_Z18populate_functionsv: 548| 1|const auto populate_functions() { 549| 1| using I = simdutf::implementation; 550| 1| using FuzzSignature = void (*)(std::span); 551| | 552| 1|#define ADD(lenfunc, conversionfunc) \ 553| 1| FuzzSignature { \ 554| 1| +[](std::span chardata) { \ 555| 1| const auto c = \ 556| 1| Conversion{ \ 559| 1| &I::lenfunc, &I::conversionfunc, \ 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ 562| 1| c.fuzz(chardata); \ 563| 1| } \ 564| 1| } 565| | 566| 1| return std::array{ 567| | // all these cases require valid input for invoking the convert function 568| | 569| | // see #493 570| | // IGNORE(latin1_length_from_utf16, convert_valid_utf16be_to_latin1), 571| 1| ADD(utf32_length_from_utf16be, convert_valid_utf16be_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 572| 1| ADD(utf8_length_from_utf16be, convert_valid_utf16be_to_utf8), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 573| | 574| | // see #493 575| | // IGNORE(latin1_length_from_utf16, convert_valid_utf16le_to_latin1), 576| 1| ADD(utf32_length_from_utf16le, convert_valid_utf16le_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 577| 1| ADD(utf8_length_from_utf16le, convert_valid_utf16le_to_utf8), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 578| | 579| | // see #493 580| | // IGNORE(latin1_length_from_utf32, convert_valid_utf32_to_latin1), 581| 1| ADD(utf16_length_from_utf32, convert_valid_utf32_to_utf16be), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 582| 1| ADD(utf16_length_from_utf32, convert_valid_utf32_to_utf16le), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 583| 1| ADD(utf8_length_from_utf32, convert_valid_utf32_to_utf8), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 584| | 585| | // see #493 586| | // IGNORE(latin1_length_from_utf8, convert_valid_utf8_to_latin1), 587| 1| ADD(utf16_length_from_utf8, convert_valid_utf8_to_utf16be), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 588| 1| ADD(utf16_length_from_utf8, convert_valid_utf8_to_utf16le), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 589| 1| ADD(utf32_length_from_utf8, convert_valid_utf8_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 590| | 591| | // all these cases operate on arbitrary data 592| 1| ADD(latin1_length_from_utf16, convert_utf16be_to_latin1), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 593| 1| ADD(utf32_length_from_utf16be, convert_utf16be_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 594| 1| ADD(utf8_length_from_utf16be, convert_utf16be_to_utf8), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 595| | 596| 1| ADD(latin1_length_from_utf16, convert_utf16le_to_latin1), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 597| 1| ADD(utf32_length_from_utf16le, convert_utf16le_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 598| 1| ADD(utf8_length_from_utf16le, convert_utf16le_to_utf8), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 599| | 600| 1| ADD(latin1_length_from_utf32, convert_utf32_to_latin1), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 601| 1| ADD(utf16_length_from_utf32, convert_utf32_to_utf16be), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 602| 1| ADD(utf16_length_from_utf32, convert_utf32_to_utf16le), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 603| 1| ADD(utf8_length_from_utf32, convert_utf32_to_utf8), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 604| | 605| 1| ADD(latin1_length_from_utf8, convert_utf8_to_latin1), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 606| 1| ADD(utf16_length_from_utf8, convert_utf8_to_utf16be), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 607| 1| ADD(utf16_length_from_utf8, convert_utf8_to_utf16le), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 608| 1| ADD(utf32_length_from_utf8, convert_utf8_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 609| | 610| | // all these cases operate on arbitrary data and use the _with_errors 611| | // variant 612| 1| ADD(latin1_length_from_utf16, convert_utf16be_to_latin1_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 613| 1| ADD(utf32_length_from_utf16be, convert_utf16be_to_utf32_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 614| 1| ADD(utf8_length_from_utf16be, convert_utf16be_to_utf8_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 615| | 616| 1| ADD(latin1_length_from_utf16, convert_utf16le_to_latin1_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 617| 1| ADD(utf32_length_from_utf16le, convert_utf16le_to_utf32_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 618| 1| ADD(utf8_length_from_utf16le, convert_utf16le_to_utf8_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 619| | 620| 1| ADD(latin1_length_from_utf32, convert_utf32_to_latin1_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 621| 1| ADD(utf16_length_from_utf32, convert_utf32_to_utf16be_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 622| 1| ADD(utf16_length_from_utf32, convert_utf32_to_utf16le_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 623| 1| ADD(utf8_length_from_utf32, convert_utf32_to_utf8_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 624| | 625| 1| ADD(latin1_length_from_utf8, convert_utf8_to_latin1_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 626| 1| ADD(utf16_length_from_utf8, convert_utf8_to_utf16be_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 627| 1| ADD(utf16_length_from_utf8, convert_utf8_to_utf16le_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 628| 1| ADD(utf32_length_from_utf8, convert_utf8_to_utf32_with_errors), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 629| | 630| | // these are a bit special since all input is valid 631| 1| ADD(utf32_length_from_latin1, convert_latin1_to_utf32), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 632| 1| ADD(utf16_length_from_latin1, convert_latin1_to_utf16be), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 633| 1| ADD(utf16_length_from_latin1, convert_latin1_to_utf16le), ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 634| 1| ADD(utf8_length_from_latin1, convert_latin1_to_utf8)}; ------------------ | | 553| 1| FuzzSignature { \ | | 554| 1| +[](std::span chardata) { \ | | 555| 1| const auto c = \ | | 556| 1| Conversion{ \ | | 559| 1| &I::lenfunc, &I::conversionfunc, \ | | 560| 1| std::string{NAMEOF(&I::lenfunc)}, \ | | 561| 1| std::string{NAMEOF(&I::conversionfunc)}}; \ | | 562| 1| c.fuzz(chardata); \ | | 563| 1| } \ | | 564| 1| } ------------------ 635| | 636| 1|#undef ADD 637| 1|} LLVMFuzzerTestOneInput: 639| 9.22k|extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size) { 640| 9.22k| static const auto fptrs = populate_functions(); 641| 9.22k| constexpr std::size_t Ncases = fptrs.size(); 642| | 643| | // pick one of the function pointers, based on the fuzz data 644| | // the first byte is which action to take. step forward 645| | // several bytes so the input is aligned. 646| 9.22k| if (size < 4) { ------------------ | Branch (646:7): [True: 3, False: 9.21k] ------------------ 647| 3| return 0; 648| 3| } 649| | 650| 9.21k| constexpr auto actionmask = std::bit_ceil(Ncases) - 1; 651| 9.21k| const auto action = data[0] & actionmask; 652| 9.21k| data += 4; 653| 9.21k| size -= 4; 654| | 655| 9.21k| if (action >= Ncases) { ------------------ | Branch (655:7): [True: 1, False: 9.21k] ------------------ 656| 1| return 0; 657| 1| } 658| | 659| 9.21k| if constexpr (use_separate_allocation) { 660| | // this is better at exercising null input and catch buffer underflows 661| 9.21k| const std::vector separate{data, data + size}; 662| 9.21k| fptrs[action](std::span(separate)); 663| | } else { 664| | std::span chardata{(const char*)data, size}; 665| | fptrs[action](chardata); 666| | } 667| | 668| 9.21k| return 0; 669| 9.21k|} conversion.cpp:_ZZ18populate_functionsvENK3$_0clENSt3__14spanIKcLm18446744073709551615EEE: 554| 96| +[](std::span chardata) { \ 555| 96| const auto c = \ 556| 96| Conversion{ \ 559| 96| &I::lenfunc, &I::conversionfunc, \ 560| 96| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 96|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 96| ::std::void_t(); \ | | 1180| 96| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 96| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 96| constexpr auto _size = _name.size(); \ | | 1183| 96| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 96| return _nameof; }() ------------------ 561| 96| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 96|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 96| ::std::void_t(); \ | | 1180| 96| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 96| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 96| constexpr auto _size = _name.size(); \ | | 1183| 96| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 96| return _nameof; }() ------------------ 562| 96| c.fuzz(chardata); \ 563| 96| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 237| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 237| const FromSpan from{reinterpret_cast(chardata.data()), 178| 237| chardata.size() / sizeof(FromType)}; 179| | 180| 237| static const bool do_print_testcase = 181| 237| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 237| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 237] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 237| do { 189| | // step 0 - is the input valid? 190| 237| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 237| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 237] | 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| 237| From == UtfEncodings::UTF8) { 198| 237| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 237] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 237| } 201| | 202| | // step 2 - what is the required size of the output? 203| 237| const auto [output_length, length_agree] = 204| 237| calculate_length(from, inputisvalid); 205| 237| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 237] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 237| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 101, False: 136] | Branch (208:28): [True: 13, False: 88] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 13| return; 211| 13| } 212| | 213| | // step 3 - run the conversion 214| 224| const auto [written, outputs_agree] = 215| 224| do_conversion(from, output_length, inputisvalid); 216| 224| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 224] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 224| return; 221| 224| } 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| 237| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 237| validation_result verify_valid_input(FromSpan src) const { 231| 237| validation_result ret{}; 232| | 233| 237| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 237| const auto implementations = get_supported_implementations(); 235| 237| std::vector results; 236| 237| results.reserve(implementations.size()); 237| | 238| 711| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 711, False: 237] ------------------ 239| 711| results.push_back( 240| 711| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 711| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 711| const bool validation2 = 245| 711| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 711| src.data(), src.size()); 247| 711| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 711] ------------------ 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| 711| } 258| | 259| 237| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 237| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 237] ------------------ 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| 237| } else { 273| 237| ret.implementations_agree = true; 274| 237| } 275| 237| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 237| return r.error == simdutf::SUCCESS; 277| 237| }); 278| 237| return ret; 279| 237| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 474| 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| 509| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 509| return r.error == simdutf::SUCCESS; 277| 509| }); _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 237| bool count_the_input(FromSpan src) const { 290| 237| const auto implementations = get_supported_implementations(); 291| 237| std::vector results; 292| 237| results.reserve(implementations.size()); 293| | 294| 711| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 711, False: 237] ------------------ 295| 711| std::size_t ret; 296| 711| if constexpr (From == UtfEncodings::UTF16BE) { 297| 711| 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| 711| results.push_back(ret); 304| 711| } 305| 237| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 237| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 237] ------------------ 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| 237| return true; 321| 237| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 474| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 237| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 237| length_result ret{}; 344| | 345| 237| const auto implementations = get_supported_implementations(); 346| 237| std::vector results; 347| 237| results.reserve(implementations.size()); 348| | 349| 711| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 711, False: 237] ------------------ 350| 711| const auto len = invoke_lengthcalc(impl, src); 351| 711| results.push_back(len); 352| 711| ret.length.push_back(len); 353| 711| } 354| | 355| 237| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 237| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 237] ------------------ 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| 237| } else { 375| 237| ret.implementations_agree = true; 376| 237| } 377| 237| return ret; 378| 237| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 711| FromSpan src) const { 331| 711| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 711| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 474| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 224| const bool inputisvalid) const { 383| 224| conversion_result ret{}; 384| | 385| 224| const auto implementations = get_supported_implementations(); 386| | 387| 224| std::vector> results; 388| 224| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 224| std::vector> outputbuffers; 393| 224| outputbuffers.reserve(implementations.size()); 394| 896| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 672, False: 224] ------------------ 395| 672| auto impl = implementations[i]; 396| 672| const ToType canary1{42}; 397| 672| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 672| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 672| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 672| const auto success = [](const ConversionResult& r) -> bool { 402| 672| if constexpr (std::is_same_v) { 403| 672| return r != 0; 404| 672| } else { 405| 672| return r.error == simdutf::error_code::SUCCESS; 406| 672| } 407| 672| }(implret1); 408| 672| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 672| 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| 672| const ToType canary2{25}; 414| 672| const auto outputbuffer_first_run = outputbuffer; 415| 672| std::ranges::fill(outputbuffer, canary2); 416| 672| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 672| src.size(), outputbuffer.data()); 418| | 419| 672| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 672] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 672| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 408, False: 264] | Branch (423:29): [True: 396, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 396| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 396| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 396] ------------------ 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| 396| } 440| 672| } 441| 672| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 396, False: 276] ------------------ 442| 672| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 224| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 88, False: 136] ------------------ 447| 264| for (auto& e : results) { ------------------ | Branch (447:20): [True: 264, False: 88] ------------------ 448| 264| e.outputhash.clear(); 449| 264| } 450| 88| } 451| | 452| 224| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 224| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 224] ------------------ 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| 224| } else { 474| 224| ret.implementations_agree = true; 475| 224| } 476| 224| return ret; 477| 224| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 672| const auto success = [](const ConversionResult& r) -> bool { 402| 672| if constexpr (std::is_same_v) { 403| 672| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 672| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 448| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_1clENSt3__14spanIKcLm18446744073709551615EEE: 554| 224| +[](std::span chardata) { \ 555| 224| const auto c = \ 556| 224| Conversion{ \ 559| 224| &I::lenfunc, &I::conversionfunc, \ 560| 224| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 224|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 224| ::std::void_t(); \ | | 1180| 224| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 224| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 224| constexpr auto _size = _name.size(); \ | | 1183| 224| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 224| return _nameof; }() ------------------ 561| 224| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 224|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 224| ::std::void_t(); \ | | 1180| 224| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 224| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 224| constexpr auto _size = _name.size(); \ | | 1183| 224| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 224| return _nameof; }() ------------------ 562| 224| c.fuzz(chardata); \ 563| 224| } \ _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: 86, False: 317] | Branch (208:28): [True: 17, False: 69] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 17| return; 211| 17| } 212| | 213| | // step 3 - run the conversion 214| 386| const auto [written, outputs_agree] = 215| 386| do_conversion(from, output_length, inputisvalid); 216| 386| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 386] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 386| return; 221| 386| } 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| 386| const bool inputisvalid) const { 383| 386| conversion_result ret{}; 384| | 385| 386| const auto implementations = get_supported_implementations(); 386| | 387| 386| std::vector> results; 388| 386| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 386| std::vector> outputbuffers; 393| 386| outputbuffers.reserve(implementations.size()); 394| 1.54k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.15k, False: 386] ------------------ 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: 951, False: 207] | Branch (423:29): [True: 942, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 942| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 942| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 942] ------------------ 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| 942| } 440| 1.15k| } 441| 1.15k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 942, False: 216] ------------------ 442| 1.15k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 386| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 69, False: 317] ------------------ 447| 207| for (auto& e : results) { ------------------ | Branch (447:20): [True: 207, False: 69] ------------------ 448| 207| e.outputhash.clear(); 449| 207| } 450| 69| } 451| | 452| 386| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 386| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 386] ------------------ 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| 386| } else { 474| 386| ret.implementations_agree = true; 475| 386| } 476| 386| return ret; 477| 386| } _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| 772| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_2clENSt3__14spanIKcLm18446744073709551615EEE: 554| 65| +[](std::span chardata) { \ 555| 65| const auto c = \ 556| 65| Conversion{ \ 559| 65| &I::lenfunc, &I::conversionfunc, \ 560| 65| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 65|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 65| ::std::void_t(); \ | | 1180| 65| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 65| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 65| constexpr auto _size = _name.size(); \ | | 1183| 65| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 65| return _nameof; }() ------------------ 561| 65| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 65|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 65| ::std::void_t(); \ | | 1180| 65| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 65| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 65| constexpr auto _size = _name.size(); \ | | 1183| 65| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 65| return _nameof; }() ------------------ 562| 65| c.fuzz(chardata); \ 563| 65| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 219| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 219| const FromSpan from{reinterpret_cast(chardata.data()), 178| 219| chardata.size() / sizeof(FromType)}; 179| | 180| 219| static const bool do_print_testcase = 181| 219| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 219| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 219] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 219| do { 189| | // step 0 - is the input valid? 190| 219| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 219| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 219] | 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| 219| From == UtfEncodings::UTF8) { 198| 219| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 219] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 219| } 201| | 202| | // step 2 - what is the required size of the output? 203| 219| const auto [output_length, length_agree] = 204| 219| calculate_length(from, inputisvalid); 205| 219| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 219] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 219| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 101, False: 118] | Branch (208:28): [True: 5, False: 96] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 5| return; 211| 5| } 212| | 213| | // step 3 - run the conversion 214| 214| const auto [written, outputs_agree] = 215| 214| do_conversion(from, output_length, inputisvalid); 216| 214| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 214] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 214| return; 221| 214| } 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| 219| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 219| validation_result verify_valid_input(FromSpan src) const { 231| 219| validation_result ret{}; 232| | 233| 219| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 219| const auto implementations = get_supported_implementations(); 235| 219| std::vector results; 236| 219| results.reserve(implementations.size()); 237| | 238| 657| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 657, False: 219] ------------------ 239| 657| results.push_back( 240| 657| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 657| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 657| const bool validation2 = 245| 657| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 657| src.data(), src.size()); 247| 657| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 657] ------------------ 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| 657| } 258| | 259| 219| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 219| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 219] ------------------ 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| 219| } else { 273| 219| ret.implementations_agree = true; 274| 219| } 275| 219| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 219| return r.error == simdutf::SUCCESS; 277| 219| }); 278| 219| return ret; 279| 219| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 438| 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| 455| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 455| return r.error == simdutf::SUCCESS; 277| 455| }); _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 219| bool count_the_input(FromSpan src) const { 290| 219| const auto implementations = get_supported_implementations(); 291| 219| std::vector results; 292| 219| results.reserve(implementations.size()); 293| | 294| 657| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 657, False: 219] ------------------ 295| 657| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 657| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 657| 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| 657| results.push_back(ret); 304| 657| } 305| 219| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 219| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 219] ------------------ 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| 219| return true; 321| 219| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 438| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 219| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 219| length_result ret{}; 344| | 345| 219| const auto implementations = get_supported_implementations(); 346| 219| std::vector results; 347| 219| results.reserve(implementations.size()); 348| | 349| 657| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 657, False: 219] ------------------ 350| 657| const auto len = invoke_lengthcalc(impl, src); 351| 657| results.push_back(len); 352| 657| ret.length.push_back(len); 353| 657| } 354| | 355| 219| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 219| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 219] ------------------ 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| 219| } else { 375| 219| ret.implementations_agree = true; 376| 219| } 377| 219| return ret; 378| 219| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 657| FromSpan src) const { 331| 657| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 657| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 438| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 214| const bool inputisvalid) const { 383| 214| conversion_result ret{}; 384| | 385| 214| const auto implementations = get_supported_implementations(); 386| | 387| 214| std::vector> results; 388| 214| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 214| std::vector> outputbuffers; 393| 214| outputbuffers.reserve(implementations.size()); 394| 856| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 642, False: 214] ------------------ 395| 642| auto impl = implementations[i]; 396| 642| const ToType canary1{42}; 397| 642| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 642| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 642| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 642| const auto success = [](const ConversionResult& r) -> bool { 402| 642| if constexpr (std::is_same_v) { 403| 642| return r != 0; 404| 642| } else { 405| 642| return r.error == simdutf::error_code::SUCCESS; 406| 642| } 407| 642| }(implret1); 408| 642| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 642| 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| 642| const ToType canary2{25}; 414| 642| const auto outputbuffer_first_run = outputbuffer; 415| 642| std::ranges::fill(outputbuffer, canary2); 416| 642| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 642| src.size(), outputbuffer.data()); 418| | 419| 642| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 642] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 642| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 354, False: 288] | Branch (423:29): [True: 342, False: 12] ------------------ 424| | // only care about the output if the input is valid 425| 342| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 342| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 342] ------------------ 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| 342| } 440| 642| } 441| 642| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 342, False: 300] ------------------ 442| 642| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 214| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 96, False: 118] ------------------ 447| 288| for (auto& e : results) { ------------------ | Branch (447:20): [True: 288, False: 96] ------------------ 448| 288| e.outputhash.clear(); 449| 288| } 450| 96| } 451| | 452| 214| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 214| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 214] ------------------ 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| 214| } else { 474| 214| ret.implementations_agree = true; 475| 214| } 476| 214| return ret; 477| 214| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 642| const auto success = [](const ConversionResult& r) -> bool { 402| 642| if constexpr (std::is_same_v) { 403| 642| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 642| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 428| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_3clENSt3__14spanIKcLm18446744073709551615EEE: 554| 205| +[](std::span chardata) { \ 555| 205| const auto c = \ 556| 205| Conversion{ \ 559| 205| &I::lenfunc, &I::conversionfunc, \ 560| 205| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 205|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 205| ::std::void_t(); \ | | 1180| 205| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 205| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 205| constexpr auto _size = _name.size(); \ | | 1183| 205| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 205| return _nameof; }() ------------------ 561| 205| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 205|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 205| ::std::void_t(); \ | | 1180| 205| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 205| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 205| constexpr auto _size = _name.size(); \ | | 1183| 205| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 205| return _nameof; }() ------------------ 562| 205| c.fuzz(chardata); \ 563| 205| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 451| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 451| const FromSpan from{reinterpret_cast(chardata.data()), 178| 451| chardata.size() / sizeof(FromType)}; 179| | 180| 451| static const bool do_print_testcase = 181| 451| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 451| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 451] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 451| do { 189| | // step 0 - is the input valid? 