32_t)0x10ffff); 296| | 297| 830| const size_t safety_margin = 298| 830| 12; // to avoid overruns, see issue 299| | // https://github.com/simdutf/simdutf/issues/92 300| | 301| 32.2k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (301:10): [True: 31.6k, False: 590] ------------------ 302| 31.6k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 303| 31.6k| __m256i nextin = _mm256_loadu_si256((__m256i *)buf + 1); 304| | // Check for too large input 305| 31.6k| const __m256i max_input = 306| 31.6k| _mm256_max_epu32(_mm256_max_epu32(in, nextin), v_10ffff); 307| 31.6k| if (static_cast(_mm256_movemask_epi8( ------------------ | Branch (307:9): [True: 160, False: 31.4k] ------------------ 308| 31.6k| _mm256_cmpeq_epi32(max_input, v_10ffff))) != 0xffffffff) { 309| 160| return std::make_pair(result(error_code::TOO_LARGE, buf - start), 310| 160| utf8_output); 311| 160| } 312| | 313| | // Pack 32-bit UTF-32 code units to 16-bit UTF-16 code units with unsigned 314| | // saturation 315| 31.4k| __m256i in_16 = _mm256_packus_epi32(_mm256_and_si256(in, v_7fffffff), 316| 31.4k| _mm256_and_si256(nextin, v_7fffffff)); 317| 31.4k| 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| 31.4k| if (_mm256_testz_si256(in_16, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (322:9): [True: 26.8k, False: 4.62k] ------------------ 323| | // 1. pack the bytes 324| 26.8k| const __m128i utf8_packed = _mm_packus_epi16( 325| 26.8k| _mm256_castsi256_si128(in_16), _mm256_extractf128_si256(in_16, 1)); 326| | // 2. store (16 bytes) 327| 26.8k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 328| | // 3. adjust pointers 329| 26.8k| buf += 16; 330| 26.8k| utf8_output += 16; 331| 26.8k| continue; // we are done for this round! 332| 26.8k| } 333| | // no bits set above 7th bit 334| 4.62k| const __m256i one_byte_bytemask = 335| 4.62k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_ff80), v_0000); 336| 4.62k| const uint32_t one_byte_bitmask = 337| 4.62k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 338| | 339| | // no bits set above 11th bit 340| 4.62k| const __m256i one_or_two_bytes_bytemask = 341| 4.62k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_0000); 342| 4.62k| const uint32_t one_or_two_bytes_bitmask = 343| 4.62k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 344| 4.62k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (344:9): [True: 2.51k, False: 2.10k] ------------------ 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| 2.51k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 349| 2.51k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 350| | 351| | // t0 = [000a|aaaa|bbbb|bb00] 352| 2.51k| const __m256i t0 = _mm256_slli_epi16(in_16, 2); 353| | // t1 = [000a|aaaa|0000|0000] 354| 2.51k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 355| | // t2 = [0000|0000|00bb|bbbb] 356| 2.51k| const __m256i t2 = _mm256_and_si256(in_16, v_003f); 357| | // t3 = [000a|aaaa|00bb|bbbb] 358| 2.51k| const __m256i t3 = _mm256_or_si256(t1, t2); 359| | // t4 = [110a|aaaa|10bb|bbbb] 360| 2.51k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 361| | 362| | // 2. merge ASCII and 2-byte codewords 363| 2.51k| const __m256i utf8_unpacked = 364| 2.51k| _mm256_blendv_epi8(t4, in_16, one_byte_bytemask); 365| | 366| | // 3. prepare bitmask for 8-bit lookup 367| 2.51k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 368| 2.51k| const uint32_t M1 = M0 >> 7; 369| 2.51k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 370| | // 4. pack the bytes 371| | 372| 2.51k| const uint8_t *row = 373| 2.51k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 374| 2.51k| const uint8_t *row_2 = 375| 2.51k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 376| 2.51k| 16)][0]; 377| | 378| 2.51k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 379| 2.51k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 380| | 381| 2.51k| const __m256i utf8_packed = _mm256_shuffle_epi8( 382| 2.51k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 2.51k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 2.51k| _mm256_castsi128_si256(xmm2), 2) ------------------ 383| | // 5. store bytes 384| 2.51k| _mm_storeu_si128((__m128i *)utf8_output, 385| 2.51k| _mm256_castsi256_si128(utf8_packed)); 386| 2.51k| utf8_output += row[0]; 387| 2.51k| _mm_storeu_si128((__m128i *)utf8_output, 388| 2.51k| _mm256_extractf128_si256(utf8_packed, 1)); 389| 2.51k| utf8_output += row_2[0]; 390| | 391| | // 6. adjust pointers 392| 2.51k| buf += 16; 393| 2.51k| continue; 394| 2.51k| } 395| | // Must check for overflow in packing 396| 2.10k| const __m256i saturation_bytemask = _mm256_cmpeq_epi32( 397| 2.10k| _mm256_and_si256(_mm256_or_si256(in, nextin), v_ffff0000), v_0000); 398| 2.10k| const uint32_t saturation_bitmask = 399| 2.10k| static_cast(_mm256_movemask_epi8(saturation_bytemask)); 400| 2.10k| if (saturation_bitmask == 0xffffffff) { ------------------ | Branch (400:9): [True: 732, False: 1.37k] ------------------ 401| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 402| | 403| | // Check for illegal surrogate code units 404| 732| const __m256i v_d800 = _mm256_set1_epi16((uint16_t)0xd800); 405| 732| const __m256i forbidden_bytemask = 406| 732| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_d800); 407| 732| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != ------------------ | Branch (407:11): [True: 48, False: 684] ------------------ 408| 732| 0x0) { 409| 48| return std::make_pair(result(error_code::SURROGATE, buf - start), 410| 48| utf8_output); 411| 48| } 412| | 413| 684| const __m256i dup_even = _mm256_setr_epi16( 414| 684| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 415| 684| 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| 684|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 445| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 446| 684| const __m256i t0 = _mm256_shuffle_epi8(in_16, dup_even); 447| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 448| 684| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 444| 684|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 449| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 450| 684| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 444| 684|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 451| | 452| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 453| 684| const __m256i s0 = _mm256_srli_epi16(in_16, 4); 454| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 455| 684| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 444| 684|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 456| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 457| 684| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 444| 684|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 458| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 459| 684| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 444| 684|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 460| 684| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 461| 684| simdutf_vec(0b0100000000000000)); ------------------ | | 444| 684|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 462| 684| const __m256i s4 = _mm256_xor_si256(s3, m0); 463| 684|#undef simdutf_vec 464| | 465| | // 4. expand code units 16-bit => 32-bit 466| 684| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 467| 684| 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| 684| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 471| 684| (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| 684| const uint8_t mask0 = uint8_t(mask); 492| 684| const uint8_t *row0 = 493| 684| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 494| 684| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 495| 684| const __m128i utf8_0 = 496| 684| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 497| | 498| 684| const uint8_t mask1 = static_cast(mask >> 8); 499| 684| const uint8_t *row1 = 500| 684| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 501| 684| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 502| 684| const __m128i utf8_1 = 503| 684| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 504| | 505| 684| const uint8_t mask2 = static_cast(mask >> 16); 506| 684| const uint8_t *row2 = 507| 684| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 508| 684| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 509| 684| const __m128i utf8_2 = 510| 684| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 511| | 512| 684| const uint8_t mask3 = static_cast(mask >> 24); 513| 684| const uint8_t *row3 = 514| 684| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 515| 684| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 516| 684| const __m128i utf8_3 = 517| 684| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 518| | 519| 684| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 520| 684| utf8_output += row0[0]; 521| 684| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 522| 684| utf8_output += row1[0]; 523| 684| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 524| 684| utf8_output += row2[0]; 525| 684| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 526| 684| utf8_output += row3[0]; 527| 684| buf += 16; 528| 1.37k| } 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| 1.37k| size_t forward = 15; 534| 1.37k| size_t k = 0; 535| 1.37k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (535:11): [True: 0, False: 1.37k] ------------------ 536| 0| forward = size_t(end - buf - 1); 537| 0| } 538| 21.7k| for (; k < forward; k++) { ------------------ | Branch (538:14): [True: 20.4k, False: 1.34k] ------------------ 539| 20.4k| uint32_t word = buf[k]; 540| 20.4k| if ((word & 0xFFFFFF80) == 0) { // 1-byte (ASCII) ------------------ | Branch (540:13): [True: 6.25k, False: 14.1k] ------------------ 541| 6.25k| *utf8_output++ = char(word); 542| 14.1k| } else if ((word & 0xFFFFF800) == 0) { // 2-byte ------------------ | Branch (542:20): [True: 1.69k, False: 12.4k] ------------------ 543| 1.69k| *utf8_output++ = char((word >> 6) | 0b11000000); 544| 1.69k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 545| 12.4k| } else if ((word & 0xFFFF0000) == 0) { // 3-byte ------------------ | Branch (545:20): [True: 2.70k, False: 9.74k] ------------------ 546| 2.70k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (546:15): [True: 1.23k, False: 1.47k] | Branch (546:33): [True: 32, False: 1.20k] ------------------ 547| 32| return std::make_pair( 548| 32| result(error_code::SURROGATE, buf - start + k), utf8_output); 549| 32| } 550| 2.67k| *utf8_output++ = char((word >> 12) | 0b11100000); 551| 2.67k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 552| 2.67k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 553| 9.74k| } else { // 4-byte 554| 9.74k| if (word > 0x10FFFF) { ------------------ | Branch (554:15): [True: 0, False: 9.74k] ------------------ 555| 0| return std::make_pair( 556| 0| result(error_code::TOO_LARGE, buf - start + k), utf8_output); 557| 0| } 558| 9.74k| *utf8_output++ = char((word >> 18) | 0b11110000); 559| 9.74k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 560| 9.74k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 561| 9.74k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 562| 9.74k| } 563| 20.4k| } 564| 1.34k| buf += k; 565| 1.34k| } 566| 2.10k| } // while 567| | 568| 590| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 569| 830|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_129convert_masked_utf8_to_latin1EPKcmRPc: 9| 543k| 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| 543k| const __m128i in = _mm_loadu_si128((__m128i *)input); 21| | 22| 543k| const uint16_t input_utf8_end_of_code_point_mask = 23| 543k| utf8_end_of_code_point_mask & 24| 543k| 0xfff; // we are only processing 12 bytes in case it is not all ASCII 25| | 26| 543k| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (26:7): [True: 3.58k, False: 539k] ------------------ 27| | // We process the data in chunks of 12 bytes. 28| 3.58k| _mm_storeu_si128(reinterpret_cast<__m128i *>(latin1_output), in); 29| 3.58k| latin1_output += 12; // We wrote 12 characters. 30| 3.58k| return 12; // We consumed 1 bytes. 31| 3.58k| } 32| | /// We do not have a fast path available, so we fallback. 33| 539k| const uint8_t idx = 34| 539k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 35| 539k| const uint8_t consumed = 36| 539k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 37| | // this indicates an invalid input: 38| 539k| if (idx >= 64) { ------------------ | Branch (38:7): [True: 12.2k, False: 527k] ------------------ 39| 12.2k| return consumed; 40| 12.2k| } 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| 527k| const __m128i sh = 48| 527k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 49| 527k| const __m128i perm = _mm_shuffle_epi8(in, sh); 50| 527k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 51| 527k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 52| 527k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 53| 527k| 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| 527k| _mm_storel_epi64((__m128i *)latin1_output, latin1_packed); 58| 527k| latin1_output += 6; // We wrote 6 bytes. 59| 527k| return consumed; 60| 539k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE0EEEmPKcmRPDs: 10| 2.87M| 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| 2.87M| const __m128i swap = 23| 2.87M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 2.87M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 2.87M| const uint16_t input_utf8_end_of_code_point_mask = 26| 2.87M| utf8_end_of_code_point_mask & 0xfff; 27| 2.87M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 9.93k, False: 2.86M] ------------------ 28| | // We process the data in chunks of 12 bytes. 29| 9.93k| __m256i ascii = _mm256_cvtepu8_epi16(in); 30| 9.93k| if (big_endian) { ------------------ | Branch (30:9): [Folded, False: 9.93k] ------------------ 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| 9.93k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output), ascii); 37| 9.93k| utf16_output += 12; // We wrote 12 16-bit characters. 38| 9.93k| return 12; // We consumed 12 bytes. 39| 9.93k| } 40| 2.86M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (40:7): [True: 11.2k, False: 2.84M] ------------------ 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| 11.2k| const __m128i sh = 45| 11.2k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 46| 11.2k| const __m128i perm = _mm_shuffle_epi8(in, sh); 47| 11.2k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 48| 11.2k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 49| 11.2k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 50| 11.2k| if (big_endian) ------------------ | Branch (50:9): [Folded, False: 11.2k] ------------------ 51| 0| composed = _mm_shuffle_epi8(composed, swap); 52| 11.2k| _mm_storeu_si128((__m128i *)utf16_output, composed); 53| 11.2k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 54| 11.2k| return 16; 55| 11.2k| } 56| 2.84M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (56:7): [True: 2.85k, False: 2.84M] ------------------ 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| 2.85k| const __m128i sh = 61| 2.85k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 62| 2.85k| const __m128i perm = _mm_shuffle_epi8(in, sh); 63| 2.85k| const __m128i ascii = 64| 2.85k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 65| 2.85k| const __m128i middlebyte = 66| 2.85k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 67| 2.85k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 68| 2.85k| const __m128i highbyte = 69| 2.85k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 70| 2.85k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 71| 2.85k| const __m128i composed = 72| 2.85k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 73| 2.85k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 74| 2.85k| if (big_endian) ------------------ | Branch (74:9): [Folded, False: 2.85k] ------------------ 75| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 76| 2.85k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 77| 2.85k| utf16_output += 4; 78| 2.85k| return 12; 79| 2.85k| } 80| | 81| 2.84M| const uint8_t idx = simdutf::tables::utf8_to_utf16::utf8bigindex 82| 2.84M| [input_utf8_end_of_code_point_mask][0]; 83| 2.84M| const uint8_t consumed = simdutf::tables::utf8_to_utf16::utf8bigindex 84| 2.84M| [input_utf8_end_of_code_point_mask][1]; 85| 2.84M| if (idx < 64) { ------------------ | Branch (85:7): [True: 2.40M, False: 439k] ------------------ 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| 2.40M| const __m128i sh = _mm_loadu_si128( 93| 2.40M| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 94| 2.40M| const __m128i perm = _mm_shuffle_epi8(in, sh); 95| 2.40M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 96| 2.40M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 97| 2.40M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 98| 2.40M| if (big_endian) ------------------ | Branch (98:9): [Folded, False: 2.40M] ------------------ 99| 0| composed = _mm_shuffle_epi8(composed, swap); 100| 2.40M| _mm_storeu_si128((__m128i *)utf16_output, composed); 101| 2.40M| utf16_output += 6; // We wrote 12 bytes, 6 code points. There is a potential 102| | // overflow of 4 bytes. 103| 2.40M| } else if (idx < 145) { ------------------ | Branch (103:14): [True: 390k, False: 48.5k] ------------------ 104| | // FOUR (4) input code-code units 105| 390k| const __m128i sh = _mm_loadu_si128( 106| 390k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 107| 390k| const __m128i perm = _mm_shuffle_epi8(in, sh); 108| 390k| const __m128i ascii = 109| 390k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 110| 390k| const __m128i middlebyte = 111| 390k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 112| 390k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 113| 390k| const __m128i highbyte = 114| 390k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 115| 390k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 116| 390k| const __m128i composed = 117| 390k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 118| 390k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 119| 390k| if (big_endian) ------------------ | Branch (119:9): [Folded, False: 390k] ------------------ 120| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 121| 390k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 122| 390k| utf16_output += 4; // Here we overflow by 8 bytes. 123| 390k| } else if (idx < 209) { ------------------ | Branch (123:14): [True: 47.7k, False: 756] ------------------ 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| 47.7k| const __m128i sh = _mm_loadu_si128( 135| 47.7k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 136| 47.7k| const __m128i perm = _mm_shuffle_epi8(in, sh); 137| 47.7k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 138| 47.7k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 139| 47.7k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 140| 47.7k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 141| | // correct for spurious high bit 142| 47.7k| const __m128i correct = 143| 47.7k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 144| 47.7k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 145| 47.7k| 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| 47.7k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 149| 47.7k| 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| 47.7k| const __m128i composed = 155| 47.7k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 156| 47.7k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 157| 47.7k| const __m128i composedminus = 158| 47.7k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 159| 47.7k| const __m128i lowtenbits = 160| 47.7k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 161| | // Notice the 0x3ff mask: 162| 47.7k| const __m128i hightenbits = 163| 47.7k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 164| 47.7k| const __m128i lowtenbitsadd = 165| 47.7k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 166| 47.7k| const __m128i hightenbitsadd = 167| 47.7k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 168| 47.7k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 169| 47.7k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 170| 47.7k| uint32_t basic_buffer[4]; 171| 47.7k| uint32_t basic_buffer_swap[4]; 172| 47.7k| if (big_endian) { ------------------ | Branch (172:9): [Folded, False: 47.7k] ------------------ 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| 47.7k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 178| 47.7k| uint32_t surrogate_buffer[4]; 179| 47.7k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 180| 191k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (180:24): [True: 143k, False: 47.7k] ------------------ 181| 143k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (181:11): [True: 40.1k, False: 103k] ------------------ 182| 40.1k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 183| 40.1k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 184| 40.1k| utf16_output += 2; 185| 103k| } else { 186| 103k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (186:27): [Folded, False: 103k] ------------------ 187| 103k| : uint16_t(basic_buffer[i]); 188| 103k| utf16_output++; 189| 103k| } 190| 143k| } 191| 47.7k| } else { 192| | // here we know that there is an error but we do not handle errors 193| 756| } 194| 2.84M| return consumed; 195| 2.84M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE1EEEmPKcmRPDs: 10| 3.09M| 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| 3.09M| const __m128i swap = 23| 3.09M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 3.09M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 3.09M| const uint16_t input_utf8_end_of_code_point_mask = 26| 3.09M| utf8_end_of_code_point_mask & 0xfff; 27| 3.09M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 8.76k, False: 3.09M] ------------------ 28| | // We process the data in chunks of 12 bytes. 29| 8.76k| __m256i ascii = _mm256_cvtepu8_epi16(in); 30| 8.76k| if (big_endian) { ------------------ | Branch (30:9): [True: 8.76k, Folded] ------------------ 31| 8.76k| const __m256i swap256 = _mm256_setr_epi8( 32| 8.76k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 33| 8.76k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 34| 8.76k| ascii = _mm256_shuffle_epi8(ascii, swap256); 35| 8.76k| } 36| 8.76k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output), ascii); 37| 8.76k| utf16_output += 12; // We wrote 12 16-bit characters. 38| 8.76k| return 12; // We consumed 12 bytes. 39| 8.76k| } 40| 3.09M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (40:7): [True: 20.4k, False: 3.07M] ------------------ 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| 20.4k| const __m128i sh = 45| 20.4k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 46| 20.4k| const __m128i perm = _mm_shuffle_epi8(in, sh); 47| 20.4k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 48| 20.4k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 49| 20.4k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 50| 20.4k| if (big_endian) ------------------ | Branch (50:9): [True: 20.4k, Folded] ------------------ 51| 20.4k| composed = _mm_shuffle_epi8(composed, swap); 52| 20.4k| _mm_storeu_si128((__m128i *)utf16_output, composed); 53| 20.4k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 54| 20.4k| return 16; 55| 20.4k| } 56| 3.07M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (56:7): [True: 4.37k, False: 3.06M] ------------------ 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| 4.37k| const __m128i sh = 61| 4.37k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 62| 4.37k| const __m128i perm = _mm_shuffle_epi8(in, sh); 63| 4.37k| const __m128i ascii = 64| 4.37k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 65| 4.37k| const __m128i middlebyte = 66| 4.37k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 67| 4.37k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 68| 4.37k| const __m128i highbyte = 69| 4.37k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 70| 4.37k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 71| 4.37k| const __m128i composed = 72| 4.37k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 73| 4.37k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 74| 4.37k| if (big_endian) ------------------ | Branch (74:9): [True: 4.37k, Folded] ------------------ 75| 4.37k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 76| 4.37k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 77| 4.37k| utf16_output += 4; 78| 4.37k| return 12; 79| 4.37k| } 80| | 81| 3.06M| const uint8_t idx = simdutf::tables::utf8_to_utf16::utf8bigindex 82| 3.06M| [input_utf8_end_of_code_point_mask][0]; 83| 3.06M| const uint8_t consumed = simdutf::tables::utf8_to_utf16::utf8bigindex 84| 3.06M| [input_utf8_end_of_code_point_mask][1]; 85| 3.06M| if (idx < 64) { ------------------ | Branch (85:7): [True: 2.75M, False: 306k] ------------------ 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| 2.75M| const __m128i sh = _mm_loadu_si128( 93| 2.75M| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 94| 2.75M| const __m128i perm = _mm_shuffle_epi8(in, sh); 95| 2.75M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 96| 2.75M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 97| 2.75M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 98| 2.75M| if (big_endian) ------------------ | Branch (98:9): [True: 2.75M, Folded] ------------------ 99| 2.75M| composed = _mm_shuffle_epi8(composed, swap); 100| 2.75M| _mm_storeu_si128((__m128i *)utf16_output, composed); 101| 2.75M| utf16_output += 6; // We wrote 12 bytes, 6 code points. There is a potential 102| | // overflow of 4 bytes. 103| 2.75M| } else if (idx < 145) { ------------------ | Branch (103:14): [True: 253k, False: 52.3k] ------------------ 104| | // FOUR (4) input code-code units 105| 253k| const __m128i sh = _mm_loadu_si128( 106| 253k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 107| 253k| const __m128i perm = _mm_shuffle_epi8(in, sh); 108| 253k| const __m128i ascii = 109| 253k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 110| 253k| const __m128i middlebyte = 111| 253k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 112| 253k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 113| 253k| const __m128i highbyte = 114| 253k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 115| 253k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 116| 253k| const __m128i composed = 117| 253k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 118| 253k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 119| 253k| if (big_endian) ------------------ | Branch (119:9): [True: 253k, Folded] ------------------ 120| 253k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 121| 253k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 122| 253k| utf16_output += 4; // Here we overflow by 8 bytes. 123| 253k| } else if (idx < 209) { ------------------ | Branch (123:14): [True: 51.0k, False: 1.31k] ------------------ 124| | // TWO (2) input code-code units 125| | ////////////// 126| | // There might be garbage inputs where a leading byte mascarades as a 127| | // four-byte leading byte (by being followed by 3 continuation byte), but is 128| | // not greater than 0xf0. This could trigger a buffer overflow if we only 129| | // counted leading bytes of the form 0xf0 as generating surrogate pairs, 130| | // without further UTF-8 validation. Thus we must be careful to ensure that 131| | // only leading bytes at least as large as 0xf0 generate surrogate pairs. We 132| | // do as at the cost of an extra mask. 133| | ///////////// 134| 51.0k| const __m128i sh = _mm_loadu_si128( 135| 51.0k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 136| 51.0k| const __m128i perm = _mm_shuffle_epi8(in, sh); 137| 51.0k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 138| 51.0k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 139| 51.0k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 140| 51.0k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 141| | // correct for spurious high bit 142| 51.0k| const __m128i correct = 143| 51.0k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 144| 51.0k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 145| 51.0k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 146| | // We deliberately carry the leading four bits in highbyte if they are 147| | // present, we remove them later when computing hightenbits. 148| 51.0k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 149| 51.0k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 150| | // When we need to generate a surrogate pair (leading byte > 0xF0), then 151| | // the corresponding 32-bit value in 'composed' will be greater than 152| | // > (0xff00000>>6) or > 0x3c00000. This can be used later to identify the 153| | // location of the surrogate pairs. 154| 51.0k| const __m128i composed = 155| 51.0k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 156| 51.0k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 157| 51.0k| const __m128i composedminus = 158| 51.0k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 159| 51.0k| const __m128i lowtenbits = 160| 51.0k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 161| | // Notice the 0x3ff mask: 162| 51.0k| const __m128i hightenbits = 163| 51.0k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 164| 51.0k| const __m128i lowtenbitsadd = 165| 51.0k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 166| 51.0k| const __m128i hightenbitsadd = 167| 51.0k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 168| 51.0k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 169| 51.0k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 170| 51.0k| uint32_t basic_buffer[4]; 171| 51.0k| uint32_t basic_buffer_swap[4]; 172| 51.0k| if (big_endian) { ------------------ | Branch (172:9): [True: 51.0k, Folded] ------------------ 173| 51.0k| _mm_storeu_si128((__m128i *)basic_buffer_swap, 174| 51.0k| _mm_shuffle_epi8(composed, swap)); 175| 51.0k| surrogates = _mm_shuffle_epi8(surrogates, swap); 176| 51.0k| } 177| 51.0k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 178| 51.0k| uint32_t surrogate_buffer[4]; 179| 51.0k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 180| 204k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (180:24): [True: 153k, False: 51.0k] ------------------ 181| 153k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (181:11): [True: 42.9k, False: 110k] ------------------ 182| 42.9k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 183| 42.9k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 184| 42.9k| utf16_output += 2; 185| 110k| } else { 186| 110k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (186:27): [True: 110k, Folded] ------------------ 187| 110k| : uint16_t(basic_buffer[i]); 188| 110k| utf16_output++; 189| 110k| } 190| 153k| } 191| 51.0k| } else { 192| | // here we know that there is an error but we do not handle errors 193| 1.31k| } 194| 3.06M| return consumed; 195| 3.07M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128convert_masked_utf8_to_utf32EPKcmRPDi: 9| 2.98M| char32_t *&utf32_output) { 10| | // we use an approach where we try to process up to 12 input bytes. 11| | // Why 12 input bytes and not 16? Because we are concerned with the size of 12| | // the lookup tables. Also 12 is nicely divisible by two and three. 13| | // 14| | // 15| | // Optimization note: our main path below is load-latency dependent. Thus it 16| | // is maybe beneficial to have fast paths that depend on branch prediction but 17| | // have less latency. This results in more instructions but, potentially, also 18| | // higher speeds. 19| | // 20| | // We first try a few fast paths. 21| 2.98M| const __m128i in = _mm_loadu_si128((__m128i *)input); 22| 2.98M| const uint16_t input_utf8_end_of_code_point_mask = 23| 2.98M| utf8_end_of_code_point_mask & 0xfff; 24| 2.98M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (24:7): [True: 8.66k, False: 2.98M] ------------------ 25| | // We process the data in chunks of 12 bytes. 26| 8.66k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 27| 8.66k| _mm256_cvtepu8_epi32(in)); 28| 8.66k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output + 8), 29| 8.66k| _mm256_cvtepu8_epi32(_mm_srli_si128(in, 8))); 30| 8.66k| utf32_output += 12; // We wrote 12 32-bit characters. 31| 8.66k| return 12; // We consumed 12 bytes. 32| 8.66k| } 33| 2.98M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (33:7): [True: 18.3k, False: 2.96M] ------------------ 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| 18.3k| const __m128i sh = 38| 18.3k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 39| 18.3k| const __m128i perm = _mm_shuffle_epi8(in, sh); 40| 18.3k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 41| 18.3k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 42| 18.3k| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 43| 18.3k| _mm256_storeu_si256((__m256i *)utf32_output, 44| 18.3k| _mm256_cvtepu16_epi32(composed)); 45| 18.3k| utf32_output += 8; // We wrote 16 bytes, 8 code points. 46| 18.3k| return 16; 47| 18.3k| } 48| 2.96M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (48:7): [True: 2.52k, False: 2.95M] ------------------ 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| 2.52k| const __m128i sh = 53| 2.52k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 54| 2.52k| const __m128i perm = _mm_shuffle_epi8(in, sh); 55| 2.52k| const __m128i ascii = 56| 2.52k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 57| 2.52k| const __m128i middlebyte = 58| 2.52k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 59| 2.52k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 60| 2.52k| const __m128i highbyte = 61| 2.52k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 62| 2.52k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 63| 2.52k| const __m128i composed = 64| 2.52k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 65| 2.52k| _mm_storeu_si128((__m128i *)utf32_output, composed); 66| 2.52k| utf32_output += 4; 67| 2.52k| return 12; 68| 2.52k| } 69| | /// We do not have a fast path available, so we fallback. 70| | 71| 2.95M| const uint8_t idx = 72| 2.95M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 73| 2.95M| const uint8_t consumed = 74| 2.95M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 75| 2.95M| if (idx < 64) { ------------------ | Branch (75:7): [True: 2.70M, False: 250k] ------------------ 76| | // SIX (6) input code-code units 77| | // this is a relatively easy scenario 78| | // we process SIX (6) input code-code units. The max length in bytes of six 79| | // code code units spanning between 1 and 2 bytes each is 12 bytes. On 80| | // processors where pdep/pext is fast, we might be able to use a small 81| | // lookup table. 82| 2.70M| const __m128i sh = 83| 2.70M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 84| 2.70M| const __m128i perm = _mm_shuffle_epi8(in, sh); 85| 2.70M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 86| 2.70M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 87| 2.70M| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 88| 2.70M| _mm256_storeu_si256((__m256i *)utf32_output, 89| 2.70M| _mm256_cvtepu16_epi32(composed)); 90| 2.70M| utf32_output += 6; // We wrote 24 bytes, 6 code points. There is a potential 91| | // overflow of 32 - 24 = 8 bytes. 92| 2.70M| } else if (idx < 145) { ------------------ | Branch (92:14): [True: 202k, False: 48.0k] ------------------ 93| | // FOUR (4) input code-code units 94| 202k| const __m128i sh = 95| 202k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 96| 202k| const __m128i perm = _mm_shuffle_epi8(in, sh); 97| 202k| const __m128i ascii = 98| 202k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 99| 202k| const __m128i middlebyte = 100| 202k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 101| 202k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 102| 202k| const __m128i highbyte = 103| 202k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 104| 202k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 105| 202k| const __m128i composed = 106| 202k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 107| 202k| _mm_storeu_si128((__m128i *)utf32_output, composed); 108| 202k| utf32_output += 4; 109| 202k| } else if (idx < 209) { ------------------ | Branch (109:14): [True: 47.3k, False: 670] ------------------ 110| | // TWO (2) input code-code units 111| 47.3k| const __m128i sh = 112| 47.3k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 113| 47.3k| const __m128i perm = _mm_shuffle_epi8(in, sh); 114| 47.3k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 115| 47.3k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 116| 47.3k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 117| 47.3k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 118| | // correct for spurious high bit 119| 47.3k| const __m128i correct = 120| 47.3k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 121| 47.3k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 122| 47.3k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 123| 47.3k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0x07000000)); 124| 47.3k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 125| 47.3k| const __m128i composed = 126| 47.3k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 127| 47.3k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 128| 47.3k| _mm_storeu_si128((__m128i *)utf32_output, composed); 129| 47.3k| utf32_output += 130| 47.3k| 3; // We wrote 3 * 4 bytes, there is a potential overflow of 4 bytes. 