190| 451| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 451| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 451] | 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| 451| From == UtfEncodings::UTF8) { 198| 451| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 451] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 451| } 201| | 202| | // step 2 - what is the required size of the output? 203| 451| const auto [output_length, length_agree] = 204| 451| calculate_length(from, inputisvalid); 205| 451| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 451] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 451| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 132, False: 319] | Branch (208:28): [True: 13, False: 119] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 13| return; 211| 13| } 212| | 213| | // step 3 - run the conversion 214| 438| const auto [written, outputs_agree] = 215| 438| do_conversion(from, output_length, inputisvalid); 216| 438| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 438] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 438| return; 221| 438| } 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| 451| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 451| validation_result verify_valid_input(FromSpan src) const { 231| 451| validation_result ret{}; 232| | 233| 451| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 451| const auto implementations = get_supported_implementations(); 235| 451| std::vector results; 236| 451| results.reserve(implementations.size()); 237| | 238| 1.35k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.35k, False: 451] ------------------ 239| 1.35k| results.push_back( 240| 1.35k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.35k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.35k| const bool validation2 = 245| 1.35k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.35k| src.data(), src.size()); 247| 1.35k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.35k] ------------------ 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| 1.35k| } 258| | 259| 451| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 451| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 451] ------------------ 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| 451| } else { 273| 451| ret.implementations_agree = true; 274| 451| } 275| 451| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 451| return r.error == simdutf::SUCCESS; 277| 451| }); 278| 451| return ret; 279| 451| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 902| 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| 1.08k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.08k| return r.error == simdutf::SUCCESS; 277| 1.08k| }); _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 451| bool count_the_input(FromSpan src) const { 290| 451| const auto implementations = get_supported_implementations(); 291| 451| std::vector results; 292| 451| results.reserve(implementations.size()); 293| | 294| 1.35k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.35k, False: 451] ------------------ 295| 1.35k| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 1.35k| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 1.35k| 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| 1.35k| results.push_back(ret); 304| 1.35k| } 305| 451| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 451| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 451] ------------------ 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| 451| return true; 321| 451| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 902| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 451| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 451| length_result ret{}; 344| | 345| 451| const auto implementations = get_supported_implementations(); 346| 451| std::vector results; 347| 451| results.reserve(implementations.size()); 348| | 349| 1.35k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.35k, False: 451] ------------------ 350| 1.35k| const auto len = invoke_lengthcalc(impl, src); 351| 1.35k| results.push_back(len); 352| 1.35k| ret.length.push_back(len); 353| 1.35k| } 354| | 355| 451| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 451| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 451] ------------------ 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| 451| } else { 375| 451| ret.implementations_agree = true; 376| 451| } 377| 451| return ret; 378| 451| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.35k| FromSpan src) const { 331| 1.35k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.35k| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 902| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 438| const bool inputisvalid) const { 383| 438| conversion_result ret{}; 384| | 385| 438| const auto implementations = get_supported_implementations(); 386| | 387| 438| std::vector> results; 388| 438| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 438| std::vector> outputbuffers; 393| 438| outputbuffers.reserve(implementations.size()); 394| 1.75k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.31k, False: 438] ------------------ 395| 1.31k| auto impl = implementations[i]; 396| 1.31k| const ToType canary1{42}; 397| 1.31k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.31k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.31k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.31k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.31k| if constexpr (std::is_same_v) { 403| 1.31k| return r != 0; 404| 1.31k| } else { 405| 1.31k| return r.error == simdutf::error_code::SUCCESS; 406| 1.31k| } 407| 1.31k| }(implret1); 408| 1.31k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.31k| 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| 1.31k| const ToType canary2{25}; 414| 1.31k| const auto outputbuffer_first_run = outputbuffer; 415| 1.31k| std::ranges::fill(outputbuffer, canary2); 416| 1.31k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.31k| src.size(), outputbuffer.data()); 418| | 419| 1.31k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.31k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.31k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 957, False: 357] | Branch (423:29): [True: 948, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 948| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 948| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 948] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 948| } 440| 1.31k| } 441| 1.31k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 948, False: 366] ------------------ 442| 1.31k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 438| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 119, False: 319] ------------------ 447| 357| for (auto& e : results) { ------------------ | Branch (447:20): [True: 357, False: 119] ------------------ 448| 357| e.outputhash.clear(); 449| 357| } 450| 119| } 451| | 452| 438| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 438| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 438] ------------------ 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| 438| } else { 474| 438| ret.implementations_agree = true; 475| 438| } 476| 438| return ret; 477| 438| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.31k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.31k| if constexpr (std::is_same_v) { 403| 1.31k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.31k| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 876| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_4clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 274| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 274| const FromSpan from{reinterpret_cast(chardata.data()), 178| 274| chardata.size() / sizeof(FromType)}; 179| | 180| 274| static const bool do_print_testcase = 181| 274| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 274| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 274] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 274| do { 189| | // step 0 - is the input valid? 190| 274| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 274| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 274] | 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| 274| const auto [output_length, length_agree] = 204| 274| calculate_length(from, inputisvalid); 205| 274| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 274] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 274| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 148, False: 126] | Branch (208:28): [True: 4, False: 144] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 4| return; 211| 4| } 212| | 213| | // step 3 - run the conversion 214| 270| const auto [written, outputs_agree] = 215| 270| do_conversion(from, output_length, inputisvalid); 216| 270| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 270] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 270| return; 221| 270| } 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| 274| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 274| validation_result verify_valid_input(FromSpan src) const { 231| 274| validation_result ret{}; 232| | 233| 274| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 274| const auto implementations = get_supported_implementations(); 235| 274| std::vector results; 236| 274| results.reserve(implementations.size()); 237| | 238| 822| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 822, False: 274] ------------------ 239| 822| results.push_back( 240| 822| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 822| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 822| const bool validation2 = 245| 822| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 822| src.data(), src.size()); 247| 822| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 822] ------------------ 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| 822| } 258| | 259| 274| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 274| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 274] ------------------ 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| 274| } else { 273| 274| ret.implementations_agree = true; 274| 274| } 275| 274| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 274| return r.error == simdutf::SUCCESS; 277| 274| }); 278| 274| return ret; 279| 274| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 548| 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| 526| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 526| return r.error == simdutf::SUCCESS; 277| 526| }); _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 274| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 274| length_result ret{}; 344| | 345| 274| const auto implementations = get_supported_implementations(); 346| 274| std::vector results; 347| 274| results.reserve(implementations.size()); 348| | 349| 822| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 822, False: 274] ------------------ 350| 822| const auto len = invoke_lengthcalc(impl, src); 351| 822| results.push_back(len); 352| 822| ret.length.push_back(len); 353| 822| } 354| | 355| 274| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 274| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 274] ------------------ 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| 274| } else { 375| 274| ret.implementations_agree = true; 376| 274| } 377| 274| return ret; 378| 274| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 822| FromSpan src) const { 331| 822| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 822| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 548| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 270| const bool inputisvalid) const { 383| 270| conversion_result ret{}; 384| | 385| 270| const auto implementations = get_supported_implementations(); 386| | 387| 270| std::vector> results; 388| 270| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 270| std::vector> outputbuffers; 393| 270| outputbuffers.reserve(implementations.size()); 394| 1.08k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 810, False: 270] ------------------ 395| 810| auto impl = implementations[i]; 396| 810| const ToType canary1{42}; 397| 810| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 810| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 810| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 810| const auto success = [](const ConversionResult& r) -> bool { 402| 810| if constexpr (std::is_same_v) { 403| 810| return r != 0; 404| 810| } else { 405| 810| return r.error == simdutf::error_code::SUCCESS; 406| 810| } 407| 810| }(implret1); 408| 810| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 810| 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| 810| const ToType canary2{25}; 414| 810| const auto outputbuffer_first_run = outputbuffer; 415| 810| std::ranges::fill(outputbuffer, canary2); 416| 810| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 810| src.size(), outputbuffer.data()); 418| | 419| 810| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 810] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 810| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 378, False: 432] | Branch (423:29): [True: 369, False: 9] ------------------ 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| 810| } 441| 810| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 369, False: 441] ------------------ 442| 810| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 270| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 144, False: 126] ------------------ 447| 432| for (auto& e : results) { ------------------ | Branch (447:20): [True: 432, False: 144] ------------------ 448| 432| e.outputhash.clear(); 449| 432| } 450| 144| } 451| | 452| 270| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 270| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 270] ------------------ 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| 270| } else { 474| 270| ret.implementations_agree = true; 475| 270| } 476| 270| return ret; 477| 270| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 810| const auto success = [](const ConversionResult& r) -> bool { 402| 810| if constexpr (std::is_same_v) { 403| 810| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 810| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 540| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_5clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 376| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 376| const FromSpan from{reinterpret_cast(chardata.data()), 178| 376| chardata.size() / sizeof(FromType)}; 179| | 180| 376| static const bool do_print_testcase = 181| 376| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 376| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 376] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 376| do { 189| | // step 0 - is the input valid? 190| 376| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 376| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 376] | 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| 376| const auto [output_length, length_agree] = 204| 376| calculate_length(from, inputisvalid); 205| 376| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 376] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 376| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 185, False: 191] | Branch (208:28): [True: 10, False: 175] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 10| return; 211| 10| } 212| | 213| | // step 3 - run the conversion 214| 366| const auto [written, outputs_agree] = 215| 366| do_conversion(from, output_length, inputisvalid); 216| 366| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 366] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 366| return; 221| 366| } 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| 376| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 376| validation_result verify_valid_input(FromSpan src) const { 231| 376| validation_result ret{}; 232| | 233| 376| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 376| const auto implementations = get_supported_implementations(); 235| 376| std::vector results; 236| 376| results.reserve(implementations.size()); 237| | 238| 1.12k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.12k, False: 376] ------------------ 239| 1.12k| results.push_back( 240| 1.12k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.12k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.12k| const bool validation2 = 245| 1.12k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.12k| src.data(), src.size()); 247| 1.12k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.12k] ------------------ 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| 1.12k| } 258| | 259| 376| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 376| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 376] ------------------ 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| 376| } else { 273| 376| ret.implementations_agree = true; 274| 376| } 275| 376| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 376| return r.error == simdutf::SUCCESS; 277| 376| }); 278| 376| return ret; 279| 376| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 752| 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| 758| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 758| return r.error == simdutf::SUCCESS; 277| 758| }); _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 376| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 376| length_result ret{}; 344| | 345| 376| const auto implementations = get_supported_implementations(); 346| 376| std::vector results; 347| 376| results.reserve(implementations.size()); 348| | 349| 1.12k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.12k, False: 376] ------------------ 350| 1.12k| const auto len = invoke_lengthcalc(impl, src); 351| 1.12k| results.push_back(len); 352| 1.12k| ret.length.push_back(len); 353| 1.12k| } 354| | 355| 376| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 376| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 376] ------------------ 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| 376| } else { 375| 376| ret.implementations_agree = true; 376| 376| } 377| 376| return ret; 378| 376| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.12k| FromSpan src) const { 331| 1.12k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.12k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 752| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 366| const bool inputisvalid) const { 383| 366| conversion_result ret{}; 384| | 385| 366| const auto implementations = get_supported_implementations(); 386| | 387| 366| std::vector> results; 388| 366| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 366| std::vector> outputbuffers; 393| 366| outputbuffers.reserve(implementations.size()); 394| 1.46k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.09k, False: 366] ------------------ 395| 1.09k| auto impl = implementations[i]; 396| 1.09k| const ToType canary1{42}; 397| 1.09k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.09k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.09k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.09k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.09k| if constexpr (std::is_same_v) { 403| 1.09k| return r != 0; 404| 1.09k| } else { 405| 1.09k| return r.error == simdutf::error_code::SUCCESS; 406| 1.09k| } 407| 1.09k| }(implret1); 408| 1.09k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.09k| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 1.09k| const ToType canary2{25}; 414| 1.09k| const auto outputbuffer_first_run = outputbuffer; 415| 1.09k| std::ranges::fill(outputbuffer, canary2); 416| 1.09k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.09k| src.size(), outputbuffer.data()); 418| | 419| 1.09k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.09k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.09k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 573, False: 525] | Branch (423:29): [True: 558, False: 15] ------------------ 424| | // only care about the output if the input is valid 425| 558| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 558| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 558] ------------------ 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| 558| } 440| 1.09k| } 441| 1.09k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 558, False: 540] ------------------ 442| 1.09k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 366| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 175, False: 191] ------------------ 447| 525| for (auto& e : results) { ------------------ | Branch (447:20): [True: 525, False: 175] ------------------ 448| 525| e.outputhash.clear(); 449| 525| } 450| 175| } 451| | 452| 366| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 366| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 366] ------------------ 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| 366| } else { 474| 366| ret.implementations_agree = true; 475| 366| } 476| 366| return ret; 477| 366| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.09k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.09k| if constexpr (std::is_same_v) { 403| 1.09k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.09k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 732| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_6clENSt3__14spanIKcLm18446744073709551615EEE: 554| 100| +[](std::span chardata) { \ 555| 100| const auto c = \ 556| 100| Conversion{ \ 559| 100| &I::lenfunc, &I::conversionfunc, \ 560| 100| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 100|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 100| ::std::void_t(); \ | | 1180| 100| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 100| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 100| constexpr auto _size = _name.size(); \ | | 1183| 100| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 100| return _nameof; }() ------------------ 561| 100| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 100|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 100| ::std::void_t(); \ | | 1180| 100| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 100| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 100| constexpr auto _size = _name.size(); \ | | 1183| 100| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 100| return _nameof; }() ------------------ 562| 100| c.fuzz(chardata); \ 563| 100| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 442| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 442| const FromSpan from{reinterpret_cast(chardata.data()), 178| 442| chardata.size() / sizeof(FromType)}; 179| | 180| 442| static const bool do_print_testcase = 181| 442| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 442| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 442] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 442| do { 189| | // step 0 - is the input valid? 190| 442| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 442| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 442] | 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| 442| const auto [output_length, length_agree] = 204| 442| calculate_length(from, inputisvalid); 205| 442| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 442] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 442| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 293, False: 149] | Branch (208:28): [True: 6, False: 287] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 6| return; 211| 6| } 212| | 213| | // step 3 - run the conversion 214| 436| const auto [written, outputs_agree] = 215| 436| do_conversion(from, output_length, inputisvalid); 216| 436| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 436] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 436| return; 221| 436| } 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| 442| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 442| validation_result verify_valid_input(FromSpan src) const { 231| 442| validation_result ret{}; 232| | 233| 442| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 442| const auto implementations = get_supported_implementations(); 235| 442| std::vector results; 236| 442| results.reserve(implementations.size()); 237| | 238| 1.32k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.32k, False: 442] ------------------ 239| 1.32k| results.push_back( 240| 1.32k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.32k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.32k| const bool validation2 = 245| 1.32k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.32k| src.data(), src.size()); 247| 1.32k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.32k] ------------------ 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| 1.32k| } 258| | 259| 442| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 442| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 442] ------------------ 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| 442| } else { 273| 442| ret.implementations_agree = true; 274| 442| } 275| 442| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 442| return r.error == simdutf::SUCCESS; 277| 442| }); 278| 442| return ret; 279| 442| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 884| 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| 740| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 740| return r.error == simdutf::SUCCESS; 277| 740| }); _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 442| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 442| length_result ret{}; 344| | 345| 442| const auto implementations = get_supported_implementations(); 346| 442| std::vector results; 347| 442| results.reserve(implementations.size()); 348| | 349| 1.32k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.32k, False: 442] ------------------ 350| 1.32k| const auto len = invoke_lengthcalc(impl, src); 351| 1.32k| results.push_back(len); 352| 1.32k| ret.length.push_back(len); 353| 1.32k| } 354| | 355| 442| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 442| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 442] ------------------ 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| 442| } else { 375| 442| ret.implementations_agree = true; 376| 442| } 377| 442| return ret; 378| 442| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.32k| FromSpan src) const { 331| 1.32k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.32k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 884| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 436| const bool inputisvalid) const { 383| 436| conversion_result ret{}; 384| | 385| 436| const auto implementations = get_supported_implementations(); 386| | 387| 436| std::vector> results; 388| 436| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 436| std::vector> outputbuffers; 393| 436| outputbuffers.reserve(implementations.size()); 394| 1.74k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.30k, False: 436] ------------------ 395| 1.30k| auto impl = implementations[i]; 396| 1.30k| const ToType canary1{42}; 397| 1.30k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.30k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.30k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.30k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.30k| if constexpr (std::is_same_v) { 403| 1.30k| return r != 0; 404| 1.30k| } else { 405| 1.30k| return r.error == simdutf::error_code::SUCCESS; 406| 1.30k| } 407| 1.30k| }(implret1); 408| 1.30k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.30k| 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| 1.30k| const ToType canary2{25}; 414| 1.30k| const auto outputbuffer_first_run = outputbuffer; 415| 1.30k| std::ranges::fill(outputbuffer, canary2); 416| 1.30k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.30k| src.size(), outputbuffer.data()); 418| | 419| 1.30k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.30k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.30k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 447, False: 861] | Branch (423:29): [True: 438, False: 9] ------------------ 424| | // only care about the output if the input is valid 425| 438| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 438| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 438] ------------------ 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| 438| } 440| 1.30k| } 441| 1.30k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 438, False: 870] ------------------ 442| 1.30k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 436| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 287, False: 149] ------------------ 447| 861| for (auto& e : results) { ------------------ | Branch (447:20): [True: 861, False: 287] ------------------ 448| 861| e.outputhash.clear(); 449| 861| } 450| 287| } 451| | 452| 436| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 436| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 436] ------------------ 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| 436| } else { 474| 436| ret.implementations_agree = true; 475| 436| } 476| 436| return ret; 477| 436| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.30k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.30k| if constexpr (std::is_same_v) { 403| 1.30k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.30k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 872| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_7clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 619| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 619| const FromSpan from{reinterpret_cast(chardata.data()), 178| 619| chardata.size() / sizeof(FromType)}; 179| | 180| 619| static const bool do_print_testcase = 181| 619| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 619| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 619] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 619| do { 189| | // step 0 - is the input valid? 190| 619| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 619| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 619] | 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| 619| From == UtfEncodings::UTF8) { 198| 619| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 619] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 619| } 201| | 202| | // step 2 - what is the required size of the output? 203| 619| const auto [output_length, length_agree] = 204| 619| calculate_length(from, inputisvalid); 205| 619| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 619] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 619| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 287, False: 332] | Branch (208:28): [True: 10, False: 277] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 10| return; 211| 10| } 212| | 213| | // step 3 - run the conversion 214| 609| const auto [written, outputs_agree] = 215| 609| do_conversion(from, output_length, inputisvalid); 216| 609| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 609] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 609| return; 221| 609| } 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| 619| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 619| validation_result verify_valid_input(FromSpan src) const { 231| 619| validation_result ret{}; 232| | 233| 619| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 619| const auto implementations = get_supported_implementations(); 235| 619| std::vector results; 236| 619| results.reserve(implementations.size()); 237| | 238| 1.85k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.85k, False: 619] ------------------ 239| 1.85k| results.push_back( 240| 1.85k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.85k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.85k| const bool validation2 = 245| 1.85k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.85k| src.data(), src.size()); 247| 1.85k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.85k] ------------------ 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| 1.85k| } 258| | 259| 619| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 619| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 619] ------------------ 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| 619| } else { 273| 619| ret.implementations_agree = true; 274| 619| } 275| 619| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 619| return r.error == simdutf::SUCCESS; 277| 619| }); 278| 619| return ret; 279| 619| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 1.23k| 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| 1.28k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.28k| return r.error == simdutf::SUCCESS; 277| 1.28k| }); _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 619| bool count_the_input(FromSpan src) const { 290| 619| const auto implementations = get_supported_implementations(); 291| 619| std::vector results; 292| 619| results.reserve(implementations.size()); 293| | 294| 1.85k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.85k, False: 619] ------------------ 295| 1.85k| 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| 1.85k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.85k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.85k| } 303| 1.85k| results.push_back(ret); 304| 1.85k| } 305| 619| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 619| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 619] ------------------ 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| 619| return true; 321| 619| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 1.23k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 619| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 619| length_result ret{}; 344| | 345| 619| const auto implementations = get_supported_implementations(); 346| 619| std::vector results; 347| 619| results.reserve(implementations.size()); 348| | 349| 1.85k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.85k, False: 619] ------------------ 350| 1.85k| const auto len = invoke_lengthcalc(impl, src); 351| 1.85k| results.push_back(len); 352| 1.85k| ret.length.push_back(len); 353| 1.85k| } 354| | 355| 619| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 619| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 619] ------------------ 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| 619| } else { 375| 619| ret.implementations_agree = true; 376| 619| } 377| 619| return ret; 378| 619| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.85k| FromSpan src) const { 331| 1.85k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.85k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 1.23k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 609| const bool inputisvalid) const { 383| 609| conversion_result ret{}; 384| | 385| 609| const auto implementations = get_supported_implementations(); 386| | 387| 609| std::vector> results; 388| 609| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 609| std::vector> outputbuffers; 393| 609| outputbuffers.reserve(implementations.size()); 394| 2.43k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.82k, False: 609] ------------------ 395| 1.82k| auto impl = implementations[i]; 396| 1.82k| const ToType canary1{42}; 397| 1.82k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.82k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.82k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.82k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.82k| if constexpr (std::is_same_v) { 403| 1.82k| return r != 0; 404| 1.82k| } else { 405| 1.82k| return r.error == simdutf::error_code::SUCCESS; 406| 1.82k| } 407| 1.82k| }(implret1); 408| 1.82k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.82k| 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| 1.82k| const ToType canary2{25}; 414| 1.82k| const auto outputbuffer_first_run = outputbuffer; 415| 1.82k| std::ranges::fill(outputbuffer, canary2); 416| 1.82k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.82k| src.size(), outputbuffer.data()); 418| | 419| 1.82k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.82k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.82k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 996, False: 831] | Branch (423:29): [True: 990, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 990| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 990| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 990] ------------------ 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| 990| } 440| 1.82k| } 441| 1.82k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 990, False: 837] ------------------ 442| 1.82k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 609| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 277, False: 332] ------------------ 447| 831| for (auto& e : results) { ------------------ | Branch (447:20): [True: 831, False: 277] ------------------ 448| 831| e.outputhash.clear(); 449| 831| } 450| 277| } 451| | 452| 609| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 609| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 609] ------------------ 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| 609| } else { 474| 609| ret.implementations_agree = true; 475| 609| } 476| 609| return ret; 477| 609| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.82k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.82k| if constexpr (std::is_same_v) { 403| 1.82k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.82k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 1.21k| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_8clENSt3__14spanIKcLm18446744073709551615EEE: 554| 231| +[](std::span chardata) { \ 555| 231| const auto c = \ 556| 231| Conversion{ \ 559| 231| &I::lenfunc, &I::conversionfunc, \ 560| 231| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 231|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 231| ::std::void_t(); \ | | 1180| 231| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 231| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 231| constexpr auto _size = _name.size(); \ | | 1183| 231| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 231| return _nameof; }() ------------------ 561| 231| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 231|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 231| ::std::void_t(); \ | | 1180| 231| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 231| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 231| constexpr auto _size = _name.size(); \ | | 1183| 231| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 231| return _nameof; }() ------------------ 562| 231| c.fuzz(chardata); \ 563| 231| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 623| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 623| const FromSpan from{reinterpret_cast(chardata.data()), 178| 623| chardata.size() / sizeof(FromType)}; 179| | 180| 623| static const bool do_print_testcase = 181| 623| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 623| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 623] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 623| do { 189| | // step 0 - is the input valid? 190| 623| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 623| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 623] | 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| 623| From == UtfEncodings::UTF8) { 198| 623| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 623] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 623| } 201| | 202| | // step 2 - what is the required size of the output? 203| 623| const auto [output_length, length_agree] = 204| 623| calculate_length(from, inputisvalid); 205| 623| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 623] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 623| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 283, False: 340] | Branch (208:28): [True: 15, False: 268] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 15| return; 211| 15| } 212| | 213| | // step 3 - run the conversion 214| 608| const auto [written, outputs_agree] = 215| 608| do_conversion(from, output_length, inputisvalid); 216| 608| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 608] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 608| return; 221| 608| } 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| 623| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 623| validation_result verify_valid_input(FromSpan src) const { 231| 623| validation_result ret{}; 232| | 233| 623| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 623| const auto implementations = get_supported_implementations(); 235| 623| std::vector results; 236| 623| results.reserve(implementations.size()); 237| | 238| 1.86k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.86k, False: 623] ------------------ 239| 1.86k| results.push_back( 240| 1.86k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.86k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.86k| const bool validation2 = 245| 1.86k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.86k| src.data(), src.size()); 247| 1.86k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.86k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.86k| } 258| | 259| 623| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 623| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 623] ------------------ 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| 623| } else { 273| 623| ret.implementations_agree = true; 274| 623| } 275| 623| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 623| return r.error == simdutf::SUCCESS; 277| 623| }); 278| 623| return ret; 279| 623| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 1.24k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 1.30k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.30k| return r.error == simdutf::SUCCESS; 277| 1.30k| }); _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 623| bool count_the_input(FromSpan src) const { 290| 623| const auto implementations = get_supported_implementations(); 291| 623| std::vector results; 292| 623| results.reserve(implementations.size()); 293| | 294| 1.86k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.86k, False: 623] ------------------ 295| 1.86k| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| 1.86k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.86k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.86k| } 303| 1.86k| results.push_back(ret); 304| 1.86k| } 305| 623| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 623| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 623] ------------------ 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| 623| return true; 321| 623| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 1.24k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 623| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 623| length_result ret{}; 344| | 345| 623| const auto implementations = get_supported_implementations(); 346| 623| std::vector results; 347| 623| results.reserve(implementations.size()); 348| | 349| 1.86k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.86k, False: 623] ------------------ 350| 1.86k| const auto len = invoke_lengthcalc(impl, src); 351| 1.86k| results.push_back(len); 352| 1.86k| ret.length.push_back(len); 353| 1.86k| } 354| | 355| 623| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 623| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 623] ------------------ 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| 623| } else { 375| 623| ret.implementations_agree = true; 376| 623| } 377| 623| return ret; 378| 623| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.86k| FromSpan src) const { 331| 1.86k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.86k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 1.24k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 608| const bool inputisvalid) const { 383| 608| conversion_result ret{}; 384| | 385| 608| const auto implementations = get_supported_implementations(); 386| | 387| 608| std::vector> results; 388| 608| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 608| std::vector> outputbuffers; 393| 608| outputbuffers.reserve(implementations.size()); 394| 2.43k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.82k, False: 608] ------------------ 395| 1.82k| auto impl = implementations[i]; 396| 1.82k| const ToType canary1{42}; 397| 1.82k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.82k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.82k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.82k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.82k| if constexpr (std::is_same_v) { 403| 1.82k| return r != 0; 404| 1.82k| } else { 405| 1.82k| return r.error == simdutf::error_code::SUCCESS; 406| 1.82k| } 407| 1.82k| }(implret1); 408| 1.82k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.82k| 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| 1.82k| const ToType canary2{25}; 414| 1.82k| const auto outputbuffer_first_run = outputbuffer; 415| 1.82k| std::ranges::fill(outputbuffer, canary2); 416| 1.82k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.82k| src.size(), outputbuffer.data()); 418| | 419| 1.82k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.82k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.82k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 1.02k, False: 804] | Branch (423:29): [True: 1.01k, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 1.01k| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 1.01k| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 1.01k] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 1.01k| } 440| 1.82k| } 441| 1.82k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 1.01k, False: 810] ------------------ 442| 1.82k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 608| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 268, False: 340] ------------------ 447| 804| for (auto& e : results) { ------------------ | Branch (447:20): [True: 804, False: 268] ------------------ 448| 804| e.outputhash.clear(); 449| 804| } 450| 268| } 451| | 452| 608| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 608| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 608] ------------------ 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| 608| } else { 474| 608| ret.implementations_agree = true; 475| 608| } 476| 608| return ret; 477| 608| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.82k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.82k| if constexpr (std::is_same_v) { 403| 1.82k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.82k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 1.21k| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK3$_9clENSt3__14spanIKcLm18446744073709551615EEE: 554| 226| +[](std::span chardata) { \ 555| 226| const auto c = \ 556| 226| Conversion{ \ 559| 226| &I::lenfunc, &I::conversionfunc, \ 560| 226| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 226|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 226| ::std::void_t(); \ | | 1180| 226| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 226| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 226| constexpr auto _size = _name.size(); \ | | 1183| 226| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 226| return _nameof; }() ------------------ 561| 226| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 226|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 226| ::std::void_t(); \ | | 1180| 226| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 226| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 226| constexpr auto _size = _name.size(); \ | | 1183| 226| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 226| return _nameof; }() ------------------ 562| 226| c.fuzz(chardata); \ 563| 226| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 635| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 635| const FromSpan from{reinterpret_cast(chardata.data()), 178| 635| chardata.size() / sizeof(FromType)}; 179| | 180| 635| static const bool do_print_testcase = 181| 635| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 635| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 635] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 635| do { 189| | // step 0 - is the input valid? 190| 635| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 635| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 635] | 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| 635| From == UtfEncodings::UTF8) { 198| 635| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 635] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 635| } 201| | 202| | // step 2 - what is the required size of the output? 203| 635| const auto [output_length, length_agree] = 204| 635| calculate_length(from, inputisvalid); 205| 635| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 635] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 635| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 279, False: 356] | Branch (208:28): [True: 10, False: 269] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 10| return; 211| 10| } 212| | 213| | // step 3 - run the conversion 214| 625| const auto [written, outputs_agree] = 215| 625| do_conversion(from, output_length, inputisvalid); 216| 625| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 625] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 625| return; 221| 625| } 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| 635| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 635| validation_result verify_valid_input(FromSpan src) const { 231| 635| validation_result ret{}; 232| | 233| 635| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 635| const auto implementations = get_supported_implementations(); 235| 635| std::vector results; 236| 635| results.reserve(implementations.size()); 237| | 238| 1.90k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.90k, False: 635] ------------------ 239| 1.90k| results.push_back( 240| 1.90k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.90k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.90k| const bool validation2 = 245| 1.90k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.90k| src.data(), src.size()); 247| 1.90k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.90k] ------------------ 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| 1.90k| } 258| | 259| 635| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 635| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 635] ------------------ 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| 635| } else { 273| 635| ret.implementations_agree = true; 274| 635| } 275| 635| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 635| return r.error == simdutf::SUCCESS; 277| 635| }); 278| 635| return ret; 279| 635| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 1.27k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKNS1_6resultEE_clESI_: 275| 1.34k| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 1.34k| return r.error == simdutf::SUCCESS; 277| 1.34k| }); _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 635| bool count_the_input(FromSpan src) const { 290| 635| const auto implementations = get_supported_implementations(); 291| 635| std::vector results; 292| 635| results.reserve(implementations.size()); 293| | 294| 1.90k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.90k, False: 635] ------------------ 295| 1.90k| 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| 1.90k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.90k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.90k| } 303| 1.90k| results.push_back(ret); 304| 1.90k| } 305| 635| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 635| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 635] ------------------ 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| 635| return true; 321| 635| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 1.27k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 635| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 635| length_result ret{}; 344| | 345| 635| const auto implementations = get_supported_implementations(); 346| 635| std::vector results; 347| 635| results.reserve(implementations.size()); 348| | 349| 1.90k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.90k, False: 635] ------------------ 350| 1.90k| const auto len = invoke_lengthcalc(impl, src); 351| 1.90k| results.push_back(len); 352| 1.90k| ret.length.push_back(len); 353| 1.90k| } 354| | 355| 635| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 635| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 635] ------------------ 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| 635| } else { 375| 635| ret.implementations_agree = true; 376| 635| } 377| 635| return ret; 378| 635| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.90k| FromSpan src) const { 331| 1.90k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.90k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 1.27k| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 625| const bool inputisvalid) const { 383| 625| conversion_result ret{}; 384| | 385| 625| const auto implementations = get_supported_implementations(); 386| | 387| 625| std::vector> results; 388| 625| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 625| std::vector> outputbuffers; 393| 625| outputbuffers.reserve(implementations.size()); 394| 2.50k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.87k, False: 625] ------------------ 395| 1.87k| auto impl = implementations[i]; 396| 1.87k| const ToType canary1{42}; 397| 1.87k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.87k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.87k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.87k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.87k| if constexpr (std::is_same_v) { 403| 1.87k| return r != 0; 404| 1.87k| } else { 405| 1.87k| return r.error == simdutf::error_code::SUCCESS; 406| 1.87k| } 407| 1.87k| }(implret1); 408| 1.87k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.87k| 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| 1.87k| const ToType canary2{25}; 414| 1.87k| const auto outputbuffer_first_run = outputbuffer; 415| 1.87k| std::ranges::fill(outputbuffer, canary2); 416| 1.87k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.87k| src.size(), outputbuffer.data()); 418| | 419| 1.87k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.87k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.87k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 1.06k, False: 807] | Branch (423:29): [True: 1.06k, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 1.06k| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 1.06k| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 1.06k] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 1.06k| } 440| 1.87k| } 441| 1.87k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 1.06k, False: 813] ------------------ 442| 1.87k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 625| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 269, False: 356] ------------------ 447| 807| for (auto& e : results) { ------------------ | Branch (447:20): [True: 807, False: 269] ------------------ 448| 807| e.outputhash.clear(); 449| 807| } 450| 269| } 451| | 452| 625| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 625| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 625] ------------------ 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| 625| } else { 474| 625| ret.implementations_agree = true; 475| 625| } 476| 625| return ret; 477| 625| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 1.87k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.87k| if constexpr (std::is_same_v) { 403| 1.87k| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 1.87k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 1.25k| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_10clENSt3__14spanIKcLm18446744073709551615EEE: 554| 56| +[](std::span chardata) { \ 555| 56| const auto c = \ 556| 56| Conversion{ \ 559| 56| &I::lenfunc, &I::conversionfunc, \ 560| 56| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 56|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 56| ::std::void_t(); \ | | 1180| 56| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 56| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 56| constexpr auto _size = _name.size(); \ | | 1183| 56| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 56| return _nameof; }() ------------------ 561| 56| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 56|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 56| ::std::void_t(); \ | | 1180| 56| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 56| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 56| constexpr auto _size = _name.size(); \ | | 1183| 56| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 56| return _nameof; }() ------------------ 562| 56| c.fuzz(chardata); \ 563| 56| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 56| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 56| const FromSpan from{reinterpret_cast(chardata.data()), 178| 56| chardata.size() / sizeof(FromType)}; 179| | 180| 56| static const bool do_print_testcase = 181| 56| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 56| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 56] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 56| do { 189| | // step 0 - is the input valid? 