131| 47.3k| } else { 132| | // here we know that there is an error but we do not handle errors 133| 670| } 134| 2.95M| return consumed; 135| 2.96M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf16_gather_high_bytesILNS_10endiannessE0EEENS1_4simd5simd8IhEERKNS5_6simd16ItEESB_: 3| 2.70M| const simd16 &in1) { 4| 2.70M| if (big_endian) { ------------------ | Branch (4:7): [Folded, False: 2.70M] ------------------ 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| 2.70M| } else { 12| 2.70M| const auto t0 = in0.shr<8>(); 13| 2.70M| const auto t1 = in1.shr<8>(); 14| | 15| 2.70M| return simd16::pack(t0, t1); 16| 2.70M| } 17| 2.70M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf16_gather_high_bytesILNS_10endiannessE1EEENS1_4simd5simd8IhEERKNS5_6simd16ItEESB_: 3| 2.80M| const simd16 &in1) { 4| 2.80M| if (big_endian) { ------------------ | Branch (4:7): [True: 2.80M, Folded] ------------------ 5| | // we want lower bytes 6| 2.80M| const auto mask = simd16(0x00ff); 7| 2.80M| const auto t0 = in0 & mask; 8| 2.80M| const auto t1 = in1 & mask; 9| | 10| 2.80M| return simd16::pack(t0, t1); 11| 2.80M| } 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| 2.80M|} _ZNK7simdutf7haswell14implementation13validate_utf8EPKcm: 266| 3.22k|implementation::validate_utf8(const char *buf, size_t len) const noexcept { 267| 3.22k| return haswell::utf8_validation::generic_validate_utf8(buf, len); 268| 3.22k|} _ZNK7simdutf7haswell14implementation25validate_utf8_with_errorsEPKcm: 273| 3.22k| const char *buf, size_t len) const noexcept { 274| 3.22k| return haswell::utf8_validation::generic_validate_utf8_with_errors(buf, len); 275| 3.22k|} _ZNK7simdutf7haswell14implementation16validate_utf16leEPKDsm: 312| 1.30k| size_t len) const noexcept { 313| 1.30k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.30k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 17, False: 1.29k] | | ------------------ ------------------ 314| | // empty input is valid UTF-16. protect the implementation from 315| | // handling nullptr 316| 17| return true; 317| 17| } 318| 1.29k| const auto res = 319| 1.29k| haswell::utf16::validate_utf16_with_errors(buf, len); 320| 1.29k| if (res.is_err()) { ------------------ | Branch (320:7): [True: 79, False: 1.21k] ------------------ 321| 79| return false; 322| 79| } 323| | 324| 1.21k| if (res.count == len) { ------------------ | Branch (324:7): [True: 0, False: 1.21k] ------------------ 325| 0| return true; 326| 0| } 327| | 328| 1.21k| return scalar::utf16::validate(buf + res.count, 329| 1.21k| len - res.count); 330| 1.21k|} _ZNK7simdutf7haswell14implementation16validate_utf16beEPKDsm: 336| 1.23k| size_t len) const noexcept { 337| 1.23k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.23k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 15, False: 1.22k] | | ------------------ ------------------ 338| | // empty input is valid UTF-16. protect the implementation from 339| | // handling nullptr 340| 15| return true; 341| 15| } 342| 1.22k| const auto res = 343| 1.22k| haswell::utf16::validate_utf16_with_errors(buf, len); 344| 1.22k| if (res.is_err()) { ------------------ | Branch (344:7): [True: 98, False: 1.12k] ------------------ 345| 98| return false; 346| 98| } 347| | 348| 1.12k| if (res.count == len) { ------------------ | Branch (348:7): [True: 0, False: 1.12k] ------------------ 349| 0| return true; 350| 0| } 351| | 352| 1.12k| return scalar::utf16::validate(buf + res.count, 353| 1.12k| len - res.count); 354| 1.12k|} _ZNK7simdutf7haswell14implementation28validate_utf16le_with_errorsEPKDsm: 357| 1.30k| const char16_t *buf, size_t len) const noexcept { 358| | 359| 1.30k| const result res = 360| 1.30k| haswell::utf16::validate_utf16_with_errors(buf, len); 361| 1.30k| if (res.count != len) { ------------------ | Branch (361:7): [True: 1.29k, False: 17] ------------------ 362| 1.29k| const result scalar_res = 363| 1.29k| scalar::utf16::validate_with_errors( 364| 1.29k| buf + res.count, len - res.count); 365| 1.29k| return result(scalar_res.error, res.count + scalar_res.count); 366| 1.29k| } else { 367| 17| return res; 368| 17| } 369| 1.30k|} _ZNK7simdutf7haswell14implementation28validate_utf16be_with_errorsEPKDsm: 372| 1.23k| const char16_t *buf, size_t len) const noexcept { 373| 1.23k| const result res = 374| 1.23k| haswell::utf16::validate_utf16_with_errors(buf, len); 375| 1.23k| if (res.count != len) { ------------------ | Branch (375:7): [True: 1.22k, False: 15] ------------------ 376| 1.22k| const result scalar_res = 377| 1.22k| scalar::utf16::validate_with_errors(buf + res.count, 378| 1.22k| len - res.count); 379| 1.22k| return result(scalar_res.error, res.count + scalar_res.count); 380| 1.22k| } else { 381| 15| return res; 382| 15| } 383| 1.23k|} _ZNK7simdutf7haswell14implementation14validate_utf32EPKDim: 398| 2.41k|implementation::validate_utf32(const char32_t *buf, size_t len) const noexcept { 399| 2.41k| return utf32::validate(buf, len); 400| 2.41k|} _ZNK7simdutf7haswell14implementation26validate_utf32_with_errorsEPKDim: 405| 2.41k| const char32_t *buf, size_t len) const noexcept { 406| 2.41k| return utf32::validate_with_errors(buf, len); 407| 2.41k|} _ZNK7simdutf7haswell14implementation22convert_latin1_to_utf8EPKcmPc: 412| 538| const char *buf, size_t len, char *utf8_output) const noexcept { 413| 538| std::pair ret = 414| 538| avx2_convert_latin1_to_utf8(buf, len, utf8_output); 415| 538| size_t converted_chars = ret.second - utf8_output; 416| | 417| 538| if (ret.first != buf + len) { ------------------ | Branch (417:7): [True: 534, False: 4] ------------------ 418| 534| const size_t scalar_converted_chars = scalar::latin1_to_utf8::convert( 419| 534| ret.first, len - (ret.first - buf), ret.second); 420| 534| converted_chars += scalar_converted_chars; 421| 534| } 422| | 423| 538| return converted_chars; 424| 538|} _ZNK7simdutf7haswell14implementation25convert_latin1_to_utf16leEPKcmPDs: 429| 80| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 430| 80| std::pair ret = 431| 80| avx2_convert_latin1_to_utf16(buf, len, utf16_output); 432| 80| if (ret.first == nullptr) { ------------------ | Branch (432:7): [True: 4, False: 76] ------------------ 433| 4| return 0; 434| 4| } 435| 76| size_t converted_chars = ret.second - utf16_output; 436| 76| if (ret.first != buf + len) { ------------------ | Branch (436:7): [True: 62, False: 14] ------------------ 437| 62| const size_t scalar_converted_chars = 438| 62| scalar::latin1_to_utf16::convert( 439| 62| ret.first, len - (ret.first - buf), ret.second); 440| 62| if (scalar_converted_chars == 0) { ------------------ | Branch (440:9): [True: 0, False: 62] ------------------ 441| 0| return 0; 442| 0| } 443| 62| converted_chars += scalar_converted_chars; 444| 62| } 445| 76| return converted_chars; 446| 76|} _ZNK7simdutf7haswell14implementation25convert_latin1_to_utf16beEPKcmPDs: 449| 80| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 450| 80| std::pair ret = 451| 80| avx2_convert_latin1_to_utf16(buf, len, utf16_output); 452| 80| if (ret.first == nullptr) { ------------------ | Branch (452:7): [True: 4, False: 76] ------------------ 453| 4| return 0; 454| 4| } 455| 76| size_t converted_chars = ret.second - utf16_output; 456| 76| if (ret.first != buf + len) { ------------------ | Branch (456:7): [True: 62, False: 14] ------------------ 457| 62| const size_t scalar_converted_chars = 458| 62| scalar::latin1_to_utf16::convert( 459| 62| ret.first, len - (ret.first - buf), ret.second); 460| 62| if (scalar_converted_chars == 0) { ------------------ | Branch (460:9): [True: 0, False: 62] ------------------ 461| 0| return 0; 462| 0| } 463| 62| converted_chars += scalar_converted_chars; 464| 62| } 465| 76| return converted_chars; 466| 76|} _ZNK7simdutf7haswell14implementation23convert_latin1_to_utf32EPKcmPDi: 471| 90| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 472| 90| std::pair ret = 473| 90| avx2_convert_latin1_to_utf32(buf, len, utf32_output); 474| 90| if (ret.first == nullptr) { ------------------ | Branch (474:7): [True: 4, False: 86] ------------------ 475| 4| return 0; 476| 4| } 477| 86| size_t converted_chars = ret.second - utf32_output; 478| 86| if (ret.first != buf + len) { ------------------ | Branch (478:7): [True: 58, False: 28] ------------------ 479| 58| const size_t scalar_converted_chars = scalar::latin1_to_utf32::convert( 480| 58| ret.first, len - (ret.first - buf), ret.second); 481| 58| if (scalar_converted_chars == 0) { ------------------ | Branch (481:9): [True: 0, False: 58] ------------------ 482| 0| return 0; 483| 0| } 484| 58| converted_chars += scalar_converted_chars; 485| 58| } 486| 86| return converted_chars; 487| 86|} _ZNK7simdutf7haswell14implementation22convert_utf8_to_latin1EPKcmPc: 492| 562| const char *buf, size_t len, char *latin1_output) const noexcept { 493| 562| utf8_to_latin1::validating_transcoder converter; 494| 562| return converter.convert(buf, len, latin1_output); 495| 562|} _ZNK7simdutf7haswell14implementation34convert_utf8_to_latin1_with_errorsEPKcmPc: 498| 516| const char *buf, size_t len, char *latin1_output) const noexcept { 499| 516| utf8_to_latin1::validating_transcoder converter; 500| 516| return converter.convert_with_errors(buf, len, latin1_output); 501| 516|} _ZNK7simdutf7haswell14implementation23convert_utf8_to_utf16leEPKcmPDs: 511| 660| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 512| 660| utf8_to_utf16::validating_transcoder converter; 513| 660| return converter.convert(buf, len, utf16_output); 514| 660|} _ZNK7simdutf7haswell14implementation23convert_utf8_to_utf16beEPKcmPDs: 517| 712| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 518| 712| utf8_to_utf16::validating_transcoder converter; 519| 712| return converter.convert(buf, len, utf16_output); 520| 712|} _ZNK7simdutf7haswell14implementation35convert_utf8_to_utf16le_with_errorsEPKcmPDs: 523| 574| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 524| 574| utf8_to_utf16::validating_transcoder converter; 525| 574| return converter.convert_with_errors(buf, len, 526| 574| utf16_output); 527| 574|} _ZNK7simdutf7haswell14implementation35convert_utf8_to_utf16be_with_errorsEPKcmPDs: 530| 702| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 531| 702| utf8_to_utf16::validating_transcoder converter; 532| 702| return converter.convert_with_errors(buf, len, utf16_output); 533| 702|} _ZNK7simdutf7haswell14implementation29convert_valid_utf8_to_utf16leEPKcmPDs: 536| 368| const char *input, size_t size, char16_t *utf16_output) const noexcept { 537| 368| return utf8_to_utf16::convert_valid(input, size, 538| 368| utf16_output); 539| 368|} _ZNK7simdutf7haswell14implementation29convert_valid_utf8_to_utf16beEPKcmPDs: 542| 438| const char *input, size_t size, char16_t *utf16_output) const noexcept { 543| 438| return utf8_to_utf16::convert_valid(input, size, 544| 438| utf16_output); 545| 438|} _ZNK7simdutf7haswell14implementation21convert_utf8_to_utf32EPKcmPDi: 550| 712| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 551| 712| utf8_to_utf32::validating_transcoder converter; 552| 712| return converter.convert(buf, len, utf32_output); 553| 712|} _ZNK7simdutf7haswell14implementation33convert_utf8_to_utf32_with_errorsEPKcmPDi: 556| 738| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 557| 738| utf8_to_utf32::validating_transcoder converter; 558| 738| return converter.convert_with_errors(buf, len, utf32_output); 559| 738|} _ZNK7simdutf7haswell14implementation27convert_valid_utf8_to_utf32EPKcmPDi: 562| 394| const char *input, size_t size, char32_t *utf32_output) const noexcept { 563| 394| return utf8_to_utf32::convert_valid(input, size, utf32_output); 564| 394|} _ZNK7simdutf7haswell14implementation25convert_utf16le_to_latin1EPKDsmPc: 569| 124| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 570| 124| std::pair ret = 571| 124| haswell::avx2_convert_utf16_to_latin1(buf, len, 572| 124| latin1_output); 573| 124| if (ret.first == nullptr) { ------------------ | Branch (573:7): [True: 34, False: 90] ------------------ 574| 34| return 0; 575| 34| } 576| 90| size_t saved_bytes = ret.second - latin1_output; 577| 90| if (ret.first != buf + len) { ------------------ | Branch (577:7): [True: 76, False: 14] ------------------ 578| 76| const size_t scalar_saved_bytes = 579| 76| scalar::utf16_to_latin1::convert( 580| 76| ret.first, len - (ret.first - buf), ret.second); 581| 76| if (scalar_saved_bytes == 0) { ------------------ | Branch (581:9): [True: 50, False: 26] ------------------ 582| 50| return 0; 583| 50| } 584| 26| saved_bytes += scalar_saved_bytes; 585| 26| } 586| 40| return saved_bytes; 587| 90|} _ZNK7simdutf7haswell14implementation25convert_utf16be_to_latin1EPKDsmPc: 590| 126| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 591| 126| std::pair ret = 592| 126| haswell::avx2_convert_utf16_to_latin1(buf, len, 593| 126| latin1_output); 594| 126| if (ret.first == nullptr) { ------------------ | Branch (594:7): [True: 26, False: 100] ------------------ 595| 26| return 0; 596| 26| } 597| 100| size_t saved_bytes = ret.second - latin1_output; 598| 100| if (ret.first != buf + len) { ------------------ | Branch (598:7): [True: 86, False: 14] ------------------ 599| 86| const size_t scalar_saved_bytes = 600| 86| scalar::utf16_to_latin1::convert( 601| 86| ret.first, len - (ret.first - buf), ret.second); 602| 86| if (scalar_saved_bytes == 0) { ------------------ | Branch (602:9): [True: 36, False: 50] ------------------ 603| 36| return 0; 604| 36| } 605| 50| saved_bytes += scalar_saved_bytes; 606| 50| } 607| 64| return saved_bytes; 608| 100|} _ZNK7simdutf7haswell14implementation37convert_utf16le_to_latin1_with_errorsEPKDsmPc: 612| 230| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 613| 230| std::pair ret = 614| 230| avx2_convert_utf16_to_latin1_with_errors( 615| 230| buf, len, latin1_output); 616| 230| if (ret.first.error) { ------------------ | Branch (616:7): [True: 92, False: 138] ------------------ 617| 92| return ret.first; 618| 92| } // Can return directly since scalar fallback already found correct 619| | // ret.first.count 620| 138| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (620:7): [True: 102, False: 36] ------------------ 621| 102| result scalar_res = 622| 102| scalar::utf16_to_latin1::convert_with_errors( 623| 102| buf + ret.first.count, len - ret.first.count, ret.second); 624| 102| if (scalar_res.error) { ------------------ | Branch (624:9): [True: 56, False: 46] ------------------ 625| 56| scalar_res.count += ret.first.count; 626| 56| return scalar_res; 627| 56| } else { 628| 46| ret.second += scalar_res.count; 629| 46| } 630| 102| } 631| 82| ret.first.count = 632| 82| ret.second - 633| 82| latin1_output; // Set count to the number of 8-bit code units written 634| 82| return ret.first; 635| 138|} _ZNK7simdutf7haswell14implementation37convert_utf16be_to_latin1_with_errorsEPKDsmPc: 639| 252| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 640| 252| std::pair ret = 641| 252| avx2_convert_utf16_to_latin1_with_errors(buf, len, 642| 252| latin1_output); 643| 252| if (ret.first.error) { ------------------ | Branch (643:7): [True: 102, False: 150] ------------------ 644| 102| return ret.first; 645| 102| } // Can return directly since scalar fallback already found correct 646| | // ret.first.count 647| 150| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (647:7): [True: 118, False: 32] ------------------ 648| 118| result scalar_res = 649| 118| scalar::utf16_to_latin1::convert_with_errors( 650| 118| buf + ret.first.count, len - ret.first.count, ret.second); 651| 118| if (scalar_res.error) { ------------------ | Branch (651:9): [True: 54, False: 64] ------------------ 652| 54| scalar_res.count += ret.first.count; 653| 54| return scalar_res; 654| 64| } else { 655| 64| ret.second += scalar_res.count; 656| 64| } 657| 118| } 658| 96| ret.first.count = 659| 96| ret.second - 660| 96| latin1_output; // Set count to the number of 8-bit code units written 661| 96| return ret.first; 662| 150|} _ZNK7simdutf7haswell14implementation23convert_utf16le_to_utf8EPKDsmPc: 679| 802| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 680| 802| std::pair ret = 681| 802| haswell::avx2_convert_utf16_to_utf8(buf, len, 682| 802| utf8_output); 683| 802| if (ret.first == nullptr) { ------------------ | Branch (683:7): [True: 106, False: 696] ------------------ 684| 106| return 0; 685| 106| } 686| 696| size_t saved_bytes = ret.second - utf8_output; 687| 696| if (ret.first != buf + len) { ------------------ | Branch (687:7): [True: 694, False: 2] ------------------ 688| 694| const size_t scalar_saved_bytes = 689| 694| scalar::utf16_to_utf8::convert( 690| 694| ret.first, len - (ret.first - buf), ret.second); 691| 694| if (scalar_saved_bytes == 0) { ------------------ | Branch (691:9): [True: 60, False: 634] ------------------ 692| 60| return 0; 693| 60| } 694| 634| saved_bytes += scalar_saved_bytes; 695| 634| } 696| 636| return saved_bytes; 697| 696|} _ZNK7simdutf7haswell14implementation23convert_utf16be_to_utf8EPKDsmPc: 700| 768| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 701| 768| std::pair ret = 702| 768| haswell::avx2_convert_utf16_to_utf8(buf, len, 703| 768| utf8_output); 704| 768| if (ret.first == nullptr) { ------------------ | Branch (704:7): [True: 90, False: 678] ------------------ 705| 90| return 0; 706| 90| } 707| 678| size_t saved_bytes = ret.second - utf8_output; 708| 678| if (ret.first != buf + len) { ------------------ | Branch (708:7): [True: 676, False: 2] ------------------ 709| 676| const size_t scalar_saved_bytes = 710| 676| scalar::utf16_to_utf8::convert( 711| 676| ret.first, len - (ret.first - buf), ret.second); 712| 676| if (scalar_saved_bytes == 0) { ------------------ | Branch (712:9): [True: 46, False: 630] ------------------ 713| 46| return 0; 714| 46| } 715| 630| saved_bytes += scalar_saved_bytes; 716| 630| } 717| 632| return saved_bytes; 718| 678|} _ZNK7simdutf7haswell14implementation35convert_utf16le_to_utf8_with_errorsEPKDsmPc: 742| 630| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 743| 630| return convert_utf16_to_utf8_with_details(buf, len, 744| 630| utf8_output); 745| 630|} _ZNK7simdutf7haswell14implementation35convert_utf16be_to_utf8_with_errorsEPKDsmPc: 748| 586| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 749| 586| return convert_utf16_to_utf8_with_details(buf, len, 750| 586| utf8_output); 751| 586|} _ZNK7simdutf7haswell14implementation29convert_valid_utf16le_to_utf8EPKDsmPc: 754| 416| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 755| 416| return convert_utf16le_to_utf8(buf, len, utf8_output); 756| 416|} _ZNK7simdutf7haswell14implementation29convert_valid_utf16be_to_utf8EPKDsmPc: 759| 442| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 760| 442| return convert_utf16be_to_utf8(buf, len, utf8_output); 761| 442|} _ZNK7simdutf7haswell14implementation21convert_utf32_to_utf8EPKDimPc: 766| 796| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 767| 796| std::pair ret = 768| 796| avx2_convert_utf32_to_utf8(buf, len, utf8_output); 769| 796| if (ret.first == nullptr) { ------------------ | Branch (769:7): [True: 278, False: 518] ------------------ 770| 278| return 0; 771| 278| } 772| 518| size_t saved_bytes = ret.second - utf8_output; 773| 518| if (ret.first != buf + len) { ------------------ | Branch (773:7): [True: 516, False: 2] ------------------ 774| 516| const size_t scalar_saved_bytes = scalar::utf32_to_utf8::convert( 775| 516| ret.first, len - (ret.first - buf), ret.second); 776| 516| if (scalar_saved_bytes == 0) { ------------------ | Branch (776:9): [True: 226, False: 290] ------------------ 777| 226| return 0; 778| 226| } 779| 290| saved_bytes += scalar_saved_bytes; 780| 290| } 781| 292| return saved_bytes; 782| 518|} _ZNK7simdutf7haswell14implementation23convert_utf32_to_latin1EPKDimPc: 787| 188| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 788| 188| std::pair ret = 789| 188| avx2_convert_utf32_to_latin1(buf, len, latin1_output); 790| 188| if (ret.first == nullptr) { ------------------ | Branch (790:7): [True: 32, False: 156] ------------------ 791| 32| return 0; 792| 32| } 793| 156| size_t saved_bytes = ret.second - latin1_output; 794| 156| if (ret.first != buf + len) { ------------------ | Branch (794:7): [True: 142, False: 14] ------------------ 795| 142| const size_t scalar_saved_bytes = scalar::utf32_to_latin1::convert( 796| 142| ret.first, len - (ret.first - buf), ret.second); 797| 142| if (scalar_saved_bytes == 0) { ------------------ | Branch (797:9): [True: 98, False: 44] ------------------ 798| 98| return 0; 799| 98| } 800| 44| saved_bytes += scalar_saved_bytes; 801| 44| } 802| 58| return saved_bytes; 803| 156|} _ZNK7simdutf7haswell14implementation35convert_utf32_to_latin1_with_errorsEPKDimPc: 806| 420| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 807| | // ret.first.count is always the position in the buffer, not the number of 808| | // code units written even if finished 809| 420| std::pair ret = 810| 420| avx2_convert_utf32_to_latin1_with_errors(buf, len, latin1_output); 811| 420| if (ret.first.count != len) { ------------------ | Branch (811:7): [True: 404, False: 16] ------------------ 812| 404| result scalar_res = scalar::utf32_to_latin1::convert_with_errors( 813| 404| buf + ret.first.count, len - ret.first.count, ret.second); 814| 404| if (scalar_res.error) { ------------------ | Branch (814:9): [True: 370, False: 34] ------------------ 815| 370| scalar_res.count += ret.first.count; 816| 370| return scalar_res; 817| 370| } else { 818| 34| ret.second += scalar_res.count; 819| 34| } 820| 404| } 821| 50| ret.first.count = 822| 50| ret.second - 823| 50| latin1_output; // Set count to the number of 8-bit code units written 824| 50| return ret.first; 825| 420|} _ZNK7simdutf7haswell14implementation33convert_utf32_to_utf8_with_errorsEPKDimPc: 835| 830| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 836| | // ret.first.count is always the position in the buffer, not the number of 837| | // code units written even if finished 838| 830| std::pair ret = 839| 830| haswell::avx2_convert_utf32_to_utf8_with_errors(buf, len, utf8_output); 840| 830| if (ret.first.count != len) { ------------------ | Branch (840:7): [True: 826, False: 4] ------------------ 841| 826| result scalar_res = scalar::utf32_to_utf8::convert_with_errors( 842| 826| buf + ret.first.count, len - ret.first.count, ret.second); 843| 826| if (scalar_res.error) { ------------------ | Branch (843:9): [True: 632, False: 194] ------------------ 844| 632| scalar_res.count += ret.first.count; 845| 632| return scalar_res; 846| 632| } else { 847| 194| ret.second += scalar_res.count; 848| 194| } 849| 826| } 850| 198| ret.first.count = 851| 198| ret.second - 852| 198| utf8_output; // Set count to the number of 8-bit code units written 853| 198| return ret.first; 854| 830|} _ZNK7simdutf7haswell14implementation24convert_utf16le_to_utf32EPKDsmPDi: 859| 384| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 860| 384| std::pair ret = 861| 384| haswell::avx2_convert_utf16_to_utf32(buf, len, 862| 384| utf32_output); 863| 384| if (ret.first == nullptr) { ------------------ | Branch (863:7): [True: 102, False: 282] ------------------ 864| 102| return 0; 865| 102| } 866| 282| size_t saved_bytes = ret.second - utf32_output; 867| 282| if (ret.first != buf + len) { ------------------ | Branch (867:7): [True: 238, False: 44] ------------------ 868| 238| const size_t scalar_saved_bytes = 869| 238| scalar::utf16_to_utf32::convert( 870| 238| ret.first, len - (ret.first - buf), ret.second); 871| 238| if (scalar_saved_bytes == 0) { ------------------ | Branch (871:9): [True: 74, False: 164] ------------------ 872| 74| return 0; 873| 74| } 874| 164| saved_bytes += scalar_saved_bytes; 875| 164| } 876| 208| return saved_bytes; 877| 282|} _ZNK7simdutf7haswell14implementation24convert_utf16be_to_utf32EPKDsmPDi: 880| 366| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 881| 366| std::pair ret = 882| 366| haswell::avx2_convert_utf16_to_utf32(buf, len, 883| 366| utf32_output); 884| 366| if (ret.first == nullptr) { ------------------ | Branch (884:7): [True: 60, False: 306] ------------------ 885| 60| return 0; 886| 60| } 887| 306| size_t saved_bytes = ret.second - utf32_output; 888| 306| if (ret.first != buf + len) { ------------------ | Branch (888:7): [True: 264, False: 42] ------------------ 889| 264| const size_t scalar_saved_bytes = 890| 264| scalar::utf16_to_utf32::convert( 891| 264| ret.first, len - (ret.first - buf), ret.second); 892| 264| if (scalar_saved_bytes == 0) { ------------------ | Branch (892:9): [True: 92, False: 172] ------------------ 893| 92| return 0; 894| 92| } 895| 172| saved_bytes += scalar_saved_bytes; 896| 172| } 897| 214| return saved_bytes; 898| 306|} _ZNK7simdutf7haswell14implementation36convert_utf16le_to_utf32_with_errorsEPKDsmPDi: 901| 414| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 902| | // ret.first.count is always the position in the buffer, not the number of 903| | // code units written even if finished 904| 414| std::pair ret = 905| 414| haswell::avx2_convert_utf16_to_utf32_with_errors( 906| 414| buf, len, utf32_output); 907| 414| if (ret.first.error) { ------------------ | Branch (907:7): [True: 124, False: 290] ------------------ 908| 124| return ret.first; 909| 124| } // Can return directly since scalar fallback already found correct 910| | // ret.first.count 911| 290| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (911:7): [True: 260, False: 30] ------------------ 912| 260| result scalar_res = 913| 260| scalar::utf16_to_utf32::convert_with_errors( 914| 260| buf + ret.first.count, len - ret.first.count, ret.second); 915| 260| if (scalar_res.error) { ------------------ | Branch (915:9): [True: 100, False: 160] ------------------ 916| 100| scalar_res.count += ret.first.count; 917| 100| return scalar_res; 918| 160| } else { 919| 160| ret.second += scalar_res.count; 920| 160| } 921| 260| } 922| 190| ret.first.count = 923| 190| ret.second - 924| 190| utf32_output; // Set count to the number of 8-bit code units written 925| 190| return ret.first; 926| 290|} _ZNK7simdutf7haswell14implementation36convert_utf16be_to_utf32_with_errorsEPKDsmPDi: 929| 300| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 930| | // ret.first.count is always the position in the buffer, not the number of 931| | // code units written even if finished 932| 300| std::pair ret = 933| 300| haswell::avx2_convert_utf16_to_utf32_with_errors( 934| 300| buf, len, utf32_output); 935| 300| if (ret.first.error) { ------------------ | Branch (935:7): [True: 54, False: 246] ------------------ 936| 54| return ret.first; 937| 54| } // Can return directly since scalar fallback already found correct 938| | // ret.first.count 939| 246| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (939:7): [True: 216, False: 30] ------------------ 940| 216| result scalar_res = 941| 216| scalar::utf16_to_utf32::convert_with_errors( 942| 216| buf + ret.first.count, len - ret.first.count, ret.second); 943| 216| if (scalar_res.error) { ------------------ | Branch (943:9): [True: 86, False: 130] ------------------ 944| 86| scalar_res.count += ret.first.count; 945| 86| return scalar_res; 946| 130| } else { 947| 130| ret.second += scalar_res.count; 948| 130| } 949| 216| } 950| 160| ret.first.count = 951| 160| ret.second - 952| 160| utf32_output; // Set count to the number of 8-bit code units written 953| 160| return ret.first; 954| 246|} _ZNK7simdutf7haswell14implementation27convert_valid_utf32_to_utf8EPKDimPc: 959| 192| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 960| 192| return convert_utf32_to_utf8(buf, len, utf8_output); 961| 192|} _ZNK7simdutf7haswell14implementation24convert_utf32_to_utf16leEPKDimPDs: 966| 722| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 967| 722| std::pair ret = 968| 722| avx2_convert_utf32_to_utf16(buf, len, utf16_output); 969| 722| if (ret.first == nullptr) { ------------------ | Branch (969:7): [True: 120, False: 602] ------------------ 970| 120| return 0; 971| 120| } 972| 602| size_t saved_bytes = ret.second - utf16_output; 973| 602| if (ret.first != buf + len) { ------------------ | Branch (973:7): [True: 598, False: 4] ------------------ 974| 598| const size_t scalar_saved_bytes = 975| 598| scalar::utf32_to_utf16::convert( 976| 598| ret.first, len - (ret.first - buf), ret.second); 977| 598| if (scalar_saved_bytes == 0) { ------------------ | Branch (977:9): [True: 166, False: 432] ------------------ 978| 166| return 0; 979| 166| } 980| 432| saved_bytes += scalar_saved_bytes; 981| 432| } 982| 436| return saved_bytes; 983| 602|} _ZNK7simdutf7haswell14implementation24convert_utf32_to_utf16beEPKDimPDs: 986| 528| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 987| 528| std::pair ret = 988| 528| avx2_convert_utf32_to_utf16(buf, len, utf16_output); 989| 528| if (ret.first == nullptr) { ------------------ | Branch (989:7): [True: 168, False: 360] ------------------ 990| 168| return 0; 991| 168| } 992| 360| size_t saved_bytes = ret.second - utf16_output; 993| 360| if (ret.first != buf + len) { ------------------ | Branch (993:7): [True: 358, False: 2] ------------------ 994| 358| const size_t scalar_saved_bytes = 995| 358| scalar::utf32_to_utf16::convert( 996| 358| ret.first, len - (ret.first - buf), ret.second); 997| 358| if (scalar_saved_bytes == 0) { ------------------ | Branch (997:9): [True: 164, False: 194] ------------------ 998| 164| return 0; 999| 164| } 1000| 194| saved_bytes += scalar_saved_bytes; 1001| 194| } 1002| 196| return saved_bytes; 1003| 360|} _ZNK7simdutf7haswell14implementation36convert_utf32_to_utf16le_with_errorsEPKDimPDs: 1006| 716| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1007| | // ret.first.count is always the position in the buffer, not the number of 1008| | // code units written even if finished 1009| 716| std::pair ret = 1010| 716| haswell::avx2_convert_utf32_to_utf16_with_errors( 1011| 716| buf, len, utf16_output); 1012| 716| if (ret.first.count != len) { ------------------ | Branch (1012:7): [True: 712, False: 4] ------------------ 1013| 712| result scalar_res = 1014| 712| scalar::utf32_to_utf16::convert_with_errors( 1015| 712| buf + ret.first.count, len - ret.first.count, ret.second); 1016| 712| if (scalar_res.error) { ------------------ | Branch (1016:9): [True: 530, False: 182] ------------------ 1017| 530| scalar_res.count += ret.first.count; 1018| 530| return scalar_res; 1019| 530| } else { 1020| 182| ret.second += scalar_res.count; 1021| 182| } 1022| 712| } 1023| 186| ret.first.count = 1024| 186| ret.second - 1025| 186| utf16_output; // Set count to the number of 8-bit code units written 1026| 186| return ret.first; 1027| 716|} _ZNK7simdutf7haswell14implementation36convert_utf32_to_utf16be_with_errorsEPKDimPDs: 1030| 606| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1031| | // ret.first.count is always the position in the buffer, not the number of 1032| | // code units written even if finished 1033| 606| std::pair ret = 1034| 606| haswell::avx2_convert_utf32_to_utf16_with_errors( 1035| 606| buf, len, utf16_output); 1036| 606| if (ret.first.count != len) { ------------------ | Branch (1036:7): [True: 602, False: 4] ------------------ 1037| 602| result scalar_res = 1038| 602| scalar::utf32_to_utf16::convert_with_errors( 1039| 602| buf + ret.first.count, len - ret.first.count, ret.second); 1040| 602| if (scalar_res.error) { ------------------ | Branch (1040:9): [True: 434, False: 168] ------------------ 1041| 434| scalar_res.count += ret.first.count; 1042| 434| return scalar_res; 1043| 434| } else { 1044| 168| ret.second += scalar_res.count; 1045| 168| } 1046| 602| } 1047| 172| ret.first.count = 1048| 172| ret.second - 1049| 172| utf16_output; // Set count to the number of 8-bit code units written 1050| 172| return ret.first; 1051| 606|} _ZNK7simdutf7haswell14implementation30convert_valid_utf32_to_utf16leEPKDimPDs: 1054| 290| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1055| 290| return convert_utf32_to_utf16le(buf, len, utf16_output); 1056| 290|} _ZNK7simdutf7haswell14implementation30convert_valid_utf32_to_utf16beEPKDimPDs: 1059| 86| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1060| 86| return convert_utf32_to_utf16be(buf, len, utf16_output); 1061| 86|} _ZNK7simdutf7haswell14implementation30convert_valid_utf16le_to_utf32EPKDsmPDi: 1064| 112| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1065| 112| return convert_utf16le_to_utf32(buf, len, utf32_output); 1066| 112|} _ZNK7simdutf7haswell14implementation30convert_valid_utf16be_to_utf32EPKDsmPDi: 1069| 140| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1070| 140| return convert_utf16be_to_utf32(buf, len, utf32_output); 1071| 140|} _ZNK7simdutf7haswell14implementation13count_utf16leEPKDsm: 1082| 1.30k| const char16_t *input, size_t length) const noexcept { 1083| 1.30k| return utf16::count_code_points(input, length); 1084| 1.30k|} _ZNK7simdutf7haswell14implementation13count_utf16beEPKDsm: 1087| 1.23k| const char16_t *input, size_t length) const noexcept { 1088| 1.23k| return utf16::count_code_points(input, length); 1089| 1.23k|} _ZNK7simdutf7haswell14implementation10count_utf8EPKcm: 1094| 3.76k|implementation::count_utf8(const char *in, size_t size) const noexcept { 1095| 3.76k| return utf8::count_code_points_bytemask(in, size); 1096| 3.76k|} _ZNK7simdutf7haswell14implementation23latin1_length_from_utf8EPKcm: 1101| 539| const char *buf, size_t len) const noexcept { 1102| 539| return count_utf8(buf, len); 1103| 539|} _ZNK7simdutf7haswell14implementation24utf8_length_from_utf16leEPKDsm: 1108| 727| const char16_t *input, size_t length) const noexcept { 1109| 727| return utf16::utf8_length_from_utf16_bytemask(input, 1110| 727| length); 1111| 727|} _ZNK7simdutf7haswell14implementation24utf8_length_from_utf16beEPKDsm: 1114| 696| const char16_t *input, size_t length) const noexcept { 1115| 696| return utf16::utf8_length_from_utf16_bytemask(input, length); 1116| 696|} _ZNK7simdutf7haswell14implementation25utf32_length_from_utf16leEPKDsm: 1121| 404| const char16_t *input, size_t length) const noexcept { 1122| 404| return utf16::utf32_length_from_utf16(input, length); 1123| 404|} _ZNK7simdutf7haswell14implementation25utf32_length_from_utf16beEPKDsm: 1126| 351| const char16_t *input, size_t length) const noexcept { 1127| 351| return utf16::utf32_length_from_utf16(input, length); 1128| 351|} _ZNK7simdutf7haswell14implementation22utf16_length_from_utf8EPKcm: 1133| 1.74k| const char *input, size_t length) const noexcept { 1134| 1.74k| return utf8::utf16_length_from_utf8_bytemask(input, length); 1135| 1.74k|} _ZNK7simdutf7haswell14implementation23utf8_length_from_latin1EPKcm: 1174| 269| const char *input, size_t len) const noexcept { 1175| 269| const uint8_t *data = reinterpret_cast(input); 1176| 269| size_t answer = len / sizeof(__m256i) * sizeof(__m256i); 1177| 269| size_t i = 0; 1178| 269| if (answer >= 2048) { // long strings optimization ------------------ | Branch (1178:7): [True: 118, False: 151] ------------------ 1179| 118| __m256i four_64bits = _mm256_setzero_si256(); 1180| 2.61k| while (i + sizeof(__m256i) <= len) { ------------------ | Branch (1180:12): [True: 2.50k, False: 118] ------------------ 1181| 2.50k| __m256i runner = _mm256_setzero_si256(); 1182| | // We can do up to 255 loops without overflow. 1183| 2.50k| size_t iterations = (len - i) / sizeof(__m256i); 1184| 2.50k| if (iterations > 255) { ------------------ | Branch (1184:11): [True: 2.38k, False: 118] ------------------ 1185| 2.38k| iterations = 255; 1186| 2.38k| } 1187| 2.50k| size_t max_i = i + iterations * sizeof(__m256i) - sizeof(__m256i); 1188| 154k| for (; i + 4 * sizeof(__m256i) <= max_i; i += 4 * sizeof(__m256i)) { ------------------ | Branch (1188:14): [True: 152k, False: 2.50k] ------------------ 1189| 152k| __m256i input1 = _mm256_loadu_si256((const __m256i *)(data + i)); 1190| 152k| __m256i input2 = 1191| 152k| _mm256_loadu_si256((const __m256i *)(data + i + sizeof(__m256i))); 1192| 152k| __m256i input3 = _mm256_loadu_si256( 1193| 152k| (const __m256i *)(data + i + 2 * sizeof(__m256i))); 1194| 152k| __m256i input4 = _mm256_loadu_si256( 1195| 152k| (const __m256i *)(data + i + 3 * sizeof(__m256i))); 1196| 152k| __m256i input12 = 1197| 152k| _mm256_add_epi8(_mm256_cmpgt_epi8(_mm256_setzero_si256(), input1), 1198| 152k| _mm256_cmpgt_epi8(_mm256_setzero_si256(), input2)); 1199| 152k| __m256i input23 = 1200| 152k| _mm256_add_epi8(_mm256_cmpgt_epi8(_mm256_setzero_si256(), input3), 1201| 152k| _mm256_cmpgt_epi8(_mm256_setzero_si256(), input4)); 1202| 152k| __m256i input1234 = _mm256_add_epi8(input12, input23); 1203| 152k| runner = _mm256_sub_epi8(runner, input1234); 1204| 152k| } 1205| 9.92k| for (; i <= max_i; i += sizeof(__m256i)) { ------------------ | Branch (1205:14): [True: 7.42k, False: 2.50k] ------------------ 1206| 7.42k| __m256i input_256_chunk = 1207| 7.42k| _mm256_loadu_si256((const __m256i *)(data + i)); 1208| 7.42k| runner = _mm256_sub_epi8( 1209| 7.42k| runner, _mm256_cmpgt_epi8(_mm256_setzero_si256(), input_256_chunk)); 1210| 7.42k| } 1211| 2.50k| four_64bits = _mm256_add_epi64( 1212| 2.50k| four_64bits, _mm256_sad_epu8(runner, _mm256_setzero_si256())); 1213| 2.50k| } 1214| 118| answer += _mm256_extract_epi64(four_64bits, 0) + 1215| 118| _mm256_extract_epi64(four_64bits, 1) + 1216| 118| _mm256_extract_epi64(four_64bits, 2) + 1217| 118| _mm256_extract_epi64(four_64bits, 3); 1218| 151| } else if (answer > 0) { ------------------ | Branch (1218:14): [True: 80, False: 71] ------------------ 1219| 1.36k| for (; i + sizeof(__m256i) <= len; i += sizeof(__m256i)) { ------------------ | Branch (1219:12): [True: 1.28k, False: 80] ------------------ 1220| 1.28k| __m256i latin = _mm256_loadu_si256((const __m256i *)(data + i)); 1221| 1.28k| uint32_t non_ascii = _mm256_movemask_epi8(latin); 1222| 1.28k| answer += count_ones(non_ascii); 1223| 1.28k| } 1224| 80| } 1225| 269| return answer + scalar::latin1::utf8_length_from_latin1( 1226| 269| reinterpret_cast(data + i), len - i); 1227| 269|} _ZNK7simdutf7haswell14implementation22utf8_length_from_utf32EPKDim: 1232| 819| const char32_t *input, size_t length) const noexcept { 1233| 819| return utf32::utf8_length_from_utf32(input, length); 1234| 819|} _ZNK7simdutf7haswell14implementation23utf16_length_from_utf32EPKDim: 1239| 1.29k| const char32_t *input, size_t length) const noexcept { 1240| 1.29k| const __m256i v_00000000 = _mm256_setzero_si256(); 1241| 1.29k| const __m256i v_ffff0000 = _mm256_set1_epi32((uint32_t)0xffff0000); 1242| 1.29k| size_t pos = 0; 1243| 1.29k| size_t count = 0; 1244| 672k| for (; pos + 8 <= length; pos += 8) { ------------------ | Branch (1244:10): [True: 671k, False: 1.29k] ------------------ 1245| 671k| __m256i in = _mm256_loadu_si256((__m256i *)(input + pos)); 1246| 671k| const __m256i surrogate_bytemask = 1247| 671k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_ffff0000), v_00000000); 1248| 671k| const uint32_t surrogate_bitmask = 1249| 671k| static_cast(_mm256_movemask_epi8(surrogate_bytemask)); 1250| 671k| size_t surrogate_count = (32 - count_ones(surrogate_bitmask)) / 4; 1251| 671k| count += 8 + surrogate_count; 1252| 671k| } 1253| 1.29k| return count + 1254| 1.29k| scalar::utf32::utf16_length_from_utf32(input + pos, length - pos); 1255| 1.29k|} _ZNK7simdutf7haswell14implementation22utf32_length_from_utf8EPKcm: 1260| 933| const char *input, size_t length) const noexcept { 1261| 933| return utf8::count_code_points(input, length); 1262| 933|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_18is_asciiERKNS1_4simd8simd8x64IhEE: 13| 4.64M|simdutf_really_inline bool is_ascii(const simd8x64 &input) { 14| 4.64M| return input.reduce_or().is_ascii(); 15| 4.64M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_124must_be_2_3_continuationENS1_4simd5simd8IhEES4_: 22| 3.20M| const simd8 prev3) { 23| 3.20M| simd8 is_third_byte = 24| 3.20M| prev2.saturating_sub(0xe0u - 0x80); // Only 111_____ will be > 0x80 25| 3.20M| simd8 is_fourth_byte = 26| 3.20M| prev3.saturating_sub(0xf0u - 0x80); // Only 1111____ will be > 0x80 27| 3.20M| return simd8(is_third_byte | is_fourth_byte); 28| 3.20M|} _ZN7simdutf7haswell34convert_utf16_to_utf8_with_detailsILNS_10endiannessE0EEENS_11full_resultEPKDsmPc: 722| 630| const char16_t *buf, size_t len, char *utf8_output) { 723| 630| std::pair ret = 724| 630| haswell::avx2_convert_utf16_to_utf8_with_errors(buf, len, 725| 630| utf8_output); 726| 630| if (ret.first.error) { ------------------ | Branch (726:7): [True: 128, False: 502] ------------------ 727| 128| return full_result(ret.first.error, ret.first.count, 728| 128| size_t(ret.second - utf8_output)); 729| 128| } 730| 502| if (ret.first.count != len) { ------------------ | Branch (730:7): [True: 498, False: 4] ------------------ 731| 498| full_result sres = 732| 498| scalar::utf16_to_utf8::convert_with_errors( 733| 498| buf + ret.first.count, len - ret.first.count, ret.second, 0); 734| 498| return full_result(sres.error, ret.first.count + sres.input_count, 735| 498| size_t(ret.second - utf8_output) + sres.output_count); 736| 498| } 737| 4| return full_result(error_code::SUCCESS, len, 738| 4| size_t(ret.second - utf8_output)); 739| 502|} _ZN7simdutf7haswell34convert_utf16_to_utf8_with_detailsILNS_10endiannessE1EEENS_11full_resultEPKDsmPc: 