190| 56| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 56| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 56] | 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| 56| From == UtfEncodings::UTF8) { 198| 56| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 56] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 56| } 201| | 202| | // step 2 - what is the required size of the output? 203| 56| const auto [output_length, length_agree] = 204| 56| calculate_length(from, inputisvalid); 205| 56| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 56] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 56| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 12, False: 44] | Branch (208:28): [True: 0, False: 12] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 56| const auto [written, outputs_agree] = 215| 56| do_conversion(from, output_length, inputisvalid); 216| 56| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 56] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 56| return; 221| 56| } 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| 56| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 230| 56| validation_result verify_valid_input(FromSpan src) const { 231| 56| validation_result ret{}; 232| | 233| 56| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 56| const auto implementations = get_supported_implementations(); 235| 56| std::vector results; 236| 56| results.reserve(implementations.size()); 237| | 238| 168| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 168, False: 56] ------------------ 239| 168| results.push_back( 240| 168| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 168| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 168| const bool validation2 = 245| 168| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 168| src.data(), src.size()); 247| 168| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 168] ------------------ 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| 168| } 258| | 259| 56| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 56| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 56] ------------------ 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| 56| } else { 273| 56| ret.implementations_agree = true; 274| 56| } 275| 56| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 56| return r.error == simdutf::SUCCESS; 277| 56| }); 278| 56| return ret; 279| 56| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 112| 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| 144| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 144| return r.error == simdutf::SUCCESS; 277| 144| }); _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEE: 289| 56| bool count_the_input(FromSpan src) const { 290| 56| const auto implementations = get_supported_implementations(); 291| 56| std::vector results; 292| 56| results.reserve(implementations.size()); 293| | 294| 168| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 168, False: 56] ------------------ 295| 168| std::size_t ret; 296| 168| if constexpr (From == UtfEncodings::UTF16BE) { 297| 168| 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| 168| results.push_back(ret); 304| 168| } 305| 56| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 56| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 56] ------------------ 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| 56| return true; 321| 56| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 112| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 56| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 56| length_result ret{}; 344| | 345| 56| const auto implementations = get_supported_implementations(); 346| 56| std::vector results; 347| 56| results.reserve(implementations.size()); 348| | 349| 168| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 168, False: 56] ------------------ 350| 168| const auto len = invoke_lengthcalc(impl, src); 351| 168| results.push_back(len); 352| 168| ret.length.push_back(len); 353| 168| } 354| | 355| 56| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 56| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 56] ------------------ 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| 56| } else { 375| 56| ret.implementations_agree = true; 376| 56| } 377| 56| return ret; 378| 56| } _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 168| FromSpan src) const { 339| 168| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 168| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 112| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 56| const bool inputisvalid) const { 383| 56| conversion_result ret{}; 384| | 385| 56| const auto implementations = get_supported_implementations(); 386| | 387| 56| std::vector> results; 388| 56| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 56| std::vector> outputbuffers; 393| 56| outputbuffers.reserve(implementations.size()); 394| 224| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 168, False: 56] ------------------ 395| 168| auto impl = implementations[i]; 396| 168| const ToType canary1{42}; 397| 168| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 168| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 168| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 168| const auto success = [](const ConversionResult& r) -> bool { 402| 168| if constexpr (std::is_same_v) { 403| 168| return r != 0; 404| 168| } else { 405| 168| return r.error == simdutf::error_code::SUCCESS; 406| 168| } 407| 168| }(implret1); 408| 168| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 168| 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| 168| const ToType canary2{25}; 414| 168| const auto outputbuffer_first_run = outputbuffer; 415| 168| std::ranges::fill(outputbuffer, canary2); 416| 168| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 168| src.size(), outputbuffer.data()); 418| | 419| 168| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 168] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 168| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 132, False: 36] | Branch (423:29): [True: 63, False: 69] ------------------ 424| | // only care about the output if the input is valid 425| 63| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 63| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 63] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 63| } 440| 168| } 441| 168| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 63, False: 105] ------------------ 442| 168| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 56| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 12, False: 44] ------------------ 447| 36| for (auto& e : results) { ------------------ | Branch (447:20): [True: 36, False: 12] ------------------ 448| 36| e.outputhash.clear(); 449| 36| } 450| 12| } 451| | 452| 56| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 56| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 56] ------------------ 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| 56| } else { 474| 56| ret.implementations_agree = true; 475| 56| } 476| 56| return ret; 477| 56| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 168| const auto success = [](const ConversionResult& r) -> bool { 402| 168| if constexpr (std::is_same_v) { 403| 168| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 168| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 112| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_11clENSt3__14spanIKcLm18446744073709551615EEE: 554| 141| +[](std::span chardata) { \ 555| 141| const auto c = \ 556| 141| Conversion{ \ 559| 141| &I::lenfunc, &I::conversionfunc, \ 560| 141| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 141|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 141| ::std::void_t(); \ | | 1180| 141| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 141| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 141| constexpr auto _size = _name.size(); \ | | 1183| 141| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 141| return _nameof; }() ------------------ 561| 141| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 141|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 141| ::std::void_t(); \ | | 1180| 141| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 141| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 141| constexpr auto _size = _name.size(); \ | | 1183| 141| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 141| return _nameof; }() ------------------ 562| 141| c.fuzz(chardata); \ 563| 141| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_12clENSt3__14spanIKcLm18446744073709551615EEE: 554| 179| +[](std::span chardata) { \ 555| 179| const auto c = \ 556| 179| Conversion{ \ 559| 179| &I::lenfunc, &I::conversionfunc, \ 560| 179| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 179|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 179| ::std::void_t(); \ | | 1180| 179| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 179| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 179| constexpr auto _size = _name.size(); \ | | 1183| 179| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 179| return _nameof; }() ------------------ 561| 179| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 179|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 179| ::std::void_t(); \ | | 1180| 179| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 179| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 179| constexpr auto _size = _name.size(); \ | | 1183| 179| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 179| return _nameof; }() ------------------ 562| 179| c.fuzz(chardata); \ 563| 179| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_13clENSt3__14spanIKcLm18446744073709551615EEE: 554| 75| +[](std::span chardata) { \ 555| 75| const auto c = \ 556| 75| Conversion{ \ 559| 75| &I::lenfunc, &I::conversionfunc, \ 560| 75| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 75|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 75| ::std::void_t(); \ | | 1180| 75| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 75| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 75| constexpr auto _size = _name.size(); \ | | 1183| 75| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 75| return _nameof; }() ------------------ 561| 75| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 75|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 75| ::std::void_t(); \ | | 1180| 75| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 75| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 75| constexpr auto _size = _name.size(); \ | | 1183| 75| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 75| return _nameof; }() ------------------ 562| 75| c.fuzz(chardata); \ 563| 75| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 75| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 75| const FromSpan from{reinterpret_cast(chardata.data()), 178| 75| chardata.size() / sizeof(FromType)}; 179| | 180| 75| static const bool do_print_testcase = 181| 75| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 75| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 75] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 75| do { 189| | // step 0 - is the input valid? 190| 75| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 75| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 75] | 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| 75| From == UtfEncodings::UTF8) { 198| 75| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 75] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 75| } 201| | 202| | // step 2 - what is the required size of the output? 203| 75| const auto [output_length, length_agree] = 204| 75| calculate_length(from, inputisvalid); 205| 75| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 75] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 75| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 23, False: 52] | Branch (208:28): [True: 0, False: 23] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 75| const auto [written, outputs_agree] = 215| 75| do_conversion(from, output_length, inputisvalid); 216| 75| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 75] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 75| return; 221| 75| } 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| 75| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 230| 75| validation_result verify_valid_input(FromSpan src) const { 231| 75| validation_result ret{}; 232| | 233| 75| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 75| const auto implementations = get_supported_implementations(); 235| 75| std::vector results; 236| 75| results.reserve(implementations.size()); 237| | 238| 225| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 225, False: 75] ------------------ 239| 225| results.push_back( 240| 225| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 225| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 225| const bool validation2 = 245| 225| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 225| src.data(), src.size()); 247| 225| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 225] ------------------ 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| 225| } 258| | 259| 75| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 75| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 75] ------------------ 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| 75| } else { 273| 75| ret.implementations_agree = true; 274| 75| } 275| 75| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 75| return r.error == simdutf::SUCCESS; 277| 75| }); 278| 75| return ret; 279| 75| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 150| 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| 179| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 179| return r.error == simdutf::SUCCESS; 277| 179| }); _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEE: 289| 75| bool count_the_input(FromSpan src) const { 290| 75| const auto implementations = get_supported_implementations(); 291| 75| std::vector results; 292| 75| results.reserve(implementations.size()); 293| | 294| 225| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 225, False: 75] ------------------ 295| 225| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 225| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 225| 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| 225| results.push_back(ret); 304| 225| } 305| 75| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 75| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 75] ------------------ 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| 75| return true; 321| 75| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE15count_the_inputENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 150| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 75| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 75| length_result ret{}; 344| | 345| 75| const auto implementations = get_supported_implementations(); 346| 75| std::vector results; 347| 75| results.reserve(implementations.size()); 348| | 349| 225| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 225, False: 75] ------------------ 350| 225| const auto len = invoke_lengthcalc(impl, src); 351| 225| results.push_back(len); 352| 225| ret.length.push_back(len); 353| 225| } 354| | 355| 75| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 75| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 75] ------------------ 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| 75| } else { 375| 75| ret.implementations_agree = true; 376| 75| } 377| 75| return ret; 378| 75| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 225| FromSpan src) const { 339| 225| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 225| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 150| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 75| const bool inputisvalid) const { 383| 75| conversion_result ret{}; 384| | 385| 75| const auto implementations = get_supported_implementations(); 386| | 387| 75| std::vector> results; 388| 75| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 75| std::vector> outputbuffers; 393| 75| outputbuffers.reserve(implementations.size()); 394| 300| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 225, False: 75] ------------------ 395| 225| auto impl = implementations[i]; 396| 225| const ToType canary1{42}; 397| 225| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 225| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 225| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 225| const auto success = [](const ConversionResult& r) -> bool { 402| 225| if constexpr (std::is_same_v) { 403| 225| return r != 0; 404| 225| } else { 405| 225| return r.error == simdutf::error_code::SUCCESS; 406| 225| } 407| 225| }(implret1); 408| 225| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 225| 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| 225| const ToType canary2{25}; 414| 225| const auto outputbuffer_first_run = outputbuffer; 415| 225| std::ranges::fill(outputbuffer, canary2); 416| 225| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 225| src.size(), outputbuffer.data()); 418| | 419| 225| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 225] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 225| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 156, False: 69] | Branch (423:29): [True: 69, False: 87] ------------------ 424| | // only care about the output if the input is valid 425| 69| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 69| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 69] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 69| } 440| 225| } 441| 225| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 69, False: 156] ------------------ 442| 225| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 75| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 23, False: 52] ------------------ 447| 69| for (auto& e : results) { ------------------ | Branch (447:20): [True: 69, False: 23] ------------------ 448| 69| e.outputhash.clear(); 449| 69| } 450| 23| } 451| | 452| 75| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 75| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 75] ------------------ 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| 75| } else { 474| 75| ret.implementations_agree = true; 475| 75| } 476| 75| return ret; 477| 75| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 225| const auto success = [](const ConversionResult& r) -> bool { 402| 225| if constexpr (std::is_same_v) { 403| 225| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 225| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDsmPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 150| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_14clENSt3__14spanIKcLm18446744073709551615EEE: 554| 154| +[](std::span chardata) { \ 555| 154| const auto c = \ 556| 154| Conversion{ \ 559| 154| &I::lenfunc, &I::conversionfunc, \ 560| 154| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 154|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 154| ::std::void_t(); \ | | 1180| 154| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 154| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 154| constexpr auto _size = _name.size(); \ | | 1183| 154| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 154| return _nameof; }() ------------------ 561| 154| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 154|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 154| ::std::void_t(); \ | | 1180| 154| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 154| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 154| constexpr auto _size = _name.size(); \ | | 1183| 154| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 154| return _nameof; }() ------------------ 562| 154| c.fuzz(chardata); \ 563| 154| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_15clENSt3__14spanIKcLm18446744073709551615EEE: 554| 246| +[](std::span chardata) { \ 555| 246| const auto c = \ 556| 246| Conversion{ \ 559| 246| &I::lenfunc, &I::conversionfunc, \ 560| 246| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 246|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 246| ::std::void_t(); \ | | 1180| 246| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 246| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 246| constexpr auto _size = _name.size(); \ | | 1183| 246| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 246| return _nameof; }() ------------------ 561| 246| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 246|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 246| ::std::void_t(); \ | | 1180| 246| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 246| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 246| constexpr auto _size = _name.size(); \ | | 1183| 246| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 246| return _nameof; }() ------------------ 562| 246| c.fuzz(chardata); \ 563| 246| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_16clENSt3__14spanIKcLm18446744073709551615EEE: 554| 103| +[](std::span chardata) { \ 555| 103| const auto c = \ 556| 103| Conversion{ \ 559| 103| &I::lenfunc, &I::conversionfunc, \ 560| 103| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 103|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 103| ::std::void_t(); \ | | 1180| 103| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 103| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 103| constexpr auto _size = _name.size(); \ | | 1183| 103| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 103| return _nameof; }() ------------------ 561| 103| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 103|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 103| ::std::void_t(); \ | | 1180| 103| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 103| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 103| constexpr auto _size = _name.size(); \ | | 1183| 103| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 103| return _nameof; }() ------------------ 562| 103| c.fuzz(chardata); \ 563| 103| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 103| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 103| const FromSpan from{reinterpret_cast(chardata.data()), 178| 103| chardata.size() / sizeof(FromType)}; 179| | 180| 103| static const bool do_print_testcase = 181| 103| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 103| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 103] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 103| do { 189| | // step 0 - is the input valid? 190| 103| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 103| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 103] | 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| 103| const auto [output_length, length_agree] = 204| 103| calculate_length(from, inputisvalid); 205| 103| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 103] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 103| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 72, False: 31] | Branch (208:28): [True: 0, False: 72] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 103| const auto [written, outputs_agree] = 215| 103| do_conversion(from, output_length, inputisvalid); 216| 103| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 103] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 103| return; 221| 103| } 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| 103| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 230| 103| validation_result verify_valid_input(FromSpan src) const { 231| 103| validation_result ret{}; 232| | 233| 103| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 103| const auto implementations = get_supported_implementations(); 235| 103| std::vector results; 236| 103| results.reserve(implementations.size()); 237| | 238| 309| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 309, False: 103] ------------------ 239| 309| results.push_back( 240| 309| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 309| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 309| const bool validation2 = 245| 309| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 309| src.data(), src.size()); 247| 309| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 309] ------------------ 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| 309| } 258| | 259| 103| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 103| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 103] ------------------ 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| 103| } else { 273| 103| ret.implementations_agree = true; 274| 103| } 275| 103| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 103| return r.error == simdutf::SUCCESS; 277| 103| }); 278| 103| return ret; 279| 103| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 206| 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| 165| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 165| return r.error == simdutf::SUCCESS; 277| 165| }); _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 103| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 103| length_result ret{}; 344| | 345| 103| const auto implementations = get_supported_implementations(); 346| 103| std::vector results; 347| 103| results.reserve(implementations.size()); 348| | 349| 309| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 309, False: 103] ------------------ 350| 309| const auto len = invoke_lengthcalc(impl, src); 351| 309| results.push_back(len); 352| 309| ret.length.push_back(len); 353| 309| } 354| | 355| 103| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 103| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 103] ------------------ 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| 103| } else { 375| 103| ret.implementations_agree = true; 376| 103| } 377| 103| return ret; 378| 103| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 309| FromSpan src) const { 339| 309| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 309| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 206| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 103| const bool inputisvalid) const { 383| 103| conversion_result ret{}; 384| | 385| 103| const auto implementations = get_supported_implementations(); 386| | 387| 103| std::vector> results; 388| 103| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 103| std::vector> outputbuffers; 393| 103| outputbuffers.reserve(implementations.size()); 394| 412| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 309, False: 103] ------------------ 395| 309| auto impl = implementations[i]; 396| 309| const ToType canary1{42}; 397| 309| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 309| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 309| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 309| const auto success = [](const ConversionResult& r) -> bool { 402| 309| if constexpr (std::is_same_v) { 403| 309| return r != 0; 404| 309| } else { 405| 309| return r.error == simdutf::error_code::SUCCESS; 406| 309| } 407| 309| }(implret1); 408| 309| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 309| 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| 309| const ToType canary2{25}; 414| 309| const auto outputbuffer_first_run = outputbuffer; 415| 309| std::ranges::fill(outputbuffer, canary2); 416| 309| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 309| src.size(), outputbuffer.data()); 418| | 419| 309| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 309] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 309| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 93, False: 216] | Branch (423:29): [True: 72, False: 21] ------------------ 424| | // only care about the output if the input is valid 425| 72| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 72| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 72] ------------------ 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| 72| } 440| 309| } 441| 309| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 72, False: 237] ------------------ 442| 309| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 103| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 72, False: 31] ------------------ 447| 216| for (auto& e : results) { ------------------ | Branch (447:20): [True: 216, False: 72] ------------------ 448| 216| e.outputhash.clear(); 449| 216| } 450| 72| } 451| | 452| 103| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 103| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 103] ------------------ 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| 103| } else { 474| 103| ret.implementations_agree = true; 475| 103| } 476| 103| return ret; 477| 103| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 309| const auto success = [](const ConversionResult& r) -> bool { 402| 309| if constexpr (std::is_same_v) { 403| 309| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 309| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKDimPcEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 206| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_17clENSt3__14spanIKcLm18446744073709551615EEE: 554| 213| +[](std::span chardata) { \ 555| 213| const auto c = \ 556| 213| Conversion{ \ 559| 213| &I::lenfunc, &I::conversionfunc, \ 560| 213| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 213|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 213| ::std::void_t(); \ | | 1180| 213| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 213| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 213| constexpr auto _size = _name.size(); \ | | 1183| 213| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 213| return _nameof; }() ------------------ 561| 213| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 213|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 213| ::std::void_t(); \ | | 1180| 213| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 213| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 213| constexpr auto _size = _name.size(); \ | | 1183| 213| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 213| return _nameof; }() ------------------ 562| 213| c.fuzz(chardata); \ 563| 213| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_18clENSt3__14spanIKcLm18446744073709551615EEE: 554| 240| +[](std::span chardata) { \ 555| 240| const auto c = \ 556| 240| Conversion{ \ 559| 240| &I::lenfunc, &I::conversionfunc, \ 560| 240| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 240|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 240| ::std::void_t(); \ | | 1180| 240| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 240| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 240| constexpr auto _size = _name.size(); \ | | 1183| 240| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 240| return _nameof; }() ------------------ 561| 240| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 240|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 240| ::std::void_t(); \ | | 1180| 240| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 240| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 240| constexpr auto _size = _name.size(); \ | | 1183| 240| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 240| return _nameof; }() ------------------ 562| 240| c.fuzz(chardata); \ 563| 240| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_19clENSt3__14spanIKcLm18446744073709551615EEE: 554| 342| +[](std::span chardata) { \ 555| 342| const auto c = \ 556| 342| Conversion{ \ 559| 342| &I::lenfunc, &I::conversionfunc, \ 560| 342| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 342|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 342| ::std::void_t(); \ | | 1180| 342| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 342| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 342| constexpr auto _size = _name.size(); \ | | 1183| 342| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 342| return _nameof; }() ------------------ 561| 342| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 342|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 342| ::std::void_t(); \ | | 1180| 342| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 342| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 342| constexpr auto _size = _name.size(); \ | | 1183| 342| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 342| return _nameof; }() ------------------ 562| 342| c.fuzz(chardata); \ 563| 342| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_20clENSt3__14spanIKcLm18446744073709551615EEE: 554| 306| +[](std::span chardata) { \ 555| 306| const auto c = \ 556| 306| Conversion{ \ 559| 306| &I::lenfunc, &I::conversionfunc, \ 560| 306| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 306|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 306| ::std::void_t(); \ | | 1180| 306| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 306| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 306| constexpr auto _size = _name.size(); \ | | 1183| 306| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 306| return _nameof; }() ------------------ 561| 306| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 306|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 306| ::std::void_t(); \ | | 1180| 306| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 306| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 306| constexpr auto _size = _name.size(); \ | | 1183| 306| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 306| return _nameof; }() ------------------ 562| 306| c.fuzz(chardata); \ 563| 306| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 306| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 306| const FromSpan from{reinterpret_cast(chardata.data()), 178| 306| chardata.size() / sizeof(FromType)}; 179| | 180| 306| static const bool do_print_testcase = 181| 306| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 306| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 306] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 306| do { 189| | // step 0 - is the input valid? 190| 306| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 306| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 306] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 306| From == UtfEncodings::UTF8) { 198| 306| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 306] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 306| } 201| | 202| | // step 2 - what is the required size of the output? 203| 306| const auto [output_length, length_agree] = 204| 306| calculate_length(from, inputisvalid); 205| 306| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 306] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 306| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 204, False: 102] | Branch (208:28): [True: 0, False: 204] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 306| const auto [written, outputs_agree] = 215| 306| do_conversion(from, output_length, inputisvalid); 216| 306| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 306] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 306| return; 221| 306| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 306| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 306| validation_result verify_valid_input(FromSpan src) const { 231| 306| validation_result ret{}; 232| | 233| 306| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 306| const auto implementations = get_supported_implementations(); 235| 306| std::vector results; 236| 306| results.reserve(implementations.size()); 237| | 238| 918| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 918, False: 306] ------------------ 239| 918| results.push_back( 240| 918| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 918| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 918| const bool validation2 = 245| 918| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 918| src.data(), src.size()); 247| 918| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 918] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 918| } 258| | 259| 306| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 306| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 306] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 306| } else { 273| 306| ret.implementations_agree = true; 274| 306| } 275| 306| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 306| return r.error == simdutf::SUCCESS; 277| 306| }); 278| 306| return ret; 279| 306| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clINS1_6resultESO_EEDaSI_SL_: 259| 612| 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| 510| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 510| return r.error == simdutf::SUCCESS; 277| 510| }); _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 306| bool count_the_input(FromSpan src) const { 290| 306| const auto implementations = get_supported_implementations(); 291| 306| std::vector results; 292| 306| results.reserve(implementations.size()); 293| | 294| 918| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 918, False: 306] ------------------ 295| 918| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| | } else if constexpr (From == UtfEncodings::UTF16LE) { 299| | ret = impl->count_utf16le(src.data(), src.size()); 300| 918| } else if constexpr (From == UtfEncodings::UTF8) { 301| 918| ret = impl->count_utf8(src.data(), src.size()); 302| 918| } 303| 918| results.push_back(ret); 304| 918| } 305| 306| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 306| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 306] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 306| return true; 321| 306| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 305| 612| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 306| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 306| length_result ret{}; 344| | 345| 306| const auto implementations = get_supported_implementations(); 346| 306| std::vector results; 347| 306| results.reserve(implementations.size()); 348| | 349| 918| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 918, False: 306] ------------------ 350| 918| const auto len = invoke_lengthcalc(impl, src); 351| 918| results.push_back(len); 352| 918| ret.length.push_back(len); 353| 918| } 354| | 355| 306| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 306| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 306] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 306| } else { 375| 306| ret.implementations_agree = true; 376| 306| } 377| 306| return ret; 378| 306| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 918| FromSpan src) const { 331| 918| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 918| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 612| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 306| const bool inputisvalid) const { 383| 306| conversion_result ret{}; 384| | 385| 306| const auto implementations = get_supported_implementations(); 386| | 387| 306| std::vector> results; 388| 306| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 306| std::vector> outputbuffers; 393| 306| outputbuffers.reserve(implementations.size()); 394| 1.22k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 918, False: 306] ------------------ 395| 918| auto impl = implementations[i]; 396| 918| const ToType canary1{42}; 397| 918| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 918| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 918| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 918| const auto success = [](const ConversionResult& r) -> bool { 402| 918| if constexpr (std::is_same_v) { 403| 918| return r != 0; 404| 918| } else { 405| 918| return r.error == simdutf::error_code::SUCCESS; 406| 918| } 407| 918| }(implret1); 408| 918| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 918| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 918| const ToType canary2{25}; 414| 918| const auto outputbuffer_first_run = outputbuffer; 415| 918| std::ranges::fill(outputbuffer, canary2); 416| 918| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 918| src.size(), outputbuffer.data()); 418| | 419| 918| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 918] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 918| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 306, False: 612] | Branch (423:29): [True: 276, False: 30] ------------------ 424| | // only care about the output if the input is valid 425| 276| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 276| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 276] ------------------ 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| 276| } 440| 918| } 441| 918| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 276, False: 642] ------------------ 442| 918| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 306| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 204, False: 102] ------------------ 447| 612| for (auto& e : results) { ------------------ | Branch (447:20): [True: 612, False: 204] ------------------ 448| 612| e.outputhash.clear(); 449| 612| } 450| 204| } 451| | 452| 306| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 306| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 306] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 306| } else { 474| 306| ret.implementations_agree = true; 475| 306| } 476| 306| return ret; 477| 306| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 918| const auto success = [](const ConversionResult& r) -> bool { 402| 918| if constexpr (std::is_same_v) { 403| 918| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 918| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 612| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_21clENSt3__14spanIKcLm18446744073709551615EEE: 554| 379| +[](std::span chardata) { \ 555| 379| const auto c = \ 556| 379| Conversion{ \ 559| 379| &I::lenfunc, &I::conversionfunc, \ 560| 379| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 379|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 379| ::std::void_t(); \ | | 1180| 379| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 379| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 379| constexpr auto _size = _name.size(); \ | | 1183| 379| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 379| return _nameof; }() ------------------ 561| 379| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 379|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 379| ::std::void_t(); \ | | 1180| 379| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 379| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 379| constexpr auto _size = _name.size(); \ | | 1183| 379| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 379| return _nameof; }() ------------------ 562| 379| c.fuzz(chardata); \ 563| 379| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_22clENSt3__14spanIKcLm18446744073709551615EEE: 554| 392| +[](std::span chardata) { \ 555| 392| const auto c = \ 556| 392| Conversion{ \ 559| 392| &I::lenfunc, &I::conversionfunc, \ 560| 392| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 392|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 392| ::std::void_t(); \ | | 1180| 392| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 392| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 392| constexpr auto _size = _name.size(); \ | | 1183| 392| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 392| return _nameof; }() ------------------ 561| 392| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 392|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 392| ::std::void_t(); \ | | 1180| 392| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 392| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 392| constexpr auto _size = _name.size(); \ | | 1183| 392| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 392| return _nameof; }() ------------------ 562| 392| c.fuzz(chardata); \ 563| 392| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_23clENSt3__14spanIKcLm18446744073709551615EEE: 554| 409| +[](std::span chardata) { \ 555| 409| const auto c = \ 556| 409| Conversion{ \ 559| 409| &I::lenfunc, &I::conversionfunc, \ 560| 409| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 409|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 409| ::std::void_t(); \ | | 1180| 409| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 409| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 409| constexpr auto _size = _name.size(); \ | | 1183| 409| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 409| return _nameof; }() ------------------ 561| 409| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 409|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 409| ::std::void_t(); \ | | 1180| 409| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 409| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 409| constexpr auto _size = _name.size(); \ | | 1183| 409| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 409| return _nameof; }() ------------------ 562| 409| c.fuzz(chardata); \ 563| 409| } \ conversion.cpp:_ZZ18populate_functionsvENK4$_24clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_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: 20, False: 95] | Branch (208:28): [True: 0, False: 20] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 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| } _ZNK10ConversionIL12UtfEncodings0ELS0_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| } _ZZNK10ConversionIL12UtfEncodings0ELS0_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; }; _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKS5_E_clESI_: 275| 305| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 305| return r.error == simdutf::SUCCESS; 277| 305| }); _ZNK10ConversionIL12UtfEncodings0ELS0_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| 345| if constexpr (From == UtfEncodings::UTF16BE) { 297| 345| 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| 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| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 230| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_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| } _ZNK10ConversionIL12UtfEncodings0ELS0_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| } _ZZNK10ConversionIL12UtfEncodings0ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 230| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_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: 285, False: 60] | Branch (423:29): [True: 111, False: 174] ------------------ 424| | // only care about the output if the input is valid 425| 111| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 111| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 111] ------------------ 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| 111| } 440| 345| } 441| 345| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 111, False: 234] ------------------ 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: 20, False: 95] ------------------ 447| 60| for (auto& e : results) { ------------------ | Branch (447:20): [True: 60, False: 20] ------------------ 448| 60| e.outputhash.clear(); 449| 60| } 450| 20| } 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| } _ZZNK10ConversionIL12UtfEncodings0ELS0_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); _ZZNK10ConversionIL12UtfEncodings0ELS0_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$_25clENSt3__14spanIKcLm18446744073709551615EEE: 554| 211| +[](std::span chardata) { \ 555| 211| const auto c = \ 556| 211| Conversion{ \ 559| 211| &I::lenfunc, &I::conversionfunc, \ 560| 211| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 211|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 211| ::std::void_t(); \ | | 1180| 211| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 211| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 211| constexpr auto _size = _name.size(); \ | | 1183| 211| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 211| return _nameof; }() ------------------ 561| 211| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 211|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 211| ::std::void_t(); \ | | 1180| 211| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 211| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 211| constexpr auto _size = _name.size(); \ | | 1183| 211| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 211| return _nameof; }() ------------------ 562| 211| c.fuzz(chardata); \ 563| 211| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 211| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 211| const FromSpan from{reinterpret_cast(chardata.data()), 178| 211| chardata.size() / sizeof(FromType)}; 179| | 180| 211| static const bool do_print_testcase = 181| 211| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 211| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 211] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 211| do { 189| | // step 0 - is the input valid? 190| 211| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 211| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 211] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 211| From == UtfEncodings::UTF8) { 198| 211| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 211] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 211| } 201| | 202| | // step 2 - what is the required size of the output? 203| 211| const auto [output_length, length_agree] = 204| 211| calculate_length(from, inputisvalid); 205| 211| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 211] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 211| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 123, False: 88] | Branch (208:28): [True: 0, False: 123] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 211| const auto [written, outputs_agree] = 215| 211| do_conversion(from, output_length, inputisvalid); 216| 211| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 211] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 211| return; 221| 211| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 211| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 211| validation_result verify_valid_input(FromSpan src) const { 231| 211| validation_result ret{}; 232| | 233| 211| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 211| const auto implementations = get_supported_implementations(); 235| 211| std::vector results; 236| 211| results.reserve(implementations.size()); 237| | 238| 633| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 633, False: 211] ------------------ 239| 633| results.push_back( 240| 633| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 633| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 633| const bool validation2 = 245| 633| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 633| src.data(), src.size()); 247| 633| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 633] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 633| } 258| | 259| 211| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 211| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 211] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 211| } else { 273| 211| ret.implementations_agree = true; 274| 211| } 275| 211| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 211| return r.error == simdutf::SUCCESS; 277| 211| }); 278| 211| return ret; 279| 211| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 422| 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| 387| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 387| return r.error == simdutf::SUCCESS; 277| 387| }); _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 211| bool count_the_input(FromSpan src) const { 290| 211| const auto implementations = get_supported_implementations(); 291| 211| std::vector results; 292| 211| results.reserve(implementations.size()); 293| | 294| 633| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 633, False: 211] ------------------ 295| 633| std::size_t ret; 296| 633| if constexpr (From == UtfEncodings::UTF16BE) { 297| 633| 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| 633| results.push_back(ret); 304| 633| } 305| 211| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 211| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 211] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 211| return true; 321| 211| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 422| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 211| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 211| length_result ret{}; 344| | 345| 211| const auto implementations = get_supported_implementations(); 346| 211| std::vector results; 347| 211| results.reserve(implementations.size()); 348| | 349| 633| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 633, False: 211] ------------------ 350| 633| const auto len = invoke_lengthcalc(impl, src); 351| 633| results.push_back(len); 352| 633| ret.length.push_back(len); 353| 633| } 354| | 355| 211| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 211| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 211] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 211| } else { 375| 211| ret.implementations_agree = true; 376| 211| } 377| 211| return ret; 378| 211| } _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 633| FromSpan src) const { 331| 633| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 633| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 422| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 211| const bool inputisvalid) const { 383| 211| conversion_result ret{}; 384| | 385| 211| const auto implementations = get_supported_implementations(); 386| | 387| 211| std::vector> results; 388| 211| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 211| std::vector> outputbuffers; 393| 211| outputbuffers.reserve(implementations.size()); 394| 844| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 633, False: 211] ------------------ 395| 633| auto impl = implementations[i]; 396| 633| const ToType canary1{42}; 397| 633| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 633| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 633| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 633| const auto success = [](const ConversionResult& r) -> bool { 402| 633| if constexpr (std::is_same_v) { 403| 633| return r != 0; 404| 633| } else { 405| 633| return r.error == simdutf::error_code::SUCCESS; 406| 633| } 407| 633| }(implret1); 408| 633| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 633| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 633| const ToType canary2{25}; 414| 633| const auto outputbuffer_first_run = outputbuffer; 415| 633| std::ranges::fill(outputbuffer, canary2); 416| 633| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 633| src.size(), outputbuffer.data()); 418| | 419| 633| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 633] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 633| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 264, False: 369] | Branch (423:29): [True: 264, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 264| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 264| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 264] ------------------ 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| 264| } 440| 633| } 441| 633| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 264, False: 369] ------------------ 442| 633| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 211| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 123, False: 88] ------------------ 447| 369| for (auto& e : results) { ------------------ | Branch (447:20): [True: 369, False: 123] ------------------ 448| 369| e.outputhash.clear(); 449| 369| } 450| 123| } 451| | 452| 211| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 211| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 211] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 211| } else { 474| 211| ret.implementations_agree = true; 475| 211| } 476| 211| return ret; 477| 211| } _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 633| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 633| } else { 405| 633| return r.error == simdutf::error_code::SUCCESS; 406| 633| } 407| 633| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 422| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_26clENSt3__14spanIKcLm18446744073709551615EEE: 554| 324| +[](std::span chardata) { \ 555| 324| const auto c = \ 556| 324| Conversion{ \ 559| 324| &I::lenfunc, &I::conversionfunc, \ 560| 324| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 324|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 324| ::std::void_t(); \ | | 1180| 324| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 324| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 324| constexpr auto _size = _name.size(); \ | | 1183| 324| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 324| return _nameof; }() ------------------ 561| 324| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 324|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 324| ::std::void_t(); \ | | 1180| 324| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 324| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 324| constexpr auto _size = _name.size(); \ | | 1183| 324| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 324| return _nameof; }() ------------------ 562| 324| c.fuzz(chardata); \ 563| 324| } \ _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 324| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 324| const FromSpan from{reinterpret_cast(chardata.data()), 178| 324| chardata.size() / sizeof(FromType)}; 179| | 180| 324| static const bool do_print_testcase = 181| 324| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 324| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 324] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 324| do { 189| | // step 0 - is the input valid? 