722| 586| const char16_t *buf, size_t len, char *utf8_output) { 723| 586| std::pair ret = 724| 586| haswell::avx2_convert_utf16_to_utf8_with_errors(buf, len, 725| 586| utf8_output); 726| 586| if (ret.first.error) { ------------------ | Branch (726:7): [True: 164, False: 422] ------------------ 727| 164| return full_result(ret.first.error, ret.first.count, 728| 164| size_t(ret.second - utf8_output)); 729| 164| } 730| 422| if (ret.first.count != len) { ------------------ | Branch (730:7): [True: 418, False: 4] ------------------ 731| 418| full_result sres = 732| 418| scalar::utf16_to_utf8::convert_with_errors( 733| 418| buf + ret.first.count, len - ret.first.count, ret.second, 0); 734| 418| return full_result(sres.error, ret.first.count + sres.input_count, 735| 418| size_t(ret.second - utf8_output) + sres.output_count); 736| 418| } 737| 4| return full_result(error_code::SUCCESS, len, 738| 4| size_t(ret.second - utf8_output)); 739| 422|} _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| 3.65M|simdutf_really_inline long long int count_ones(uint64_t input_num) { 15| | return _popcnt64(input_num); 16| 3.65M|} _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| 19.3k| simdutf_really_inline simd8() : base8_numeric() {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEC2Ev: 124| 19.3k| simdutf_really_inline base8_numeric() : base8() {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2Ev: 71| 19.3k| simdutf_really_inline base8() : base>() {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2Ev: 14| 19.3k| simdutf_really_inline base() : value{__m256i()} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IhE9reduce_orEv: 294| 4.64M| simdutf_really_inline simd8 reduce_or() const { 295| 4.64M| return this->chunks[0] | this->chunks[1]; 296| 4.64M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEorES5_: 50| 14.9M| simdutf_really_inline Child operator|(const Child other) const { 51| 14.9M| return _mm256_or_si256(*this, other); 52| 14.9M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2EDv4_x: 193| 163M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEC2EDv4_x: 126| 163M| : base8(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2EDv4_x: 73| 163M| simdutf_really_inline base8(const __m256i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2EDv4_x: 17| 163M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE8is_asciiEv: 234| 4.64M| simdutf_really_inline bool is_ascii() const { 235| 4.64M| return _mm256_movemask_epi8(*this) == 0; 236| 4.64M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEoRES5_: 59| 7.14M| simdutf_really_inline Child &operator|=(const Child other) { 60| 7.14M| auto this_cast = static_cast(this); 61| 7.14M| *this_cast = *this_cast | other; 62| 7.14M| return *this_cast; 63| 7.14M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi1EEENS4_IhEES8_: 83| 3.32M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 84| | return _mm256_alignr_epi8( 85| 3.32M| *this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N); 86| 3.32M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE3shrILi4EEES4_v: 245| 6.64M| template simdutf_really_inline simd8 shr() const { 246| 6.64M| return simd8(_mm256_srli_epi16(*this, N)) & uint8_t(0xFFu >> N); 247| 6.64M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_: 157| 9.96M| L replace15) const { 158| 9.96M| return lookup_16(simd8::repeat_16( 159| 9.96M| replace0, replace1, replace2, replace3, replace4, replace5, replace6, 160| 9.96M| replace7, replace8, replace9, replace10, replace11, replace12, 161| 9.96M| replace13, replace14, replace15)); 162| 9.96M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES8_: 148| 9.96M| simdutf_really_inline simd8 lookup_16(simd8 lookup_table) const { 149| 9.96M| return _mm256_shuffle_epi8(lookup_table, *this); 150| 9.96M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE9repeat_16Ehhhhhhhhhhhhhhhh: 118| 9.96M| T v14, T v15) { 119| 9.96M| return simd8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, 120| 9.96M| v14, v15, v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, 121| 9.96M| v12, v13, v14, v15); 122| 9.96M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2Ehhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh: 207| 9.96M| : simd8(_mm256_setr_epi8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, 208| 9.96M| v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, 209| 9.96M| v22, v23, v24, v25, v26, v27, v28, v29, v30, 210| 9.96M| v31)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2Eh: 195| 23.3M| simdutf_really_inline simd8(uint8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi2EEENS4_IhEES8_: 83| 3.20M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 84| | return _mm256_alignr_epi8( 85| 3.20M| *this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N); 86| 3.20M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi3EEENS4_IhEES8_: 83| 3.20M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 84| | return _mm256_alignr_epi8( 85| 3.20M| *this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N); 86| 3.20M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE14saturating_subES4_: 214| 7.22M| saturating_sub(const simd8 other) const { 215| 7.22M| return _mm256_subs_epu8(*this, other); 216| 7.22M| } 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| 15.8M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEeoES5_: 56| 3.20M| simdutf_really_inline Child operator^(const Child other) const { 57| 3.20M| return _mm256_xor_si256(*this, other); 58| 3.20M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2EPKh: 197| 828k| simdutf_really_inline simd8(const uint8_t values[32]) : simd8(load(values)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE4loadEPKh: 111| 10.1M| static simdutf_really_inline simd8 load(const T values[32]) { 112| 10.1M| return _mm256_loadu_si256(reinterpret_cast(values)); 113| 10.1M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE7gt_bitsES4_: 225| 828k| gt_bits(const simd8 other) const { 226| 828k| return this->saturating_sub(other); 227| 828k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEC2EDv4_x: 17| 79.1M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEcvRKDv4_xEv: 19| 111M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IbE10to_bitmaskEv: 99| 9.01M| simdutf_really_inline uint32_t to_bitmask() const { 100| 9.01M| return uint32_t(_mm256_movemask_epi8(value)); 101| 9.01M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE21any_bits_set_anywhereEv: 241| 2.19M| simdutf_really_inline bool any_bits_set_anywhere() const { 242| 2.19M| return !bits_not_set_anywhere(); 243| 2.19M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE21bits_not_set_anywhereEv: 237| 2.19M| simdutf_really_inline bool bits_not_set_anywhere() const { 238| 2.19M| return _mm256_testz_si256(*this, *this); 239| 2.19M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaEC2EPKa: 274| 5.92M| : chunks{simd8::load(ptr), 275| 5.92M| simd8::load(ptr + sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaE4loadEPKa: 111| 21.5M| static simdutf_really_inline simd8 load(const T values[32]) { 112| 21.5M| return _mm256_loadu_si256(reinterpret_cast(values)); 113| 21.5M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEC2EDv4_x: 169| 38.1M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaEC2EDv4_x: 126| 38.1M| : base8(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IaNS2_5simd8IbEEEC2EDv4_x: 73| 38.1M| simdutf_really_inline base8(const __m256i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEC2EDv4_x: 17| 38.1M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE8is_asciiEv: 298| 5.25M| simdutf_really_inline bool is_ascii() const { 299| 5.25M| return this->reduce_or().is_ascii(); 300| 5.25M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE9reduce_orEv: 294| 5.25M| simdutf_really_inline simd8 reduce_or() const { 295| 5.25M| return this->chunks[0] | this->chunks[1]; 296| 5.25M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEorES5_: 50| 5.25M| simdutf_really_inline Child operator|(const Child other) const { 51| 5.25M| return _mm256_or_si256(*this, other); 52| 5.25M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEcvRKDv4_xEv: 19| 60.7M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaE8is_asciiEv: 177| 5.25M| simdutf_really_inline bool is_ascii() const { 178| 5.25M| return _mm256_movemask_epi8(*this) == 0; 179| 5.25M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE5storeEPa: 277| 1.00M| simdutf_really_inline void store(T *ptr) const { 278| 1.00M| this->chunks[0].store(ptr + sizeof(simd8) * 0 / sizeof(T)); 279| 1.00M| this->chunks[1].store(ptr + sizeof(simd8) * 1 / sizeof(T)); 280| 1.00M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaE5storeEPa: 129| 2.00M| simdutf_really_inline void store(T dst[32]) const { 130| 2.00M| return _mm256_storeu_si256(reinterpret_cast<__m256i *>(dst), *this); 131| 2.00M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEcvNS3_IhEEEv: 256| 2.49M|simdutf_really_inline simd8::operator simd8() const { 257| 2.49M| return this->value; 258| 2.49M|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE2ltEa: 325| 1.07M| simdutf_really_inline uint64_t lt(const T m) const { 326| 1.07M| const simd8 mask = simd8::splat(m); 327| 1.07M| return simd8x64(this->chunks[0] < mask, this->chunks[1] < mask) 328| 1.07M| .to_bitmask(); 329| 1.07M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaE5splatEa: 105| 11.3M| static simdutf_really_inline simd8 splat(T _value) { 106| 11.3M| return _mm256_set1_epi8(_value); 107| 11.3M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEltES4_: 184| 2.14M| simdutf_really_inline simd8 operator<(const simd8 other) const { 185| 2.14M| return _mm256_cmpgt_epi8(other, *this); 186| 2.14M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IbEC2ENS2_5simd8IbEES6_: 272| 1.73M| : chunks{chunk0, chunk1} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IbE10to_bitmaskEv: 282| 1.73M| simdutf_really_inline uint64_t to_bitmask() const { 283| 1.73M| uint64_t r_lo = uint32_t(this->chunks[0].to_bitmask()); 284| 1.73M| uint64_t r_hi = this->chunks[1].to_bitmask(); 285| 1.73M| return r_lo | (r_hi << 32); 286| 1.73M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf32EPDi: 310| 1.00M| simdutf_really_inline void store_ascii_as_utf32(char32_t *ptr) const { 311| 1.00M| this->chunks[0].store_ascii_as_utf32(ptr + sizeof(simd8) * 0); 312| 1.00M| this->chunks[1].store_ascii_as_utf32(ptr + sizeof(simd8) * 1); 313| 1.00M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf32EPDi: 36| 2.01M| simdutf_really_inline void store_ascii_as_utf32(char32_t *ptr) const { 37| 2.01M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), 38| 2.01M| _mm256_cvtepu8_epi32(_mm256_castsi256_si128(*this))); 39| 2.01M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 8), 40| 2.01M| _mm256_cvtepu8_epi32(_mm256_castsi256_si128( 41| 2.01M| _mm256_srli_si256(*this, 8)))); 42| 2.01M| _mm256_storeu_si256( 43| 2.01M| reinterpret_cast<__m256i *>(ptr + 16), 44| 2.01M| _mm256_cvtepu8_epi32(_mm256_extractf128_si256(*this, 1))); 45| 2.01M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 24), 46| | _mm256_cvtepu8_epi32(_mm_srli_si128( 47| 2.01M| _mm256_extractf128_si256(*this, 1), 8))); 48| 2.01M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE4zeroEv: 108| 104k| static simdutf_really_inline simd8 zero() { 109| 104k| return _mm256_setzero_si256(); 110| 104k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEgtES4_: 181| 10.9M| simdutf_really_inline simd8 operator>(const simd8 other) const { 182| 10.9M| return _mm256_cmpgt_epi8(*this, other); 183| 10.9M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEC2Ea: 172| 9.65M| simdutf_really_inline simd8(int8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEmIENS2_5simd8IhEE: 137| 12.6M| simdutf_really_inline simd8 &operator-=(const simd8 other) { 138| 12.6M| *this = *this - other; 139| 12.6M| return *static_cast *>(this); 140| 12.6M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEmiENS2_5simd8IhEE: 134| 12.6M| simdutf_really_inline simd8 operator-(const simd8 other) const { 135| 12.6M| return _mm256_sub_epi8(*this, other); 136| 12.6M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd10sum_8bytesENS2_5simd8IhEE: 353| 49.6k|simdutf_really_inline simd64 sum_8bytes(const simd8 v) { 354| 49.6k| return _mm256_sad_epu8(v.value, simd8::zero()); 355| 49.6k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE9sum_bytesEv: 249| 2.54k| simdutf_really_inline uint64_t sum_bytes() const { 250| 2.54k| const auto tmp = _mm256_sad_epu8(value, _mm256_setzero_si256()); 251| | 252| 2.54k| return _mm256_extract_epi64(tmp, 0) + _mm256_extract_epi64(tmp, 1) + 253| 2.54k| _mm256_extract_epi64(tmp, 2) + _mm256_extract_epi64(tmp, 3); 254| 2.54k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEgeES4_: 229| 2.98M| operator>=(const simd8 other) const { 230| 2.98M| return other.min_val(*this) == other; 231| 2.98M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE7min_valES4_: 220| 2.98M| min_val(const simd8 other) const { 221| 2.98M| return _mm256_min_epu8(other, *this); 222| 2.98M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE2gtEa: 331| 665k| simdutf_really_inline uint64_t gt(const T m) const { 332| 665k| const simd8 mask = simd8::splat(m); 333| 665k| return simd8x64(this->chunks[0] > mask, this->chunks[1] > mask) 334| 665k| .to_bitmask(); 335| 665k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEC2EDv4_x: 17| 16.1M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEcvRKDv4_xEv: 19| 16.1M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE5splatEh: 105| 23.4M| static simdutf_really_inline simd8 splat(T _value) { 106| 23.4M| return _mm256_set1_epi8(_value); 107| 23.4M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEcvRKDv4_xEv: 19| 217M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdeqENS2_5simd8IhEES4_: 76| 8.52M| const simd8 rhs) { 77| 8.52M| return _mm256_cmpeq_epi8(lhs, rhs); 78| 8.52M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEanES5_: 53| 25.3M| simdutf_really_inline Child operator&(const Child other) const { 54| 25.3M| return _mm256_and_si256(*this, other); 55| 25.3M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IhEC2EPKh: 274| 4.64M| : chunks{simd8::load(ptr), 275| 4.64M| simd8::load(ptr + sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEanES5_: 53| 12.5M| simdutf_really_inline Child operator&(const Child other) const { 54| 12.5M| return _mm256_and_si256(*this, other); 55| 12.5M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 303| 1.06M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 304| 1.06M| this->chunks[0].template store_ascii_as_utf16(ptr + 305| 1.06M| sizeof(simd8) * 0); 306| 1.06M| this->chunks[1].template store_ascii_as_utf16(ptr + 307| 1.06M| sizeof(simd8) * 1); 308| 1.06M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 22| 2.12M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 23| 2.12M| __m256i first = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(*this)); 24| 2.12M| __m256i second = _mm256_cvtepu8_epi16(_mm256_extractf128_si256(*this, 1)); 25| 2.12M| if (big_endian) { ------------------ | Branch (25:9): [Folded, False: 2.12M] ------------------ 26| 0| const __m256i swap = _mm256_setr_epi8( 27| 0| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 28| 0| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 29| 0| first = _mm256_shuffle_epi8(first, swap); 30| 0| second = _mm256_shuffle_epi8(second, swap); 31| 0| } 32| 2.12M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), first); 33| 2.12M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 16), second); 34| 2.12M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 303| 1.11M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 304| 1.11M| this->chunks[0].template store_ascii_as_utf16(ptr + 305| 1.11M| sizeof(simd8) * 0); 306| 1.11M| this->chunks[1].template store_ascii_as_utf16(ptr + 307| 1.11M| sizeof(simd8) * 1); 308| 1.11M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 22| 2.22M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 23| 2.22M| __m256i first = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(*this)); 24| 2.22M| __m256i second = _mm256_cvtepu8_epi16(_mm256_extractf128_si256(*this, 1)); 25| 2.22M| if (big_endian) { ------------------ | Branch (25:9): [True: 2.22M, Folded] ------------------ 26| 2.22M| const __m256i swap = _mm256_setr_epi8( 27| 2.22M| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 28| 2.22M| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 29| 2.22M| first = _mm256_shuffle_epi8(first, swap); 30| 2.22M| second = _mm256_shuffle_epi8(second, swap); 31| 2.22M| } 32| 2.22M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), first); 33| 2.22M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 16), second); 34| 2.22M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEorES5_: 50| 4.62M| simdutf_really_inline Child operator|(const Child other) const { 51| 4.62M| return _mm256_or_si256(*this, other); 52| 4.62M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEC2EDv4_x: 97| 79.1M| : base16_numeric(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItEC2EDv4_x: 73| 79.1M| : base16(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6base16ItNS2_6simd16IbEEEC2EDv4_x: 20| 79.1M| : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE4packERKS4_S6_: 149| 5.50M| 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| 5.50M| const __m128i lo_0 = _mm256_extracti128_si256(v0, 0); 156| 5.50M| const __m128i lo_1 = _mm256_extracti128_si256(v1, 0); 157| | 158| | // get the 1st lanes 159| 5.50M| const __m128i hi_0 = _mm256_extracti128_si256(v0, 1); 160| 5.50M| const __m128i hi_1 = _mm256_extracti128_si256(v1, 1); 161| | 162| | // build new vectors (shuffle lanes) 163| 5.50M| const __m256i t0 = _mm256_set_m128i(lo_1, lo_0); ------------------ | | 4| 5.50M| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 5.50M| _mm256_castsi128_si256(xmm2), 2) ------------------ 164| 5.50M| const __m256i t1 = _mm256_set_m128i(hi_1, hi_0); ------------------ | | 4| 5.50M| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 5.50M| _mm256_castsi128_si256(xmm2), 2) ------------------ 165| | 166| | // pack code units in linear order from v0 and v1 167| 5.50M| return _mm256_packus_epi16(t0, t1); 168| 5.50M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd9simd16x32ItEC2EPKt: 202| 2.31M| : chunks{simd16::load(ptr), 203| 2.31M| simd16::load(ptr + sizeof(simd16) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItE4loadEPKt: 66| 17.9M| static simdutf_really_inline simd16 load(const T values[8]) { 67| 17.9M| return _mm256_loadu_si256(reinterpret_cast(values)); 68| 17.9M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd9simd16x32ItE10swap_bytesEv: 229| 1.20M| simdutf_really_inline void swap_bytes() { 230| 1.20M| this->chunks[0] = this->chunks[0].swap_bytes(); 231| 1.20M| this->chunks[1] = this->chunks[1].swap_bytes(); 232| 1.20M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE10swap_bytesEv: 139| 3.52M| simdutf_really_inline simd16 swap_bytes() const { 140| 3.52M| const __m256i swap = _mm256_setr_epi8( 141| 3.52M| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 142| 3.52M| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 143| 3.52M| return _mm256_shuffle_epi8(*this, swap); 144| 3.52M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItE5splatEt: 58| 16.6M| static simdutf_really_inline simd16 splat(T _value) { 59| 16.6M| return _mm256_set1_epi16(_value); 60| 16.6M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdeqENS2_6simd16ItEES4_: 26| 11.5M| const simd16 rhs) { 27| 11.5M| return _mm256_cmpeq_epi16(lhs, rhs); 28| 11.5M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16IbEC2EDv4_x: 45| 16.1M| simdutf_really_inline simd16(const __m256i _value) : base16(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6base16IbNS2_6simd16IbEEEC2EDv4_x: 20| 16.1M| : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd9simd16x32IbEC2ENS2_6simd16IbEES6_: 200| 2.31M| : chunks{chunk0, chunk1} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd9simd16x32IbE10to_bitmaskEv: 210| 2.31M| simdutf_really_inline uint64_t to_bitmask() const { 211| 2.31M| uint64_t r_lo = uint32_t(this->chunks[0].to_bitmask()); 212| 2.31M| uint64_t r_hi = this->chunks[1].to_bitmask(); 213| 2.31M| return r_lo | (r_hi << 32); 214| 2.31M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16IbE10to_bitmaskEv: 50| 4.62M| simdutf_really_inline bitmask_type to_bitmask() const { 51| 4.62M| return _mm256_movemask_epi8(*this); 52| 4.62M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE3shrILi8EEES4_v: 134| 5.40M| template simdutf_really_inline simd16 shr() const { 135| 5.40M| return simd16(_mm256_srli_epi16(*this, N)); 136| 5.40M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEleES4_: 117| 4.62M| operator<=(const simd16 other) const { 118| 4.62M| return other.max_val(*this) == other; 119| 4.62M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE7max_valES4_: 108| 4.62M| max_val(const simd16 other) const { 109| 4.62M| return _mm256_max_epu16(*this, other); 110| 4.62M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEC2EPKDs: 104| 11.0M| : simd16(load(reinterpret_cast(values))) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEC2Et: 100| 12.0M| simdutf_really_inline simd16(uint16_t _value) : simd16(splat(_value)) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd9simd16x32ItE12not_in_rangeEtt: 249| 2.31M| simdutf_really_inline uint64_t not_in_range(const T low, const T high) const { 250| 2.31M| const simd16 mask_low = simd16::splat(static_cast(low - 1)); 251| 2.31M| const simd16 mask_high = simd16::splat(static_cast(high + 1)); 252| 2.31M| return simd16x32( 253| 2.31M| (this->chunks[0] >= mask_high) | (this->chunks[0] <= mask_low), 254| 2.31M| (this->chunks[1] >= mask_high) | (this->chunks[1] <= mask_low)) 255| 2.31M| .to_bitmask(); 256| 2.31M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEgeES4_: 121| 4.62M| operator>=(const simd16 other) const { 122| 4.62M| return other.min_val(*this) == other; 123| 4.62M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE7min_valES4_: 112| 4.62M| min_val(const simd16 other) const { 113| 4.62M| return _mm256_min_epu16(*this, other); 114| 4.62M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItE4zeroEv: 62| 1.45k| static simdutf_really_inline simd16 zero() { 63| 1.45k| return _mm256_setzero_si256(); 64| 1.45k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd3minENS2_6simd16ItEES4_: 259| 4.61M|simd16 min(const simd16 a, simd16 b) { 260| 4.61M| return _mm256_min_epu16(a.value, b.value); 261| 4.61M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItEpLENS2_6simd16ItEE: 87| 6.92M| simdutf_really_inline simd16 &operator+=(const simd16 other) { 88| 6.92M| *this = *this + other; 89| 6.92M| return *static_cast *>(this); 90| 6.92M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItEplENS2_6simd16ItEE: 84| 6.92M| simdutf_really_inline simd16 operator+(const simd16 other) const { 85| 6.92M| return _mm256_add_epi16(*this, other); 86| 6.92M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE3sumEv: 170| 1.45k| simdutf_really_inline uint64_t sum() const { 171| 1.45k| const auto lo_u16 = _mm256_and_si256(value, _mm256_set1_epi32(0x0000ffff)); 172| 1.45k| const auto hi_u16 = _mm256_srli_epi32(value, 16); 173| 1.45k| const auto sum_u32 = _mm256_add_epi32(lo_u16, hi_u16); 174| | 175| 1.45k| const auto lo_u32 = 176| 1.45k| _mm256_and_si256(sum_u32, _mm256_set1_epi64x(0xffffffff)); 177| 1.45k| const auto hi_u32 = _mm256_srli_epi64(sum_u32, 32); 178| 1.45k| const auto sum_u64 = _mm256_add_epi64(lo_u32, hi_u32); 179| | 180| 1.45k| return uint64_t(_mm256_extract_epi64(sum_u64, 0)) + 181| 1.45k| uint64_t(_mm256_extract_epi64(sum_u64, 1)) + 182| 1.45k| uint64_t(_mm256_extract_epi64(sum_u64, 2)) + 183| | uint64_t(_mm256_extract_epi64(sum_u64, 3)); 184| 1.45k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE5splatEj: 46| 17.6k| simdutf_really_inline static simd32 splat(uint32_t v) { 47| 17.6k| return _mm256_set1_epi32(v); 48| 17.6k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjEC2EDv4_x: 9| 6.54M| simdutf_really_inline simd32(const __m256i v) : value(v) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE4zeroEv: 42| 5.39k| simdutf_really_inline static simd32 zero() { 43| 5.39k| return _mm256_setzero_si256(); 44| 5.39k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjEC2IDiEEPKT_: 13| 1.54M| : value(_mm256_loadu_si256(reinterpret_cast(ptr))) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE10swap_bytesEv: 27| 1.25M| simdutf_really_inline simd32 swap_bytes() const { 28| 1.25M| const __m256i shuffle = 29| 1.25M| _mm256_setr_epi8(3, 2, 1, 0, 7, 6, 5, 4, 8, 9, 10, 11, 15, 14, 13, 12, 30| 1.25M| 3, 2, 1, 0, 7, 6, 5, 4, 8, 9, 10, 11, 15, 14, 13, 12); 31| | 32| 1.25M| return _mm256_shuffle_epi8(value, shuffle); 33| 1.25M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd3maxENS2_6simd32IjEES4_: 80| 2.26M| const simd32 b) { 81| 2.26M| return _mm256_max_epu32(a.value, b.value); 82| 2.26M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdplENS2_6simd32IjEES4_: 90| 1.13M| const simd32 b) { 91| 1.13M| return _mm256_add_epi32(a.value, b.value); 92| 1.13M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdgtENS2_6simd32IjEES4_: 109| 3.94k| const simd32 b) { 110| 3.94k| return !(b >= a); 111| 3.94k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdntENS2_6simd32IbEE: 104| 3.94k|simdutf_really_inline simd32 operator!(const simd32 v) { 105| 3.94k| return _mm256_xor_si256(v.value, _mm256_set1_epi8(-1)); 106| 3.94k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IbEC2EDv4_x: 59| 388k| simdutf_really_inline simd32(const __m256i v) : value(v) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdgeENS2_6simd32IjEES4_: 100| 130k| const simd32 b) { 101| 130k| return _mm256_cmpeq_epi32(_mm256_max_epu32(a.value, b.value), a.value); 102| 130k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd32IbE3anyEv: 61| 130k| simdutf_really_inline bool any() const { 62| 130k| return _mm256_movemask_epi8(value) != 0; 63| 130k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdeqENS2_6simd32IjEES4_: 95| 126k| const simd32 b) { 96| 126k| return _mm256_cmpeq_epi32(a.value, b.value); 97| 126k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdanENS2_6simd32IjEES4_: 85| 996k| const simd32 a) { 86| 996k| return _mm256_and_si256(a.value, b.value); 87| 996k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdorIbEENS2_6simd32IT_EES6_S6_: 70| 126k| const simd32 b) { 71| 126k| return _mm256_or_si256(a.value, b.value); 72| 126k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd3minENS2_6simd32IjEES4_: 75| 869k| const simd32 a) { 76| 869k| return _mm256_min_epu32(a.value, b.value); 77| 869k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjEpLES4_: 36| 869k| simdutf_really_inline simd32 &operator+=(const simd32 other) { 37| 869k| value = _mm256_add_epi32(value, other.value); 38| 869k| return *this; 39| 869k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE3sumEv: 15| 599| simdutf_really_inline uint64_t sum() const { 16| 599| const __m256i mask = _mm256_set1_epi64x(0xffffffff); 17| 599| const __m256i t0 = _mm256_and_si256(value, mask); 18| 599| const __m256i t1 = _mm256_srli_epi64(value, 32); 19| 599| const __m256i t2 = _mm256_add_epi64(t0, t1); 20| | 21| 599| return uint64_t(_mm256_extract_epi64(t2, 0)) + 22| 599| uint64_t(_mm256_extract_epi64(t2, 1)) + 23| 599| uint64_t(_mm256_extract_epi64(t2, 2)) + 24| | uint64_t(_mm256_extract_epi64(t2, 3)); 25| 599| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd64ImE4zeroEv: 27| 5.50k| simdutf_really_inline static simd64 zero() { 28| 5.50k| return _mm256_setzero_si256(); 29| 5.50k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd64ImEC2EDv4_x: 9| 55.1k| simdutf_really_inline simd64(const __m256i v) : value(v) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd64ImEpLES4_: 21| 49.6k| simdutf_really_inline simd64 &operator+=(const simd64 other) { 22| 49.6k| value = _mm256_add_epi64(value, other.value); 23| 49.6k| return *this; 24| 49.6k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd64ImE3sumEv: 15| 5.50k| simdutf_really_inline uint64_t sum() const { 16| 5.50k| return _mm256_extract_epi64(value, 0) + _mm256_extract_epi64(value, 1) + 17| 5.50k| _mm256_extract_epi64(value, 2) + _mm256_extract_epi64(value, 3); 18| 5.50k| } _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| 4.33M|simdutf_really_inline long long int count_ones(uint64_t input_num) { 15| | return _popcnt64(input_num); 16| 4.33M|} _ZN7simdutf8westmere14implementationC2Ev: 18| 1| : simdutf::implementation("westmere", "Intel/AMD SSE4.2", 19| 1| internal::instruction_set::SSE42) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2Ev: 183| 19.3k| simdutf_really_inline simd8() : base8_numeric() {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEC2Ev: 118| 19.3k| simdutf_really_inline base8_numeric() : base8() {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2Ev: 68| 19.3k| simdutf_really_inline base8() : base>() {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2Ev: 13| 19.3k| simdutf_really_inline base() : value{__m128i()} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IhE9reduce_orEv: 290| 4.64M| simdutf_really_inline simd8 reduce_or() const { 291| 4.64M| return (this->chunks[0] | this->chunks[1]) | 292| 4.64M| (this->chunks[2] | this->chunks[3]); 293| 4.64M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEorES5_: 42| 30.7M| simdutf_really_inline Child operator|(const Child other) const { 43| 30.7M| return _mm_or_si128(*this, other); 44| 30.7M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2EDv2_x: 185| 314M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEC2EDv2_x: 120| 314M| : base8(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2EDv2_x: 69| 314M| simdutf_really_inline base8(const __m128i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2EDv2_x: 16| 314M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE8is_asciiEv: 228| 4.64M| simdutf_really_inline bool is_ascii() const { 229| 4.64M| return _mm_movemask_epi8(*this) == 0; 230| 4.64M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEoRES5_: 51| 10.4M| simdutf_really_inline Child &operator|=(const Child other) { 52| 10.4M| auto this_cast = static_cast(this); 53| 10.4M| *this_cast = *this_cast | other; 54| 10.4M| return *this_cast; 55| 10.4M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi1EEENS4_IhEES8_: 79| 6.64M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 80| | return _mm_alignr_epi8(*this, prev_chunk, 16 - N); 81| 6.64M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE3shrILi4EEES4_v: 238| 13.2M| template simdutf_really_inline simd8 shr() const { 239| 13.2M| return simd8(_mm_srli_epi16(*this, N)) & uint8_t(0xFFu >> N); 240| 13.2M| } 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| 19.9M| L replace15) const { 152| 19.9M| return lookup_16(simd8::repeat_16( 153| 19.9M| replace0, replace1, replace2, replace3, replace4, replace5, replace6, 154| 19.9M| replace7, replace8, replace9, replace10, replace11, replace12, 155| 19.9M| replace13, replace14, replace15)); 156| 19.9M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES8_: 142| 19.9M| simdutf_really_inline simd8 lookup_16(simd8 lookup_table) const { 143| 19.9M| return _mm_shuffle_epi8(lookup_table, *this); 144| 19.9M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE9repeat_16Ehhhhhhhhhhhhhhhh: 113| 19.9M| T v14, T v15) { 114| 19.9M| return simd8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, 115| 19.9M| v14, v15); 116| 19.9M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2Ehhhhhhhhhhhhhhhh: 196| 19.9M| : simd8(_mm_setr_epi8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, 197| 19.9M| v12, v13, v14, v15)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2Eh: 188| 45.9M| simdutf_really_inline simd8(uint8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi2EEENS4_IhEES8_: 79| 6.40M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 80| | return _mm_alignr_epi8(*this, prev_chunk, 16 - N); 81| 6.40M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi3EEENS4_IhEES8_: 79| 6.40M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 80| | return _mm_alignr_epi8(*this, prev_chunk, 16 - N); 81| 6.40M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE14saturating_subES4_: 201| 13.6M| saturating_sub(const simd8 other) const { 202| 13.6M| return _mm_subs_epu8(*this, other); 203| 13.6M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IbEC2EDv2_x: 91| 44.1M| simdutf_really_inline simd8(const __m128i _value) : base8(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IbNS2_5simd8IbEEEC2EDv2_x: 69| 44.1M| simdutf_really_inline base8(const __m128i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEC2EDv2_x: 16| 44.1M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEcvRKDv2_xEv: 18| 44.1M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEeoES5_: 48| 6.40M| simdutf_really_inline Child operator^(const Child other) const { 49| 6.40M| return _mm_xor_si128(*this, other); 50| 6.40M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2EPKh: 190| 828k| simdutf_really_inline simd8(const uint8_t *values) : simd8(load(values)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE4loadEPKh: 106| 19.4M| static simdutf_really_inline simd8 load(const T values[16]) { 107| 19.4M| return _mm_loadu_si128(reinterpret_cast(values)); 108| 19.4M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE7gt_bitsES4_: 212| 828k| gt_bits(const simd8 other) const { 213| 828k| return this->saturating_sub(other); 214| 828k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEC2EDv2_x: 16| 128M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEcvRKDv2_xEv: 18| 173M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IbE10to_bitmaskEv: 95| 12.4M| simdutf_really_inline int to_bitmask() const { 96| 12.4M| return _mm_movemask_epi8(*this); 97| 12.4M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE21any_bits_set_anywhereEv: 235| 2.19M| simdutf_really_inline bool any_bits_set_anywhere() const { 236| 2.19M| return !bits_not_set_anywhere(); 237| 2.19M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE21bits_not_set_anywhereEv: 232| 2.19M| simdutf_really_inline bool bits_not_set_anywhere() const { 233| 2.19M| return _mm_testz_si128(*this, *this); 234| 2.19M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaEC2EPKa: 270| 5.92M| : chunks{simd8::load(ptr), 271| 5.92M| simd8::load(ptr + sizeof(simd8) / sizeof(T)), 272| 5.92M| simd8::load(ptr + 2 * sizeof(simd8) / sizeof(T)), 273| 5.92M| simd8::load(ptr + 3 * sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaE4loadEPKa: 106| 43.0M| static simdutf_really_inline simd8 load(const T values[16]) { 107| 43.0M| return _mm_loadu_si128(reinterpret_cast(values)); 108| 43.0M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEC2EDv2_x: 163| 79.8M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaEC2EDv2_x: 120| 79.8M| : base8(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IaNS2_5simd8IbEEEC2EDv2_x: 69| 79.8M| simdutf_really_inline base8(const __m128i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEC2EDv2_x: 16| 79.8M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE8is_asciiEv: 295| 5.25M| simdutf_really_inline bool is_ascii() const { 296| 5.25M| return this->reduce_or().is_ascii(); 297| 5.25M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE9reduce_orEv: 290| 5.25M| simdutf_really_inline simd8 reduce_or() const { 291| 5.25M| return (this->chunks[0] | this->chunks[1]) | 292| 5.25M| (this->chunks[2] | this->chunks[3]); 293| 5.25M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEorES5_: 42| 15.7M| simdutf_really_inline Child operator|(const Child other) const { 43| 15.7M| return _mm_or_si128(*this, other); 44| 15.7M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEcvRKDv2_xEv: 18| 126M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaE8is_asciiEv: 168| 5.25M| simdutf_really_inline bool is_ascii() const { 169| 5.25M| return _mm_movemask_epi8(*this) == 0; 170| 5.25M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE5storeEPa: 275| 1.00M| simdutf_really_inline void store(T *ptr) const { 276| 1.00M| this->chunks[0].store(ptr + sizeof(simd8) * 0 / sizeof(T)); 277| 1.00M| this->chunks[1].store(ptr + sizeof(simd8) * 1 / sizeof(T)); 278| 1.00M| this->chunks[2].store(ptr + sizeof(simd8) * 2 / sizeof(T)); 279| 1.00M| this->chunks[3].store(ptr + sizeof(simd8) * 3 / sizeof(T)); 280| 1.00M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaE5storeEPa: 123| 4.01M| simdutf_really_inline void store(T dst[16]) const { 124| 4.01M| return _mm_storeu_si128(reinterpret_cast<__m128i *>(dst), *this); 125| 4.01M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEcvNS3_IhEEEv: 251| 5.82M|simdutf_really_inline simd8::operator simd8() const { 252| 5.82M| return this->value; 253| 5.82M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE2ltEa: 326| 1.07M| simdutf_really_inline uint64_t lt(const T m) const { 327| 1.07M| const simd8 mask = simd8::splat(m); 328| 1.07M| return simd8x64(this->chunks[0] < mask, this->chunks[1] < mask, 329| 1.07M| this->chunks[2] < mask, this->chunks[3] < mask) 330| 1.07M| .to_bitmask(); 331| 1.07M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaE5splatEa: 102| 21.0M| static simdutf_really_inline simd8 splat(T _value) { 103| 21.0M| return _mm_set1_epi8(_value); 104| 21.0M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEltES4_: 176| 4.28M| simdutf_really_inline simd8 operator<(const simd8 other) const { 177| 4.28M| return _mm_cmpgt_epi8(other, *this); 178| 4.28M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IbEC2ENS2_5simd8IbEES6_S6_S6_: 268| 1.73M| : chunks{chunk0, chunk1, chunk2, chunk3} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IbE10to_bitmaskEv: 318| 1.73M| simdutf_really_inline uint64_t to_bitmask() const { 319| 1.73M| uint64_t r0 = uint32_t(this->chunks[0].to_bitmask()); 320| 1.73M| uint64_t r1 = this->chunks[1].to_bitmask(); 321| 1.73M| uint64_t r2 = this->chunks[2].to_bitmask(); 322| 1.73M| uint64_t r3 = this->chunks[3].to_bitmask(); 323| 1.73M| return r0 | (r1 << 16) | (r2 << 32) | (r3 << 48); 324| 1.73M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf32EPDi: 311| 1.00M| simdutf_really_inline void store_ascii_as_utf32(char32_t *ptr) const { 312| 1.00M| this->chunks[0].store_ascii_as_utf32(ptr + sizeof(simd8) * 0); 313| 1.00M| this->chunks[1].store_ascii_as_utf32(ptr + sizeof(simd8) * 1); 314| 1.00M| this->chunks[2].store_ascii_as_utf32(ptr + sizeof(simd8) * 2); 315| 1.00M| this->chunks[3].store_ascii_as_utf32(ptr + sizeof(simd8) * 3); 316| 1.00M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf32EPDi: 32| 4.02M| simdutf_really_inline void store_ascii_as_utf32(char32_t *p) const { 33| 4.02M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p), _mm_cvtepu8_epi32(*this)); 34| 4.02M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 4), 35| 4.02M| _mm_cvtepu8_epi32(_mm_srli_si128(*this, 4))); 36| 4.02M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 8), 37| 4.02M| _mm_cvtepu8_epi32(_mm_srli_si128(*this, 8))); 38| 4.02M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 12), 39| | _mm_cvtepu8_epi32(_mm_srli_si128(*this, 12))); 40| 4.02M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE4zeroEv: 105| 204k| static simdutf_really_inline simd8 zero() { return _mm_setzero_si128(); } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEgtES4_: 173| 21.9M| simdutf_really_inline simd8 operator>(const simd8 other) const { 174| 21.9M| return _mm_cmpgt_epi8(*this, other); 175| 21.9M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEC2Ea: 165| 19.3M| simdutf_really_inline simd8(int8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEmIENS2_5simd8IhEE: 134| 25.2M| simdutf_really_inline simd8 &operator-=(const simd8 other) { 135| 25.2M| *this = *this - other; 136| 25.2M| return *static_cast *>(this); 137| 25.2M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEmiENS2_5simd8IhEE: 131| 25.2M| simdutf_really_inline simd8 operator-(const simd8 other) const { 132| 25.2M| return _mm_sub_epi8(*this, other); 133| 25.2M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd10sum_8bytesENS2_5simd8IhEE: 366| 99.5k|simdutf_really_inline simd64 sum_8bytes(const simd8 v) { 367| 99.5k| return _mm_sad_epu8(v.value, simd8::zero()); 368| 99.5k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE9sum_bytesEv: 245| 3.52k| simdutf_really_inline uint64_t sum_bytes() const { 246| 3.52k| const auto tmp = _mm_sad_epu8(value, _mm_setzero_si128()); 247| 3.52k| return _mm_extract_epi64(tmp, 0) + _mm_extract_epi64(tmp, 1); 248| 3.52k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEgeES4_: 217| 5.97M| operator>=(const simd8 other) const { 218| 5.97M| return other.min_val(*this) == other; 219| 5.97M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE7min_valES4_: 207| 5.97M| min_val(const simd8 other) const { 208| 5.97M| return _mm_min_epu8(*this, other); 209| 5.97M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE2gtEa: 333| 665k| simdutf_really_inline uint64_t gt(const T m) const { 334| 665k| const simd8 mask = simd8::splat(m); 335| 665k| return simd8x64(this->chunks[0] > mask, this->chunks[1] > mask, 336| 665k| this->chunks[2] > mask, this->chunks[3] > mask) 337| 665k| .to_bitmask(); 338| 665k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEC2EDv2_x: 16| 32.3M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEcvRKDv2_xEv: 18| 32.3M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE5splatEh: 102| 45.9M| static simdutf_really_inline simd8 splat(T _value) { 103| 45.9M| return _mm_set1_epi8(_value); 104| 45.9M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEcvRKDv2_xEv: 18| 385M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_5simd8IhEES4_: 72| 11.5M| const simd8 rhs) { 73| 11.5M| return _mm_cmpeq_epi8(lhs, rhs); 74| 11.5M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEanES5_: 45| 45.1M| simdutf_really_inline Child operator&(const Child other) const { 46| 45.1M| return _mm_and_si128(*this, other); 47| 45.1M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IhEC2EPKh: 270| 4.64M| : chunks{simd8::load(ptr), 271| 4.64M| simd8::load(ptr + sizeof(simd8) / sizeof(T)), 272| 4.64M| simd8::load(ptr + 2 * sizeof(simd8) / sizeof(T)), 273| 4.64M| simd8::load(ptr + 3 * sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEanES5_: 45| 19.4M| simdutf_really_inline Child operator&(const Child other) const { 46| 19.4M| return _mm_and_si128(*this, other); 47| 19.4M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 300| 1.06M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 301| 1.06M| this->chunks[0].template store_ascii_as_utf16(ptr + 302| 1.06M| sizeof(simd8) * 0); 303| 1.06M| this->chunks[1].template store_ascii_as_utf16(ptr + 304| 1.06M| sizeof(simd8) * 1); 305| 1.06M| this->chunks[2].template store_ascii_as_utf16(ptr + 306| 1.06M| sizeof(simd8) * 2); 307| 1.06M| this->chunks[3].template store_ascii_as_utf16(ptr + 308| 1.06M| sizeof(simd8) * 3); 309| 1.06M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 20| 4.24M| simdutf_really_inline void store_ascii_as_utf16(char16_t *p) const { 21| 4.24M| __m128i first = _mm_cvtepu8_epi16(*this); 22| 4.24M| __m128i second = _mm_cvtepu8_epi16(_mm_srli_si128(*this, 8)); 23| 4.24M| if (big_endian) { ------------------ | Branch (23:9): [Folded, False: 4.24M] ------------------ 24| 0| const __m128i swap = 25| 0| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 26| 0| first = _mm_shuffle_epi8(first, swap); 27| 0| second = _mm_shuffle_epi8(second, swap); 28| 0| } 29| 4.24M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p), first); 30| 4.24M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 8), second); 31| 4.24M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 300| 1.11M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 301| 1.11M| this->chunks[0].template store_ascii_as_utf16(ptr + 302| 1.11M| sizeof(simd8) * 0); 303| 1.11M| this->chunks[1].template store_ascii_as_utf16(ptr + 304| 1.11M| sizeof(simd8) * 1); 305| 1.11M| this->chunks[2].template store_ascii_as_utf16(ptr + 306| 1.11M| sizeof(simd8) * 2); 307| 1.11M| this->chunks[3].template store_ascii_as_utf16(ptr + 308| 1.11M| sizeof(simd8) * 3); 309| 1.11M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 20| 4.45M| simdutf_really_inline void store_ascii_as_utf16(char16_t *p) const { 21| 4.45M| __m128i first = _mm_cvtepu8_epi16(*this); 22| 4.45M| __m128i second = _mm_cvtepu8_epi16(_mm_srli_si128(*this, 8)); 23| 4.45M| if (big_endian) { ------------------ | Branch (23:9): [True: 4.45M, Folded] ------------------ 24| 4.45M| const __m128i swap = 25| 4.45M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 26| 4.45M| first = _mm_shuffle_epi8(first, swap); 27| 4.45M| second = _mm_shuffle_epi8(second, swap); 28| 4.45M| } 29| 4.45M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p), first); 30| 4.45M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 8), second); 31| 4.45M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd12as_vector_u8ENS2_6simd32IjEE: 370| 5.80k|simdutf_really_inline simd8 as_vector_u8(const simd32 v) { 371| 5.80k| return simd8(v.value); 372| 5.80k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEorES5_: 42| 9.25M| simdutf_really_inline Child operator|(const Child other) const { 43| 9.25M| return _mm_or_si128(*this, other); 44| 9.25M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEC2EDv2_x: 79| 128M| : base16_numeric(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItEC2EDv2_x: 54| 128M| : base16(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6base16ItNS2_6simd16IbEEEC2EDv2_x: 8| 128M| : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE4packERKS4_S6_: 130| 5.50M| const simd16 &v1) { 131| 5.50M| return _mm_packus_epi16(v0, v1); 132| 5.50M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd9simd16x32ItEC2EPKt: 164| 2.31M| : chunks{simd16::load(ptr), 165| 2.31M| simd16::load(ptr + sizeof(simd16) / sizeof(T)), 166| 2.31M| simd16::load(ptr + 2 * sizeof(simd16) / sizeof(T)), 167| 2.31M| simd16::load(ptr + 3 * sizeof(simd16) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItE4loadEPKt: 47| 24.8M| static simdutf_really_inline simd16 load(const T values[8]) { 48| 24.8M| return _mm_loadu_si128(reinterpret_cast(values)); 49| 24.8M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd9simd16x32ItE10swap_bytesEv: 200| 1.20M| simdutf_really_inline void swap_bytes() { 201| 1.20M| this->chunks[0] = this->chunks[0].swap_bytes(); 202| 1.20M| this->chunks[1] = this->chunks[1].swap_bytes(); 203| 1.20M| this->chunks[2] = this->chunks[2].swap_bytes(); 204| 1.20M| this->chunks[3] = this->chunks[3].swap_bytes(); 205| 1.20M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE10swap_bytesEv: 121| 7.04M| simdutf_really_inline simd16 swap_bytes() const { 122| 7.04M| const __m128i swap = 123| 7.04M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 124| 7.04M| return _mm_shuffle_epi8(*this, swap); 125| 7.04M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItE5splatEt: 41| 25.8M| static simdutf_really_inline simd16 splat(T _value) { 42| 25.8M| return _mm_set1_epi16(_value); 43| 25.8M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_6simd16ItEES4_: 11| 23.1M| const simd16 rhs) { 12| 23.1M| return _mm_cmpeq_epi16(lhs, rhs); 13| 23.1M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16IbEC2EDv2_x: 28| 32.3M| simdutf_really_inline simd16(const __m128i _value) : base16(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6base16IbNS2_6simd16IbEEEC2EDv2_x: 8| 32.3M| : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd9simd16x32IbEC2ENS2_6simd16IbEES6_S6_S6_: 162| 2.31M| : chunks{chunk0, chunk1, chunk2, chunk3} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd9simd16x32IbE10to_bitmaskEv: 192| 2.31M| simdutf_really_inline uint64_t to_bitmask() const { 193| 2.31M| uint64_t r0 = uint32_t(this->chunks[0].to_bitmask()); 194| 2.31M| uint64_t r1 = this->chunks[1].to_bitmask(); 195| 2.31M| uint64_t r2 = this->chunks[2].to_bitmask(); 196| 2.31M| uint64_t r3 = this->chunks[3].to_bitmask(); 197| 2.31M| return r0 | (r1 << 16) | (r2 << 32) | (r3 << 48); 198| 2.31M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16IbE10to_bitmaskEv: 33| 9.25M| simdutf_really_inline int to_bitmask() const { 34| 9.25M| return _mm_movemask_epi8(*this); 35| 9.25M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEgeES4_: 104| 9.25M| operator>=(const simd16 other) const { 105| 9.25M| return other.min_val(*this) == other; 106| 9.25M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE7min_valES4_: 95| 9.25M| min_val(const simd16 other) const { 96| 9.25M| return _mm_min_epu16(*this, other); 97| 9.25M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE3shrILi8EEES4_v: 116| 5.40M| template simdutf_really_inline simd16 shr() const { 117| 5.40M| return simd16(_mm_srli_epi16(*this, N)); 118| 5.40M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEleES4_: 100| 9.25M| operator<=(const simd16 other) const { 101| 9.25M| return other.max_val(*this) == other; 102| 9.25M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE7max_valES4_: 90| 9.25M| max_val(const simd16 other) const { 91| 9.25M| return _mm_max_epu16(*this, other); 92| 9.25M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEC2EPKDs: 86| 11.0M| : simd16(load(reinterpret_cast(values))) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEC2Et: 82| 21.2M| simdutf_really_inline simd16(uint16_t _value) : simd16(splat(_value)) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd9simd16x32ItE12not_in_rangeEtt: 228| 2.31M| simdutf_really_inline uint64_t not_in_range(const T low, const T high) const { 229| 2.31M| const simd16 mask_low = simd16::splat(static_cast(low - 1)); 230| 2.31M| const simd16 mask_high = simd16::splat(static_cast(high + 1)); 231| 2.31M| return simd16x32( 232| 2.31M| (this->chunks[0] >= mask_high) | (this->chunks[0] <= mask_low), 233| 2.31M| (this->chunks[1] >= mask_high) | (this->chunks[1] <= mask_low), 234| 2.31M| (this->chunks[2] >= mask_high) | (this->chunks[2] <= mask_low), 235| 2.31M| (this->chunks[3] >= mask_high) | (this->chunks[3] <= mask_low)) 236| 2.31M| .to_bitmask(); 237| 2.31M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItE4zeroEv: 45| 1.52k| static simdutf_really_inline simd16 zero() { return _mm_setzero_si128(); } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd3minENS2_6simd16ItEES4_: 240| 9.23M|simd16 min(const simd16 a, simd16 b) { 241| 9.23M| return _mm_min_epu16(a.value, b.value); 242| 9.23M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItEpLENS2_6simd16ItEE: 68| 13.8M| simdutf_really_inline simd16 &operator+=(const simd16 other) { 69| 13.8M| *this = *this + other; 70| 13.8M| return *static_cast *>(this); 71| 13.8M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItEplENS2_6simd16ItEE: 65| 13.8M| simdutf_really_inline simd16 operator+(const simd16 other) const { 66| 13.8M| return _mm_add_epi16(*this, other); 67| 13.8M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE3sumEv: 134| 1.52k| simdutf_really_inline uint64_t sum() const { 135| 1.52k| const auto lo_u16 = _mm_and_si128(value, _mm_set1_epi32(0x0000ffff)); 136| 1.52k| const auto hi_u16 = _mm_srli_epi32(value, 16); 137| 1.52k| const auto sum_u32 = _mm_add_epi32(lo_u16, hi_u16); 138| | 139| 1.52k| const auto lo_u32 = _mm_and_si128(sum_u32, _mm_set1_epi64x(0xffffffff)); 140| 1.52k| const auto hi_u32 = _mm_srli_epi64(sum_u32, 32); 141| 1.52k| const auto sum_u64 = _mm_add_epi64(lo_u32, hi_u32); 142| | 143| 1.52k| return uint64_t(_mm_extract_epi64(sum_u64, 0)) + 144| | uint64_t(_mm_extract_epi64(sum_u64, 1)); 145| 1.52k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE5splatEj: 56| 26.6k| simdutf_really_inline static simd32 splat(uint32_t v) { 57| 26.6k| return _mm_set1_epi32(v); 58| 26.6k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjEC2EDv2_x: 9| 13.1M| simdutf_really_inline simd32(const __m128i v) : value(v) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE4zeroEv: 52| 5.67k| simdutf_really_inline static simd32 zero() { 53| 5.67k| return _mm_setzero_si128(); 54| 5.67k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjEC2IDiEEPKT_: 13| 3.09M| : value(_mm_loadu_si128(reinterpret_cast(ptr))) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE10swap_bytesEv: 22| 2.51M| simdutf_really_inline simd32 swap_bytes() const { 23| 2.51M| const __m128i shuffle = 24| 2.51M| _mm_setr_epi8(3, 2, 1, 0, 7, 6, 5, 4, 8, 9, 10, 11, 15, 14, 13, 12); 25| | 26| 2.51M| return _mm_shuffle_epi8(value, shuffle); 27| 2.51M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd3maxENS2_6simd32IjEES4_: 94| 4.53M| const simd32 b) { 95| 4.53M| return _mm_max_epu32(a.value, b.value); 96| 4.53M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdplENS2_6simd32IjEES4_: 119| 2.27M| const simd32 b) { 120| 2.27M| return _mm_add_epi32(a.value, b.value); 121| 2.27M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdgtENS2_6simd32IjEES4_: 143| 12.8k| const simd32 b) { 144| 12.8k| return !(b >= a); 145| 12.8k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdntENS2_6simd32IbEE: 138| 12.8k|simdutf_really_inline simd32 operator!(const simd32 v) { 139| 12.8k| return _mm_xor_si128(v.value, _mm_set1_epi8(-1)); 140| 12.8k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IbEC2EDv2_x: 69| 797k| simdutf_really_inline simd32(const __m128i v) : value(v) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdgeENS2_6simd32IjEES4_: 134| 266k| const simd32 b) { 135| 266k| return _mm_cmpeq_epi32(_mm_max_epu32(a.value, b.value), a.value); 136| 266k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IbE3anyEv: 71| 260k| simdutf_really_inline bool any() const { 72| 260k| return _mm_movemask_epi8(value) != 0; 73| 260k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_6simd32IjEES4_: 129| 253k| const simd32 b) { 130| 253k| return _mm_cmpeq_epi32(a.value, b.value); 131| 253k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdanENS2_6simd32IjEES4_: 104| 1.99M| const simd32 b) { 105| 1.99M| return _mm_and_si128(a.value, b.value); 106| 1.99M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdorIbEENS2_6simd32IT_EES6_S6_: 84| 259k| const simd32 b) { 85| 259k| return _mm_or_si128(a.value, b.value); 86| 259k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd3minENS2_6simd32IjEES4_: 89| 1.73M| const simd32 b) { 90| 1.73M| return _mm_min_epu32(a.value, b.value); 91| 1.73M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjEpLES4_: 46| 1.73M| simdutf_really_inline simd32 &operator+=(const simd32 other) { 47| 1.73M| value = _mm_add_epi32(value, other.value); 48| 1.73M| return *this; 49| 1.73M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE3sumEv: 15| 877| simdutf_really_inline uint64_t sum() const { 16| 877| return uint64_t(_mm_extract_epi32(value, 0)) + 17| 877| uint64_t(_mm_extract_epi32(value, 1)) + 18| 877| uint64_t(_mm_extract_epi32(value, 2)) + 19| | uint64_t(_mm_extract_epi32(value, 3)); 20| 877| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_6simd32IjEEj: 99| 5.80k| uint32_t b) { 100| 5.80k| return _mm_cmpeq_epi32(a.value, _mm_set1_epi32(b)); 101| 5.80k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdanENS2_6simd32IjEEj: 109| 17.4k| uint32_t b) { 110| 17.4k| return _mm_and_si128(a.value, _mm_set1_epi32(b)); 111| 17.4k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IbE15to_4bit_bitmaskEv: 75| 5.80k| simdutf_really_inline uint8_t to_4bit_bitmask() const { 76| 5.80k| return uint8_t(_mm_movemask_ps(_mm_castsi128_ps(value))); 77| 5.80k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdmiENS2_6simd32IjEEj: 124| 5.80k| uint32_t b) { 125| 5.80k| return _mm_sub_epi32(a.value, _mm_set1_epi32(b)); 126| 5.80k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE3shrILi10EEES4_v: 29| 5.80k| template simdutf_really_inline simd32 shr() const { 30| 5.80k| return _mm_srli_epi32(value, N); 31| 5.80k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE3shlILi16EEES4_v: 33| 5.80k| template simdutf_really_inline simd32 shl() const { 34| 5.80k| return _mm_slli_epi32(value, N); 35| 5.80k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdorENS2_6simd32IjEEj: 114| 5.80k| uint32_t b) { 115| 5.80k| return _mm_or_si128(a.value, _mm_set1_epi32(b)); 116| 5.80k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdorIjEENS2_6simd32IT_EES6_S6_: 84| 5.80k| const simd32 b) { 85| 5.80k| return _mm_or_si128(a.value, b.value); 86| 5.80k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6selectENS2_6simd32IbEENS3_IjEES5_: 149| 5.80k| const simd32 v_false) { 150| 5.80k| return _mm_blendv_epi8(v_false.value, v_true.value, cond.value); 151| 5.80k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd64ImE4zeroEv: 26| 5.50k| simdutf_really_inline static simd64 zero() { 27| 5.50k| return _mm_setzero_si128(); 28| 5.50k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd64ImEC2EDv2_x: 9| 105k| simdutf_really_inline simd64(const __m128i v) : value(v) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd64ImEpLES4_: 20| 99.5k| simdutf_really_inline simd64 &operator+=(const simd64 other) { 21| 99.5k| value = _mm_add_epi64(value, other.value); 22| 99.5k| return *this; 23| 99.5k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd64ImE3sumEv: 15| 5.50k| simdutf_really_inline uint64_t sum() const { 16| 5.50k| return _mm_extract_epi64(value, 0) + _mm_extract_epi64(value, 1); 17| 5.50k| } _ZNK7simdutf8westmere14implementation13validate_utf8EPKcm: 299| 3.22k|implementation::validate_utf8(const char *buf, size_t len) const noexcept { 300| 3.22k| return westmere::utf8_validation::generic_validate_utf8(buf, len); 301| 3.22k|} _ZNK7simdutf8westmere14implementation25validate_utf8_with_errorsEPKcm: 306| 3.22k| const char *buf, size_t len) const noexcept { 307| 3.22k| return westmere::utf8_validation::generic_validate_utf8_with_errors(buf, len); 308| 3.22k|} _ZNK7simdutf8westmere14implementation16validate_utf16leEPKDsm: 345| 1.30k| size_t len) const noexcept { 346| 1.30k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.30k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 17, False: 1.29k] | | ------------------ ------------------ 347| | // empty input is valid UTF-16. protect the implementation from 348| | // handling nullptr 349| 17| return true; 350| 17| } 351| 1.29k| const auto res = 352| 1.29k| westmere::utf16::validate_utf16_with_errors(buf, len); 353| 1.29k| if (res.is_err()) { ------------------ | Branch (353:7): [True: 125, False: 1.16k] ------------------ 354| 125| return false; 355| 125| } 356| | 357| 1.16k| if (res.count == len) ------------------ | Branch (357:7): [True: 0, False: 1.16k] ------------------ 358| 0| return true; 359| | 360| 1.16k| return scalar::utf16::validate(buf + res.count, 361| 1.16k| len - res.count); 362| 1.16k|} _ZNK7simdutf8westmere14implementation16validate_utf16beEPKDsm: 368| 1.23k| size_t len) const noexcept { 369| 1.23k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.23k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 15, False: 1.22k] | | ------------------ ------------------ 370| | // empty input is valid UTF-16. protect the implementation from 371| | // handling nullptr 372| 15| return true; 373| 15| } 374| 1.22k| const auto res = 375| 1.22k| westmere::utf16::validate_utf16_with_errors(buf, len); 376| 1.22k| if (res.is_err()) { ------------------ | Branch (376:7): [True: 131, False: 1.09k] ------------------ 377| 131| return false; 378| 131| } 379| | 380| 1.09k| if (res.count == len) ------------------ | Branch (380:7): [True: 0, False: 1.09k] ------------------ 381| 0| return true; 382| | 383| 1.09k| return scalar::utf16::validate(buf + res.count, 384| 1.09k| len - res.count); 385| 1.09k|} _ZNK7simdutf8westmere14implementation28validate_utf16le_with_errorsEPKDsm: 388| 1.30k| const char16_t *buf, size_t len) const noexcept { 389| 1.30k| const result res = 390| 1.30k| westmere::utf16::validate_utf16_with_errors(buf, len); 391| 1.30k| if (res.count != len) { ------------------ | Branch (391:7): [True: 1.29k, False: 17] ------------------ 392| 1.29k| const result scalar_res = 393| 1.29k| scalar::utf16::validate_with_errors( 394| 1.29k| buf + res.count, len - res.count); 395| 1.29k| return result(scalar_res.error, res.count + scalar_res.count); 396| 1.29k| } else { 397| 17| return res; 398| 17| } 399| 1.30k|} _ZNK7simdutf8westmere14implementation28validate_utf16be_with_errorsEPKDsm: 402| 1.23k| const char16_t *buf, size_t len) const noexcept { 403| 1.23k| const result res = 404| 1.23k| westmere::utf16::validate_utf16_with_errors(buf, len); 405| 1.23k| if (res.count != len) { ------------------ | Branch (405:7): [True: 1.22k, False: 15] ------------------ 406| 1.22k| result scalar_res = scalar::utf16::validate_with_errors( 407| 1.22k| buf + res.count, len - res.count); 408| 1.22k| return result(scalar_res.error, res.count + scalar_res.count); 409| 1.22k| } else { 410| 15| return res; 411| 15| } 412| 1.23k|} _ZNK7simdutf8westmere14implementation14validate_utf32EPKDim: 427| 2.41k|implementation::validate_utf32(const char32_t *buf, size_t len) const noexcept { 428| 2.41k| return utf32::validate(buf, len); 429| 2.41k|} _ZNK7simdutf8westmere14implementation26validate_utf32_with_errorsEPKDim: 434| 2.41k| const char32_t *buf, size_t len) const noexcept { 435| 2.41k| return utf32::validate_with_errors(buf, len); 436| 2.41k|} _ZNK7simdutf8westmere14implementation22convert_latin1_to_utf8EPKcmPc: 441| 538| const char *buf, size_t len, char *utf8_output) const noexcept { 442| | 443| 538| std::pair ret = 444| 538| sse_convert_latin1_to_utf8(buf, len, utf8_output); 445| 538| size_t converted_chars = ret.second - utf8_output; 446| | 447| 538| if (ret.first != buf + len) { ------------------ | Branch (447:7): [True: 452, False: 86] ------------------ 448| 452| const size_t scalar_converted_chars = scalar::latin1_to_utf8::convert( 449| 452| ret.first, len - (ret.first - buf), ret.second); 450| 452| converted_chars += scalar_converted_chars; 451| 452| } 452| | 453| 538| return converted_chars; 454| 538|} _ZNK7simdutf8westmere14implementation25convert_latin1_to_utf16leEPKcmPDs: 459| 80| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 460| 80| std::pair ret = 461| 80| sse_convert_latin1_to_utf16(buf, len, utf16_output); 462| 80| if (ret.first == nullptr) { ------------------ | Branch (462:7): [True: 4, False: 76] ------------------ 463| 4| return 0; 464| 4| } 465| 76| size_t converted_chars = ret.second - utf16_output; 466| 76| if (ret.first != buf + len) { ------------------ | Branch (466:7): [True: 62, False: 14] ------------------ 467| 62| const size_t scalar_converted_chars = 468| 62| scalar::latin1_to_utf16::convert( 469| 62| ret.first, len - (ret.first - buf), ret.second); 470| 62| if (scalar_converted_chars == 0) { ------------------ | Branch (470:9): [True: 0, False: 62] ------------------ 471| 0| return 0; 472| 0| } 473| 62| converted_chars += scalar_converted_chars; 474| 62| } 475| 76| return converted_chars; 476| 76|} _ZNK7simdutf8westmere14implementation25convert_latin1_to_utf16beEPKcmPDs: 479| 80| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 480| 80| std::pair ret = 481| 80| sse_convert_latin1_to_utf16(buf, len, utf16_output); 482| 80| if (ret.first == nullptr) { ------------------ | Branch (482:7): [True: 4, False: 76] ------------------ 483| 4| return 0; 484| 4| } 485| 76| size_t converted_chars = ret.second - utf16_output; 486| 76| if (ret.first != buf + len) { ------------------ | Branch (486:7): [True: 62, False: 14] ------------------ 487| 62| const size_t scalar_converted_chars = 488| 62| scalar::latin1_to_utf16::convert( 489| 62| ret.first, len - (ret.first - buf), ret.second); 490| 62| if (scalar_converted_chars == 0) { ------------------ | Branch (490:9): [True: 0, False: 62] ------------------ 491| 0| return 0; 492| 0| } 493| 62| converted_chars += scalar_converted_chars; 494| 62| } 495| 76| return converted_chars; 496| 76|} _ZNK7simdutf8westmere14implementation23convert_latin1_to_utf32EPKcmPDi: 501| 90| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 502| 90| std::pair ret = 503| 90| sse_convert_latin1_to_utf32(buf, len, utf32_output); 504| 90| if (ret.first == nullptr) { ------------------ | Branch (504:7): [True: 4, False: 86] ------------------ 505| 4| return 0; 506| 4| } 507| 86| size_t converted_chars = ret.second - utf32_output; 508| 86| if (ret.first != buf + len) { ------------------ | Branch (508:7): [True: 70, False: 16] ------------------ 509| 70| const size_t scalar_converted_chars = scalar::latin1_to_utf32::convert( 510| 70| ret.first, len - (ret.first - buf), ret.second); 511| 70| if (scalar_converted_chars == 0) { ------------------ | Branch (511:9): [True: 0, False: 70] ------------------ 512| 0| return 0; 513| 0| } 514| 70| converted_chars += scalar_converted_chars; 515| 70| } 516| 86| return converted_chars; 517| 86|} _ZNK7simdutf8westmere14implementation22convert_utf8_to_latin1EPKcmPc: 522| 562| const char *buf, size_t len, char *latin1_output) const noexcept { 523| 562| utf8_to_latin1::validating_transcoder converter; 524| 562| return converter.convert(buf, len, latin1_output); 525| 562|} _ZNK7simdutf8westmere14implementation34convert_utf8_to_latin1_with_errorsEPKcmPc: 528| 516| const char *buf, size_t len, char *latin1_output) const noexcept { 529| 516| utf8_to_latin1::validating_transcoder converter; 530| 516| return converter.convert_with_errors(buf, len, latin1_output); 531| 516|} _ZNK7simdutf8westmere14implementation23convert_utf8_to_utf16leEPKcmPDs: 541| 660| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 542| 660| utf8_to_utf16::validating_transcoder converter; 543| 660| return converter.convert(buf, len, utf16_output); 544| 660|} _ZNK7simdutf8westmere14implementation23convert_utf8_to_utf16beEPKcmPDs: 547| 712| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 548| 712| utf8_to_utf16::validating_transcoder converter; 549| 712| return converter.convert(buf, len, utf16_output); 550| 712|} _ZNK7simdutf8westmere14implementation35convert_utf8_to_utf16le_with_errorsEPKcmPDs: 553| 574| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 554| 574| utf8_to_utf16::validating_transcoder converter; 555| 574| return converter.convert_with_errors(buf, len, 556| 574| utf16_output); 557| 574|} _ZNK7simdutf8westmere14implementation35convert_utf8_to_utf16be_with_errorsEPKcmPDs: 560| 702| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 561| 702| utf8_to_utf16::validating_transcoder converter; 562| 702| return converter.convert_with_errors(buf, len, utf16_output); 563| 702|} _ZNK7simdutf8westmere14implementation29convert_valid_utf8_to_utf16leEPKcmPDs: 566| 368| const char *input, size_t size, char16_t *utf16_output) const noexcept { 567| 368| return utf8_to_utf16::convert_valid(input, size, 568| 368| utf16_output); 569| 368|} _ZNK7simdutf8westmere14implementation29convert_valid_utf8_to_utf16beEPKcmPDs: 572| 438| const char *input, size_t size, char16_t *utf16_output) const noexcept { 573| 438| return utf8_to_utf16::convert_valid(input, size, 574| 438| utf16_output); 575| 438|} _ZNK7simdutf8westmere14implementation21convert_utf8_to_utf32EPKcmPDi: 580| 712| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 581| 712| utf8_to_utf32::validating_transcoder converter; 582| 712| return converter.convert(buf, len, utf32_output); 583| 712|} _ZNK7simdutf8westmere14implementation33convert_utf8_to_utf32_with_errorsEPKcmPDi: 586| 738| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 587| 738| utf8_to_utf32::validating_transcoder converter; 588| 738| return converter.convert_with_errors(buf, len, utf32_output); 589| 738|} _ZNK7simdutf8westmere14implementation27convert_valid_utf8_to_utf32EPKcmPDi: 592| 394| const char *input, size_t size, char32_t *utf32_output) const noexcept { 593| 394| return utf8_to_utf32::convert_valid(input, size, utf32_output); 594| 394|} _ZNK7simdutf8westmere14implementation25convert_utf16le_to_latin1EPKDsmPc: 599| 124| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 600| 124| std::pair ret = 601| 124| sse_convert_utf16_to_latin1(buf, len, latin1_output); 602| 124| if (ret.first == nullptr) { ------------------ | Branch (602:7): [True: 54, False: 70] ------------------ 603| 54| return 0; 604| 54| } 605| 70| size_t saved_bytes = ret.second - latin1_output; 606| | 607| 70| if (ret.first != buf + len) { ------------------ | Branch (607:7): [True: 52, False: 18] ------------------ 608| 52| const size_t scalar_saved_bytes = 609| 52| scalar::utf16_to_latin1::convert( 610| 52| ret.first, len - (ret.first - buf), ret.second); 611| 52| if (scalar_saved_bytes == 0) { ------------------ | Branch (611:9): [True: 30, False: 22] ------------------ 612| 30| return 0; 613| 30| } 614| 22| saved_bytes += scalar_saved_bytes; 615| 22| } 616| 40| return saved_bytes; 617| 70|} _ZNK7simdutf8westmere14implementation25convert_utf16be_to_latin1EPKDsmPc: 620| 126| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 621| 126| std::pair ret = 622| 126| sse_convert_utf16_to_latin1(buf, len, latin1_output); 623| 126| if (ret.first == nullptr) { ------------------ | Branch (623:7): [True: 38, False: 88] ------------------ 624| 38| return 0; 625| 38| } 626| 88| size_t saved_bytes = ret.second - latin1_output; 627| | 628| 88| if (ret.first != buf + len) { ------------------ | Branch (628:7): [True: 70, False: 18] ------------------ 629| 70| const size_t scalar_saved_bytes = 630| 70| scalar::utf16_to_latin1::convert( 631| 70| ret.first, len - (ret.first - buf), ret.second); 632| 70| if (scalar_saved_bytes == 0) { ------------------ | Branch (632:9): [True: 24, False: 46] ------------------ 633| 24| return 0; 634| 24| } 635| 46| saved_bytes += scalar_saved_bytes; 636| 46| } 637| 64| return saved_bytes; 638| 88|} _ZNK7simdutf8westmere14implementation37convert_utf16le_to_latin1_with_errorsEPKDsmPc: 642| 230| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 643| 230| std::pair ret = 644| 230| sse_convert_utf16_to_latin1_with_errors( 645| 230| buf, len, latin1_output); 646| 230| if (ret.first.error) { ------------------ | Branch (646:7): [True: 128, False: 102] ------------------ 647| 128| return ret.first; 648| 128| } // Can return directly since scalar fallback already found correct 649| | // ret.first.count 650| 102| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (650:7): [True: 64, False: 38] ------------------ 651| 64| result scalar_res = 652| 64| scalar::utf16_to_latin1::convert_with_errors( 653| 64| buf + ret.first.count, len - ret.first.count, ret.second); 654| 64| if (scalar_res.error) { ------------------ | Branch (654:9): [True: 20, False: 44] ------------------ 655| 20| scalar_res.count += ret.first.count; 656| 20| return scalar_res; 657| 44| } else { 658| 44| ret.second += scalar_res.count; 659| 44| } 660| 64| } 661| 82| ret.first.count = 662| 82| ret.second - 663| 82| latin1_output; // Set count to the number of 8-bit code units written 664| 82| return ret.first; 665| 102|} _ZNK7simdutf8westmere14implementation37convert_utf16be_to_latin1_with_errorsEPKDsmPc: 669| 252| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 670| 252| std::pair ret = 671| 252| sse_convert_utf16_to_latin1_with_errors(buf, len, 672| 252| latin1_output); 673| 252| if (ret.first.error) { ------------------ | Branch (673:7): [True: 128, False: 124] ------------------ 674| 128| return ret.first; 675| 128| } // Can return directly since scalar fallback already found correct 676| | // ret.first.count 677| 124| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (677:7): [True: 80, False: 44] ------------------ 678| 80| result scalar_res = 679| 80| scalar::utf16_to_latin1::convert_with_errors( 680| 80| buf + ret.first.count, len - ret.first.count, ret.second); 681| 80| if (scalar_res.error) { ------------------ | Branch (681:9): [True: 28, False: 52] ------------------ 682| 28| scalar_res.count += ret.first.count; 683| 28| return scalar_res; 684| 52| } else { 685| 52| ret.second += scalar_res.count; 686| 52| } 687| 80| } 688| 96| ret.first.count = 689| 96| ret.second - 690| 96| latin1_output; // Set count to the number of 8-bit code units written 691| 96| return ret.first; 692| 124|} _ZNK7simdutf8westmere14implementation23convert_utf16le_to_utf8EPKDsmPc: 709| 802| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 710| 802| std::pair ret = 711| 802| sse_convert_utf16_to_utf8(buf, len, utf8_output); 712| 802| if (ret.first == nullptr) { ------------------ | Branch (712:7): [True: 90, False: 712] ------------------ 713| 90| return 0; 714| 90| } 715| 712| size_t saved_bytes = ret.second - utf8_output; 716| 712| if (ret.first != buf + len) { ------------------ | Branch (716:7): [True: 710, False: 2] ------------------ 717| 710| const size_t scalar_saved_bytes = 718| 710| scalar::utf16_to_utf8::convert( 719| 710| ret.first, len - (ret.first - buf), ret.second); 720| 710| if (scalar_saved_bytes == 0) { ------------------ | Branch (720:9): [True: 76, False: 634] ------------------ 721| 76| return 0; 722| 76| } 723| 634| saved_bytes += scalar_saved_bytes; 724| 634| } 725| 636| return saved_bytes; 726| 712|} _ZNK7simdutf8westmere14implementation23convert_utf16be_to_utf8EPKDsmPc: 729| 768| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 730| 768| std::pair ret = 731| 768| sse_convert_utf16_to_utf8(buf, len, utf8_output); 732| 768| if (ret.first == nullptr) { ------------------ | Branch (732:7): [True: 82, False: 686] ------------------ 733| 82| return 0; 734| 82| } 735| 686| size_t saved_bytes = ret.second - utf8_output; 736| 686| if (ret.first != buf + len) { ------------------ | Branch (736:7): [True: 684, False: 2] ------------------ 737| 684| const size_t scalar_saved_bytes = 738| 684| scalar::utf16_to_utf8::convert( 739| 684| ret.first, len - (ret.first - buf), ret.second); 740| 684| if (scalar_saved_bytes == 0) { ------------------ | Branch (740:9): [True: 54, False: 630] ------------------ 741| 54| return 0; 742| 54| } 743| 630| saved_bytes += scalar_saved_bytes; 744| 630| } 745| 632| return saved_bytes; 746| 686|} _ZNK7simdutf8westmere14implementation35convert_utf16le_to_utf8_with_errorsEPKDsmPc: 770| 630| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 771| 630| return convert_utf16_to_utf8_with_details(buf, len, 772| 630| utf8_output); 773| 630|} _ZNK7simdutf8westmere14implementation35convert_utf16be_to_utf8_with_errorsEPKDsmPc: 776| 586| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 777| 586| return convert_utf16_to_utf8_with_details(buf, len, 778| 586| utf8_output); 779| 586|} _ZNK7simdutf8westmere14implementation29convert_valid_utf16le_to_utf8EPKDsmPc: 782| 416| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 783| 416| return convert_utf16le_to_utf8(buf, len, utf8_output); 784| 416|} _ZNK7simdutf8westmere14implementation29convert_valid_utf16be_to_utf8EPKDsmPc: 787| 442| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 788| 442| return convert_utf16be_to_utf8(buf, len, utf8_output); 789| 442|} _ZNK7simdutf8westmere14implementation23convert_utf32_to_latin1EPKDimPc: 794| 188| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 795| 188| std::pair ret = 796| 188| sse_convert_utf32_to_latin1(buf, len, latin1_output); 797| 188| if (ret.first == nullptr) { ------------------ | Branch (797:7): [True: 46, False: 142] ------------------ 798| 46| return 0; 799| 46| } 800| 142| size_t saved_bytes = ret.second - latin1_output; 801| | // if (ret.first != buf + len) { 802| 142| if (ret.first < buf + len) { ------------------ | Branch (802:7): [True: 126, False: 16] ------------------ 803| 126| const size_t scalar_saved_bytes = scalar::utf32_to_latin1::convert( 804| 126| ret.first, len - (ret.first - buf), ret.second); 805| 126| if (scalar_saved_bytes == 0) { ------------------ | Branch (805:9): [True: 84, False: 42] ------------------ 806| 84| return 0; 807| 84| } 808| 42| saved_bytes += scalar_saved_bytes; 809| 42| } 810| 58| return saved_bytes; 811| 142|} _ZNK7simdutf8westmere14implementation35convert_utf32_to_latin1_with_errorsEPKDimPc: 814| 420| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 815| | // ret.first.count is always the position in the buffer, not the number of 816| | // code units written even if finished 817| 420| std::pair ret = 818| 420| westmere::sse_convert_utf32_to_latin1_with_errors(buf, len, 819| 420| latin1_output); 820| 420| if (ret.first.count != len) { ------------------ | Branch (820:7): [True: 396, False: 24] ------------------ 821| 396| result scalar_res = scalar::utf32_to_latin1::convert_with_errors( 822| 396| buf + ret.first.count, len - ret.first.count, ret.second); 823| 396| if (scalar_res.error) { ------------------ | Branch (823:9): [True: 370, False: 26] ------------------ 824| 370| scalar_res.count += ret.first.count; 825| 370| return scalar_res; 826| 370| } else { 827| 26| ret.second += scalar_res.count; 828| 26| } 829| 396| } 830| 50| ret.first.count = 831| 50| ret.second - 832| 50| latin1_output; // Set count to the number of 8-bit code units written 833| 50| return ret.first; 834| 420|} _ZNK7simdutf8westmere14implementation21convert_utf32_to_utf8EPKDimPc: 845| 796| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 846| 796| std::pair ret = 847| 796| sse_convert_utf32_to_utf8(buf, len, utf8_output); 848| 796| if (ret.first == nullptr) { ------------------ | Branch (848:7): [True: 244, False: 552] ------------------ 849| 244| return 0; 850| 244| } 851| 552| size_t saved_bytes = ret.second - utf8_output; 852| 552| if (ret.first != buf + len) { ------------------ | Branch (852:7): [True: 550, False: 2] ------------------ 853| 550| const size_t scalar_saved_bytes = scalar::utf32_to_utf8::convert( 854| 550| ret.first, len - (ret.first - buf), ret.second); 855| 550| if (scalar_saved_bytes == 0) { ------------------ | Branch (855:9): [True: 260, False: 290] ------------------ 856| 260| return 0; 857| 260| } 858| 290| saved_bytes += scalar_saved_bytes; 859| 290| } 860| 292| return saved_bytes; 861| 552|} _ZNK7simdutf8westmere14implementation33convert_utf32_to_utf8_with_errorsEPKDimPc: 864| 830| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 865| | // ret.first.count is always the position in the buffer, not the number of 866| | // code units written even if finished 867| 830| std::pair ret = 868| 830| westmere::sse_convert_utf32_to_utf8_with_errors(buf, len, utf8_output); 869| 830| if (ret.first.count != len) { ------------------ | Branch (869:7): [True: 826, False: 4] ------------------ 870| 826| result scalar_res = scalar::utf32_to_utf8::convert_with_errors( 871| 826| buf + ret.first.count, len - ret.first.count, ret.second); 872| 826| if (scalar_res.error) { ------------------ | Branch (872:9): [True: 632, False: 194] ------------------ 873| 632| scalar_res.count += ret.first.count; 874| 632| return scalar_res; 875| 632| } else { 876| 194| ret.second += scalar_res.count; 877| 194| } 878| 826| } 879| 198| ret.first.count = 880| 198| ret.second - 881| 198| utf8_output; // Set count to the number of 8-bit code units written 882| 198| return ret.first; 883| 830|} _ZNK7simdutf8westmere14implementation24convert_utf16le_to_utf32EPKDsmPDi: 888| 384| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 