190| 324| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 324| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 324] | 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| 324| From == UtfEncodings::UTF8) { 198| 324| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 324] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 324| } 201| | 202| | // step 2 - what is the required size of the output? 203| 324| const auto [output_length, length_agree] = 204| 324| calculate_length(from, inputisvalid); 205| 324| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 324] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 324| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 134, False: 190] | Branch (208:28): [True: 0, False: 134] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 324| const auto [written, outputs_agree] = 215| 324| do_conversion(from, output_length, inputisvalid); 216| 324| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 324] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 324| return; 221| 324| } 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| 324| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 324| validation_result verify_valid_input(FromSpan src) const { 231| 324| validation_result ret{}; 232| | 233| 324| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 324| const auto implementations = get_supported_implementations(); 235| 324| std::vector results; 236| 324| results.reserve(implementations.size()); 237| | 238| 972| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 972, False: 324] ------------------ 239| 972| results.push_back( 240| 972| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 972| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 972| const bool validation2 = 245| 972| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 972| src.data(), src.size()); 247| 972| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 972] ------------------ 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| 972| } 258| | 259| 324| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 324| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 324] ------------------ 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| 324| } else { 273| 324| ret.implementations_agree = true; 274| 324| } 275| 324| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 324| return r.error == simdutf::SUCCESS; 277| 324| }); 278| 324| return ret; 279| 324| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 648| 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| 704| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 704| return r.error == simdutf::SUCCESS; 277| 704| }); _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 324| bool count_the_input(FromSpan src) const { 290| 324| const auto implementations = get_supported_implementations(); 291| 324| std::vector results; 292| 324| results.reserve(implementations.size()); 293| | 294| 972| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 972, False: 324] ------------------ 295| 972| std::size_t ret; 296| 972| if constexpr (From == UtfEncodings::UTF16BE) { 297| 972| 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| 972| results.push_back(ret); 304| 972| } 305| 324| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 324| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 324] ------------------ 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| 324| return true; 321| 324| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 648| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 324| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 324| length_result ret{}; 344| | 345| 324| const auto implementations = get_supported_implementations(); 346| 324| std::vector results; 347| 324| results.reserve(implementations.size()); 348| | 349| 972| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 972, False: 324] ------------------ 350| 972| const auto len = invoke_lengthcalc(impl, src); 351| 972| results.push_back(len); 352| 972| ret.length.push_back(len); 353| 972| } 354| | 355| 324| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 324| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 324] ------------------ 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| 324| } else { 375| 324| ret.implementations_agree = true; 376| 324| } 377| 324| return ret; 378| 324| } _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 972| FromSpan src) const { 331| 972| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 972| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 648| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 324| const bool inputisvalid) const { 383| 324| conversion_result ret{}; 384| | 385| 324| const auto implementations = get_supported_implementations(); 386| | 387| 324| std::vector> results; 388| 324| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 324| std::vector> outputbuffers; 393| 324| outputbuffers.reserve(implementations.size()); 394| 1.29k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 972, False: 324] ------------------ 395| 972| auto impl = implementations[i]; 396| 972| const ToType canary1{42}; 397| 972| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 972| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 972| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 972| const auto success = [](const ConversionResult& r) -> bool { 402| 972| if constexpr (std::is_same_v) { 403| 972| return r != 0; 404| 972| } else { 405| 972| return r.error == simdutf::error_code::SUCCESS; 406| 972| } 407| 972| }(implret1); 408| 972| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 972| 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| 972| const ToType canary2{25}; 414| 972| const auto outputbuffer_first_run = outputbuffer; 415| 972| std::ranges::fill(outputbuffer, canary2); 416| 972| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 972| src.size(), outputbuffer.data()); 418| | 419| 972| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 972] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 972| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 570, False: 402] | Branch (423:29): [True: 570, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 570| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 570| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 570] ------------------ 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| 570| } 440| 972| } 441| 972| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 570, False: 402] ------------------ 442| 972| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 324| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 134, False: 190] ------------------ 447| 402| for (auto& e : results) { ------------------ | Branch (447:20): [True: 402, False: 134] ------------------ 448| 402| e.outputhash.clear(); 449| 402| } 450| 134| } 451| | 452| 324| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 324| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 324] ------------------ 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| 324| } else { 474| 324| ret.implementations_agree = true; 475| 324| } 476| 324| return ret; 477| 324| } _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 972| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 972| } else { 405| 972| return r.error == simdutf::error_code::SUCCESS; 406| 972| } 407| 972| }(implret1); _ZZNK10ConversionIL12UtfEncodings0ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 648| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_27clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 113| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 113| const FromSpan from{reinterpret_cast(chardata.data()), 178| 113| chardata.size() / sizeof(FromType)}; 179| | 180| 113| static const bool do_print_testcase = 181| 113| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 113| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 113] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 113| do { 189| | // step 0 - is the input valid? 190| 113| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 113| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 113] | 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| 113| From == UtfEncodings::UTF8) { 198| 113| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 113] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 113| } 201| | 202| | // step 2 - what is the required size of the output? 203| 113| const auto [output_length, length_agree] = 204| 113| calculate_length(from, inputisvalid); 205| 113| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 113] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 113| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 19, False: 94] | Branch (208:28): [True: 0, False: 19] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 113| const auto [written, outputs_agree] = 215| 113| do_conversion(from, output_length, inputisvalid); 216| 113| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 113] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 113| return; 221| 113| } 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| 113| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEE: 230| 113| validation_result verify_valid_input(FromSpan src) const { 231| 113| validation_result ret{}; 232| | 233| 113| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 113| const auto implementations = get_supported_implementations(); 235| 113| std::vector results; 236| 113| results.reserve(implementations.size()); 237| | 238| 339| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 339, False: 113] ------------------ 239| 339| results.push_back( 240| 339| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 339| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 339| const bool validation2 = 245| 339| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 339| src.data(), src.size()); 247| 339| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 339] ------------------ 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| 339| } 258| | 259| 113| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 113| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 113] ------------------ 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| 113| } else { 273| 113| ret.implementations_agree = true; 274| 113| } 275| 113| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 113| return r.error == simdutf::SUCCESS; 277| 113| }); 278| 113| return ret; 279| 113| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS5_S5_EEDaSJ_SM_: 259| 226| 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| 301| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 301| return r.error == simdutf::SUCCESS; 277| 301| }); _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEE: 289| 113| bool count_the_input(FromSpan src) const { 290| 113| const auto implementations = get_supported_implementations(); 291| 113| std::vector results; 292| 113| results.reserve(implementations.size()); 293| | 294| 339| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 339, False: 113] ------------------ 295| 339| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 339| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 339| 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| 339| results.push_back(ret); 304| 339| } 305| 113| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 113| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 113] ------------------ 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| 113| return true; 321| 113| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE15count_the_inputENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 226| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEb: 342| 113| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 113| length_result ret{}; 344| | 345| 113| const auto implementations = get_supported_implementations(); 346| 113| std::vector results; 347| 113| results.reserve(implementations.size()); 348| | 349| 339| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 339, False: 113] ------------------ 350| 339| const auto len = invoke_lengthcalc(impl, src); 351| 339| results.push_back(len); 352| 339| ret.length.push_back(len); 353| 339| } 354| | 355| 113| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 113| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 113] ------------------ 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| 113| } else { 375| 113| ret.implementations_agree = true; 376| 113| } 377| 113| return ret; 378| 113| } _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSF_NSt3__14spanIS6_Lm18446744073709551615EEE: 338| 339| FromSpan src) const { 339| 339| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 339| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 226| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 113| const bool inputisvalid) const { 383| 113| conversion_result ret{}; 384| | 385| 113| const auto implementations = get_supported_implementations(); 386| | 387| 113| std::vector> results; 388| 113| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 113| std::vector> outputbuffers; 393| 113| outputbuffers.reserve(implementations.size()); 394| 452| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 339, False: 113] ------------------ 395| 339| auto impl = implementations[i]; 396| 339| const ToType canary1{42}; 397| 339| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 339| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 339| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 339| const auto success = [](const ConversionResult& r) -> bool { 402| 339| if constexpr (std::is_same_v) { 403| 339| return r != 0; 404| 339| } else { 405| 339| return r.error == simdutf::error_code::SUCCESS; 406| 339| } 407| 339| }(implret1); 408| 339| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 339| 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| 339| const ToType canary2{25}; 414| 339| const auto outputbuffer_first_run = outputbuffer; 415| 339| std::ranges::fill(outputbuffer, canary2); 416| 339| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 339| src.size(), outputbuffer.data()); 418| | 419| 339| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 339] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 339| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 282, False: 57] | Branch (423:29): [True: 141, False: 141] ------------------ 424| | // only care about the output if the input is valid 425| 141| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 141| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 141] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 141| } 440| 339| } 441| 339| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 141, False: 198] ------------------ 442| 339| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 113| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 19, False: 94] ------------------ 447| 57| for (auto& e : results) { ------------------ | Branch (447:20): [True: 57, False: 19] ------------------ 448| 57| e.outputhash.clear(); 449| 57| } 450| 19| } 451| | 452| 113| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 113| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 113] ------------------ 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| 113| } else { 474| 113| ret.implementations_agree = true; 475| 113| } 476| 113| return ret; 477| 113| } _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS5_E_clESM_: 401| 339| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 339| } else { 405| 339| return r.error == simdutf::error_code::SUCCESS; 406| 339| } 407| 339| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDsmPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS5_ESU_EEDaSN_SQ_: 452| 226| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_28clENSt3__14spanIKcLm18446744073709551615EEE: 554| 201| +[](std::span chardata) { \ 555| 201| const auto c = \ 556| 201| Conversion{ \ 559| 201| &I::lenfunc, &I::conversionfunc, \ 560| 201| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 201|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 201| ::std::void_t(); \ | | 1180| 201| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 201| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 201| constexpr auto _size = _name.size(); \ | | 1183| 201| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 201| return _nameof; }() ------------------ 561| 201| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 201|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 201| ::std::void_t(); \ | | 1180| 201| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 201| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 201| constexpr auto _size = _name.size(); \ | | 1183| 201| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 201| return _nameof; }() ------------------ 562| 201| c.fuzz(chardata); \ 563| 201| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_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: 101, False: 100] | Branch (208:28): [True: 0, False: 101] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 201| const auto [written, outputs_agree] = 215| 201| do_conversion(from, output_length, inputisvalid); 216| 201| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 201] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 201| return; 221| 201| } 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| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_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| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 402| 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| 401| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 401| return r.error == simdutf::SUCCESS; 277| 401| }); _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_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| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 603| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 603| 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| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 402| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_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| } _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 603| FromSpan src) const { 331| 603| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 603| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 402| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 201| const bool inputisvalid) const { 383| 201| conversion_result ret{}; 384| | 385| 201| const auto implementations = get_supported_implementations(); 386| | 387| 201| std::vector> results; 388| 201| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 201| std::vector> outputbuffers; 393| 201| outputbuffers.reserve(implementations.size()); 394| 804| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 603, False: 201] ------------------ 395| 603| auto impl = implementations[i]; 396| 603| const ToType canary1{42}; 397| 603| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 603| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 603| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 603| const auto success = [](const ConversionResult& r) -> bool { 402| 603| if constexpr (std::is_same_v) { 403| 603| return r != 0; 404| 603| } else { 405| 603| return r.error == simdutf::error_code::SUCCESS; 406| 603| } 407| 603| }(implret1); 408| 603| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 603| 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| 603| const ToType canary2{25}; 414| 603| const auto outputbuffer_first_run = outputbuffer; 415| 603| std::ranges::fill(outputbuffer, canary2); 416| 603| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 603| src.size(), outputbuffer.data()); 418| | 419| 603| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 603] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 603| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 300, False: 303] | Branch (423:29): [True: 300, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 300| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 300| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 300] ------------------ 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| 300| } 440| 603| } 441| 603| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 300, False: 303] ------------------ 442| 603| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 201| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 101, False: 100] ------------------ 447| 303| for (auto& e : results) { ------------------ | Branch (447:20): [True: 303, False: 101] ------------------ 448| 303| e.outputhash.clear(); 449| 303| } 450| 101| } 451| | 452| 201| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 201| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 201] ------------------ 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| 201| } else { 474| 201| ret.implementations_agree = true; 475| 201| } 476| 201| return ret; 477| 201| } _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 603| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 603| } else { 405| 603| return r.error == simdutf::error_code::SUCCESS; 406| 603| } 407| 603| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_3EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 402| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_29clENSt3__14spanIKcLm18446744073709551615EEE: 554| 327| +[](std::span chardata) { \ 555| 327| const auto c = \ 556| 327| Conversion{ \ 559| 327| &I::lenfunc, &I::conversionfunc, \ 560| 327| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 327|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 327| ::std::void_t(); \ | | 1180| 327| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 327| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 327| constexpr auto _size = _name.size(); \ | | 1183| 327| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 327| return _nameof; }() ------------------ 561| 327| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 327|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 327| ::std::void_t(); \ | | 1180| 327| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 327| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 327| constexpr auto _size = _name.size(); \ | | 1183| 327| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 327| return _nameof; }() ------------------ 562| 327| c.fuzz(chardata); \ 563| 327| } \ _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 327| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 327| const FromSpan from{reinterpret_cast(chardata.data()), 178| 327| chardata.size() / sizeof(FromType)}; 179| | 180| 327| static const bool do_print_testcase = 181| 327| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 327| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 327] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 327| do { 189| | // step 0 - is the input valid? 190| 327| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 327| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 327] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| 327| From == UtfEncodings::UTF8) { 198| 327| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 327] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 327| } 201| | 202| | // step 2 - what is the required size of the output? 