889| 384| std::pair ret = 890| 384| sse_convert_utf16_to_utf32(buf, len, utf32_output); 891| 384| if (ret.first == nullptr) { ------------------ | Branch (891:7): [True: 118, False: 266] ------------------ 892| 118| return 0; 893| 118| } 894| 266| size_t saved_bytes = ret.second - utf32_output; 895| 266| if (ret.first != buf + len) { ------------------ | Branch (895:7): [True: 202, False: 64] ------------------ 896| 202| const size_t scalar_saved_bytes = 897| 202| scalar::utf16_to_utf32::convert( 898| 202| ret.first, len - (ret.first - buf), ret.second); 899| 202| if (scalar_saved_bytes == 0) { ------------------ | Branch (899:9): [True: 58, False: 144] ------------------ 900| 58| return 0; 901| 58| } 902| 144| saved_bytes += scalar_saved_bytes; 903| 144| } 904| 208| return saved_bytes; 905| 266|} _ZNK7simdutf8westmere14implementation24convert_utf16be_to_utf32EPKDsmPDi: 908| 366| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 909| 366| std::pair ret = 910| 366| sse_convert_utf16_to_utf32(buf, len, utf32_output); 911| 366| if (ret.first == nullptr) { ------------------ | Branch (911:7): [True: 86, False: 280] ------------------ 912| 86| return 0; 913| 86| } 914| 280| size_t saved_bytes = ret.second - utf32_output; 915| 280| if (ret.first != buf + len) { ------------------ | Branch (915:7): [True: 218, False: 62] ------------------ 916| 218| const size_t scalar_saved_bytes = 917| 218| scalar::utf16_to_utf32::convert( 918| 218| ret.first, len - (ret.first - buf), ret.second); 919| 218| if (scalar_saved_bytes == 0) { ------------------ | Branch (919:9): [True: 66, False: 152] ------------------ 920| 66| return 0; 921| 66| } 922| 152| saved_bytes += scalar_saved_bytes; 923| 152| } 924| 214| return saved_bytes; 925| 280|} _ZNK7simdutf8westmere14implementation36convert_utf16le_to_utf32_with_errorsEPKDsmPDi: 928| 414| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 929| | // ret.first.count is always the position in the buffer, not the number of 930| | // code units written even if finished 931| 414| std::pair ret = 932| 414| westmere::sse_convert_utf16_to_utf32_with_errors( 933| 414| buf, len, utf32_output); 934| 414| if (ret.first.error) { ------------------ | Branch (934:7): [True: 148, False: 266] ------------------ 935| 148| return ret.first; 936| 148| } // Can return directly since scalar fallback already found correct 937| | // ret.first.count 938| 266| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (938:7): [True: 218, False: 48] ------------------ 939| 218| result scalar_res = 940| 218| scalar::utf16_to_utf32::convert_with_errors( 941| 218| buf + ret.first.count, len - ret.first.count, ret.second); 942| 218| if (scalar_res.error) { ------------------ | Branch (942:9): [True: 76, False: 142] ------------------ 943| 76| scalar_res.count += ret.first.count; 944| 76| return scalar_res; 945| 142| } else { 946| 142| ret.second += scalar_res.count; 947| 142| } 948| 218| } 949| 190| ret.first.count = 950| 190| ret.second - 951| 190| utf32_output; // Set count to the number of 8-bit code units written 952| 190| return ret.first; 953| 266|} _ZNK7simdutf8westmere14implementation36convert_utf16be_to_utf32_with_errorsEPKDsmPDi: 956| 300| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 957| | // ret.first.count is always the position in the buffer, not the number of 958| | // code units written even if finished 959| 300| std::pair ret = 960| 300| westmere::sse_convert_utf16_to_utf32_with_errors( 961| 300| buf, len, utf32_output); 962| 300| if (ret.first.error) { ------------------ | Branch (962:7): [True: 76, False: 224] ------------------ 963| 76| return ret.first; 964| 76| } // Can return directly since scalar fallback already found correct 965| | // ret.first.count 966| 224| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (966:7): [True: 186, False: 38] ------------------ 967| 186| result scalar_res = 968| 186| scalar::utf16_to_utf32::convert_with_errors( 969| 186| buf + ret.first.count, len - ret.first.count, ret.second); 970| 186| if (scalar_res.error) { ------------------ | Branch (970:9): [True: 64, False: 122] ------------------ 971| 64| scalar_res.count += ret.first.count; 972| 64| return scalar_res; 973| 122| } else { 974| 122| ret.second += scalar_res.count; 975| 122| } 976| 186| } 977| 160| ret.first.count = 978| 160| ret.second - 979| 160| utf32_output; // Set count to the number of 8-bit code units written 980| 160| return ret.first; 981| 224|} _ZNK7simdutf8westmere14implementation27convert_valid_utf32_to_utf8EPKDimPc: 986| 192| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 987| 192| return convert_utf32_to_utf8(buf, len, utf8_output); 988| 192|} _ZNK7simdutf8westmere14implementation24convert_utf32_to_utf16leEPKDimPDs: 993| 722| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 994| 722| std::pair ret = 995| 722| sse_convert_utf32_to_utf16(buf, len, utf16_output); 996| 722| if (ret.first == nullptr) { ------------------ | Branch (996:7): [True: 52, False: 670] ------------------ 997| 52| return 0; 998| 52| } 999| 670| size_t saved_bytes = ret.second - utf16_output; 1000| 670| if (ret.first != buf + len) { ------------------ | Branch (1000:7): [True: 666, False: 4] ------------------ 1001| 666| const size_t scalar_saved_bytes = 1002| 666| scalar::utf32_to_utf16::convert( 1003| 666| ret.first, len - (ret.first - buf), ret.second); 1004| 666| if (scalar_saved_bytes == 0) { ------------------ | Branch (1004:9): [True: 234, False: 432] ------------------ 1005| 234| return 0; 1006| 234| } 1007| 432| saved_bytes += scalar_saved_bytes; 1008| 432| } 1009| 436| return saved_bytes; 1010| 670|} _ZNK7simdutf8westmere14implementation24convert_utf32_to_utf16beEPKDimPDs: 1013| 528| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1014| 528| std::pair ret = 1015| 528| sse_convert_utf32_to_utf16(buf, len, utf16_output); 1016| 528| if (ret.first == nullptr) { ------------------ | Branch (1016:7): [True: 44, False: 484] ------------------ 1017| 44| return 0; 1018| 44| } 1019| 484| size_t saved_bytes = ret.second - utf16_output; 1020| 484| if (ret.first != buf + len) { ------------------ | Branch (1020:7): [True: 482, False: 2] ------------------ 1021| 482| const size_t scalar_saved_bytes = 1022| 482| scalar::utf32_to_utf16::convert( 1023| 482| ret.first, len - (ret.first - buf), ret.second); 1024| 482| if (scalar_saved_bytes == 0) { ------------------ | Branch (1024:9): [True: 288, False: 194] ------------------ 1025| 288| return 0; 1026| 288| } 1027| 194| saved_bytes += scalar_saved_bytes; 1028| 194| } 1029| 196| return saved_bytes; 1030| 484|} _ZNK7simdutf8westmere14implementation36convert_utf32_to_utf16le_with_errorsEPKDimPDs: 1033| 716| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1034| | // ret.first.count is always the position in the buffer, not the number of 1035| | // code units written even if finished 1036| 716| std::pair ret = 1037| 716| westmere::sse_convert_utf32_to_utf16_with_errors( 1038| 716| buf, len, utf16_output); 1039| 716| if (ret.first.count != len) { ------------------ | Branch (1039:7): [True: 692, False: 24] ------------------ 1040| 692| result scalar_res = 1041| 692| scalar::utf32_to_utf16::convert_with_errors( 1042| 692| buf + ret.first.count, len - ret.first.count, ret.second); 1043| 692| if (scalar_res.error) { ------------------ | Branch (1043:9): [True: 530, False: 162] ------------------ 1044| 530| scalar_res.count += ret.first.count; 1045| 530| return scalar_res; 1046| 530| } else { 1047| 162| ret.second += scalar_res.count; 1048| 162| } 1049| 692| } 1050| 186| ret.first.count = 1051| 186| ret.second - 1052| 186| utf16_output; // Set count to the number of 8-bit code units written 1053| 186| return ret.first; 1054| 716|} _ZNK7simdutf8westmere14implementation36convert_utf32_to_utf16be_with_errorsEPKDimPDs: 1057| 606| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1058| | // ret.first.count is always the position in the buffer, not the number of 1059| | // code units written even if finished 1060| 606| std::pair ret = 1061| 606| westmere::sse_convert_utf32_to_utf16_with_errors( 1062| 606| buf, len, utf16_output); 1063| 606| if (ret.first.count != len) { ------------------ | Branch (1063:7): [True: 584, False: 22] ------------------ 1064| 584| result scalar_res = 1065| 584| scalar::utf32_to_utf16::convert_with_errors( 1066| 584| buf + ret.first.count, len - ret.first.count, ret.second); 1067| 584| if (scalar_res.error) { ------------------ | Branch (1067:9): [True: 434, False: 150] ------------------ 1068| 434| scalar_res.count += ret.first.count; 1069| 434| return scalar_res; 1070| 434| } else { 1071| 150| ret.second += scalar_res.count; 1072| 150| } 1073| 584| } 1074| 172| ret.first.count = 1075| 172| ret.second - 1076| 172| utf16_output; // Set count to the number of 8-bit code units written 1077| 172| return ret.first; 1078| 606|} _ZNK7simdutf8westmere14implementation30convert_valid_utf32_to_utf16leEPKDimPDs: 1081| 290| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1082| 290| return convert_utf32_to_utf16le(buf, len, utf16_output); 1083| 290|} _ZNK7simdutf8westmere14implementation30convert_valid_utf32_to_utf16beEPKDimPDs: 1086| 86| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1087| 86| return convert_utf32_to_utf16be(buf, len, utf16_output); 1088| 86|} _ZNK7simdutf8westmere14implementation30convert_valid_utf16le_to_utf32EPKDsmPDi: 1091| 112| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1092| 112| return convert_utf16le_to_utf32(buf, len, utf32_output); 1093| 112|} _ZNK7simdutf8westmere14implementation30convert_valid_utf16be_to_utf32EPKDsmPDi: 1096| 140| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1097| 140| return convert_utf16be_to_utf32(buf, len, utf32_output); 1098| 140|} _ZNK7simdutf8westmere14implementation13count_utf16leEPKDsm: 1109| 1.30k| const char16_t *input, size_t length) const noexcept { 1110| 1.30k| return utf16::count_code_points(input, length); 1111| 1.30k|} _ZNK7simdutf8westmere14implementation13count_utf16beEPKDsm: 1114| 1.23k| const char16_t *input, size_t length) const noexcept { 1115| 1.23k| return utf16::count_code_points(input, length); 1116| 1.23k|} _ZNK7simdutf8westmere14implementation10count_utf8EPKcm: 1121| 3.76k|implementation::count_utf8(const char *input, size_t length) const noexcept { 1122| 3.76k| return utf8::count_code_points_bytemask(input, length); 1123| 3.76k|} _ZNK7simdutf8westmere14implementation23latin1_length_from_utf8EPKcm: 1128| 539| const char *buf, size_t len) const noexcept { 1129| 539| return count_utf8(buf, len); 1130| 539|} _ZNK7simdutf8westmere14implementation24utf8_length_from_utf16leEPKDsm: 1135| 727| const char16_t *input, size_t length) const noexcept { 1136| 727| return utf16::utf8_length_from_utf16_bytemask(input, 1137| 727| length); 1138| 727|} _ZNK7simdutf8westmere14implementation24utf8_length_from_utf16beEPKDsm: 1141| 696| const char16_t *input, size_t length) const noexcept { 1142| 696| return utf16::utf8_length_from_utf16_bytemask(input, length); 1143| 696|} _ZNK7simdutf8westmere14implementation23utf8_length_from_latin1EPKcm: 1148| 269| const char *input, size_t len) const noexcept { 1149| 269| const uint8_t *str = reinterpret_cast(input); 1150| 269| size_t answer = len / sizeof(__m128i) * sizeof(__m128i); 1151| 269| size_t i = 0; 1152| 269| if (answer >= 2048) { // long strings optimization ------------------ | Branch (1152:7): [True: 118, False: 151] ------------------ 1153| 118| __m128i two_64bits = _mm_setzero_si128(); 1154| 5.04k| while (i + sizeof(__m128i) <= len) { ------------------ | Branch (1154:12): [True: 4.92k, False: 118] ------------------ 1155| 4.92k| __m128i runner = _mm_setzero_si128(); 1156| 4.92k| size_t iterations = (len - i) / sizeof(__m128i); 1157| 4.92k| if (iterations > 255) { ------------------ | Branch (1157:11): [True: 4.81k, False: 118] ------------------ 1158| 4.81k| iterations = 255; 1159| 4.81k| } 1160| 4.92k| size_t max_i = i + iterations * sizeof(__m128i) - sizeof(__m128i); 1161| 309k| for (; i + 4 * sizeof(__m128i) <= max_i; i += 4 * sizeof(__m128i)) { ------------------ | Branch (1161:14): [True: 304k, False: 4.92k] ------------------ 1162| 304k| __m128i input1 = _mm_loadu_si128((const __m128i *)(str + i)); 1163| 304k| __m128i input2 = 1164| 304k| _mm_loadu_si128((const __m128i *)(str + i + sizeof(__m128i))); 1165| 304k| __m128i input3 = 1166| 304k| _mm_loadu_si128((const __m128i *)(str + i + 2 * sizeof(__m128i))); 1167| 304k| __m128i input4 = 1168| 304k| _mm_loadu_si128((const __m128i *)(str + i + 3 * sizeof(__m128i))); 1169| 304k| __m128i input12 = 1170| 304k| _mm_add_epi8(_mm_cmpgt_epi8(_mm_setzero_si128(), input1), 1171| 304k| _mm_cmpgt_epi8(_mm_setzero_si128(), input2)); 1172| 304k| __m128i input34 = 1173| 304k| _mm_add_epi8(_mm_cmpgt_epi8(_mm_setzero_si128(), input3), 1174| 304k| _mm_cmpgt_epi8(_mm_setzero_si128(), input4)); 1175| 304k| __m128i input1234 = _mm_add_epi8(input12, input34); 1176| 304k| runner = _mm_sub_epi8(runner, input1234); 1177| 304k| } 1178| 19.6k| for (; i <= max_i; i += sizeof(__m128i)) { ------------------ | Branch (1178:14): [True: 14.6k, False: 4.92k] ------------------ 1179| 14.6k| __m128i more_input = _mm_loadu_si128((const __m128i *)(str + i)); 1180| 14.6k| runner = _mm_sub_epi8(runner, 1181| 14.6k| _mm_cmpgt_epi8(_mm_setzero_si128(), more_input)); 1182| 14.6k| } 1183| 4.92k| two_64bits = 1184| 4.92k| _mm_add_epi64(two_64bits, _mm_sad_epu8(runner, _mm_setzero_si128())); 1185| 4.92k| } 1186| 118| answer += 1187| 118| _mm_extract_epi64(two_64bits, 0) + _mm_extract_epi64(two_64bits, 1); 1188| 151| } else if (answer > 0) { // short string optimization ------------------ | Branch (1188:14): [True: 101, False: 50] ------------------ 1189| 1.38k| for (; i + 2 * sizeof(__m128i) <= len; i += 2 * sizeof(__m128i)) { ------------------ | Branch (1189:12): [True: 1.28k, False: 101] ------------------ 1190| 1.28k| __m128i latin = _mm_loadu_si128((const __m128i *)(input + i)); 1191| 1.28k| uint16_t non_ascii = (uint16_t)_mm_movemask_epi8(latin); 1192| 1.28k| answer += count_ones(non_ascii); 1193| 1.28k| latin = _mm_loadu_si128((const __m128i *)(input + i) + 1); 1194| 1.28k| non_ascii = (uint16_t)_mm_movemask_epi8(latin); 1195| 1.28k| answer += count_ones(non_ascii); 1196| 1.28k| } 1197| 151| for (; i + sizeof(__m128i) <= len; i += sizeof(__m128i)) { ------------------ | Branch (1197:12): [True: 50, False: 101] ------------------ 1198| 50| __m128i latin = _mm_loadu_si128((const __m128i *)(input + i)); 1199| 50| uint16_t non_ascii = (uint16_t)_mm_movemask_epi8(latin); 1200| 50| answer += count_ones(non_ascii); 1201| 50| } 1202| 101| } 1203| 269| return answer + scalar::latin1::utf8_length_from_latin1( 1204| 269| reinterpret_cast(str + i), len - i); 1205| 269|} _ZNK7simdutf8westmere14implementation25utf32_length_from_utf16leEPKDsm: 1210| 404| const char16_t *input, size_t length) const noexcept { 1211| 404| return utf16::utf32_length_from_utf16(input, length); 1212| 404|} _ZNK7simdutf8westmere14implementation25utf32_length_from_utf16beEPKDsm: 1215| 351| const char16_t *input, size_t length) const noexcept { 1216| 351| return utf16::utf32_length_from_utf16(input, length); 1217| 351|} _ZNK7simdutf8westmere14implementation22utf16_length_from_utf8EPKcm: 1222| 1.74k| const char *input, size_t length) const noexcept { 1223| 1.74k| return utf8::utf16_length_from_utf8_bytemask(input, length); 1224| 1.74k|} _ZNK7simdutf8westmere14implementation22utf8_length_from_utf32EPKDim: 1263| 819| const char32_t *input, size_t length) const noexcept { 1264| 819| return utf32::utf8_length_from_utf32(input, length); 1265| 819|} _ZNK7simdutf8westmere14implementation23utf16_length_from_utf32EPKDim: 1270| 1.29k| const char32_t *input, size_t length) const noexcept { 1271| 1.29k| const __m128i v_00000000 = _mm_setzero_si128(); 1272| 1.29k| const __m128i v_ffff0000 = _mm_set1_epi32((uint32_t)0xffff0000); 1273| 1.29k| size_t pos = 0; 1274| 1.29k| size_t count = 0; 1275| 1.34M| for (; pos + 4 <= length; pos += 4) { ------------------ | Branch (1275:10): [True: 1.34M, False: 1.29k] ------------------ 1276| 1.34M| __m128i in = _mm_loadu_si128((__m128i *)(input + pos)); 1277| 1.34M| const __m128i surrogate_bytemask = 1278| 1.34M| _mm_cmpeq_epi32(_mm_and_si128(in, v_ffff0000), v_00000000); 1279| 1.34M| const uint16_t surrogate_bitmask = 1280| 1.34M| static_cast(_mm_movemask_epi8(surrogate_bytemask)); 1281| 1.34M| size_t surrogate_count = (16 - count_ones(surrogate_bitmask)) / 4; 1282| 1.34M| count += 4 + surrogate_count; 1283| 1.34M| } 1284| 1.29k| return count + 1285| 1.29k| scalar::utf32::utf16_length_from_utf32(input + pos, length - pos); 1286| 1.29k|} _ZNK7simdutf8westmere14implementation22utf32_length_from_utf8EPKcm: 1291| 933| const char *input, size_t length) const noexcept { 1292| 933| return utf8::count_code_points(input, length); 1293| 933|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_18is_asciiERKNS1_4simd8simd8x64IhEE: 13| 4.64M|simdutf_really_inline bool is_ascii(const simd8x64 &input) { 14| 4.64M| return input.reduce_or().is_ascii(); 15| 4.64M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_124must_be_2_3_continuationENS1_4simd5simd8IhEES4_: 22| 6.40M| const simd8 prev3) { 23| 6.40M| simd8 is_third_byte = 24| 6.40M| prev2.saturating_sub(0xe0u - 0x80); // Only 111_____ will be >= 0x80 25| 6.40M| simd8 is_fourth_byte = 26| 6.40M| prev3.saturating_sub(0xf0u - 0x80); // Only 1111____ will be >= 0x80 27| 6.40M| return simd8(is_third_byte | is_fourth_byte); 28| 6.40M|} _ZN7simdutf8westmere34convert_utf16_to_utf8_with_detailsILNS_10endiannessE0EEENS_11full_resultEPKDsmPc: 750| 630| const char16_t *buf, size_t len, char *utf8_output) { 751| 630| std::pair ret = 752| 630| westmere::sse_convert_utf16_to_utf8_with_errors(buf, len, 753| 630| utf8_output); 754| 630| if (ret.first.error) { ------------------ | Branch (754:7): [True: 112, False: 518] ------------------ 755| 112| return full_result(ret.first.error, ret.first.count, 756| 112| size_t(ret.second - utf8_output)); 757| 112| } 758| 518| if (ret.first.count != len) { ------------------ | Branch (758:7): [True: 514, False: 4] ------------------ 759| 514| full_result sres = 760| 514| scalar::utf16_to_utf8::convert_with_errors( 761| 514| buf + ret.first.count, len - ret.first.count, ret.second, 0); 762| 514| return full_result(sres.error, ret.first.count + sres.input_count, 763| 514| size_t(ret.second - utf8_output) + sres.output_count); 764| 514| } 765| 4| return full_result(error_code::SUCCESS, len, 766| 4| size_t(ret.second - utf8_output)); 767| 518|} _ZN7simdutf8westmere34convert_utf16_to_utf8_with_detailsILNS_10endiannessE1EEENS_11full_resultEPKDsmPc: 750| 586| const char16_t *buf, size_t len, char *utf8_output) { 751| 586| std::pair ret = 752| 586| westmere::sse_convert_utf16_to_utf8_with_errors(buf, len, 753| 586| utf8_output); 754| 586| if (ret.first.error) { ------------------ | Branch (754:7): [True: 142, False: 444] ------------------ 755| 142| return full_result(ret.first.error, ret.first.count, 756| 142| size_t(ret.second - utf8_output)); 757| 142| } 758| 444| if (ret.first.count != len) { ------------------ | Branch (758:7): [True: 440, False: 4] ------------------ 759| 440| full_result sres = 760| 440| scalar::utf16_to_utf8::convert_with_errors( 761| 440| buf + ret.first.count, len - ret.first.count, ret.second, 0); 762| 440| return full_result(sres.error, ret.first.count + sres.input_count, 763| 440| size_t(ret.second - utf8_output) + sres.output_count); 764| 440| } 765| 4| return full_result(error_code::SUCCESS, len, 766| 4| size_t(ret.second - utf8_output)); 767| 444|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_18internal8westmere26write_v_u16_11bits_to_utf8EDv2_xRPcS4_S4_: 57| 410k| const __m128i v_ff80) { 58| | // no bits set above 7th bit 59| 410k| const __m128i one_byte_bytemask = 60| 410k| _mm_cmpeq_epi16(_mm_and_si128(v_u16, v_ff80), v_0000); 61| 410k| const uint16_t one_byte_bitmask = 62| 410k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 63| | 64| 410k| write_v_u16_11bits_to_utf8(v_u16, utf8_output, one_byte_bytemask, 65| 410k| one_byte_bitmask); 66| 410k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_18internal8westmere26write_v_u16_11bits_to_utf8EDv2_xRPcS4_t: 10| 724k| const uint16_t one_byte_bitmask) { 11| | // 0b1100_0000_1000_0000 12| 724k| const __m128i v_c080 = _mm_set1_epi16((int16_t)0xc080); 13| | // 0b0001_1111_0000_0000 14| 724k| const __m128i v_1f00 = _mm_set1_epi16((int16_t)0x1f00); 15| | // 0b0000_0000_0011_1111 16| 724k| 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| 724k| const __m128i t0 = _mm_slli_epi16(v_u16, 2); 24| | // t1 = [000a|aaaa|0000|0000] 25| 724k| const __m128i t1 = _mm_and_si128(t0, v_1f00); 26| | // t2 = [0000|0000|00bb|bbbb] 27| 724k| const __m128i t2 = _mm_and_si128(v_u16, v_003f); 28| | // t3 = [000a|aaaa|00bb|bbbb] 29| 724k| const __m128i t3 = _mm_or_si128(t1, t2); 30| | // t4 = [110a|aaaa|10bb|bbbb] 31| 724k| const __m128i t4 = _mm_or_si128(t3, v_c080); 32| | 33| | // 2. merge ASCII and 2-byte codewords 34| 724k| 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| 724k| const uint16_t m0 = one_byte_bitmask & 0x5555; // m0 = 0h0g0f0e0d0c0b0a 40| 724k| const uint16_t m1 = static_cast(m0 >> 7); // m1 = 00000000h0g0f0e0 41| 724k| const uint8_t m2 = static_cast((m0 | m1) & 0xff); // m2 = hdgcfbea 42| | // 4. pack the bytes 43| 724k| const uint8_t *row = 44| 724k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[m2][0]; 45| 724k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 46| 724k| const __m128i utf8_packed = _mm_shuffle_epi8(utf8_unpacked, shuffle); 47| | 48| | // 5. store bytes 49| 724k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 50| | 51| | // 6. adjust pointers 52| 724k| utf8_output += row[0]; 53| 724k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_latin1_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 80| char16_t *utf16_output) { 5| 80| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| 159k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (6:22): [True: 159k, False: 80] ------------------ 7| | // Load 16 Latin1 characters into a 128-bit register 8| 159k| __m128i in = 9| 159k| _mm_loadu_si128(reinterpret_cast(&latin1_input[i])); 10| 159k| __m128i out1 = big_endian ? _mm_unpacklo_epi8(_mm_setzero_si128(), in) ------------------ | Branch (10:20): [Folded, False: 159k] ------------------ 11| 159k| : _mm_unpacklo_epi8(in, _mm_setzero_si128()); 12| 159k| __m128i out2 = big_endian ? _mm_unpackhi_epi8(_mm_setzero_si128(), in) ------------------ | Branch (12:20): [Folded, False: 159k] ------------------ 13| 159k| : _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| 159k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i]), out1); 17| 159k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i + 8]), out2); 18| 159k| } 19| | // return pointers pointing to where we left off 20| 80| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 21| 80|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_latin1_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 80| char16_t *utf16_output) { 5| 80| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| 147k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (6:22): [True: 147k, False: 80] ------------------ 7| | // Load 16 Latin1 characters into a 128-bit register 8| 147k| __m128i in = 9| 147k| _mm_loadu_si128(reinterpret_cast(&latin1_input[i])); 10| 147k| __m128i out1 = big_endian ? _mm_unpacklo_epi8(_mm_setzero_si128(), in) ------------------ | Branch (10:20): [True: 147k, Folded] ------------------ 11| 147k| : _mm_unpacklo_epi8(in, _mm_setzero_si128()); 12| 147k| __m128i out2 = big_endian ? _mm_unpackhi_epi8(_mm_setzero_si128(), in) ------------------ | Branch (12:20): [True: 147k, Folded] ------------------ 13| 147k| : _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| 147k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i]), out1); 17| 147k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i + 8]), out2); 18| 147k| } 19| | // return pointers pointing to where we left off 20| 80| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 21| 80|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_latin1_to_utf32EPKcmPDi: 3| 90| char32_t *utf32_output) { 4| 90| const char *end = buf + len; 5| | 6| 168k| while (end - buf >= 16) { ------------------ | Branch (6:10): [True: 168k, False: 90] ------------------ 7| | // Load 16 Latin1 characters (16 bytes) into a 128-bit register 8| 168k| __m128i in = _mm_loadu_si128((__m128i *)buf); 9| | 10| | // Shift input to process next 4 bytes 11| 168k| __m128i in_shifted1 = _mm_srli_si128(in, 4); 12| 168k| __m128i in_shifted2 = _mm_srli_si128(in, 8); 13| 168k| __m128i in_shifted3 = _mm_srli_si128(in, 12); 14| | 15| | // expand 8-bit to 32-bit unit 16| 168k| __m128i out1 = _mm_cvtepu8_epi32(in); 17| 168k| __m128i out2 = _mm_cvtepu8_epi32(in_shifted1); 18| 168k| __m128i out3 = _mm_cvtepu8_epi32(in_shifted2); 19| 168k| __m128i out4 = _mm_cvtepu8_epi32(in_shifted3); 20| | 21| 168k| _mm_storeu_si128((__m128i *)utf32_output, out1); 22| 168k| _mm_storeu_si128((__m128i *)(utf32_output + 4), out2); 23| 168k| _mm_storeu_si128((__m128i *)(utf32_output + 8), out3); 24| 168k| _mm_storeu_si128((__m128i *)(utf32_output + 12), out4); 25| | 26| 168k| utf32_output += 16; 27| 168k| buf += 16; 28| 168k| } 29| | 30| 90| return std::make_pair(buf, utf32_output); 31| 90|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_latin1_to_utf8EPKcmPc: 3| 538| const size_t latin_input_length, char *utf8_output) { 4| 538| const char *end = latin_input + latin_input_length; 5| | 6| 538| const __m128i v_0000 = _mm_setzero_si128(); 7| | // 0b1000_0000 8| 538| const __m128i v_80 = _mm_set1_epi8((uint8_t)0x80); 9| | // 0b1111_1111_1000_0000 10| 538| const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80); 11| | 12| 538| const __m128i latin_1_half_into_u16_byte_mask = 13| 538| _mm_setr_epi8(0, '\x80', 1, '\x80', 2, '\x80', 3, '\x80', 4, '\x80', 5, 14| 538| '\x80', 6, '\x80', 7, '\x80'); 15| | 16| 538| const __m128i latin_2_half_into_u16_byte_mask = 17| 538| _mm_setr_epi8(8, '\x80', 9, '\x80', 10, '\x80', 11, '\x80', 12, '\x80', 18| 538| 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| 2.47M| while (end - latin_input >= 16 + 8) { ------------------ | Branch (25:10): [True: 2.47M, False: 538] ------------------ 26| | // Load 16 Latin1 characters (16 bytes) into a 128-bit register 27| 2.47M| __m128i v_latin = _mm_loadu_si128((__m128i *)latin_input); 28| | 29| 2.47M| if (_mm_testz_si128(v_latin, v_80)) { // ASCII fast path!!!! ------------------ | Branch (29:9): [True: 2.26M, False: 205k] ------------------ 30| 2.26M| _mm_storeu_si128((__m128i *)utf8_output, v_latin); 31| 2.26M| latin_input += 16; 32| 2.26M| utf8_output += 16; 33| 2.26M| continue; 34| 2.26M| } 35| | 36| | // assuming a/b are bytes and A/B are uint16 of the same value 37| | // aaaa_aaaa_bbbb_bbbb -> AAAA_AAAA 38| 205k| __m128i v_u16_latin_1_half = 39| 205k| _mm_shuffle_epi8(v_latin, latin_1_half_into_u16_byte_mask); 40| | // aaaa_aaaa_bbbb_bbbb -> BBBB_BBBB 41| 205k| __m128i v_u16_latin_2_half = 42| 205k| _mm_shuffle_epi8(v_latin, latin_2_half_into_u16_byte_mask); 43| | 44| 205k| internal::westmere::write_v_u16_11bits_to_utf8(v_u16_latin_1_half, 45| 205k| utf8_output, v_0000, v_ff80); 46| 205k| internal::westmere::write_v_u16_11bits_to_utf8(v_u16_latin_2_half, 47| 205k| utf8_output, v_0000, v_ff80); 48| 205k| latin_input += 16; 49| 205k| } 50| | 51| 538| if (end - latin_input >= 16) { ------------------ | Branch (51:7): [True: 294, False: 244] ------------------ 52| | // Load 16 Latin1 characters (16 bytes) into a 128-bit register 53| 294| __m128i v_latin = _mm_loadu_si128((__m128i *)latin_input); 54| | 55| 294| if (_mm_testz_si128(v_latin, v_80)) { // ASCII fast path!!!! ------------------ | Branch (55:9): [True: 110, False: 184] ------------------ 56| 110| _mm_storeu_si128((__m128i *)utf8_output, v_latin); 57| 110| latin_input += 16; 58| 110| utf8_output += 16; 59| 184| } else { 60| | // assuming a/b are bytes and A/B are uint16 of the same value 61| | // aaaa_aaaa_bbbb_bbbb -> AAAA_AAAA 62| 184| __m128i v_u16_latin_1_half = 63| 184| _mm_shuffle_epi8(v_latin, latin_1_half_into_u16_byte_mask); 64| 184| internal::westmere::write_v_u16_11bits_to_utf8( 65| 184| v_u16_latin_1_half, utf8_output, v_0000, v_ff80); 66| 184| latin_input += 8; 67| 184| } 68| 294| } 69| | 70| 538| return std::make_pair(latin_input, utf8_output); 71| 538|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_utf16_to_latin1ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 124| char *latin1_output) { 5| 124| const char16_t *end = buf + len; 6| 39.1k| while (end - buf >= 8) { ------------------ | Branch (6:10): [True: 39.1k, False: 74] ------------------ 7| | // Load 8 UTF-16 characters into 128-bit SSE register 8| 39.1k| __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| 39.1k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 17| 39.1k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (17:9): [True: 39.0k, False: 50] ------------------ 18| | // Pack 16-bit characters into 8-bit and store in latin1_output 19| 39.0k| __m128i latin1_packed = _mm_packus_epi16(in, in); 20| 39.0k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 21| 39.0k| latin1_packed); 22| | // Adjust pointers for next iteration 23| 39.0k| buf += 8; 24| 39.0k| latin1_output += 8; 25| 39.0k| } else { 26| 50| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 27| 50| } 28| 39.1k| } // while 29| 74| return std::make_pair(buf, latin1_output); 30| 124|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_utf16_to_latin1ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 126| char *latin1_output) { 5| 126| const char16_t *end = buf + len; 6| 69.3k| while (end - buf >= 8) { ------------------ | Branch (6:10): [True: 69.2k, False: 92] ------------------ 7| | // Load 8 UTF-16 characters into 128-bit SSE register 8| 69.2k| __m128i in = _mm_loadu_si128(reinterpret_cast(buf)); 9| | 10| 69.2k| if constexpr (!match_system(big_endian)) { 11| 69.2k| const __m128i swap = 12| 69.2k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 13| 69.2k| in = _mm_shuffle_epi8(in, swap); 14| 69.2k| } 15| | 16| 69.2k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 17| 69.2k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (17:9): [True: 69.2k, False: 34] ------------------ 18| | // Pack 16-bit characters into 8-bit and store in latin1_output 19| 69.2k| __m128i latin1_packed = _mm_packus_epi16(in, in); 20| 69.2k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 21| 69.2k| latin1_packed); 22| | // Adjust pointers for next iteration 23| 69.2k| buf += 8; 24| 69.2k| latin1_output += 8; 25| 69.2k| } else { 26| 34| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 27| 34| } 28| 69.2k| } // while 29| 92| return std::make_pair(buf, latin1_output); 30| 126|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_139sse_convert_utf16_to_latin1_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 35| 230| char *latin1_output) { 36| 230| const char16_t *start = buf; 37| 230| const char16_t *end = buf + len; 38| 903k| while (end - buf >= 8) { ------------------ | Branch (38:10): [True: 903k, False: 102] ------------------ 39| 903k| __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| 903k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 48| 903k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (48:9): [True: 903k, False: 128] ------------------ 49| 903k| __m128i latin1_packed = _mm_packus_epi16(in, in); 50| 903k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 51| 903k| latin1_packed); 52| 903k| buf += 8; 53| 903k| latin1_output += 8; 54| 903k| } else { 55| | // Fallback to scalar code for handling errors 56| 458| for (int k = 0; k < 8; k++) { ------------------ | Branch (56:23): [True: 458, False: 0] ------------------ 57| 458| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 58| 458| if (word <= 0xff) { ------------------ | Branch (58:13): [True: 330, False: 128] ------------------ 59| 330| *latin1_output++ = char(word); 60| 330| } else { 61| 128| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 62| 128| latin1_output); 63| 128| } 64| 458| } 65| 0| buf += 8; 66| 0| } 67| 903k| } // while 68| 102| return std::make_pair(result(error_code::SUCCESS, buf - start), 69| 102| latin1_output); 70| 230|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_139sse_convert_utf16_to_latin1_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 35| 252| char *latin1_output) { 36| 252| const char16_t *start = buf; 37| 252| const char16_t *end = buf + len; 38| 1.04M| while (end - buf >= 8) { ------------------ | Branch (38:10): [True: 1.04M, False: 124] ------------------ 39| 1.04M| __m128i in = _mm_loadu_si128(reinterpret_cast(buf)); 40| | 41| 1.04M| if constexpr (!match_system(big_endian)) { 42| 1.04M| const __m128i swap = 43| 1.04M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 44| 1.04M| in = _mm_shuffle_epi8(in, swap); 45| 1.04M| } 46| | 47| 1.04M| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 48| 1.04M| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (48:9): [True: 1.04M, False: 128] ------------------ 49| 1.04M| __m128i latin1_packed = _mm_packus_epi16(in, in); 50| 1.04M| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 51| 1.04M| latin1_packed); 52| 1.04M| buf += 8; 53| 1.04M| latin1_output += 8; 54| 1.04M| } else { 55| | // Fallback to scalar code for handling errors 56| 448| for (int k = 0; k < 8; k++) { ------------------ | Branch (56:23): [True: 448, False: 0] ------------------ 57| 448| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 58| 448| if (word <= 0xff) { ------------------ | Branch (58:13): [True: 320, False: 128] ------------------ 59| 320| *latin1_output++ = char(word); 60| 320| } else { 61| 128| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 62| 128| latin1_output); 63| 128| } 64| 448| } 65| 0| buf += 8; 66| 0| } 67| 1.04M| } // while 68| 124| return std::make_pair(result(error_code::SUCCESS, buf - start), 69| 124| latin1_output); 70| 252|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf16_to_utf32ILNS_10endiannessE0EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 384| char32_t *utf32_output) { 58| 384| const char16_t *end = buf + len; 59| | 60| 384| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 61| 384| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 62| | 63| 321k| while (end - buf >= 8) { ------------------ | Branch (63:10): [True: 320k, False: 272] ------------------ 64| 320k| __m128i in = _mm_loadu_si128((__m128i *)buf); 65| | 66| 320k| if (big_endian) { ------------------ | Branch (66:9): [Folded, False: 320k] ------------------ 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| 320k| const __m128i surrogates_bytemask = 76| 320k| _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| 320k| const uint16_t surrogates_bitmask = 81| 320k| 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| 320k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (84:9): [True: 316k, False: 3.82k] ------------------ 85| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 86| 316k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 87| 316k| _mm_cvtepu16_epi32(in)); 88| 316k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 89| 316k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 90| 316k| utf32_output += 8; 91| 316k| buf += 8; 92| | // surrogate pair(s) in a register 93| 316k| } 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| 3.82k| size_t forward = 15; 98| 3.82k| size_t k = 0; 99| 3.82k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (99:11): [True: 70, False: 3.75k] ------------------ 100| 70| forward = size_t(end - buf - 1); 101| 70| } 102| 55.0k| for (; k < forward; k++) { ------------------ | Branch (102:14): [True: 51.3k, False: 3.71k] ------------------ 103| 51.3k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 104| 51.3k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (104:13): [True: 46.3k, False: 4.96k] ------------------ 105| 46.3k| *utf32_output++ = char32_t(word); 106| 46.3k| } else { 107| | // must be a surrogate pair 108| 4.96k| uint16_t diff = uint16_t(word - 0xD800); 109| 4.96k| uint16_t next_word = 110| 4.96k| scalar::utf16::swap_if_needed(buf[k + 1]); 111| 4.96k| k++; 112| 4.96k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 113| 4.96k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (113:15): [True: 112, False: 4.84k] ------------------ 114| 112| return std::make_pair(nullptr, utf32_output); 115| 112| } 116| 4.84k| uint32_t value = (diff << 10) + diff2 + 0x10000; 117| 4.84k| *utf32_output++ = char32_t(value); 118| 4.84k| } 119| 51.3k| } 120| 3.71k| buf += k; 121| 3.71k| } 122| 320k| } // while 123| 272| return std::make_pair(buf, utf32_output); 124| 384|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf16_to_utf32ILNS_10endiannessE1EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 366| char32_t *utf32_output) { 58| 366| const char16_t *end = buf + len; 59| | 60| 366| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 61| 366| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 62| | 63| 643k| while (end - buf >= 8) { ------------------ | Branch (63:10): [True: 642k, False: 284] ------------------ 64| 642k| __m128i in = _mm_loadu_si128((__m128i *)buf); 65| | 66| 642k| if (big_endian) { ------------------ | Branch (66:9): [True: 642k, Folded] ------------------ 67| 642k| const __m128i swap = 68| 642k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 69| 642k| in = _mm_shuffle_epi8(in, swap); 70| 642k| } 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| 642k| const __m128i surrogates_bytemask = 76| 642k| _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| 642k| const uint16_t surrogates_bitmask = 81| 642k| 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| 642k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (84:9): [True: 641k, False: 1.09k] ------------------ 85| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 86| 641k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 87| 641k| _mm_cvtepu16_epi32(in)); 88| 641k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 89| 641k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 90| 641k| utf32_output += 8; 91| 641k| buf += 8; 92| | // surrogate pair(s) in a register 93| 641k| } 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| 1.09k| size_t forward = 15; 98| 1.09k| size_t k = 0; 99| 1.09k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (99:11): [True: 82, False: 1.01k] ------------------ 100| 82| forward = size_t(end - buf - 1); 101| 82| } 102| 14.5k| for (; k < forward; k++) { ------------------ | Branch (102:14): [True: 13.5k, False: 1.01k] ------------------ 103| 13.5k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 104| 13.5k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (104:13): [True: 11.7k, False: 1.88k] ------------------ 105| 11.7k| *utf32_output++ = char32_t(word); 106| 11.7k| } else { 107| | // must be a surrogate pair 108| 1.88k| uint16_t diff = uint16_t(word - 0xD800); 109| 1.88k| uint16_t next_word = 110| 1.88k| scalar::utf16::swap_if_needed(buf[k + 1]); 111| 1.88k| k++; 112| 1.88k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 113| 1.88k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (113:15): [True: 82, False: 1.79k] ------------------ 114| 82| return std::make_pair(nullptr, utf32_output); 115| 82| } 116| 1.79k| uint32_t value = (diff << 10) + diff2 + 0x10000; 117| 1.79k| *utf32_output++ = char32_t(value); 118| 1.79k| } 119| 13.5k| } 120| 1.01k| buf += k; 121| 1.01k| } 122| 642k| } // while 123| 284| return std::make_pair(buf, utf32_output); 124| 366|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf16_to_utf32_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 136| 414| char32_t *utf32_output) { 137| 414| const char16_t *start = buf; 138| 414| const char16_t *end = buf + len; 139| | 140| 414| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 141| 414| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 142| | 143| 308k| while (end - buf >= 8) { ------------------ | Branch (143:10): [True: 307k, False: 266] ------------------ 144| 307k| __m128i in = _mm_loadu_si128((__m128i *)buf); 145| | 146| 307k| if (big_endian) { ------------------ | Branch (146:9): [Folded, False: 307k] ------------------ 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| 307k| const __m128i surrogates_bytemask = 156| 307k| _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| 307k| const uint16_t surrogates_bitmask = 161| 307k| 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| 307k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (164:9): [True: 301k, False: 6.57k] ------------------ 165| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 166| 301k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 167| 301k| _mm_cvtepu16_epi32(in)); 168| 301k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 169| 301k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 170| 301k| utf32_output += 8; 171| 301k| buf += 8; 172| | // surrogate pair(s) in a register 173| 301k| } 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| 6.57k| size_t forward = 15; 178| 6.57k| size_t k = 0; 179| 6.57k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (179:11): [True: 88, False: 6.48k] ------------------ 180| 88| forward = size_t(end - buf - 1); 181| 88| } 182| 92.4k| for (; k < forward; k++) { ------------------ | Branch (182:14): [True: 86.0k, False: 6.42k] ------------------ 183| 86.0k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 184| 86.0k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (184:13): [True: 74.4k, False: 11.5k] ------------------ 185| 74.4k| *utf32_output++ = char32_t(word); 186| 74.4k| } else { 187| | // must be a surrogate pair 188| 11.5k| uint16_t diff = uint16_t(word - 0xD800); 189| 11.5k| uint16_t next_word = 190| 11.5k| scalar::utf16::swap_if_needed(buf[k + 1]); 191| 11.5k| k++; 192| 11.5k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 193| 11.5k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (193:15): [True: 148, False: 11.3k] ------------------ 194| 148| return std::make_pair( 195| 148| result(error_code::SURROGATE, buf - start + k - 1), 196| 148| utf32_output); 197| 148| } 198| 11.3k| uint32_t value = (diff << 10) + diff2 + 0x10000; 199| 11.3k| *utf32_output++ = char32_t(value); 200| 11.3k| } 201| 86.0k| } 202| 6.42k| buf += k; 203| 6.42k| } 204| 307k| } // while 205| 266| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 206| 414|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf16_to_utf32_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 136| 300| char32_t *utf32_output) { 137| 300| const char16_t *start = buf; 138| 300| const char16_t *end = buf + len; 139| | 140| 300| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 141| 300| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 142| | 143| 422k| while (end - buf >= 8) { ------------------ | Branch (143:10): [True: 422k, False: 224] ------------------ 144| 422k| __m128i in = _mm_loadu_si128((__m128i *)buf); 145| | 146| 422k| if (big_endian) { ------------------ | Branch (146:9): [True: 422k, Folded] ------------------ 147| 422k| const __m128i swap = 148| 422k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 149| 422k| in = _mm_shuffle_epi8(in, swap); 150| 422k| } 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| 422k| const __m128i surrogates_bytemask = 156| 422k| _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| 422k| const uint16_t surrogates_bitmask = 161| 422k| 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| 422k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (164:9): [True: 418k, False: 3.87k] ------------------ 165| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 166| 418k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 167| 418k| _mm_cvtepu16_epi32(in)); 168| 418k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 169| 418k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 170| 418k| utf32_output += 8; 171| 418k| buf += 8; 172| | // surrogate pair(s) in a register 173| 418k| } 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| 3.87k| size_t forward = 15; 178| 3.87k| size_t k = 0; 179| 3.87k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (179:11): [True: 54, False: 3.82k] ------------------ 180| 54| forward = size_t(end - buf - 1); 181| 54| } 182| 51.8k| for (; k < forward; k++) { ------------------ | Branch (182:14): [True: 48.0k, False: 3.80k] ------------------ 183| 48.0k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 184| 48.0k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (184:13): [True: 37.9k, False: 10.0k] ------------------ 185| 37.9k| *utf32_output++ = char32_t(word); 186| 37.9k| } else { 187| | // must be a surrogate pair 188| 10.0k| uint16_t diff = uint16_t(word - 0xD800); 189| 10.0k| uint16_t next_word = 190| 10.0k| scalar::utf16::swap_if_needed(buf[k + 1]); 191| 10.0k| k++; 192| 10.0k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 193| 10.0k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (193:15): [True: 76, False: 10.0k] ------------------ 194| 76| return std::make_pair( 195| 76| result(error_code::SURROGATE, buf - start + k - 1), 196| 76| utf32_output); 197| 76| } 198| 10.0k| uint32_t value = (diff << 10) + diff2 + 0x10000; 199| 10.0k| *utf32_output++ = char32_t(value); 200| 10.0k| } 201| 48.0k| } 202| 3.80k| buf += k; 203| 3.80k| } 204| 422k| } // while 205| 224| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 206| 300|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_125sse_convert_utf16_to_utf8ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 802|sse_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| | 58| 802| const char16_t *end = buf + len; 59| | 60| 802| const __m128i v_0000 = _mm_setzero_si128(); 61| 802| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 62| 802| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 63| 802| const size_t safety_margin = 64| 802| 12; // to avoid overruns, see issue 65| | // https://github.com/simdutf/simdutf/issues/92 66| | 67| 1.16M| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (67:10): [True: 1.16M, False: 718] ------------------ 68| 1.16M| __m128i in = _mm_loadu_si128((__m128i *)buf); 69| 1.16M| if (big_endian) { ------------------ | Branch (69:9): [Folded, False: 1.16M] ------------------ 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| 1.16M| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 76| 1.16M| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 1.02M, False: 139k] ------------------ 77| 1.02M| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 78| 1.02M| if (big_endian) { ------------------ | Branch (78:11): [Folded, False: 1.02M] ------------------ 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| 1.02M| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (83:11): [True: 30.5k, False: 991k] ------------------ 84| | // 1. pack the bytes 85| | // obviously suboptimal. 86| 30.5k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 87| | // 2. store (16 bytes) 88| 30.5k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 89| | // 3. adjust pointers 90| 30.5k| buf += 8; 91| 30.5k| utf8_output += 8; 92| 30.5k| in = nextin; 93| 991k| } else { 94| | // 1. pack the bytes 95| | // obviously suboptimal. 96| 991k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 97| | // 2. store (16 bytes) 98| 991k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 99| | // 3. adjust pointers 100| 991k| buf += 16; 101| 991k| utf8_output += 16; 102| 991k| continue; // we are done for this round! 103| 991k| } 104| 1.02M| } 105| | 106| | // no bits set above 7th bit 107| 169k| const __m128i one_byte_bytemask = 108| 169k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 109| 169k| const uint16_t one_byte_bitmask = 110| 169k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 111| | 112| | // no bits set above 11th bit 113| 169k| const __m128i one_or_two_bytes_bytemask = 114| 169k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 115| 169k| const uint16_t one_or_two_bytes_bitmask = 116| 169k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 117| | 118| 169k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (118:9): [True: 50.9k, False: 118k] ------------------ 119| 50.9k| internal::westmere::write_v_u16_11bits_to_utf8( 120| 50.9k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 121| 50.9k| buf += 8; 122| 50.9k| continue; 123| 50.9k| } 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| 118k| const __m128i surrogates_bytemask = 129| 118k| _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| 118k| const uint16_t surrogates_bitmask = 134| 118k| 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| 118k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (137:9): [True: 111k, False: 6.99k] ------------------ 138| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 139| 111k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 140| 111k| 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| 111k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 170| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 171| 111k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 172| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 173| 111k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 169| 111k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 174| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 175| 111k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 169| 111k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 176| | 177| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 178| 111k| const __m128i s0 = _mm_srli_epi16(in, 4); 179| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 180| 111k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 169| 111k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 181| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 182| 111k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 169| 111k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 183| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 184| 111k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 169| 111k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 185| 111k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 186| 111k| simdutf_vec(0b0100000000000000)); ------------------ | | 169| 111k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 187| 111k| const __m128i s4 = _mm_xor_si128(s3, m0); 188| 111k|#undef simdutf_vec 189| | 190| | // 4. expand code units 16-bit => 32-bit 191| 111k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 192| 111k| 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| 111k| const uint16_t mask = 196| 111k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 197| 111k| if (mask == 0) { ------------------ | Branch (197:11): [True: 26.9k, False: 84.6k] ------------------ 198| | // We only have three-byte code units. Use fast path. 199| 26.9k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 200| 26.9k| 15, 13, -1, -1, -1, -1); 201| 26.9k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 202| 26.9k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 203| 26.9k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 204| 26.9k| utf8_output += 12; 205| 26.9k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 206| 26.9k| utf8_output += 12; 207| 26.9k| buf += 8; 208| 26.9k| continue; 209| 26.9k| } 210| 84.6k| const uint8_t mask0 = uint8_t(mask); 211| | 212| 84.6k| const uint8_t *row0 = 213| 84.6k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 214| 84.6k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 215| 84.6k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 216| | 217| 84.6k| const uint8_t mask1 = static_cast(mask >> 8); 218| | 219| 84.6k| const uint8_t *row1 = 220| 84.6k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 221| 84.6k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 222| 84.6k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 223| | 224| 84.6k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 225| 84.6k| utf8_output += row0[0]; 226| 84.6k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 227| 84.6k| utf8_output += row1[0]; 228| | 229| 84.6k| buf += 8; 230| | // surrogate pair(s) in a register 231| 84.6k| } else { 232| | // Let us do a scalar fallback. 233| | // It may seem wasteful to use scalar code, but being efficient with SIMD 234| | // in the presence of surrogate pairs may require non-trivial tables. 235| 6.99k| size_t forward = 15; 236| 6.99k| size_t k = 0; 237| 6.99k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (237:11): [True: 0, False: 6.99k] ------------------ 238| 0| forward = size_t(end - buf - 1); 239| 0| } 240| 103k| for (; k < forward; k++) { ------------------ | Branch (240:14): [True: 96.1k, False: 6.90k] ------------------ 241| 96.1k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 242| 96.1k| if ((word & 0xFF80) == 0) { ------------------ | Branch (242:13): [True: 13.0k, False: 83.1k] ------------------ 243| 13.0k| *utf8_output++ = char(word); 244| 83.1k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (244:20): [True: 4.49k, False: 78.6k] ------------------ 245| 4.49k| *utf8_output++ = char((word >> 6) | 0b11000000); 246| 4.49k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 247| 78.6k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (247:20): [True: 70.4k, False: 8.13k] ------------------ 248| 70.4k| *utf8_output++ = char((word >> 12) | 0b11100000); 249| 70.4k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 250| 70.4k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 251| 70.4k| } else { 252| | // must be a surrogate pair 253| 8.13k| uint16_t diff = uint16_t(word - 0xD800); 254| 8.13k| uint16_t next_word = 255| 8.13k| scalar::utf16::swap_if_needed(buf[k + 1]); 256| 8.13k| k++; 257| 8.13k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 258| 8.13k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (258:15): [True: 84, False: 8.05k] ------------------ 259| 84| return std::make_pair(nullptr, utf8_output); 260| 84| } 261| 8.05k| uint32_t value = (diff << 10) + diff2 + 0x10000; 262| 8.05k| *utf8_output++ = char((value >> 18) | 0b11110000); 263| 8.05k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 264| 8.05k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 265| 8.05k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 266| 8.05k| } 267| 96.1k| } 268| 6.90k| buf += k; 269| 6.90k| } 270| 118k| } // while 271| | 272| 718| return std::make_pair(buf, utf8_output); 273| 802|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_125sse_convert_utf16_to_utf8ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 768|sse_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| | 58| 768| const char16_t *end = buf + len; 59| | 60| 768| const __m128i v_0000 = _mm_setzero_si128(); 61| 768| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 62| 768| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 63| 768| const size_t safety_margin = 64| 768| 12; // to avoid overruns, see issue 65| | // https://github.com/simdutf/simdutf/issues/92 66| | 67| 1.35M| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (67:10): [True: 1.35M, False: 690] ------------------ 68| 1.35M| __m128i in = _mm_loadu_si128((__m128i *)buf); 69| 1.35M| if (big_endian) { ------------------ | Branch (69:9): [True: 1.35M, Folded] ------------------ 70| 1.35M| const __m128i swap = 71| 1.35M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 72| 1.35M| in = _mm_shuffle_epi8(in, swap); 73| 1.35M| } 74| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 75| 1.35M| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 76| 1.35M| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 847k, False: 504k] ------------------ 77| 847k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 78| 847k| if (big_endian) { ------------------ | Branch (78:11): [True: 847k, Folded] ------------------ 79| 847k| const __m128i swap = 80| 847k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 81| 847k| nextin = _mm_shuffle_epi8(nextin, swap); 82| 847k| } 83| 847k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (83:11): [True: 45.0k, False: 802k] ------------------ 84| | // 1. pack the bytes 85| | // obviously suboptimal. 86| 45.0k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 87| | // 2. store (16 bytes) 88| 45.0k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 89| | // 3. adjust pointers 90| 45.0k| buf += 8; 91| 45.0k| utf8_output += 8; 92| 45.0k| in = nextin; 93| 802k| } else { 94| | // 1. pack the bytes 95| | // obviously suboptimal. 96| 802k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 97| | // 2. store (16 bytes) 98| 802k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 99| | // 3. adjust pointers 100| 802k| buf += 16; 101| 802k| utf8_output += 16; 102| 802k| continue; // we are done for this round! 103| 802k| } 104| 847k| } 105| | 106| | // no bits set above 7th bit 107| 549k| const __m128i one_byte_bytemask = 108| 549k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 109| 549k| const uint16_t one_byte_bitmask = 110| 549k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 111| | 112| | // no bits set above 11th bit 113| 549k| const __m128i one_or_two_bytes_bytemask = 114| 549k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 115| 549k| const uint16_t one_or_two_bytes_bitmask = 116| 549k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 117| | 118| 549k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (118:9): [True: 34.5k, False: 514k] ------------------ 119| 34.5k| internal::westmere::write_v_u16_11bits_to_utf8( 120| 34.5k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 121| 34.5k| buf += 8; 122| 34.5k| continue; 123| 34.5k| } 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| 514k| const __m128i surrogates_bytemask = 129| 514k| _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| 514k| const uint16_t surrogates_bitmask = 134| 514k| 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| 514k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (137:9): [True: 508k, False: 6.71k] ------------------ 138| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 139| 508k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 140| 508k| 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| 508k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 170| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 171| 508k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 172| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 173| 508k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 169| 508k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 174| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 175| 508k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 169| 508k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 176| | 177| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 178| 508k| const __m128i s0 = _mm_srli_epi16(in, 4); 179| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 180| 508k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 169| 508k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 181| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 182| 508k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 169| 508k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 183| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 184| 508k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 169| 508k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 185| 508k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 186| 508k| simdutf_vec(0b0100000000000000)); ------------------ | | 169| 508k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 187| 508k| const __m128i s4 = _mm_xor_si128(s3, m0); 188| 508k|#undef simdutf_vec 189| | 190| | // 4. expand code units 16-bit => 32-bit 191| 508k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 192| 508k| 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| 508k| const uint16_t mask = 196| 508k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 197| 508k| if (mask == 0) { ------------------ | Branch (197:11): [True: 274k, False: 233k] ------------------ 198| | // We only have three-byte code units. Use fast path. 199| 274k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 200| 274k| 15, 13, -1, -1, -1, -1); 201| 274k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 202| 274k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 203| 274k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 204| 274k| utf8_output += 12; 205| 274k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 206| 274k| utf8_output += 12; 207| 274k| buf += 8; 208| 274k| continue; 209| 274k| } 210| 233k| const uint8_t mask0 = uint8_t(mask); 211| | 212| 233k| const uint8_t *row0 = 213| 233k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 214| 233k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 215| 233k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 216| | 217| 233k| const uint8_t mask1 = static_cast(mask >> 8); 218| | 219| 233k| const uint8_t *row1 = 220| 233k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 221| 233k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 222| 233k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 223| | 224| 233k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 225| 233k| utf8_output += row0[0]; 226| 233k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 227| 233k| utf8_output += row1[0]; 228| | 229| 233k| buf += 8; 230| | // surrogate pair(s) in a register 231| 233k| } else { 232| | // Let us do a scalar fallback. 233| | // It may seem wasteful to use scalar code, but being efficient with SIMD 234| | // in the presence of surrogate pairs may require non-trivial tables. 235| 6.71k| size_t forward = 15; 236| 6.71k| size_t k = 0; 237| 6.71k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (237:11): [True: 0, False: 6.71k] ------------------ 238| 0| forward = size_t(end - buf - 1); 239| 0| } 240| 98.6k| for (; k < forward; k++) { ------------------ | Branch (240:14): [True: 91.9k, False: 6.63k] ------------------ 241| 91.9k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 242| 91.9k| if ((word & 0xFF80) == 0) { ------------------ | Branch (242:13): [True: 10.7k, False: 81.1k] ------------------ 243| 10.7k| *utf8_output++ = char(word); 244| 81.1k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (244:20): [True: 2.13k, False: 79.0k] ------------------ 245| 2.13k| *utf8_output++ = char((word >> 6) | 0b11000000); 246| 2.13k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 247| 79.0k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (247:20): [True: 70.8k, False: 8.15k] ------------------ 248| 70.8k| *utf8_output++ = char((word >> 12) | 0b11100000); 249| 70.8k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 250| 70.8k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 251| 70.8k| } else { 252| | // must be a surrogate pair 253| 8.15k| uint16_t diff = uint16_t(word - 0xD800); 254| 8.15k| uint16_t next_word = 255| 8.15k| scalar::utf16::swap_if_needed(buf[k + 1]); 256| 8.15k| k++; 257| 8.15k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 258| 8.15k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (258:15): [True: 78, False: 8.07k] ------------------ 259| 78| return std::make_pair(nullptr, utf8_output); 260| 78| } 261| 8.07k| uint32_t value = (diff << 10) + diff2 + 0x10000; 262| 8.07k| *utf8_output++ = char((value >> 18) | 0b11110000); 263| 8.07k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 264| 8.07k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 265| 8.07k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 266| 8.07k| } 267| 91.9k| } 268| 6.63k| buf += k; 269| 6.63k| } 270| 514k| } // while 271| | 272| 690| return std::make_pair(buf, utf8_output); 273| 768|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_137sse_convert_utf16_to_utf8_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 285| 630| char *utf8_output) { 286| 630| const char16_t *start = buf; 287| 630| const char16_t *end = buf + len; 288| | 289| 630| const __m128i v_0000 = _mm_setzero_si128(); 290| 630| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 291| 630| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 292| 630| const size_t safety_margin = 293| 630| 12; // to avoid overruns, see issue 294| | // https://github.com/simdutf/simdutf/issues/92 295| | 296| 1.06M| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (296:10): [True: 1.06M, False: 518] ------------------ 297| 1.06M| __m128i in = _mm_loadu_si128((__m128i *)buf); 298| 1.06M| if (big_endian) { ------------------ | Branch (298:9): [Folded, False: 1.06M] ------------------ 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| 1.06M| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 305| 1.06M| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (305:9): [True: 479k, False: 589k] ------------------ 306| 479k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 307| 479k| if (big_endian) { ------------------ | Branch (307:11): [Folded, False: 479k] ------------------ 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| 479k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (312:11): [True: 15.2k, False: 464k] ------------------ 313| | // 1. pack the bytes 314| | // obviously suboptimal. 315| 15.2k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 316| | // 2. store (16 bytes) 317| 15.2k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 318| | // 3. adjust pointers 319| 15.2k| buf += 8; 320| 15.2k| utf8_output += 8; 321| 15.2k| in = nextin; 322| 464k| } else { 323| | // 1. pack the bytes 324| | // obviously suboptimal. 325| 464k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 326| | // 2. store (16 bytes) 327| 464k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 328| | // 3. adjust pointers 329| 464k| buf += 16; 330| 464k| utf8_output += 16; 331| 464k| continue; // we are done for this round! 332| 464k| } 333| 479k| } 334| | 335| | // no bits set above 7th bit 336| 605k| const __m128i one_byte_bytemask = 337| 605k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 338| 605k| const uint16_t one_byte_bitmask = 339| 605k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 340| | 341| | // no bits set above 11th bit 342| 605k| const __m128i one_or_two_bytes_bytemask = 343| 605k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 344| 605k| const uint16_t one_or_two_bytes_bitmask = 345| 605k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 346| | 347| 605k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (347:9): [True: 188k, False: 416k] ------------------ 348| 188k| internal::westmere::write_v_u16_11bits_to_utf8( 349| 188k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 350| 188k| buf += 8; 351| 188k| continue; 352| 188k| } 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| 416k| const __m128i surrogates_bytemask = 358| 416k| _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| 416k| const uint16_t surrogates_bitmask = 363| 416k| 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| 416k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (366:9): [True: 412k, False: 3.55k] ------------------ 367| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 368| 412k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 369| 412k| 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| 412k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 399| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 400| 412k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 401| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 402| 412k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 398| 412k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 403| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 404| 412k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 398| 412k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 405| | 406| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 407| 412k| const __m128i s0 = _mm_srli_epi16(in, 4); 408| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 409| 412k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 398| 412k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 410| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 411| 412k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 398| 412k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 412| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 413| 412k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 398| 412k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 414| 412k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 415| 412k| simdutf_vec(0b0100000000000000)); ------------------ | | 398| 412k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 416| 412k| const __m128i s4 = _mm_xor_si128(s3, m0); 417| 412k|#undef simdutf_vec 418| | 419| | // 4. expand code units 16-bit => 32-bit 420| 412k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 421| 412k| 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| 412k| const uint16_t mask = 425| 412k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 426| 412k| if (mask == 0) { ------------------ | Branch (426:11): [True: 318k, False: 94.5k] ------------------ 427| | // We only have three-byte code units. Use fast path. 428| 318k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 429| 318k| 15, 13, -1, -1, -1, -1); 430| 318k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 431| 318k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 432| 318k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 433| 318k| utf8_output += 12; 434| 318k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 435| 318k| utf8_output += 12; 436| 318k| buf += 8; 437| 318k| continue; 438| 318k| } 439| 94.5k| const uint8_t mask0 = uint8_t(mask); 440| | 441| 94.5k| const uint8_t *row0 = 442| 94.5k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 443| 94.5k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 444| 94.5k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 445| | 446| 94.5k| const uint8_t mask1 = static_cast(mask >> 8); 447| | 448| 94.5k| const uint8_t *row1 = 449| 94.5k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 450| 94.5k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 451| 94.5k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 452| | 453| 94.5k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 454| 94.5k| utf8_output += row0[0]; 455| 94.5k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 456| 94.5k| utf8_output += row1[0]; 457| | 458| 94.5k| buf += 8; 459| | // surrogate pair(s) in a register 460| 94.5k| } 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| 3.55k| size_t forward = 15; 465| 3.55k| size_t k = 0; 466| 3.55k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (466:11): [True: 0, False: 3.55k] ------------------ 467| 0| forward = size_t(end - buf - 1); 468| 0| } 469| 51.6k| for (; k < forward; k++) { ------------------ | Branch (469:14): [True: 48.1k, False: 3.44k] ------------------ 470| 48.1k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 471| 48.1k| if ((word & 0xFF80) == 0) { ------------------ | Branch (471:13): [True: 8.91k, False: 39.2k] ------------------ 472| 8.91k| *utf8_output++ = char(word); 473| 39.2k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (473:20): [True: 2.02k, False: 37.2k] ------------------ 474| 2.02k| *utf8_output++ = char((word >> 6) | 0b11000000); 475| 2.02k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 476| 37.2k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (476:20): [True: 33.0k, False: 4.19k] ------------------ 477| 33.0k| *utf8_output++ = char((word >> 12) | 0b11100000); 478| 33.0k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 479| 33.0k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 480| 33.0k| } else { 481| | // must be a surrogate pair 482| 4.19k| uint16_t diff = uint16_t(word - 0xD800); 483| 4.19k| uint16_t next_word = 484| 4.19k| scalar::utf16::swap_if_needed(buf[k + 1]); 485| 4.19k| k++; 486| 4.19k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 487| 4.19k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (487:15): [True: 112, False: 4.08k] ------------------ 488| 112| return std::make_pair( 489| 112| result(error_code::SURROGATE, buf - start + k - 1), 490| 112| utf8_output); 491| 112| } 492| 4.08k| uint32_t value = (diff << 10) + diff2 + 0x10000; 493| 4.08k| *utf8_output++ = char((value >> 18) | 0b11110000); 494| 4.08k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 495| 4.08k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 496| 4.08k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 497| 4.08k| } 498| 48.1k| } 499| 3.44k| buf += k; 500| 3.44k| } 501| 416k| } // while 502| | 503| 518| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 504| 630|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_137sse_convert_utf16_to_utf8_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 285| 586| char *utf8_output) { 286| 586| const char16_t *start = buf; 287| 586| const char16_t *end = buf + len; 288| | 289| 586| const __m128i v_0000 = _mm_setzero_si128(); 290| 586| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 291| 586| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 292| 586| const size_t safety_margin = 293| 586| 12; // to avoid overruns, see issue 294| | // https://github.com/simdutf/simdutf/issues/92 295| | 296| 523k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (296:10): [True: 522k, False: 444] ------------------ 297| 522k| __m128i in = _mm_loadu_si128((__m128i *)buf); 298| 522k| if (big_endian) { ------------------ | Branch (298:9): [True: 522k, Folded] ------------------ 299| 522k| const __m128i swap = 300| 522k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 301| 522k| in = _mm_shuffle_epi8(in, swap); 302| 522k| } 303| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 304| 522k| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 305| 522k| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (305:9): [True: 405k, False: 117k] ------------------ 306| 405k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 307| 405k| if (big_endian) { ------------------ | Branch (307:11): [True: 405k, Folded] ------------------ 308| 405k| const __m128i swap = 309| 405k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 310| 405k| nextin = _mm_shuffle_epi8(nextin, swap); 311| 405k| } 312| 405k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (312:11): [True: 31.3k, False: 373k] ------------------ 313| | // 1. pack the bytes 314| | // obviously suboptimal. 315| 31.3k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 316| | // 2. store (16 bytes) 317| 31.3k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 318| | // 3. adjust pointers 319| 31.3k| buf += 8; 320| 31.3k| utf8_output += 8; 321| 31.3k| in = nextin; 322| 373k| } else { 323| | // 1. pack the bytes 324| | // obviously suboptimal. 325| 373k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 326| | // 2. store (16 bytes) 327| 373k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 328| | // 3. adjust pointers 329| 373k| buf += 16; 330| 373k| utf8_output += 16; 331| 373k| continue; // we are done for this round! 332| 373k| } 333| 405k| } 334| | 335| | // no bits set above 7th bit 336| 148k| const __m128i one_byte_bytemask = 337| 148k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 338| 148k| const uint16_t one_byte_bitmask = 339| 148k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 340| | 341| | // no bits set above 11th bit 342| 148k| const __m128i one_or_two_bytes_bytemask = 343| 148k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 344| 148k| const uint16_t one_or_two_bytes_bitmask = 345| 148k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 346| | 347| 148k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (347:9): [True: 39.7k, False: 109k] ------------------ 348| 39.7k| internal::westmere::write_v_u16_11bits_to_utf8( 349| 39.7k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 350| 39.7k| buf += 8; 351| 39.7k| continue; 352| 39.7k| } 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| 109k| const __m128i surrogates_bytemask = 358| 109k| _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| 109k| const uint16_t surrogates_bitmask = 363| 109k| 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| 109k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (366:9): [True: 102k, False: 6.74k] ------------------ 367| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 368| 102k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 369| 102k| 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| 102k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 399| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 400| 102k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 401| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 402| 102k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 398| 102k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 403| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 404| 102k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 398| 102k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 405| | 406| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 407| 102k| const __m128i s0 = _mm_srli_epi16(in, 4); 408| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 409| 102k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 398| 102k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 410| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 411| 102k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 398| 102k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 412| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 413| 102k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 398| 102k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 414| 102k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 415| 102k| simdutf_vec(0b0100000000000000)); ------------------ | | 398| 102k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 416| 102k| const __m128i s4 = _mm_xor_si128(s3, m0); 417| 102k|#undef simdutf_vec 418| | 419| | // 4. expand code units 16-bit => 32-bit 420| 102k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 421| 102k| 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| 102k| const uint16_t mask = 425| 102k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 426| 102k| if (mask == 0) { ------------------ | Branch (426:11): [True: 27.1k, False: 75.1k] ------------------ 427| | // We only have three-byte code units. Use fast path. 428| 27.1k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 429| 27.1k| 15, 13, -1, -1, -1, -1); 430| 27.1k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 431| 27.1k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 432| 27.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 433| 27.1k| utf8_output += 12; 434| 27.