203| 327| const auto [output_length, length_agree] = 204| 327| calculate_length(from, inputisvalid); 205| 327| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 327] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 327| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 119, False: 208] | Branch (208:28): [True: 0, False: 119] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 327| const auto [written, outputs_agree] = 215| 327| do_conversion(from, output_length, inputisvalid); 216| 327| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 327] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 327| return; 221| 327| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 327| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 327| validation_result verify_valid_input(FromSpan src) const { 231| 327| validation_result ret{}; 232| | 233| 327| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 327| const auto implementations = get_supported_implementations(); 235| 327| std::vector results; 236| 327| results.reserve(implementations.size()); 237| | 238| 981| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 981, False: 327] ------------------ 239| 981| results.push_back( 240| 981| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 981| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 981| const bool validation2 = 245| 981| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 981| src.data(), src.size()); 247| 981| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 981] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 981| } 258| | 259| 327| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 327| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 327] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 327| } else { 273| 327| ret.implementations_agree = true; 274| 327| } 275| 327| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 327| return r.error == simdutf::SUCCESS; 277| 327| }); 278| 327| return ret; 279| 327| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 654| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKS7_E_clESI_: 275| 743| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 743| return r.error == simdutf::SUCCESS; 277| 743| }); _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 327| bool count_the_input(FromSpan src) const { 290| 327| const auto implementations = get_supported_implementations(); 291| 327| std::vector results; 292| 327| results.reserve(implementations.size()); 293| | 294| 981| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 981, False: 327] ------------------ 295| 981| std::size_t ret; 296| | if constexpr (From == UtfEncodings::UTF16BE) { 297| | ret = impl->count_utf16be(src.data(), src.size()); 298| 981| } else if constexpr (From == UtfEncodings::UTF16LE) { 299| 981| ret = impl->count_utf16le(src.data(), src.size()); 300| | } else if constexpr (From == UtfEncodings::UTF8) { 301| | ret = impl->count_utf8(src.data(), src.size()); 302| | } 303| 981| results.push_back(ret); 304| 981| } 305| 327| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 327| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 327] ------------------ 307| 0| std::cerr << "begin errormessage for count_the_input()\n"; 308| 0| std::cerr << "in fuzz case for " 309| 0| << ValidationFunctionTrait::ValidationWithErrorsName 310| 0| << " invoked with " << src.size() << " elements:\n"; 311| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (311:31): [True: 0, False: 0] ------------------ 312| 0| std::cerr << "got return " << std::dec << results[i] 313| 0| << " from implementation " << implementations[i]->name() 314| 0| << '\n'; 315| 0| } 316| 0| std::cerr << "end errormessage\n"; 317| 0| return false; 318| 0| } 319| | 320| 327| return true; 321| 327| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 654| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 327| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 327| length_result ret{}; 344| | 345| 327| const auto implementations = get_supported_implementations(); 346| 327| std::vector results; 347| 327| results.reserve(implementations.size()); 348| | 349| 981| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 981, False: 327] ------------------ 350| 981| const auto len = invoke_lengthcalc(impl, src); 351| 981| results.push_back(len); 352| 981| ret.length.push_back(len); 353| 981| } 354| | 355| 327| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 327| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 327] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 327| } else { 375| 327| ret.implementations_agree = true; 376| 327| } 377| 327| return ret; 378| 327| } _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 981| FromSpan src) const { 331| 981| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 981| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 654| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 327| const bool inputisvalid) const { 383| 327| conversion_result ret{}; 384| | 385| 327| const auto implementations = get_supported_implementations(); 386| | 387| 327| std::vector> results; 388| 327| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 327| std::vector> outputbuffers; 393| 327| outputbuffers.reserve(implementations.size()); 394| 1.30k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 981, False: 327] ------------------ 395| 981| auto impl = implementations[i]; 396| 981| const ToType canary1{42}; 397| 981| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 981| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 981| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 981| const auto success = [](const ConversionResult& r) -> bool { 402| 981| if constexpr (std::is_same_v) { 403| 981| return r != 0; 404| 981| } else { 405| 981| return r.error == simdutf::error_code::SUCCESS; 406| 981| } 407| 981| }(implret1); 408| 981| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 981| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 981| const ToType canary2{25}; 414| 981| const auto outputbuffer_first_run = outputbuffer; 415| 981| std::ranges::fill(outputbuffer, canary2); 416| 981| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 981| src.size(), outputbuffer.data()); 418| | 419| 981| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 981] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 981| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 624, False: 357] | Branch (423:29): [True: 624, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 624| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 624| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 624] ------------------ 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| 624| } 440| 981| } 441| 981| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 624, False: 357] ------------------ 442| 981| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 327| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 119, False: 208] ------------------ 447| 357| for (auto& e : results) { ------------------ | Branch (447:20): [True: 357, False: 119] ------------------ 448| 357| e.outputhash.clear(); 449| 357| } 450| 119| } 451| | 452| 327| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 327| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 327] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 327| } else { 474| 327| ret.implementations_agree = true; 475| 327| } 476| 327| return ret; 477| 327| } _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 981| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 981| } else { 405| 981| return r.error == simdutf::error_code::SUCCESS; 406| 981| } 407| 981| }(implret1); _ZZNK10ConversionIL12UtfEncodings1ELS0_2EMN7simdutf14implementationEKDoFmPKDsmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 654| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_30clENSt3__14spanIKcLm18446744073709551615EEE: 554| 227| +[](std::span chardata) { \ 555| 227| const auto c = \ 556| 227| Conversion{ \ 559| 227| &I::lenfunc, &I::conversionfunc, \ 560| 227| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 227|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 227| ::std::void_t(); \ | | 1180| 227| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 227| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 227| constexpr auto _size = _name.size(); \ | | 1183| 227| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 227| return _nameof; }() ------------------ 561| 227| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 227|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 227| ::std::void_t(); \ | | 1180| 227| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 227| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 227| constexpr auto _size = _name.size(); \ | | 1183| 227| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 227| return _nameof; }() ------------------ 562| 227| c.fuzz(chardata); \ 563| 227| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 227| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 227| const FromSpan from{reinterpret_cast(chardata.data()), 178| 227| chardata.size() / sizeof(FromType)}; 179| | 180| 227| static const bool do_print_testcase = 181| 227| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 227| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 227] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 227| do { 189| | // step 0 - is the input valid? 190| 227| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 227| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 227] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | 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| 227| const auto [output_length, length_agree] = 204| 227| calculate_length(from, inputisvalid); 205| 227| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 227] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 227| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 195, False: 32] | Branch (208:28): [True: 0, False: 195] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 227| const auto [written, outputs_agree] = 215| 227| do_conversion(from, output_length, inputisvalid); 216| 227| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 227] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 227| return; 221| 227| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 227| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEE: 230| 227| validation_result verify_valid_input(FromSpan src) const { 231| 227| validation_result ret{}; 232| | 233| 227| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 227| const auto implementations = get_supported_implementations(); 235| 227| std::vector results; 236| 227| results.reserve(implementations.size()); 237| | 238| 681| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 681, False: 227] ------------------ 239| 681| results.push_back( 240| 681| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 681| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 681| const bool validation2 = 245| 681| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 681| src.data(), src.size()); 247| 681| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 681] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 681| } 258| | 259| 227| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 227| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 227] ------------------ 261| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 262| 0| std::cerr << "in fuzz case for " 263| 0| << ValidationFunctionTrait::ValidationWithErrorsName 264| 0| << " invoked with " << src.size() << " elements:\n"; 265| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (265:31): [True: 0, False: 0] ------------------ 266| 0| std::cerr << "got return " << std::dec << results[i] 267| 0| << " from implementation " << implementations[i]->name() 268| 0| << '\n'; 269| 0| } 270| 0| std::cerr << "end errormessage\n"; 271| 0| ret.implementations_agree = false; 272| 227| } else { 273| 227| ret.implementations_agree = true; 274| 227| } 275| 227| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 227| return r.error == simdutf::SUCCESS; 277| 227| }); 278| 227| return ret; 279| 227| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS6_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS5_S5_EEDaSJ_SM_: 259| 454| 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| 291| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 291| return r.error == simdutf::SUCCESS; 277| 291| }); _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEb: 342| 227| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 227| length_result ret{}; 344| | 345| 227| const auto implementations = get_supported_implementations(); 346| 227| std::vector results; 347| 227| results.reserve(implementations.size()); 348| | 349| 681| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 681, False: 227] ------------------ 350| 681| const auto len = invoke_lengthcalc(impl, src); 351| 681| results.push_back(len); 352| 681| ret.length.push_back(len); 353| 681| } 354| | 355| 227| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 227| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 227] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 227| } else { 375| 227| ret.implementations_agree = true; 376| 227| } 377| 227| return ret; 378| 227| } _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSF_NSt3__14spanIS6_Lm18446744073709551615EEE: 338| 681| FromSpan src) const { 339| 681| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 681| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE16calculate_lengthENSt3__14spanIS6_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 454| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 227| const bool inputisvalid) const { 383| 227| conversion_result ret{}; 384| | 385| 227| const auto implementations = get_supported_implementations(); 386| | 387| 227| std::vector> results; 388| 227| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 227| std::vector> outputbuffers; 393| 227| outputbuffers.reserve(implementations.size()); 394| 908| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 681, False: 227] ------------------ 395| 681| auto impl = implementations[i]; 396| 681| const ToType canary1{42}; 397| 681| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 681| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 681| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 681| const auto success = [](const ConversionResult& r) -> bool { 402| 681| if constexpr (std::is_same_v) { 403| 681| return r != 0; 404| 681| } else { 405| 681| return r.error == simdutf::error_code::SUCCESS; 406| 681| } 407| 681| }(implret1); 408| 681| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 681| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 681| const ToType canary2{25}; 414| 681| const auto outputbuffer_first_run = outputbuffer; 415| 681| std::ranges::fill(outputbuffer, canary2); 416| 681| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 681| src.size(), outputbuffer.data()); 418| | 419| 681| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 681] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 681| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 96, False: 585] | Branch (423:29): [True: 75, False: 21] ------------------ 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| 681| } 441| 681| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 75, False: 606] ------------------ 442| 681| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 227| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 195, False: 32] ------------------ 447| 585| for (auto& e : results) { ------------------ | Branch (447:20): [True: 585, False: 195] ------------------ 448| 585| e.outputhash.clear(); 449| 585| } 450| 195| } 451| | 452| 227| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 227| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 227] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 227| } else { 474| 227| ret.implementations_agree = true; 475| 227| } 476| 227| return ret; 477| 227| } _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS5_E_clESM_: 401| 681| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 681| } else { 405| 681| return r.error == simdutf::error_code::SUCCESS; 406| 681| } 407| 681| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_4EMN7simdutf14implementationEKDoFmmEMS2_KDoFNS1_6resultEPKDimPcEE13do_conversionENSt3__14spanIS6_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS5_ESU_EEDaSN_SQ_: 452| 454| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_31clENSt3__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| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 330| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 330| const FromSpan from{reinterpret_cast(chardata.data()), 178| 330| chardata.size() / sizeof(FromType)}; 179| | 180| 330| static const bool do_print_testcase = 181| 330| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 330| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 330] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 330| do { 189| | // step 0 - is the input valid? 190| 330| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 330| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 330] | 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| 330| const auto [output_length, length_agree] = 204| 330| calculate_length(from, inputisvalid); 205| 330| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 330] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 330| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 239, False: 91] | Branch (208:28): [True: 0, False: 239] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 330| const auto [written, outputs_agree] = 215| 330| do_conversion(from, output_length, inputisvalid); 216| 330| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 330] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 330| return; 221| 330| } 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| 330| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 330| validation_result verify_valid_input(FromSpan src) const { 231| 330| validation_result ret{}; 232| | 233| 330| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 330| const auto implementations = get_supported_implementations(); 235| 330| std::vector results; 236| 330| results.reserve(implementations.size()); 237| | 238| 990| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 990, False: 330] ------------------ 239| 990| results.push_back( 240| 990| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 990| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 990| const bool validation2 = 245| 990| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 990| src.data(), src.size()); 247| 990| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 990] ------------------ 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| 990| } 258| | 259| 330| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 330| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 330] ------------------ 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| 330| } else { 273| 330| ret.implementations_agree = true; 274| 330| } 275| 330| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 330| return r.error == simdutf::SUCCESS; 277| 330| }); 278| 330| return ret; 279| 330| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 660| 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| 512| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 512| return r.error == simdutf::SUCCESS; 277| 512| }); _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 330| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 330| length_result ret{}; 344| | 345| 330| const auto implementations = get_supported_implementations(); 346| 330| std::vector results; 347| 330| results.reserve(implementations.size()); 348| | 349| 990| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 990, False: 330] ------------------ 350| 990| const auto len = invoke_lengthcalc(impl, src); 351| 990| results.push_back(len); 352| 990| ret.length.push_back(len); 353| 990| } 354| | 355| 330| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 330| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 330] ------------------ 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| 330| } else { 375| 330| ret.implementations_agree = true; 376| 330| } 377| 330| return ret; 378| 330| } _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 990| FromSpan src) const { 331| 990| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 990| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 660| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 330| const bool inputisvalid) const { 383| 330| conversion_result ret{}; 384| | 385| 330| const auto implementations = get_supported_implementations(); 386| | 387| 330| std::vector> results; 388| 330| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 330| std::vector> outputbuffers; 393| 330| outputbuffers.reserve(implementations.size()); 394| 1.32k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 990, False: 330] ------------------ 395| 990| auto impl = implementations[i]; 396| 990| const ToType canary1{42}; 397| 990| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 990| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 990| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 990| const auto success = [](const ConversionResult& r) -> bool { 402| 990| if constexpr (std::is_same_v) { 403| 990| return r != 0; 404| 990| } else { 405| 990| return r.error == simdutf::error_code::SUCCESS; 406| 990| } 407| 990| }(implret1); 408| 990| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 990| 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| 990| const ToType canary2{25}; 414| 990| const auto outputbuffer_first_run = outputbuffer; 415| 990| std::ranges::fill(outputbuffer, canary2); 416| 990| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 990| src.size(), outputbuffer.data()); 418| | 419| 990| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 990] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 990| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 273, False: 717] | Branch (423:29): [True: 273, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 273| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 273| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 273] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 273| } 440| 990| } 441| 990| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 273, False: 717] ------------------ 442| 990| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 330| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 239, False: 91] ------------------ 447| 717| for (auto& e : results) { ------------------ | Branch (447:20): [True: 717, False: 239] ------------------ 448| 717| e.outputhash.clear(); 449| 717| } 450| 239| } 451| | 452| 330| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 330| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 330] ------------------ 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| 330| } else { 474| 330| ret.implementations_agree = true; 475| 330| } 476| 330| return ret; 477| 330| } _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 990| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 990| } else { 405| 990| return r.error == simdutf::error_code::SUCCESS; 406| 990| } 407| 990| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_0EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 660| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_32clENSt3__14spanIKcLm18446744073709551615EEE: 554| 357| +[](std::span chardata) { \ 555| 357| const auto c = \ 556| 357| Conversion{ \ 559| 357| &I::lenfunc, &I::conversionfunc, \ 560| 357| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 357|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 357| ::std::void_t(); \ | | 1180| 357| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 357| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 357| constexpr auto _size = _name.size(); \ | | 1183| 357| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 357| return _nameof; }() ------------------ 561| 357| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 357|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 357| ::std::void_t(); \ | | 1180| 357| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 357| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 357| constexpr auto _size = _name.size(); \ | | 1183| 357| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 357| return _nameof; }() ------------------ 562| 357| c.fuzz(chardata); \ 563| 357| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 357| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 357| const FromSpan from{reinterpret_cast(chardata.data()), 178| 357| chardata.size() / sizeof(FromType)}; 179| | 180| 357| static const bool do_print_testcase = 181| 357| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 357| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 357] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 357| do { 189| | // step 0 - is the input valid? 190| 357| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 357| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 357] | 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| 357| const auto [output_length, length_agree] = 204| 357| calculate_length(from, inputisvalid); 205| 357| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 357] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 357| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 268, False: 89] | Branch (208:28): [True: 0, False: 268] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 357| const auto [written, outputs_agree] = 215| 357| do_conversion(from, output_length, inputisvalid); 216| 357| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 357] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 357| return; 221| 357| } 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| 357| } _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 357| validation_result verify_valid_input(FromSpan src) const { 231| 357| validation_result ret{}; 232| | 233| 357| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 357| const auto implementations = get_supported_implementations(); 235| 357| std::vector results; 236| 357| results.reserve(implementations.size()); 237| | 238| 1.07k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.07k, False: 357] ------------------ 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| 357| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 357| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 357] ------------------ 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| 357| } else { 273| 357| ret.implementations_agree = true; 274| 357| } 275| 357| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 357| return r.error == simdutf::SUCCESS; 277| 357| }); 278| 357| return ret; 279| 357| } _ZZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 714| 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| 535| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 535| return r.error == simdutf::SUCCESS; 277| 535| }); _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 357| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 357| length_result ret{}; 344| | 345| 357| const auto implementations = get_supported_implementations(); 346| 357| std::vector results; 347| 357| results.reserve(implementations.size()); 348| | 349| 1.07k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.07k, False: 357] ------------------ 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| 357| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 357| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 357] ------------------ 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| 357| } else { 375| 357| ret.implementations_agree = true; 376| 357| } 377| 357| return ret; 378| 357| } _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| 714| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_1EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 357| const bool inputisvalid) const { 383| 357| conversion_result ret{}; 384| | 385| 357| const auto implementations = get_supported_implementations(); 386| | 387| 357| std::vector> results; 388| 357| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 357| std::vector> outputbuffers; 393| 357| outputbuffers.reserve(implementations.size()); 394| 1.42k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.07k, False: 357] ------------------ 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: 267, False: 804] | Branch (423:29): [True: 267, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 267| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 267| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 267] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 267| } 440| 1.07k| } 441| 1.07k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 267, False: 804] ------------------ 442| 1.07k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 357| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 268, False: 89] ------------------ 447| 804| for (auto& e : results) { ------------------ | Branch (447:20): [True: 804, False: 268] ------------------ 448| 804| e.outputhash.clear(); 449| 804| } 450| 268| } 451| | 452| 357| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 357| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 357] ------------------ 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| 357| } else { 474| 357| ret.implementations_agree = true; 475| 357| } 476| 357| return ret; 477| 357| } _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| 714| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_33clENSt3__14spanIKcLm18446744073709551615EEE: 554| 462| +[](std::span chardata) { \ 555| 462| const auto c = \ 556| 462| Conversion{ \ 559| 462| &I::lenfunc, &I::conversionfunc, \ 560| 462| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 462|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 462| ::std::void_t(); \ | | 1180| 462| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 462| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 462| constexpr auto _size = _name.size(); \ | | 1183| 462| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 462| return _nameof; }() ------------------ 561| 462| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 462|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 462| ::std::void_t(); \ | | 1180| 462| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 462| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 462| constexpr auto _size = _name.size(); \ | | 1183| 462| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 462| return _nameof; }() ------------------ 562| 462| c.fuzz(chardata); \ 563| 462| } \ _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIKcLm18446744073709551615EEE: 175| 462| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 462| const FromSpan from{reinterpret_cast(chardata.data()), 178| 462| chardata.size() / sizeof(FromType)}; 179| | 180| 462| static const bool do_print_testcase = 181| 462| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 462| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 462] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 462| do { 189| | // step 0 - is the input valid? 