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 435| 27.1k| utf8_output += 12; 436| 27.1k| buf += 8; 437| 27.1k| continue; 438| 27.1k| } 439| 75.1k| const uint8_t mask0 = uint8_t(mask); 440| | 441| 75.1k| const uint8_t *row0 = 442| 75.1k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 443| 75.1k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 444| 75.1k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 445| | 446| 75.1k| const uint8_t mask1 = static_cast(mask >> 8); 447| | 448| 75.1k| const uint8_t *row1 = 449| 75.1k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 450| 75.1k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 451| 75.1k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 452| | 453| 75.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 454| 75.1k| utf8_output += row0[0]; 455| 75.1k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 456| 75.1k| utf8_output += row1[0]; 457| | 458| 75.1k| buf += 8; 459| | // surrogate pair(s) in a register 460| 75.1k| } else { 461| | // Let us do a scalar fallback. 462| | // It may seem wasteful to use scalar code, but being efficient with SIMD 463| | // in the presence of surrogate pairs may require non-trivial tables. 464| 6.74k| size_t forward = 15; 465| 6.74k| size_t k = 0; 466| 6.74k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (466:11): [True: 0, False: 6.74k] ------------------ 467| 0| forward = size_t(end - buf - 1); 468| 0| } 469| 98.6k| for (; k < forward; k++) { ------------------ | Branch (469:14): [True: 92.0k, False: 6.59k] ------------------ 470| 92.0k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 471| 92.0k| if ((word & 0xFF80) == 0) { ------------------ | Branch (471:13): [True: 31.4k, False: 60.6k] ------------------ 472| 31.4k| *utf8_output++ = char(word); 473| 60.6k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (473:20): [True: 4.65k, False: 55.9k] ------------------ 474| 4.65k| *utf8_output++ = char((word >> 6) | 0b11000000); 475| 4.65k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 476| 55.9k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (476:20): [True: 48.1k, False: 7.83k] ------------------ 477| 48.1k| *utf8_output++ = char((word >> 12) | 0b11100000); 478| 48.1k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 479| 48.1k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 480| 48.1k| } else { 481| | // must be a surrogate pair 482| 7.83k| uint16_t diff = uint16_t(word - 0xD800); 483| 7.83k| uint16_t next_word = 484| 7.83k| scalar::utf16::swap_if_needed(buf[k + 1]); 485| 7.83k| k++; 486| 7.83k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 487| 7.83k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (487:15): [True: 142, False: 7.69k] ------------------ 488| 142| return std::make_pair( 489| 142| result(error_code::SURROGATE, buf - start + k - 1), 490| 142| utf8_output); 491| 142| } 492| 7.69k| uint32_t value = (diff << 10) + diff2 + 0x10000; 493| 7.69k| *utf8_output++ = char((value >> 18) | 0b11110000); 494| 7.69k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 495| 7.69k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 496| 7.69k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 497| 7.69k| } 498| 92.0k| } 499| 6.59k| buf += k; 500| 6.59k| } 501| 109k| } // while 502| | 503| 444| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 504| 586|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_utf32_to_latin1EPKDimPc: 3| 188| char *latin1_output) { 4| 188| const size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 5| | 6| 188| __m128i high_bytes_mask = _mm_set1_epi32(0xFFFFFF00); 7| 188| __m128i shufmask = 8| 188| _mm_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 12, 8, 4, 0); 9| | 10| 5.70k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (10:22): [True: 5.56k, False: 146] ------------------ 11| 5.56k| __m128i in1 = _mm_loadu_si128((__m128i *)buf); 12| 5.56k| __m128i in2 = _mm_loadu_si128((__m128i *)(buf + 4)); 13| 5.56k| __m128i in3 = _mm_loadu_si128((__m128i *)(buf + 8)); 14| 5.56k| __m128i in4 = _mm_loadu_si128((__m128i *)(buf + 12)); 15| | 16| 5.56k| __m128i check_combined = _mm_or_si128(in1, in2); 17| 5.56k| check_combined = _mm_or_si128(check_combined, in3); 18| 5.56k| check_combined = _mm_or_si128(check_combined, in4); 19| | 20| 5.56k| if (!_mm_testz_si128(check_combined, high_bytes_mask)) { ------------------ | Branch (20:9): [True: 42, False: 5.52k] ------------------ 21| 42| return std::make_pair(nullptr, latin1_output); 22| 42| } 23| 5.52k| __m128i pack1 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in1, shufmask), 24| 5.52k| _mm_shuffle_epi8(in2, shufmask)); 25| 5.52k| __m128i pack2 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in3, shufmask), 26| 5.52k| _mm_shuffle_epi8(in4, shufmask)); 27| 5.52k| __m128i pack = _mm_unpacklo_epi64(pack1, pack2); 28| 5.52k| _mm_storeu_si128((__m128i *)latin1_output, pack); 29| 5.52k| latin1_output += 16; 30| 5.52k| buf += 16; 31| 5.52k| } 32| | 33| 146| return std::make_pair(buf, latin1_output); 34| 188|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_139sse_convert_utf32_to_latin1_with_errorsEPKDimPc: 38| 420| char *latin1_output) { 39| 420| const char32_t *start = buf; 40| 420| const size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 41| | 42| 420| __m128i high_bytes_mask = _mm_set1_epi32(0xFFFFFF00); 43| 420| __m128i shufmask = 44| 420| _mm_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 12, 8, 4, 0); 45| | 46| 26.6k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (46:22): [True: 26.4k, False: 248] ------------------ 47| 26.4k| __m128i in1 = _mm_loadu_si128((__m128i *)buf); 48| 26.4k| __m128i in2 = _mm_loadu_si128((__m128i *)(buf + 4)); 49| 26.4k| __m128i in3 = _mm_loadu_si128((__m128i *)(buf + 8)); 50| 26.4k| __m128i in4 = _mm_loadu_si128((__m128i *)(buf + 12)); 51| | 52| 26.4k| __m128i check_combined = _mm_or_si128(in1, in2); 53| 26.4k| check_combined = _mm_or_si128(check_combined, in3); 54| 26.4k| check_combined = _mm_or_si128(check_combined, in4); 55| | 56| 26.4k| if (!_mm_testz_si128(check_combined, high_bytes_mask)) { ------------------ | Branch (56:9): [True: 172, False: 26.2k] ------------------ 57| | // Fallback to scalar code for handling errors 58| 694| for (int k = 0; k < 16; k++) { ------------------ | Branch (58:23): [True: 694, False: 0] ------------------ 59| 694| char32_t codepoint = buf[k]; 60| 694| if (codepoint <= 0xff) { ------------------ | Branch (60:13): [True: 522, False: 172] ------------------ 61| 522| *latin1_output++ = char(codepoint); 62| 522| } else { 63| 172| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 64| 172| latin1_output); 65| 172| } 66| 694| } 67| 0| buf += 16; 68| 0| continue; 69| 172| } 70| 26.2k| __m128i pack1 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in1, shufmask), 71| 26.2k| _mm_shuffle_epi8(in2, shufmask)); 72| 26.2k| __m128i pack2 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in3, shufmask), 73| 26.2k| _mm_shuffle_epi8(in4, shufmask)); 74| 26.2k| __m128i pack = _mm_unpacklo_epi64(pack1, pack2); 75| 26.2k| _mm_storeu_si128((__m128i *)latin1_output, pack); 76| 26.2k| latin1_output += 16; 77| 26.2k| buf += 16; 78| 26.2k| } 79| | 80| 248| return std::make_pair(result(error_code::SUCCESS, buf - start), 81| 248| latin1_output); 82| 420|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf32_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKDiPDsEES7_mS8_: 58| 722| char16_t *utf16_output) { 59| | 60| 722| const char32_t *end = buf + len; 61| | 62| 722| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 63| 722| __m128i forbidden_bytemask = _mm_setzero_si128(); 64| | 65| 5.66k| while (end - buf >= 16 + 8) { ------------------ | Branch (65:10): [True: 4.97k, False: 692] ------------------ 66| 4.97k| const __m128i *ptr = reinterpret_cast(buf); 67| 4.97k| const __m128i in0 = _mm_loadu_si128(ptr + 0); 68| 4.97k| const __m128i in1 = _mm_loadu_si128(ptr + 1); 69| 4.97k| const __m128i in2 = _mm_loadu_si128(ptr + 2); 70| 4.97k| const __m128i in3 = _mm_loadu_si128(ptr + 3); 71| | 72| 4.97k| const __m128i combined = 73| 4.97k| _mm_or_si128(_mm_or_si128(in2, in3), _mm_or_si128(in0, in1)); 74| 4.97k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 4.97k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.80k, False: 1.16k] | | ------------------ ------------------ 75| | // No bits set above 16th, directly pack UTF-32 to UTF-16 76| 3.80k| __m128i utf16_packed0 = _mm_packus_epi32(in0, in1); 77| 3.80k| __m128i utf16_packed1 = _mm_packus_epi32(in2, in3); 78| | 79| 3.80k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 80| 3.80k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 81| 3.80k| forbidden_bytemask = _mm_or_si128( 82| 3.80k| forbidden_bytemask, 83| 3.80k| _mm_or_si128( 84| 3.80k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed0, v_f800), v_d800), 85| 3.80k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed1, v_f800), v_d800))); 86| | 87| 3.80k| if (big_endian) { ------------------ | Branch (87:11): [Folded, False: 3.80k] ------------------ 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| 3.80k| _mm_storeu_si128((__m128i *)utf16_output + 0, utf16_packed0); 95| 3.80k| _mm_storeu_si128((__m128i *)utf16_output + 1, utf16_packed1); 96| 3.80k| utf16_output += 16; 97| 3.80k| buf += 16; 98| 3.80k| } else { 99| 1.16k| if (!validate_utf32(in0, in1) || !validate_utf32(in2, in3)) { ------------------ | Branch (99:11): [True: 12, False: 1.15k] | Branch (99:40): [True: 18, False: 1.13k] ------------------ 100| 30| return std::make_pair(nullptr, utf16_output); 101| 30| } 102| | 103| 1.13k| const auto ret0 = sse_expand_surrogate(in0); 104| 1.13k| _mm_storeu_si128((__m128i *)utf16_output, ret0.compressed); 105| 1.13k| utf16_output += ret0.u16count; 106| | 107| 1.13k| const auto ret1 = sse_expand_surrogate(in1); 108| 1.13k| _mm_storeu_si128((__m128i *)utf16_output, ret1.compressed); 109| 1.13k| utf16_output += ret1.u16count; 110| | 111| 1.13k| const auto ret2 = sse_expand_surrogate(in2); 112| 1.13k| _mm_storeu_si128((__m128i *)utf16_output, ret2.compressed); 113| 1.13k| utf16_output += ret2.u16count; 114| | 115| 1.13k| const auto ret3 = sse_expand_surrogate(in3); 116| 1.13k| _mm_storeu_si128((__m128i *)utf16_output, ret3.compressed); 117| 1.13k| utf16_output += ret3.u16count; 118| | 119| 1.13k| buf += 16; 120| 1.13k| } 121| 4.97k| } 122| | 123| | // check for invalid input 124| 692| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (124:7): [True: 18, False: 674] ------------------ 125| 18| return std::make_pair(nullptr, utf16_output); 126| 18| } 127| | 128| 674| return std::make_pair(buf, utf16_output); 129| 692|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114validate_utf32EDv2_xS2_: 37| 3.00k|simdutf_really_inline bool validate_utf32(const __m128i a, const __m128i b) { 38| 3.00k| using vector_u32 = simd32; 39| | 40| 3.00k| const auto in0 = vector_u32(a); 41| 3.00k| const auto in1 = vector_u32(b); 42| | 43| 3.00k| const auto standardmax = vector_u32::splat(0x10ffff); 44| 3.00k| const auto offset = vector_u32::splat(0xffff2000); 45| 3.00k| const auto standardoffsetmax = vector_u32::splat(0xfffff7ff); 46| | 47| 3.00k| const auto too_large = max(in0, in1) > standardmax; 48| 3.00k| const auto surrogate0 = (in0 + offset) > standardoffsetmax; 49| 3.00k| const auto surrogate1 = (in1 + offset) > standardoffsetmax; 50| | 51| 3.00k| const auto combined = too_large | surrogate0 | surrogate1; 52| 3.00k| return !combined.any(); 53| 3.00k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_120sse_expand_surrogateILNS_10endiannessE0EEENS1_18expansion_result_tEDv2_x: 9| 4.53k|expansion_result_t sse_expand_surrogate(const __m128i x) { 10| 4.53k| using vector_u32 = simd32; 11| 4.53k| using vector_u8 = simd8; 12| | 13| 4.53k| const auto in = vector_u32(x); 14| | 15| 4.53k| const auto non_surrogate_mask = (in & uint32_t(0xffff0000)) == uint32_t(0); 16| 4.53k| const auto mask = (~non_surrogate_mask.to_4bit_bitmask()) & 0xf; 17| | 18| 4.53k| const auto t0 = in - uint32_t(0x00010000); 19| 4.53k| const auto hi = t0.shr<10>() & uint32_t(0x000003ff); 20| 4.53k| const auto lo = t0.shl<16>() & uint32_t(0x03ff0000); 21| 4.53k| const auto surrogates = (lo | hi) | uint32_t(0xdc00d800); 22| | 23| 4.53k| const auto merged = as_vector_u8(select(non_surrogate_mask, in, surrogates)); 24| | 25| 4.53k| const auto shuffle = vector_u8::load( 26| 4.53k| (byte_order == endianness::LITTLE) ------------------ | Branch (26:7): [True: 4.53k, Folded] ------------------ 27| 4.53k| ? tables::utf32_to_utf16::pack_utf32_to_utf16le[mask] 28| 4.53k| : tables::utf32_to_utf16::pack_utf32_to_utf16be[mask]); 29| | 30| 4.53k| const size_t u16count = (4 + count_ones(mask)); 31| 4.53k| const auto compressed = shuffle.lookup_16(merged); 32| | 33| 4.53k| return {u16count, compressed}; 34| 4.53k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf32_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKDiPDsEES7_mS8_: 58| 528| char16_t *utf16_output) { 59| | 60| 528| const char32_t *end = buf + len; 61| | 62| 528| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 63| 528| __m128i forbidden_bytemask = _mm_setzero_si128(); 64| | 65| 4.06k| while (end - buf >= 16 + 8) { ------------------ | Branch (65:10): [True: 3.57k, False: 494] ------------------ 66| 3.57k| const __m128i *ptr = reinterpret_cast(buf); 67| 3.57k| const __m128i in0 = _mm_loadu_si128(ptr + 0); 68| 3.57k| const __m128i in1 = _mm_loadu_si128(ptr + 1); 69| 3.57k| const __m128i in2 = _mm_loadu_si128(ptr + 2); 70| 3.57k| const __m128i in3 = _mm_loadu_si128(ptr + 3); 71| | 72| 3.57k| const __m128i combined = 73| 3.57k| _mm_or_si128(_mm_or_si128(in2, in3), _mm_or_si128(in0, in1)); 74| 3.57k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 3.57k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.21k, False: 352] | | ------------------ ------------------ 75| | // No bits set above 16th, directly pack UTF-32 to UTF-16 76| 3.21k| __m128i utf16_packed0 = _mm_packus_epi32(in0, in1); 77| 3.21k| __m128i utf16_packed1 = _mm_packus_epi32(in2, in3); 78| | 79| 3.21k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 80| 3.21k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 81| 3.21k| forbidden_bytemask = _mm_or_si128( 82| 3.21k| forbidden_bytemask, 83| 3.21k| _mm_or_si128( 84| 3.21k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed0, v_f800), v_d800), 85| 3.21k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed1, v_f800), v_d800))); 86| | 87| 3.21k| if (big_endian) { ------------------ | Branch (87:11): [True: 3.21k, Folded] ------------------ 88| 3.21k| const __m128i swap = 89| 3.21k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 90| 3.21k| utf16_packed0 = _mm_shuffle_epi8(utf16_packed0, swap); 91| 3.21k| utf16_packed1 = _mm_shuffle_epi8(utf16_packed1, swap); 92| 3.21k| } 93| | 94| 3.21k| _mm_storeu_si128((__m128i *)utf16_output + 0, utf16_packed0); 95| 3.21k| _mm_storeu_si128((__m128i *)utf16_output + 1, utf16_packed1); 96| 3.21k| utf16_output += 16; 97| 3.21k| buf += 16; 98| 3.21k| } else { 99| 352| if (!validate_utf32(in0, in1) || !validate_utf32(in2, in3)) { ------------------ | Branch (99:11): [True: 14, False: 338] | Branch (99:40): [True: 20, False: 318] ------------------ 100| 34| return std::make_pair(nullptr, utf16_output); 101| 34| } 102| | 103| 318| const auto ret0 = sse_expand_surrogate(in0); 104| 318| _mm_storeu_si128((__m128i *)utf16_output, ret0.compressed); 105| 318| utf16_output += ret0.u16count; 106| | 107| 318| const auto ret1 = sse_expand_surrogate(in1); 108| 318| _mm_storeu_si128((__m128i *)utf16_output, ret1.compressed); 109| 318| utf16_output += ret1.u16count; 110| | 111| 318| const auto ret2 = sse_expand_surrogate(in2); 112| 318| _mm_storeu_si128((__m128i *)utf16_output, ret2.compressed); 113| 318| utf16_output += ret2.u16count; 114| | 115| 318| const auto ret3 = sse_expand_surrogate(in3); 116| 318| _mm_storeu_si128((__m128i *)utf16_output, ret3.compressed); 117| 318| utf16_output += ret3.u16count; 118| | 119| 318| buf += 16; 120| 318| } 121| 3.57k| } 122| | 123| | // check for invalid input 124| 494| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (124:7): [True: 6, False: 488] ------------------ 125| 6| return std::make_pair(nullptr, utf16_output); 126| 6| } 127| | 128| 488| return std::make_pair(buf, utf16_output); 129| 494|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_120sse_expand_surrogateILNS_10endiannessE1EEENS1_18expansion_result_tEDv2_x: 9| 1.27k|expansion_result_t sse_expand_surrogate(const __m128i x) { 10| 1.27k| using vector_u32 = simd32; 11| 1.27k| using vector_u8 = simd8; 12| | 13| 1.27k| const auto in = vector_u32(x); 14| | 15| 1.27k| const auto non_surrogate_mask = (in & uint32_t(0xffff0000)) == uint32_t(0); 16| 1.27k| const auto mask = (~non_surrogate_mask.to_4bit_bitmask()) & 0xf; 17| | 18| 1.27k| const auto t0 = in - uint32_t(0x00010000); 19| 1.27k| const auto hi = t0.shr<10>() & uint32_t(0x000003ff); 20| 1.27k| const auto lo = t0.shl<16>() & uint32_t(0x03ff0000); 21| 1.27k| const auto surrogates = (lo | hi) | uint32_t(0xdc00d800); 22| | 23| 1.27k| const auto merged = as_vector_u8(select(non_surrogate_mask, in, surrogates)); 24| | 25| 1.27k| const auto shuffle = vector_u8::load( 26| 1.27k| (byte_order == endianness::LITTLE) ------------------ | Branch (26:7): [Folded, False: 1.27k] ------------------ 27| 1.27k| ? tables::utf32_to_utf16::pack_utf32_to_utf16le[mask] 28| 1.27k| : tables::utf32_to_utf16::pack_utf32_to_utf16be[mask]); 29| | 30| 1.27k| const size_t u16count = (4 + count_ones(mask)); 31| 1.27k| const auto compressed = shuffle.lookup_16(merged); 32| | 33| 1.27k| return {u16count, compressed}; 34| 1.27k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf32_to_utf16_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 134| 716| char16_t *utf16_output) { 135| 716| const char32_t *start = buf; 136| 716| const char32_t *end = buf + len; 137| | 138| 716| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 139| | 140| 88.8k| while (end - buf >= 8) { ------------------ | Branch (140:10): [True: 88.4k, False: 326] ------------------ 141| 88.4k| const __m128i in = _mm_loadu_si128((__m128i *)buf); 142| 88.4k| const __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 143| | 144| 88.4k| const __m128i combined = _mm_or_si128(in, nextin); 145| 88.4k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 88.4k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 85.4k, False: 3.03k] | | ------------------ ------------------ 146| | // No bits set above 16th, directly pack UTF-32 to UTF-16 147| 85.4k| __m128i utf16_packed = _mm_packus_epi32(in, nextin); 148| | 149| 85.4k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 150| 85.4k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 151| 85.4k| const __m128i forbidden_bytemask = 152| 85.4k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed, v_f800), v_d800); 153| 85.4k| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (153:11): [True: 52, False: 85.3k] ------------------ 154| 52| return std::make_pair(result(error_code::SURROGATE, buf - start), 155| 52| utf16_output); 156| 52| } 157| | 158| 85.3k| if (big_endian) { ------------------ | Branch (158:11): [Folded, False: 85.3k] ------------------ 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| 85.3k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 165| 85.3k| utf16_output += 8; 166| 85.3k| buf += 8; 167| 85.3k| } else { 168| 3.03k| size_t forward = 7; 169| 3.03k| size_t k = 0; 170| 3.03k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (170:11): [True: 0, False: 3.03k] ------------------ 171| 0| forward = size_t(end - buf - 1); 172| 0| } 173| 22.7k| for (; k < forward; k++) { ------------------ | Branch (173:14): [True: 20.0k, False: 2.69k] ------------------ 174| 20.0k| uint32_t word = buf[k]; 175| 20.0k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (175:13): [True: 9.99k, False: 10.0k] ------------------ 176| | // will not generate a surrogate pair 177| 9.99k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (177:15): [True: 986, False: 9.00k] | Branch (177:33): [True: 40, False: 946] ------------------ 178| 40| return std::make_pair( 179| 40| result(error_code::SURROGATE, buf - start + k), utf16_output); 180| 40| } 181| 9.95k| *utf16_output++ = 182| 9.95k| big_endian ------------------ | Branch (182:15): [Folded, False: 9.95k] ------------------ 183| 9.95k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 184| 9.95k| : char16_t(word); 185| 10.0k| } else { 186| | // will generate a surrogate pair 187| 10.0k| if (word > 0x10FFFF) { ------------------ | Branch (187:15): [True: 298, False: 9.71k] ------------------ 188| 298| return std::make_pair( 189| 298| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 190| 298| } 191| 9.71k| word -= 0x10000; 192| 9.71k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 193| 9.71k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 194| 9.71k| if (big_endian) { ------------------ | Branch (194:15): [Folded, False: 9.71k] ------------------ 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| 9.71k| *utf16_output++ = char16_t(high_surrogate); 201| 9.71k| *utf16_output++ = char16_t(low_surrogate); 202| 9.71k| } 203| 20.0k| } 204| 2.69k| buf += k; 205| 2.69k| } 206| 88.4k| } 207| | 208| 326| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 209| 716|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf32_to_utf16_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 134| 606| char16_t *utf16_output) { 135| 606| const char32_t *start = buf; 136| 606| const char32_t *end = buf + len; 137| | 138| 606| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 139| | 140| 13.0k| while (end - buf >= 8) { ------------------ | Branch (140:10): [True: 12.7k, False: 306] ------------------ 141| 12.7k| const __m128i in = _mm_loadu_si128((__m128i *)buf); 142| 12.7k| const __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 143| | 144| 12.7k| const __m128i combined = _mm_or_si128(in, nextin); 145| 12.7k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 12.7k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 11.4k, False: 1.24k] | | ------------------ ------------------ 146| | // No bits set above 16th, directly pack UTF-32 to UTF-16 147| 11.4k| __m128i utf16_packed = _mm_packus_epi32(in, nextin); 148| | 149| 11.4k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 150| 11.4k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 151| 11.4k| const __m128i forbidden_bytemask = 152| 11.4k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed, v_f800), v_d800); 153| 11.4k| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (153:11): [True: 24, False: 11.4k] ------------------ 154| 24| return std::make_pair(result(error_code::SURROGATE, buf - start), 155| 24| utf16_output); 156| 24| } 157| | 158| 11.4k| if (big_endian) { ------------------ | Branch (158:11): [True: 11.4k, Folded] ------------------ 159| 11.4k| const __m128i swap = 160| 11.4k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 161| 11.4k| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 162| 11.4k| } 163| | 164| 11.4k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 165| 11.4k| utf16_output += 8; 166| 11.4k| buf += 8; 167| 11.4k| } else { 168| 1.24k| size_t forward = 7; 169| 1.24k| size_t k = 0; 170| 1.24k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (170:11): [True: 0, False: 1.24k] ------------------ 171| 0| forward = size_t(end - buf - 1); 172| 0| } 173| 8.65k| for (; k < forward; k++) { ------------------ | Branch (173:14): [True: 7.68k, False: 972] ------------------ 174| 7.68k| uint32_t word = buf[k]; 175| 7.68k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (175:13): [True: 4.35k, False: 3.33k] ------------------ 176| | // will not generate a surrogate pair 177| 4.35k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (177:15): [True: 968, False: 3.38k] | Branch (177:33): [True: 18, False: 950] ------------------ 178| 18| return std::make_pair( 179| 18| result(error_code::SURROGATE, buf - start + k), utf16_output); 180| 18| } 181| 4.33k| *utf16_output++ = 182| 4.33k| big_endian ------------------ | Branch (182:15): [True: 4.33k, Folded] ------------------ 183| 4.33k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 184| 4.33k| : char16_t(word); 185| 4.33k| } else { 186| | // will generate a surrogate pair 187| 3.33k| if (word > 0x10FFFF) { ------------------ | Branch (187:15): [True: 258, False: 3.07k] ------------------ 188| 258| return std::make_pair( 189| 258| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 190| 258| } 191| 3.07k| word -= 0x10000; 192| 3.07k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 193| 3.07k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 194| 3.07k| if (big_endian) { ------------------ | Branch (194:15): [True: 3.07k, Folded] ------------------ 195| 3.07k| high_surrogate = 196| 3.07k| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 197| 3.07k| low_surrogate = 198| 3.07k| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 199| 3.07k| } 200| 3.07k| *utf16_output++ = char16_t(high_surrogate); 201| 3.07k| *utf16_output++ = char16_t(low_surrogate); 202| 3.07k| } 203| 7.68k| } 204| 972| buf += k; 205| 972| } 206| 12.7k| } 207| | 208| 306| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 209| 606|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_125sse_convert_utf32_to_utf8EPKDimPc: 2| 796|sse_convert_utf32_to_utf8(const char32_t *buf, size_t len, char *utf8_output) { 3| 796| const char32_t *end = buf + len; 4| | 5| 796| const __m128i v_0000 = _mm_setzero_si128(); //__m128 = 128 bits 6| 796| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); // 1111 1000 0000 7| | // 0000 8| 796| const __m128i v_c080 = _mm_set1_epi16((uint16_t)0xc080); // 1100 0000 1000 9| | // 0000 10| 796| const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80); // 1111 1111 1000 11| | // 0000 12| 796| const __m128i v_ffff0000 = _mm_set1_epi32( 13| 796| (uint32_t)0xffff0000); // 1111 1111 1111 1111 0000 0000 0000 0000 14| 796| const __m128i v_7fffffff = _mm_set1_epi32( 15| 796| (uint32_t)0x7fffffff); // 0111 1111 1111 1111 1111 1111 1111 1111 16| 796| __m128i running_max = _mm_setzero_si128(); 17| 796| __m128i forbidden_bytemask = _mm_setzero_si128(); 18| 796| const size_t safety_margin = 19| 796| 12; // to avoid overruns, see issue 20| | // https://github.com/simdutf/simdutf/issues/92 21| | 22| 6.41k| while (end - buf >= ------------------ | Branch (22:10): [True: 5.77k, False: 642] ------------------ 23| 6.41k| std::ptrdiff_t( 24| 6.41k| 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| 5.77k| __m128i in = _mm_loadu_si128((__m128i *)buf); 29| 5.77k| __m128i nextin = _mm_loadu_si128( 30| 5.77k| (__m128i *)buf + 1); // These two values can hold only 8 UTF32 chars 31| 5.77k| running_max = _mm_max_epu32( 32| 5.77k| _mm_max_epu32(in, running_max), // take element-wise max char32_t from 33| | // in and running_max vector 34| 5.77k| 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| 5.77k| __m128i in_16 = _mm_packus_epi32( 40| 5.77k| _mm_and_si128(in, v_7fffffff), 41| 5.77k| _mm_and_si128( 42| 5.77k| nextin, 43| 5.77k| 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| 5.77k| if (_mm_testz_si128(in_16, v_ff80)) { // if the first two blocks are ASCII ------------------ | Branch (60:9): [True: 1.60k, False: 4.16k] ------------------ 61| 1.60k| __m128i thirdin = _mm_loadu_si128((__m128i *)buf + 2); 62| 1.60k| __m128i fourthin = _mm_loadu_si128((__m128i *)buf + 3); 63| 1.60k| running_max = _mm_max_epu32( 64| 1.60k| _mm_max_epu32(thirdin, running_max), 65| 1.60k| fourthin); // take the running max of all 4 vectors thus far 66| 1.60k| __m128i nextin_16 = _mm_packus_epi32( 67| 1.60k| _mm_and_si128(thirdin, v_7fffffff), 68| 1.60k| _mm_and_si128(fourthin, 69| 1.60k| v_7fffffff)); // pack into 1 vector, now you have two 70| 1.60k| if (!_mm_testz_si128( ------------------ | Branch (70:11): [True: 718, False: 890] ------------------ 71| 1.60k| nextin_16, 72| 1.60k| v_ff80)) { // checks if the second packed vector is ASCII, if not: 73| | // 1. pack the bytes 74| | // obviously suboptimal. 75| 718| const __m128i utf8_packed = _mm_packus_epi16( 76| 718| in_16, in_16); // creates two copy of in_16 in 1 vector 77| | // 2. store (16 bytes) 78| 718| _mm_storeu_si128((__m128i *)utf8_output, 79| 718| utf8_packed); // put them into the output 80| | // 3. adjust pointers 81| 718| buf += 8; // the char32_t buffer pointer goes up 8 char32_t chars* 32 82| | // bits = 256 bits 83| 718| utf8_output += 84| 718| 8; // same with output, e.g. lift the first two blocks alone. 85| | // Proceed with next input 86| 718| in_16 = nextin_16; 87| | // We need to update in and nextin because they are used later. 88| 718| in = thirdin; 89| 718| nextin = fourthin; 90| 890| } else { 91| | // 1. pack the bytes 92| 890| const __m128i utf8_packed = _mm_packus_epi16(in_16, nextin_16); 93| | // 2. store (16 bytes) 94| 890| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 95| | // 3. adjust pointers 96| 890| buf += 16; 97| 890| utf8_output += 16; 98| 890| continue; // we are done for this round! 99| 890| } 100| 1.60k| } 101| | 102| | // no bits set above 7th bit -- find out all the ASCII characters 103| 4.88k| const __m128i one_byte_bytemask = 104| 4.88k| _mm_cmpeq_epi16( // this takes four bytes at a time and compares: 105| 4.88k| _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| 4.88k| v_0000 // 108| 4.88k| ); // 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| 4.88k| const uint16_t one_byte_bitmask = static_cast(_mm_movemask_epi8( 113| 4.88k| 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| 4.88k| const __m128i one_or_two_bytes_bytemask = 118| 4.88k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_0000); 119| 4.88k| const uint16_t one_or_two_bytes_bitmask = 120| 4.88k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 121| | 122| 4.88k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (122:9): [True: 2.44k, False: 2.44k] ------------------ 123| | // case: all code units either produce 1 or 2 UTF-8 bytes (at least one 124| | // produces 2 bytes) 125| | // 1. prepare 2-byte values 126| | // input 16-bit word : [0000|0aaa|aabb|bbbb] x 8 127| | // expected output : [110a|aaaa|10bb|bbbb] x 8 128| 2.44k| const __m128i v_1f00 = 129| 2.44k| _mm_set1_epi16((int16_t)0x1f00); // 0001 1111 0000 0000 130| 2.44k| const __m128i v_003f = 131| 2.44k| _mm_set1_epi16((int16_t)0x003f); // 0000 0000 0011 1111 132| | 133| | // t0 = [000a|aaaa|bbbb|bb00] 134| 2.44k| const __m128i t0 = _mm_slli_epi16(in_16, 2); // shift packed vector by two 135| | // t1 = [000a|aaaa|0000|0000] 136| 2.44k| const __m128i t1 = _mm_and_si128(t0, v_1f00); // potential first utf8 byte 137| | // t2 = [0000|0000|00bb|bbbb] 138| 2.44k| const __m128i t2 = 139| 2.44k| _mm_and_si128(in_16, v_003f); // potential second utf8 byte 140| | // t3 = [000a|aaaa|00bb|bbbb] 141| 2.44k| const __m128i t3 = 142| 2.44k| _mm_or_si128(t1, t2); // first and second potential utf8 byte together 143| | // t4 = [110a|aaaa|10bb|bbbb] 144| 2.44k| const __m128i t4 = _mm_or_si128( 145| 2.44k| t3, 146| 2.44k| v_c080); // t3 | 1100 0000 1000 0000 = full potential 2-byte utf8 unit 147| | 148| | // 2. merge ASCII and 2-byte codewords 149| 2.44k| const __m128i utf8_unpacked = 150| 2.44k| _mm_blendv_epi8(t4, in_16, one_byte_bytemask); 151| | 152| | // 3. prepare bitmask for 8-bit lookup 153| | // one_byte_bitmask = hhggffeeddccbbaa -- the bits are doubled (h - 154| | // MSB, a - LSB) 155| 2.44k| const uint16_t m0 = one_byte_bitmask & 0x5555; // m0 = 0h0g0f0e0d0c0b0a 156| 2.44k| const uint16_t m1 = 157| 2.44k| static_cast(m0 >> 7); // m1 = 00000000h0g0f0e0 158| 2.44k| const uint8_t m2 = 159| 2.44k| static_cast((m0 | m1) & 0xff); // m2 = hdgcfbea 160| | // 4. pack the bytes 161| 2.44k| const uint8_t *row = 162| 2.44k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[m2][0]; 163| 2.44k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 164| 2.44k| const __m128i utf8_packed = _mm_shuffle_epi8(utf8_unpacked, shuffle); 165| | 166| | // 5. store bytes 167| 2.44k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 168| | 169| | // 6. adjust pointers 170| 2.44k| buf += 8; 171| 2.44k| utf8_output += row[0]; 172| 2.44k| continue; 173| 2.44k| } 174| | 175| | // Check for overflow in packing 176| | 177| 2.44k| const __m128i saturation_bytemask = _mm_cmpeq_epi32( 178| 2.44k| _mm_and_si128(_mm_or_si128(in, nextin), v_ffff0000), v_0000); 179| 2.44k| const uint32_t saturation_bitmask = 180| 2.44k| static_cast(_mm_movemask_epi8(saturation_bytemask)); 181| 2.44k| if (saturation_bitmask == 0xffff) { ------------------ | Branch (181:9): [True: 1.25k, False: 1.18k] ------------------ 182| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 183| 1.25k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 184| 1.25k| forbidden_bytemask = 185| 1.25k| _mm_or_si128(forbidden_bytemask, 186| 1.25k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_d800)); 187| | 188| 1.25k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 189| 1.25k| 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| 1.25k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 219| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 220| 1.25k| const __m128i t0 = _mm_shuffle_epi8(in_16, dup_even); 221| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 222| 1.25k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 218| 1.25k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 223| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 224| 1.25k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 218| 1.25k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 225| | 226| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 227| 1.25k| const __m128i s0 = _mm_srli_epi16(in_16, 4); 228| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 229| 1.25k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 218| 1.25k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 230| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 231| 1.25k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 218| 1.25k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 232| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 233| 1.25k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 218| 1.25k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 234| 1.25k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 235| 1.25k| simdutf_vec(0b0100000000000000)); ------------------ | | 218| 1.25k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 236| 1.25k| const __m128i s4 = _mm_xor_si128(s3, m0); 237| 1.25k|#undef simdutf_vec 238| | 239| | // 4. expand code units 16-bit => 32-bit 240| 1.25k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 241| 1.25k| 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| 1.25k| const uint16_t mask = 245| 1.25k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 246| 1.25k| if (mask == 0) { ------------------ | Branch (246:11): [True: 610, False: 648] ------------------ 247| | // We only have three-byte code units. Use fast path. 248| 610| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 249| 610| 15, 13, -1, -1, -1, -1); 250| 610| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 251| 610| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 252| 610| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 253| 610| utf8_output += 12; 254| 610| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 255| 610| utf8_output += 12; 256| 610| buf += 8; 257| 610| continue; 258| 610| } 259| 648| const uint8_t mask0 = uint8_t(mask); 260| | 261| 648| const uint8_t *row0 = 262| 648| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 263| 648| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 264| 648| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 265| | 266| 648| const uint8_t mask1 = static_cast(mask >> 8); 267| | 268| 648| const uint8_t *row1 = 269| 648| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 270| 648| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 271| 648| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 272| | 273| 648| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 274| 648| utf8_output += row0[0]; 275| 648| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 276| 648| utf8_output += row1[0]; 277| | 278| 648| buf += 8; 279| 1.18k| } 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| 1.18k| size_t forward = 15; 285| 1.18k| size_t k = 0; 286| 1.18k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (286:11): [True: 0, False: 1.18k] ------------------ 287| 0| forward = size_t(end - buf - 1); 288| 0| } 289| 17.2k| for (; k < forward; k++) { ------------------ | Branch (289:14): [True: 16.2k, False: 1.02k] ------------------ 290| 16.2k| uint32_t word = buf[k]; 291| 16.2k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (291:13): [True: 7.63k, False: 8.58k] ------------------ 292| 7.63k| *utf8_output++ = char(word); 293| 8.58k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (293:20): [True: 928, False: 7.65k] ------------------ 294| 928| *utf8_output++ = char((word >> 6) | 0b11000000); 295| 928| *utf8_output++ = char((word & 0b111111) | 0b10000000); 296| 7.65k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (296:20): [True: 2.40k, False: 5.24k] ------------------ 297| 2.40k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (297:15): [True: 998, False: 1.40k] | Branch (297:33): [True: 8, False: 990] ------------------ 298| 8| return std::make_pair(nullptr, utf8_output); 299| 8| } 300| 2.39k| *utf8_output++ = char((word >> 12) | 0b11100000); 301| 2.39k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 302| 2.39k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 303| 5.24k| } else { 304| 5.24k| if (word > 0x10FFFF) { ------------------ | Branch (304:15): [True: 146, False: 5.10k] ------------------ 305| 146| return std::make_pair(nullptr, utf8_output); 306| 146| } 307| 5.10k| *utf8_output++ = char((word >> 18) | 0b11110000); 308| 5.10k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 309| 5.10k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 310| 5.10k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 311| 5.10k| } 312| 16.2k| } 313| 1.02k| buf += k; 314| 1.02k| } 315| 2.44k| } // while 316| | 317| | // check for invalid input 318| 642| const __m128i v_10ffff = _mm_set1_epi32((uint32_t)0x10ffff); 319| 642| if (static_cast(_mm_movemask_epi8(_mm_cmpeq_epi32( ------------------ | Branch (319:7): [True: 28, False: 614] ------------------ 320| 642| _mm_max_epu32(running_max, v_10ffff), v_10ffff))) != 0xffff) { 321| 28| return std::make_pair(nullptr, utf8_output); 322| 28| } 323| | 324| 614| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (324:7): [True: 58, False: 556] ------------------ 325| 58| return std::make_pair(nullptr, utf8_output); 326| 58| } 327| | 328| 556| return std::make_pair(buf, utf8_output); 329| 614|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_137sse_convert_utf32_to_utf8_with_errorsEPKDimPc: 333| 830| char *utf8_output) { 334| 830| const char32_t *end = buf + len; 335| 830| const char32_t *start = buf; 336| | 337| 830| const __m128i v_0000 = _mm_setzero_si128(); 338| 830| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 339| 830| const __m128i v_c080 = _mm_set1_epi16((uint16_t)0xc080); 340| 830| const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80); 341| 830| const __m128i v_ffff0000 = _mm_set1_epi32((uint32_t)0xffff0000); 342| 830| const __m128i v_7fffffff = _mm_set1_epi32((uint32_t)0x7fffffff); 343| 830| const __m128i v_10ffff = _mm_set1_epi32((uint32_t)0x10ffff); 344| | 345| 830| const size_t safety_margin = 346| 830| 12; // to avoid overruns, see issue 347| | // https://github.com/simdutf/simdutf/issues/92 348| | 349| 62.3k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (349:10): [True: 61.7k, False: 620] ------------------ 350| | // We load two 16 bytes registers for a total of 32 bytes or 8 characters. 351| 61.7k| __m128i in = _mm_loadu_si128((__m128i *)buf); 352| 61.7k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 353| | // Check for too large input 354| 61.7k| __m128i max_input = _mm_max_epu32(_mm_max_epu32(in, nextin), v_10ffff); 355| 61.7k| if (static_cast(_mm_movemask_epi8( ------------------ | Branch (355:9): [True: 70, False: 61.6k] ------------------ 356| 61.7k| _mm_cmpeq_epi32(max_input, v_10ffff))) != 0xffff) { 357| 70| return std::make_pair(result(error_code::TOO_LARGE, buf - start), 358| 70| utf8_output); 359| 70| } 360| | 361| | // Pack 32-bit UTF-32 code units to 16-bit UTF-16 code units with unsigned 362| | // saturation 363| 61.6k| __m128i in_16 = _mm_packus_epi32(_mm_and_si128(in, v_7fffffff), 364| 61.6k| _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| 61.6k| if (_mm_testz_si128(in_16, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (369:9): [True: 54.2k, False: 7.40k] ------------------ 370| | // 1. pack the bytes 371| | // obviously suboptimal. 372| 54.2k| const __m128i utf8_packed = _mm_packus_epi16(in_16, in_16); 373| | // 2. store (16 bytes) 374| 54.2k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 375| | // 3. adjust pointers 376| 54.2k| buf += 8; 377| 54.2k| utf8_output += 8; 378| 54.2k| continue; 379| 54.2k| } 380| | 381| | // no bits set above 7th bit 382| 7.40k| const __m128i one_byte_bytemask = 383| 7.40k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_ff80), v_0000); 384| 7.40k| const uint16_t one_byte_bitmask = 385| 7.40k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 386| | 387| | // no bits set above 11th bit 388| 7.40k| const __m128i one_or_two_bytes_bytemask = 389| 7.40k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_0000); 390| 7.40k| const uint16_t one_or_two_bytes_bitmask = 391| 7.40k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 392| | 393| 7.40k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (393:9): [True: 4.86k, False: 2.54k] ------------------ 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| 4.86k| const __m128i v_1f00 = _mm_set1_epi16((int16_t)0x1f00); 400| 4.86k| const __m128i v_003f = _mm_set1_epi16((int16_t)0x003f); 401| | 402| | // t0 = [000a|aaaa|bbbb|bb00] 403| 4.86k| const __m128i t0 = _mm_slli_epi16(in_16, 2); 404| | // t1 = [000a|aaaa|0000|0000] 405| 4.86k| const __m128i t1 = _mm_and_si128(t0, v_1f00); 406| | // t2 = [0000|0000|00bb|bbbb] 407| 4.86k| const __m128i t2 = _mm_and_si128(in_16, v_003f); 408| | // t3 = [000a|aaaa|00bb|bbbb] 409| 4.86k| const __m128i t3 = _mm_or_si128(t1, t2); 410| | // t4 = [110a|aaaa|10bb|bbbb] 411| 4.86k| const __m128i t4 = _mm_or_si128(t3, v_c080); 412| | 413| | // 2. merge ASCII and 2-byte codewords 414| 4.86k| const __m128i utf8_unpacked = 415| 4.86k| _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| 4.86k| const uint16_t m0 = one_byte_bitmask & 0x5555; // m0 = 0h0g0f0e0d0c0b0a 421| 4.86k| const uint16_t m1 = 422| 4.86k| static_cast(m0 >> 7); // m1 = 00000000h0g0f0e0 423| 4.86k| const uint8_t m2 = 424| 4.86k| static_cast((m0 | m1) & 0xff); // m2 = hdgcfbea 425| | // 4. pack the bytes 426| 4.86k| const uint8_t *row = 427| 4.86k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[m2][0]; 428| 4.86k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 429| 4.86k| const __m128i utf8_packed = _mm_shuffle_epi8(utf8_unpacked, shuffle); 430| | 431| | // 5. store bytes 432| 4.86k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 433| | 434| | // 6. adjust pointers 435| 4.86k| buf += 8; 436| 4.86k| utf8_output += row[0]; 437| 4.86k| continue; 438| 4.86k| } 439| | 440| | // Check for overflow in packing 441| 2.54k| const __m128i saturation_bytemask = _mm_cmpeq_epi32( 442| 2.54k| _mm_and_si128(_mm_or_si128(in, nextin), v_ffff0000), v_0000); 443| 2.54k| const uint32_t saturation_bitmask = 444| 2.54k| static_cast(_mm_movemask_epi8(saturation_bytemask)); 445| | 446| 2.54k| if (saturation_bitmask == 0xffff) { ------------------ | Branch (446:9): [True: 1.16k, False: 1.37k] ------------------ 447| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 448| | 449| | // Check for illegal surrogate code units 450| 1.16k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 451| 1.16k| const __m128i forbidden_bytemask = 452| 1.16k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_d800); 453| 1.16k| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (453:11): [True: 32, False: 1.13k] ------------------ 454| 32| return std::make_pair(result(error_code::SURROGATE, buf - start), 455| 32| utf8_output); 456| 32| } 457| | 458| 1.13k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 459| 1.13k| 0x0808, 0x0a0a, 0x0c0c, 0x0e0e); 460| | 461| | /* In this branch we handle three cases: 462| | 1. [0000|0000|0ccc|cccc] => [0ccc|cccc] - 463| | single UFT-8 byte 464| | 2. [0000|0bbb|bbcc|cccc] => [110b|bbbb], [10cc|cccc] - 465| | two UTF-8 bytes 466| | 3. [aaaa|bbbb|bbcc|cccc] => [1110|aaaa], [10bb|bbbb], [10cc|cccc] - 467| | three UTF-8 bytes 468| | 469| | We expand the input word (16-bit) into two code units (32-bit), thus 470| | we have room for four bytes. However, we need five distinct bit 471| | layouts. Note that the last byte in cases #2 and #3 is the same. 472| | 473| | We precompute byte 1 for case #1 and the common byte for cases #2 & #3 474| | in register t2. 475| | 476| | We precompute byte 1 for case #3 and -- **conditionally** -- precompute 477| | either byte 1 for case #2 or byte 2 for case #3. Note that they 478| | differ by exactly one bit. 479| | 480| | Finally from these two code units we build proper UTF-8 sequence, taking 481| | into account the case (i.e, the number of bytes to write). 482| | */ 483| | /** 484| | * Given [aaaa|bbbb|bbcc|cccc] our goal is to produce: 485| | * t2 => [0ccc|cccc] [10cc|cccc] 486| | * s4 => [1110|aaaa] ([110b|bbbb] OR [10bb|bbbb]) 487| | */ 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 489| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 490| 1.13k| const __m128i t0 = _mm_shuffle_epi8(in_16, dup_even); 491| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 492| 1.13k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 493| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 494| 1.13k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 495| | 496| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 497| 1.13k| const __m128i s0 = _mm_srli_epi16(in_16, 4); 498| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 499| 1.13k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 500| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 501| 1.13k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 502| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 503| 1.13k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 504| 1.13k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 505| 1.13k| simdutf_vec(0b0100000000000000)); ------------------ | | 488| 1.13k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 506| 1.13k| const __m128i s4 = _mm_xor_si128(s3, m0); 507| 1.13k|#undef simdutf_vec 508| | 509| | // 4. expand code units 16-bit => 32-bit 510| 1.13k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 511| 1.13k| const __m128i out1 = _mm_unpackhi_epi16(t2, s4); 512| | 513| | // 5. compress 32-bit code units into 1, 2 or 3 bytes -- 2 x shuffle 514| 1.13k| const uint16_t mask = 515| 1.13k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 516| 1.13k| if (mask == 0) { ------------------ | Branch (516:11): [True: 452, False: 684] ------------------ 517| | // We only have three-byte code units. Use fast path. 518| 452| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 519| 452| 15, 13, -1, -1, -1, -1); 520| 452| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 521| 452| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 522| 452| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 523| 452| utf8_output += 12; 524| 452| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 525| 452| utf8_output += 12; 526| 452| buf += 8; 527| 452| continue; 528| 452| } 529| 684| const uint8_t mask0 = uint8_t(mask); 530| | 531| 684| const uint8_t *row0 = 532| 684| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 533| 684| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 534| 684| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 535| | 536| 684| const uint8_t mask1 = static_cast(mask >> 8); 537| | 538| 684| const uint8_t *row1 = 539| 684| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 540| 684| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 541| 684| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 542| | 543| 684| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 544| 684| utf8_output += row0[0]; 545| 684| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 546| 684| utf8_output += row1[0]; 547| | 548| 684| buf += 8; 549| 1.37k| } 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| 1.37k| size_t forward = 15; 555| 1.37k| size_t k = 0; 556| 1.37k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (556:11): [True: 0, False: 1.37k] ------------------ 557| 0| forward = size_t(end - buf - 1); 558| 0| } 559| 21.3k| for (; k < forward; k++) { ------------------ | Branch (559:14): [True: 20.0k, False: 1.26k] ------------------ 560| 20.0k| uint32_t word = buf[k]; 561| 20.0k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (561:13): [True: 5.79k, False: 14.2k] ------------------ 562| 5.79k| *utf8_output++ = char(word); 563| 14.2k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (563:20): [True: 1.49k, False: 12.7k] ------------------ 564| 1.49k| *utf8_output++ = char((word >> 6) | 0b11000000); 565| 1.49k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 566| 12.7k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (566:20): [True: 2.75k, False: 10.0k] ------------------ 567| 2.75k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (567:15): [True: 1.24k, False: 1.51k] | Branch (567:33): [True: 28, False: 1.21k] ------------------ 568| 28| return std::make_pair( 569| 28| result(error_code::SURROGATE, buf - start + k), utf8_output); 570| 28| } 571| 2.72k| *utf8_output++ = char((word >> 12) | 0b11100000); 572| 2.72k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 573| 2.72k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 574| 10.0k| } else { 575| 10.0k| if (word > 0x10FFFF) { ------------------ | Branch (575:15): [True: 80, False: 9.92k] ------------------ 576| 80| return std::make_pair( 577| 80| result(error_code::TOO_LARGE, buf - start + k), utf8_output); 578| 80| } 579| 9.92k| *utf8_output++ = char((word >> 18) | 0b11110000); 580| 9.92k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 581| 9.92k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 582| 9.92k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 583| 9.92k| } 584| 20.0k| } 585| 1.26k| buf += k; 586| 1.26k| } 587| 2.54k| } // while 588| 620| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 589| 830|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_129convert_masked_utf8_to_latin1EPKcmRPc: 9| 543k| 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| 543k| const __m128i in = _mm_loadu_si128((__m128i *)input); 21| 543k| const uint16_t input_utf8_end_of_code_point_mask = 22| 543k| utf8_end_of_code_point_mask & 23| 543k| 0xfff; // we are only processing 12 bytes in case it is not all ASCII 24| 543k| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (24:7): [True: 3.58k, False: 539k] ------------------ 25| | // We process the data in chunks of 12 bytes. 26| 3.58k| _mm_storeu_si128(reinterpret_cast<__m128i *>(latin1_output), in); 27| 3.58k| latin1_output += 12; // We wrote 12 characters. 28| 3.58k| return 12; // We consumed 12 bytes. 29| 3.58k| } 30| | /// We do not have a fast path available, so we fallback. 31| 539k| const uint8_t idx = 32| 539k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 33| 539k| const uint8_t consumed = 34| 539k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 35| | // this indicates an invalid input: 36| 539k| if (idx >= 64) { ------------------ | Branch (36:7): [True: 12.2k, False: 527k] ------------------ 37| 12.2k| return consumed; 38| 12.2k| } 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| 527k| const __m128i sh = 46| 527k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 47| 527k| const __m128i perm = _mm_shuffle_epi8(in, sh); 48| 527k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 49| 527k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 50| 527k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 51| 527k| 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| 527k| _mm_storel_epi64((__m128i *)latin1_output, latin1_packed); 56| 527k| latin1_output += 6; // We wrote 6 bytes. 57| 527k| return consumed; 58| 539k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE0EEEmPKcmRPDs: 10| 2.87M| 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| 2.87M| const __m128i swap = 23| 2.87M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 2.87M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 2.87M| const uint16_t input_utf8_end_of_code_point_mask = 26| 2.87M| utf8_end_of_code_point_mask & 0xfff; 27| 2.87M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 9.93k, False: 2.86M] ------------------ 28| | // We process the data in chunks of 12 bytes. 29| | // Note: using 16 bytes is unsafe, see issue_ossfuzz_71218 30| 9.93k| __m128i ascii_first = _mm_cvtepu8_epi16(in); 31| 9.93k| __m128i ascii_second = _mm_cvtepu8_epi16(_mm_srli_si128(in, 8)); 32| 9.93k| if (big_endian) { ------------------ | Branch (32:9): [Folded, False: 9.93k] ------------------ 33| 0| ascii_first = _mm_shuffle_epi8(ascii_first, swap); 34| 0| ascii_second = _mm_shuffle_epi8(ascii_second, swap); 35| 0| } 36| 9.93k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf16_output), ascii_first); 37| 9.93k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf16_output + 8), 38| 9.93k| ascii_second); 39| 9.93k| utf16_output += 12; // We wrote 12 16-bit characters. 40| 9.93k| return 12; // We consumed 12 bytes. 41| 9.93k| } 42| 2.86M| if (((utf8_end_of_code_point_mask & 0xFFFF) == 0xaaaa)) { ------------------ | Branch (42:7): [True: 11.2k, False: 2.84M] ------------------ 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| 11.2k| const __m128i sh = 47| 11.2k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 48| 11.2k| const __m128i perm = _mm_shuffle_epi8(in, sh); 49| 11.2k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 50| 11.2k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 51| 11.2k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 52| 11.2k| if (big_endian) ------------------ | Branch (52:9): [Folded, False: 11.2k] ------------------ 53| 0| composed = _mm_shuffle_epi8(composed, swap); 54| 11.2k| _mm_storeu_si128((__m128i *)utf16_output, composed); 55| 11.2k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 56| 11.2k| return 16; 57| 11.2k| } 58| 2.84M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (58:7): [True: 2.85k, False: 2.84M] ------------------ 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| 2.85k| const __m128i sh = 63| 2.85k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 64| 2.85k| const __m128i perm = _mm_shuffle_epi8(in, sh); 65| 2.85k| const __m128i ascii = 66| 2.85k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 67| 2.85k| const __m128i middlebyte = 68| 2.85k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 69| 2.85k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 70| 2.85k| const __m128i highbyte = 71| 2.85k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 72| 2.85k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 73| 2.85k| const __m128i composed = 74| 2.85k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 75| 2.85k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 76| 2.85k| if (big_endian) ------------------ | Branch (76:9): [Folded, False: 2.85k] ------------------ 77| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 78| 2.85k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 79| 2.85k| utf16_output += 4; 80| 2.85k| return 12; 81| 2.85k| } 82| | /// We do not have a fast path available, so we fallback. 83| | 84| 2.84M| const uint8_t idx = 85| 2.84M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 86| 2.84M| const uint8_t consumed = 87| 2.84M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 88| 2.84M| if (idx < 64) { ------------------ | Branch (88:7): [True: 2.40M, False: 439k] ------------------ 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| 2.40M| const __m128i sh = 96| 2.40M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 97| 2.40M| const __m128i perm = _mm_shuffle_epi8(in, sh); 98| 2.40M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 99| 2.40M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 100| 2.40M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 101| 2.40M| if (big_endian) ------------------ | Branch (101:9): [Folded, False: 2.40M] ------------------ 102| 0| composed = _mm_shuffle_epi8(composed, swap); 103| 2.40M| _mm_storeu_si128((__m128i *)utf16_output, composed); 104| 2.40M| utf16_output += 6; // We wrote 12 bytes, 6 code points. 105| 2.40M| } else if (idx < 145) { ------------------ | Branch (105:14): [True: 390k, False: 48.5k] ------------------ 106| | // FOUR (4) input code-code units 107| 390k| const __m128i sh = 108| 390k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 109| 390k| const __m128i perm = _mm_shuffle_epi8(in, sh); 110| 390k| const __m128i ascii = 111| 390k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 112| 390k| const __m128i middlebyte = 113| 390k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 114| 390k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 115| 390k| const __m128i highbyte = 116| 390k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 117| 390k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 118| 390k| const __m128i composed = 119| 390k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 120| 390k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 121| 390k| if (big_endian) ------------------ | Branch (121:9): [Folded, False: 390k] ------------------ 122| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 123| 390k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 124| 390k| utf16_output += 4; 125| 390k| } else if (idx < 209) { ------------------ | Branch (125:14): [True: 47.7k, False: 756] ------------------ 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| 47.7k| const __m128i sh = 137| 47.7k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 138| 47.7k| const __m128i perm = _mm_shuffle_epi8(in, sh); 139| 47.7k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 140| 47.7k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 141| 47.7k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 142| 47.7k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 143| | // correct for spurious high bit 144| 47.7k| const __m128i correct = 145| 47.7k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 146| 47.7k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 147| 47.7k| 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| 47.7k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 151| 47.7k| 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| 47.7k| const __m128i composed = 157| 47.7k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 158| 47.7k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 159| 47.7k| const __m128i composedminus = 160| 47.7k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 161| 47.7k| const __m128i lowtenbits = 162| 47.7k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 163| | // Notice the 0x3ff mask: 164| 47.7k| const __m128i hightenbits = 165| 47.7k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 166| 47.7k| const __m128i lowtenbitsadd = 167| 47.7k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 168| 47.7k| const __m128i hightenbitsadd = 169| 47.7k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 170| 47.7k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 171| 47.7k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 172| 47.7k| uint32_t basic_buffer[4]; 173| 47.7k| uint32_t basic_buffer_swap[4]; 174| 47.7k| if (big_endian) { ------------------ | Branch (174:9): [Folded, False: 47.7k] ------------------ 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| 47.7k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 180| 47.7k| uint32_t surrogate_buffer[4]; 181| 47.7k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 182| 191k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (182:24): [True: 143k, False: 47.7k] ------------------ 183| 143k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (183:11): [True: 40.1k, False: 103k] ------------------ 184| 40.1k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 185| 40.1k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 186| 40.1k| utf16_output += 2; 187| 103k| } else { 188| 103k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (188:27): [Folded, False: 103k] ------------------ 189| 103k| : uint16_t(basic_buffer[i]); 190| 103k| utf16_output++; 191| 103k| } 192| 143k| } 193| 47.7k| } else { 194| | // here we know that there is an error but we do not handle errors 195| 756| } 196| 2.84M| return consumed; 197| 2.84M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE1EEEmPKcmRPDs: 10| 3.09M| 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| 3.09M| const __m128i swap = 23| 3.09M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 3.09M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 3.09M| const uint16_t input_utf8_end_of_code_point_mask = 26| 3.09M| utf8_end_of_code_point_mask & 0xfff; 27| 3.09M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 8.76k, False: 3.09M] ------------------ 28| | // We process the data in chunks of 12 bytes. 29| | // Note: using 16 bytes is unsafe, see issue_ossfuzz_71218 30| 8.76k| __m128i ascii_first = _mm_cvtepu8_epi16(in); 31| 8.76k| __m128i ascii_second = _mm_cvtepu8_epi16(_mm_srli_si128(in, 8)); 32| 8.76k| if (big_endian) { ------------------ | Branch (32:9): [True: 8.76k, Folded] ------------------ 33| 8.76k| ascii_first = _mm_shuffle_epi8(ascii_first, swap); 34| 8.76k| ascii_second = _mm_shuffle_epi8(ascii_second, swap); 35| 8.76k| } 36| 8.76k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf16_output), ascii_first); 37| 8.76k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf16_output + 8), 38| 8.76k| ascii_second); 39| 8.76k| utf16_output += 12; // We wrote 12 16-bit characters. 40| 8.76k| return 12; // We consumed 12 bytes. 41| 8.76k| } 42| 3.09M| if (((utf8_end_of_code_point_mask & 0xFFFF) == 0xaaaa)) { ------------------ | Branch (42:7): [True: 20.4k, False: 3.07M] ------------------ 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| 20.4k| const __m128i sh = 47| 20.4k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 48| 20.4k| const __m128i perm = _mm_shuffle_epi8(in, sh); 49| 20.4k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 50| 20.4k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 51| 20.4k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 52| 20.4k| if (big_endian) ------------------ | Branch (52:9): [True: 20.4k, Folded] ------------------ 53| 20.4k| composed = _mm_shuffle_epi8(composed, swap); 54| 20.4k| _mm_storeu_si128((__m128i *)utf16_output, composed); 55| 20.4k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 56| 20.4k| return 16; 57| 20.4k| } 58| 3.07M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (58:7): [True: 4.37k, False: 3.06M] ------------------ 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| 4.37k| const __m128i sh = 63| 4.37k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 64| 4.37k| const __m128i perm = _mm_shuffle_epi8(in, sh); 65| 4.37k| const __m128i ascii = 66| 4.37k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 67| 4.37k| const __m128i middlebyte = 68| 4.37k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 69| 4.37k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 70| 4.37k| const __m128i highbyte = 71| 4.37k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 72| 4.37k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 73| 4.37k| const __m128i composed = 74| 4.37k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 75| 4.37k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 76| 4.37k| if (big_endian) ------------------ | Branch (76:9): [True: 4.37k, Folded] ------------------ 77| 4.37k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 78| 4.37k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 79| 4.37k| utf16_output += 4; 80| 4.37k| return 12; 81| 4.37k| } 82| | /// We do not have a fast path available, so we fallback. 83| | 84| 3.06M| const uint8_t idx = 85| 3.06M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 86| 3.06M| const uint8_t consumed = 87| 3.06M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 88| 3.06M| if (idx < 64) { ------------------ | Branch (88:7): [True: 2.75M, False: 306k] ------------------ 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| 2.75M| const __m128i sh = 96| 2.75M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 97| 2.75M| const __m128i perm = _mm_shuffle_epi8(in, sh); 98| 2.75M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 99| 2.75M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 100| 2.75M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 101| 2.75M| if (big_endian) ------------------ | Branch (101:9): [True: 2.75M, Folded] ------------------ 102| 2.75M| composed = _mm_shuffle_epi8(composed, swap); 103| 2.75M| _mm_storeu_si128((__m128i *)utf16_output, composed); 104| 2.75M| utf16_output += 6; // We wrote 12 bytes, 6 code points. 105| 2.75M| } else if (idx < 145) { ------------------ | Branch (105:14): [True: 253k, False: 52.3k] ------------------ 106| | // FOUR (4) input code-code units 107| 253k| const __m128i sh = 108| 253k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 109| 253k| const __m128i perm = _mm_shuffle_epi8(in, sh); 110| 253k| const __m128i ascii = 111| 253k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 112| 253k| const __m128i middlebyte = 113| 253k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 114| 253k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 115| 253k| const __m128i highbyte = 116| 253k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 117| 253k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 118| 253k| const __m128i composed = 119| 253k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 120| 253k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 121| 253k| if (big_endian) ------------------ | Branch (121:9): [True: 253k, Folded] ------------------ 122| 253k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 123| 253k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 124| 253k| utf16_output += 4; 125| 253k| } else if (idx < 209) { ------------------ | Branch (125:14): [True: 51.0k, False: 1.31k] ------------------ 126| | // TWO (2) input code-code units 127| | ////////////// 128| | // There might be garbage inputs where a leading byte mascarades as a 129| | // four-byte leading byte (by being followed by 3 continuation byte), but is 130| | // not greater than 0xf0. This could trigger a buffer overflow if we only 131| | // counted leading bytes of the form 0xf0 as generating surrogate pairs, 132| | // without further UTF-8 validation. Thus we must be careful to ensure that 133| | // only leading bytes at least as large as 0xf0 generate surrogate pairs. We 134| | // do as at the cost of an extra mask. 135| | ///////////// 136| 51.0k| const __m128i sh = 137| 51.0k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 138| 51.0k| const __m128i perm = _mm_shuffle_epi8(in, sh); 139| 51.0k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 140| 51.0k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 141| 51.0k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 142| 51.0k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 143| | // correct for spurious high bit 144| 51.0k| const __m128i correct = 145| 51.0k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 146| 51.0k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 147| 51.0k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 148| | // We deliberately carry the leading four bits in highbyte if they are 149| | // present, we remove them later when computing hightenbits. 150| 51.0k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 151| 51.0k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 152| | // When we need to generate a surrogate pair (leading byte > 0xF0), then 153| | // the corresponding 32-bit value in 'composed' will be greater than 154| | // > (0xff00000>>6) or > 0x3c00000. This can be used later to identify the 155| | // location of the surrogate pairs. 156| 51.0k| const __m128i composed = 157| 51.0k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 158| 51.0k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 159| 51.0k| const __m128i composedminus = 160| 51.0k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 161| 51.0k| const __m128i lowtenbits = 162| 51.0k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 163| | // Notice the 0x3ff mask: 164| 51.0k| const __m128i hightenbits = 165| 51.0k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 166| 51.0k| const __m128i lowtenbitsadd = 167| 51.0k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 168| 51.0k| const __m128i hightenbitsadd = 169| 51.0k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 170| 51.0k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 171| 51.0k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 172| 51.0k| uint32_t basic_buffer[4]; 173| 51.0k| uint32_t basic_buffer_swap[4]; 174| 51.0k| if (big_endian) { ------------------ | Branch (174:9): [True: 51.0k, Folded] ------------------ 175| 51.0k| _mm_storeu_si128((__m128i *)basic_buffer_swap, 176| 51.0k| _mm_shuffle_epi8(composed, swap)); 177| 51.0k| surrogates = _mm_shuffle_epi8(surrogates, swap); 178| 51.0k| } 179| 51.0k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 180| 51.0k| uint32_t surrogate_buffer[4]; 181| 51.0k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 182| 204k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (182:24): [True: 153k, False: 51.0k] ------------------ 183| 153k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (183:11): [True: 42.9k, False: 110k] ------------------ 184| 42.9k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 185| 42.9k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 186| 42.9k| utf16_output += 2; 187| 110k| } else { 188| 110k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (188:27): [True: 110k, Folded] ------------------ 189| 110k| : uint16_t(basic_buffer[i]); 190| 110k| utf16_output++; 191| 110k| } 192| 153k| } 193| 51.0k| } else { 194| | // here we know that there is an error but we do not handle errors 195| 1.31k| } 196| 3.06M| return consumed; 197| 3.07M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_128convert_masked_utf8_to_utf32EPKcmRPDi: 9| 2.98M| char32_t *&utf32_output) { 10| | // we use an approach where we try to process up to 12 input bytes. 11| | // Why 12 input bytes and not 16? Because we are concerned with the size of 12| | // the lookup tables. Also 12 is nicely divisible by two and three. 13| | // 14| | // 15| | // Optimization note: our main path below is load-latency dependent. Thus it 16| | // is maybe beneficial to have fast paths that depend on branch prediction but 17| | // have less latency. This results in more instructions but, potentially, also 18| | // higher speeds. 19| | // 20| | // We first try a few fast paths. 21| 2.98M| const __m128i in = _mm_loadu_si128((__m128i *)input); 22| 2.98M| const uint16_t input_utf8_end_of_code_point_mask = 23| 2.98M| utf8_end_of_code_point_mask & 0xfff; 24| 2.98M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (24:7): [True: 8.66k, False: 2.98M] ------------------ 25| | // We process the data in chunks of 12 bytes. 26| 8.66k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 27| 8.66k| _mm_cvtepu8_epi32(in)); 28| 8.66k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 29| 8.66k| _mm_cvtepu8_epi32(_mm_srli_si128(in, 4))); 30| 8.66k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 8), 31| 8.66k| _mm_cvtepu8_epi32(_mm_srli_si128(in, 8))); 32| 8.66k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 12), 33| 8.66k| _mm_cvtepu8_epi32(_mm_srli_si128(in, 12))); 34| 8.66k| utf32_output += 12; // We wrote 12 32-bit characters. 35| 8.66k| return 12; // We consumed 12 bytes. 36| 8.66k| } 37| 2.98M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (37:7): [True: 18.3k, False: 2.96M] ------------------ 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| 18.3k| const __m128i sh = 42| 18.3k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 43| 18.3k| const __m128i perm = _mm_shuffle_epi8(in, sh); 44| 18.3k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 45| 18.3k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 46| 18.3k| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 47| 18.3k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 48| 18.3k| _mm_cvtepu16_epi32(composed)); 49| 18.3k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 50| 18.3k| _mm_cvtepu16_epi32(_mm_srli_si128(composed, 8))); 51| 18.3k| utf32_output += 8; // We wrote 32 bytes, 8 code points. 52| 18.3k| return 16; 53| 18.3k| } 54| 2.96M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (54:7): [True: 2.52k, False: 2.95M] ------------------ 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| 2.52k| const __m128i sh = 59| 2.52k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 60| 2.52k| const __m128i perm = _mm_shuffle_epi8(in, sh); 61| 2.52k| const __m128i ascii = 62| 2.52k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 63| 2.52k| const __m128i middlebyte = 64| 2.52k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 65| 2.52k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 66| 2.52k| const __m128i highbyte = 67| 2.52k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 68| 2.52k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 69| 2.52k| const __m128i composed = 70| 2.52k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 71| 2.52k| _mm_storeu_si128((__m128i *)utf32_output, composed); 72| 2.52k| utf32_output += 4; 73| 2.52k| return 12; 74| 2.52k| } 75| | /// We do not have a fast path available, so we fallback. 76| | 77| 2.95M| const uint8_t idx = 78| 2.95M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 79| 2.95M| const uint8_t consumed = 80| 2.95M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 81| 2.95M| if (idx < 64) { ------------------ | Branch (81:7): [True: 2.70M, False: 250k] ------------------ 82| | // SIX (6) input code-code units 83| | // this is a relatively easy scenario 84| | // we process SIX (6) input code-code units. The max length in bytes of six 85| | // code code units spanning between 1 and 2 bytes each is 12 bytes. On 86| | // processors where pdep/pext is fast, we might be able to use a small 87| | // lookup table. 88| 2.70M| const __m128i sh = 89| 2.70M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 90| 2.70M| const __m128i perm = _mm_shuffle_epi8(in, sh); 91| 2.70M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 92| 2.70M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 93| 2.70M| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 94| 2.70M| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 95| 2.70M| _mm_cvtepu16_epi32(composed)); 96| 2.70M| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 97| 2.70M| _mm_cvtepu16_epi32(_mm_srli_si128(composed, 8))); 98| 2.70M| utf32_output += 6; // We wrote 12 bytes, 6 code points. 99| 2.70M| } else if (idx < 145) { ------------------ | Branch (99:14): [True: 202k, False: 48.0k] ------------------ 100| | // FOUR (4) input code-code units 101| 202k| const __m128i sh = 102| 202k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 103| 202k| const __m128i perm = _mm_shuffle_epi8(in, sh); 104| 202k| const __m128i ascii = 105| 202k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 106| 202k| const __m128i middlebyte = 107| 202k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 108| 202k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 109| 202k| const __m128i highbyte = 110| 202k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 111| 202k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 112| 202k| const __m128i composed = 113| 202k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 114| 202k| _mm_storeu_si128((__m128i *)utf32_output, composed); 115| 202k| utf32_output += 4; 116| 202k| } else if (idx < 209) { ------------------ | Branch (116:14): [True: 47.3k, False: 670] ------------------ 117| | // TWO (2) input code-code units 118| 47.3k| const __m128i sh = 119| 47.3k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 120| 47.3k| const __m128i perm = _mm_shuffle_epi8(in, sh); 121| 47.3k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 122| 47.3k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 123| 47.3k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 124| 47.3k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 125| | // correct for spurious high bit 126| 47.3k| const __m128i correct = 127| 47.3k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 128| 47.3k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 129| 47.3k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 130| 47.3k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0x07000000)); 131| 47.3k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 132| 47.3k| const __m128i composed = 133| 47.3k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 134| 47.3k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 135| 47.3k| _mm_storeu_si128((__m128i *)utf32_output, composed); 136| 47.3k| utf32_output += 3; 137| 47.3k| } else { 138| | // here we know that there is an error but we do not handle errors 139| 670| } 140| 2.95M| return consumed; 141| 2.96M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_123utf16_gather_high_bytesILNS_10endiannessE0EEENS1_4simd5simd8IhEENS4_6simd16ItEES8_: 3| 2.70M| const simd16 in1) { 4| 2.70M| if (big_endian) { ------------------ | Branch (4:7): [Folded, False: 2.70M] ------------------ 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| 2.70M| } else { 12| 2.70M| const auto t0 = in0.shr<8>(); 13| 2.70M| const auto t1 = in1.shr<8>(); 14| | 15| 2.70M| return simd16::pack(t0, t1); 16| 2.70M| } 17| 2.70M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_123utf16_gather_high_bytesILNS_10endiannessE1EEENS1_4simd5simd8IhEENS4_6simd16ItEES8_: 3| 2.80M| const simd16 in1) { 4| 2.80M| if (big_endian) { ------------------ | Branch (4:7): [True: 2.80M, Folded] ------------------ 5| | // we want lower bytes 6| 2.80M| const auto mask = simd16(0x00ff); 7| 2.80M| const auto t0 = in0 & mask; 8| 2.80M| const auto t1 = in1 & mask; 9| | 10| 2.80M| return simd16::pack(t0, t1); 11| 2.80M| } 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| 2.80M|}