190| 462| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 462| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 462] | 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| 462| const auto [output_length, length_agree] = 204| 462| calculate_length(from, inputisvalid); 205| 462| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 462] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 462| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 345, False: 117] | Branch (208:28): [True: 0, False: 345] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 462| const auto [written, outputs_agree] = 215| 462| do_conversion(from, output_length, inputisvalid); 216| 462| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 462] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 462| return; 221| 462| } 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| 462| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 462| validation_result verify_valid_input(FromSpan src) const { 231| 462| validation_result ret{}; 232| | 233| 462| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 462| const auto implementations = get_supported_implementations(); 235| 462| std::vector results; 236| 462| results.reserve(implementations.size()); 237| | 238| 1.38k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.38k, False: 462] ------------------ 239| 1.38k| results.push_back( 240| 1.38k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.38k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.38k| const bool validation2 = 245| 1.38k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.38k| src.data(), src.size()); 247| 1.38k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.38k] ------------------ 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| 1.38k| } 258| | 259| 462| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 462| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 462] ------------------ 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| 462| } else { 273| 462| ret.implementations_agree = true; 274| 462| } 275| 462| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 462| return r.error == simdutf::SUCCESS; 277| 462| }); 278| 462| return ret; 279| 462| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 924| 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| 696| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 696| return r.error == simdutf::SUCCESS; 277| 696| }); _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 462| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 462| length_result ret{}; 344| | 345| 462| const auto implementations = get_supported_implementations(); 346| 462| std::vector results; 347| 462| results.reserve(implementations.size()); 348| | 349| 1.38k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.38k, False: 462] ------------------ 350| 1.38k| const auto len = invoke_lengthcalc(impl, src); 351| 1.38k| results.push_back(len); 352| 1.38k| ret.length.push_back(len); 353| 1.38k| } 354| | 355| 462| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 462| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 462] ------------------ 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| 462| } else { 375| 462| ret.implementations_agree = true; 376| 462| } 377| 462| return ret; 378| 462| } _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.38k| FromSpan src) const { 331| 1.38k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.38k| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 924| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 462| const bool inputisvalid) const { 383| 462| conversion_result ret{}; 384| | 385| 462| const auto implementations = get_supported_implementations(); 386| | 387| 462| std::vector> results; 388| 462| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 462| std::vector> outputbuffers; 393| 462| outputbuffers.reserve(implementations.size()); 394| 1.84k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.38k, False: 462] ------------------ 395| 1.38k| auto impl = implementations[i]; 396| 1.38k| const ToType canary1{42}; 397| 1.38k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.38k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.38k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.38k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.38k| if constexpr (std::is_same_v) { 403| 1.38k| return r != 0; 404| 1.38k| } else { 405| 1.38k| return r.error == simdutf::error_code::SUCCESS; 406| 1.38k| } 407| 1.38k| }(implret1); 408| 1.38k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.38k| 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| 1.38k| const ToType canary2{25}; 414| 1.38k| const auto outputbuffer_first_run = outputbuffer; 415| 1.38k| std::ranges::fill(outputbuffer, canary2); 416| 1.38k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.38k| src.size(), outputbuffer.data()); 418| | 419| 1.38k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.38k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.38k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 351, False: 1.03k] | Branch (423:29): [True: 351, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 351| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 351| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 351] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 351| } 440| 1.38k| } 441| 1.38k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 351, False: 1.03k] ------------------ 442| 1.38k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 462| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 345, False: 117] ------------------ 447| 1.03k| for (auto& e : results) { ------------------ | Branch (447:20): [True: 1.03k, False: 345] ------------------ 448| 1.03k| e.outputhash.clear(); 449| 1.03k| } 450| 345| } 451| | 452| 462| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 462| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 462] ------------------ 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| 462| } else { 474| 462| ret.implementations_agree = true; 475| 462| } 476| 462| return ret; 477| 462| } _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.38k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.38k| } else { 405| 1.38k| return r.error == simdutf::error_code::SUCCESS; 406| 1.38k| } 407| 1.38k| }(implret1); _ZZNK10ConversionIL12UtfEncodings3ELS0_2EMN7simdutf14implementationEKDoFmPKDimEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 924| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_34clENSt3__14spanIKcLm18446744073709551615EEE: 554| 261| +[](std::span chardata) { \ 555| 261| const auto c = \ 556| 261| Conversion{ \ 559| 261| &I::lenfunc, &I::conversionfunc, \ 560| 261| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 261|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 261| ::std::void_t(); \ | | 1180| 261| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 261| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 261| constexpr auto _size = _name.size(); \ | | 1183| 261| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 261| return _nameof; }() ------------------ 561| 261| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 261|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 261| ::std::void_t(); \ | | 1180| 261| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 261| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 261| constexpr auto _size = _name.size(); \ | | 1183| 261| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 261| return _nameof; }() ------------------ 562| 261| c.fuzz(chardata); \ 563| 261| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 261| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 261| const FromSpan from{reinterpret_cast(chardata.data()), 178| 261| chardata.size() / sizeof(FromType)}; 179| | 180| 261| static const bool do_print_testcase = 181| 261| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 261| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 261] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 261| do { 189| | // step 0 - is the input valid? 190| 261| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 261| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 261] | 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| 261| From == UtfEncodings::UTF8) { 198| 261| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 261] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 261| } 201| | 202| | // step 2 - what is the required size of the output? 203| 261| const auto [output_length, length_agree] = 204| 261| calculate_length(from, inputisvalid); 205| 261| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 261] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 261| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 148, False: 113] | Branch (208:28): [True: 0, False: 148] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 261| const auto [written, outputs_agree] = 215| 261| do_conversion(from, output_length, inputisvalid); 216| 261| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 261] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 261| return; 221| 261| } 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| 261| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 261| validation_result verify_valid_input(FromSpan src) const { 231| 261| validation_result ret{}; 232| | 233| 261| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 261| const auto implementations = get_supported_implementations(); 235| 261| std::vector results; 236| 261| results.reserve(implementations.size()); 237| | 238| 783| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 783, False: 261] ------------------ 239| 783| results.push_back( 240| 783| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 783| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 783| const bool validation2 = 245| 783| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 783| src.data(), src.size()); 247| 783| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 783] ------------------ 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| 783| } 258| | 259| 261| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 261| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 261] ------------------ 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| 261| } else { 273| 261| ret.implementations_agree = true; 274| 261| } 275| 261| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 261| return r.error == simdutf::SUCCESS; 277| 261| }); 278| 261| return ret; 279| 261| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 522| 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| 487| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 487| return r.error == simdutf::SUCCESS; 277| 487| }); _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 261| bool count_the_input(FromSpan src) const { 290| 261| const auto implementations = get_supported_implementations(); 291| 261| std::vector results; 292| 261| results.reserve(implementations.size()); 293| | 294| 783| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 783, False: 261] ------------------ 295| 783| 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| 783| } else if constexpr (From == UtfEncodings::UTF8) { 301| 783| ret = impl->count_utf8(src.data(), src.size()); 302| 783| } 303| 783| results.push_back(ret); 304| 783| } 305| 261| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 261| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 261] ------------------ 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| 261| return true; 321| 261| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 522| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 261| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 261| length_result ret{}; 344| | 345| 261| const auto implementations = get_supported_implementations(); 346| 261| std::vector results; 347| 261| results.reserve(implementations.size()); 348| | 349| 783| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 783, False: 261] ------------------ 350| 783| const auto len = invoke_lengthcalc(impl, src); 351| 783| results.push_back(len); 352| 783| ret.length.push_back(len); 353| 783| } 354| | 355| 261| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 261| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 261] ------------------ 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| 261| } else { 375| 261| ret.implementations_agree = true; 376| 261| } 377| 261| return ret; 378| 261| } _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 783| FromSpan src) const { 331| 783| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 783| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 522| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 261| const bool inputisvalid) const { 383| 261| conversion_result ret{}; 384| | 385| 261| const auto implementations = get_supported_implementations(); 386| | 387| 261| std::vector> results; 388| 261| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 261| std::vector> outputbuffers; 393| 261| outputbuffers.reserve(implementations.size()); 394| 1.04k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 783, False: 261] ------------------ 395| 783| auto impl = implementations[i]; 396| 783| const ToType canary1{42}; 397| 783| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 783| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 783| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 783| const auto success = [](const ConversionResult& r) -> bool { 402| 783| if constexpr (std::is_same_v) { 403| 783| return r != 0; 404| 783| } else { 405| 783| return r.error == simdutf::error_code::SUCCESS; 406| 783| } 407| 783| }(implret1); 408| 783| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 783| 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| 783| const ToType canary2{25}; 414| 783| const auto outputbuffer_first_run = outputbuffer; 415| 783| std::ranges::fill(outputbuffer, canary2); 416| 783| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 783| src.size(), outputbuffer.data()); 418| | 419| 783| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 783] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 783| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 339, False: 444] | Branch (423:29): [True: 315, False: 24] ------------------ 424| | // only care about the output if the input is valid 425| 315| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 315| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 315] ------------------ 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| 315| } 440| 783| } 441| 783| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 315, False: 468] ------------------ 442| 783| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 261| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 148, False: 113] ------------------ 447| 444| for (auto& e : results) { ------------------ | Branch (447:20): [True: 444, False: 148] ------------------ 448| 444| e.outputhash.clear(); 449| 444| } 450| 148| } 451| | 452| 261| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 261| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 261] ------------------ 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| 261| } else { 474| 261| ret.implementations_agree = true; 475| 261| } 476| 261| return ret; 477| 261| } _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 783| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 783| } else { 405| 783| return r.error == simdutf::error_code::SUCCESS; 406| 783| } 407| 783| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_4EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 522| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_35clENSt3__14spanIKcLm18446744073709551615EEE: 554| 386| +[](std::span chardata) { \ 555| 386| const auto c = \ 556| 386| Conversion{ \ 559| 386| &I::lenfunc, &I::conversionfunc, \ 560| 386| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 386|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 386| ::std::void_t(); \ | | 1180| 386| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 386| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 386| constexpr auto _size = _name.size(); \ | | 1183| 386| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 386| return _nameof; }() ------------------ 561| 386| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 386|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 386| ::std::void_t(); \ | | 1180| 386| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 386| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 386| constexpr auto _size = _name.size(); \ | | 1183| 386| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 386| return _nameof; }() ------------------ 562| 386| c.fuzz(chardata); \ 563| 386| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 386| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 386| const FromSpan from{reinterpret_cast(chardata.data()), 178| 386| chardata.size() / sizeof(FromType)}; 179| | 180| 386| static const bool do_print_testcase = 181| 386| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 386| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 386] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 386| do { 189| | // step 0 - is the input valid? 190| 386| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 386| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 386] | 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| 386| From == UtfEncodings::UTF8) { 198| 386| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 386] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 386| } 201| | 202| | // step 2 - what is the required size of the output? 203| 386| const auto [output_length, length_agree] = 204| 386| calculate_length(from, inputisvalid); 205| 386| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 386] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 386| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 211, False: 175] | Branch (208:28): [True: 0, False: 211] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 386| const auto [written, outputs_agree] = 215| 386| do_conversion(from, output_length, inputisvalid); 216| 386| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 386] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 386| return; 221| 386| } 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| 386| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 386| validation_result verify_valid_input(FromSpan src) const { 231| 386| validation_result ret{}; 232| | 233| 386| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 386| const auto implementations = get_supported_implementations(); 235| 386| std::vector results; 236| 386| results.reserve(implementations.size()); 237| | 238| 1.15k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.15k, False: 386] ------------------ 239| 1.15k| results.push_back( 240| 1.15k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.15k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.15k| const bool validation2 = 245| 1.15k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.15k| src.data(), src.size()); 247| 1.15k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.15k] ------------------ 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| 1.15k| } 258| | 259| 386| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 386| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 386] ------------------ 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| 386| } else { 273| 386| ret.implementations_agree = true; 274| 386| } 275| 386| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 386| return r.error == simdutf::SUCCESS; 277| 386| }); 278| 386| return ret; 279| 386| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 772| 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| 736| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 736| return r.error == simdutf::SUCCESS; 277| 736| }); _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 386| bool count_the_input(FromSpan src) const { 290| 386| const auto implementations = get_supported_implementations(); 291| 386| std::vector results; 292| 386| results.reserve(implementations.size()); 293| | 294| 1.15k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.15k, False: 386] ------------------ 295| 1.15k| 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| 1.15k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.15k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.15k| } 303| 1.15k| results.push_back(ret); 304| 1.15k| } 305| 386| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 386| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 386] ------------------ 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| 386| return true; 321| 386| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 772| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 386| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 386| length_result ret{}; 344| | 345| 386| const auto implementations = get_supported_implementations(); 346| 386| std::vector results; 347| 386| results.reserve(implementations.size()); 348| | 349| 1.15k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.15k, False: 386] ------------------ 350| 1.15k| const auto len = invoke_lengthcalc(impl, src); 351| 1.15k| results.push_back(len); 352| 1.15k| ret.length.push_back(len); 353| 1.15k| } 354| | 355| 386| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 386| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 386] ------------------ 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| 386| } else { 375| 386| ret.implementations_agree = true; 376| 386| } 377| 386| return ret; 378| 386| } _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.15k| FromSpan src) const { 331| 1.15k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.15k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 772| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 386| const bool inputisvalid) const { 383| 386| conversion_result ret{}; 384| | 385| 386| const auto implementations = get_supported_implementations(); 386| | 387| 386| std::vector> results; 388| 386| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 386| std::vector> outputbuffers; 393| 386| outputbuffers.reserve(implementations.size()); 394| 1.54k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.15k, False: 386] ------------------ 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: 525, False: 633] | Branch (423:29): [True: 525, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 525| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 525| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 525] ------------------ 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| 525| } 440| 1.15k| } 441| 1.15k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 525, False: 633] ------------------ 442| 1.15k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 386| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 211, False: 175] ------------------ 447| 633| for (auto& e : results) { ------------------ | Branch (447:20): [True: 633, False: 211] ------------------ 448| 633| e.outputhash.clear(); 449| 633| } 450| 211| } 451| | 452| 386| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 386| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 386] ------------------ 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| 386| } else { 474| 386| ret.implementations_agree = true; 475| 386| } 476| 386| return ret; 477| 386| } _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.15k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.15k| } else { 405| 1.15k| return r.error == simdutf::error_code::SUCCESS; 406| 1.15k| } 407| 1.15k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_0EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 772| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_36clENSt3__14spanIKcLm18446744073709551615EEE: 554| 297| +[](std::span chardata) { \ 555| 297| const auto c = \ 556| 297| Conversion{ \ 559| 297| &I::lenfunc, &I::conversionfunc, \ 560| 297| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 297|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 297| ::std::void_t(); \ | | 1180| 297| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 297| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 297| constexpr auto _size = _name.size(); \ | | 1183| 297| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 297| return _nameof; }() ------------------ 561| 297| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 297|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 297| ::std::void_t(); \ | | 1180| 297| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 297| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 297| constexpr auto _size = _name.size(); \ | | 1183| 297| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 297| return _nameof; }() ------------------ 562| 297| c.fuzz(chardata); \ 563| 297| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 297| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 297| const FromSpan from{reinterpret_cast(chardata.data()), 178| 297| chardata.size() / sizeof(FromType)}; 179| | 180| 297| static const bool do_print_testcase = 181| 297| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 297| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 297] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 297| do { 189| | // step 0 - is the input valid? 190| 297| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 297| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 297] | 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| 297| From == UtfEncodings::UTF8) { 198| 297| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 297] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 297| } 201| | 202| | // step 2 - what is the required size of the output? 203| 297| const auto [output_length, length_agree] = 204| 297| calculate_length(from, inputisvalid); 205| 297| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 297] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 297| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 170, False: 127] | Branch (208:28): [True: 0, False: 170] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 297| const auto [written, outputs_agree] = 215| 297| do_conversion(from, output_length, inputisvalid); 216| 297| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 297] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 297| return; 221| 297| } 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| 297| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 297| validation_result verify_valid_input(FromSpan src) const { 231| 297| validation_result ret{}; 232| | 233| 297| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 297| const auto implementations = get_supported_implementations(); 235| 297| std::vector results; 236| 297| results.reserve(implementations.size()); 237| | 238| 891| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 891, False: 297] ------------------ 239| 891| results.push_back( 240| 891| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 891| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 891| const bool validation2 = 245| 891| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 891| src.data(), src.size()); 247| 891| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 891] ------------------ 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| 891| } 258| | 259| 297| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 297| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 297] ------------------ 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| 297| } else { 273| 297| ret.implementations_agree = true; 274| 297| } 275| 297| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 297| return r.error == simdutf::SUCCESS; 277| 297| }); 278| 297| return ret; 279| 297| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 594| 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| 551| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 551| return r.error == simdutf::SUCCESS; 277| 551| }); _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 297| bool count_the_input(FromSpan src) const { 290| 297| const auto implementations = get_supported_implementations(); 291| 297| std::vector results; 292| 297| results.reserve(implementations.size()); 293| | 294| 891| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 891, False: 297] ------------------ 295| 891| 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| 891| } else if constexpr (From == UtfEncodings::UTF8) { 301| 891| ret = impl->count_utf8(src.data(), src.size()); 302| 891| } 303| 891| results.push_back(ret); 304| 891| } 305| 297| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 297| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 297] ------------------ 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| 297| return true; 321| 297| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 594| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 297| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 297| length_result ret{}; 344| | 345| 297| const auto implementations = get_supported_implementations(); 346| 297| std::vector results; 347| 297| results.reserve(implementations.size()); 348| | 349| 891| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 891, False: 297] ------------------ 350| 891| const auto len = invoke_lengthcalc(impl, src); 351| 891| results.push_back(len); 352| 891| ret.length.push_back(len); 353| 891| } 354| | 355| 297| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 297| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 297] ------------------ 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| 297| } else { 375| 297| ret.implementations_agree = true; 376| 297| } 377| 297| return ret; 378| 297| } _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 891| FromSpan src) const { 331| 891| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 891| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 594| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 297| const bool inputisvalid) const { 383| 297| conversion_result ret{}; 384| | 385| 297| const auto implementations = get_supported_implementations(); 386| | 387| 297| std::vector> results; 388| 297| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 297| std::vector> outputbuffers; 393| 297| outputbuffers.reserve(implementations.size()); 394| 1.18k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 891, False: 297] ------------------ 395| 891| auto impl = implementations[i]; 396| 891| const ToType canary1{42}; 397| 891| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 891| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 891| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 891| const auto success = [](const ConversionResult& r) -> bool { 402| 891| if constexpr (std::is_same_v) { 403| 891| return r != 0; 404| 891| } else { 405| 891| return r.error == simdutf::error_code::SUCCESS; 406| 891| } 407| 891| }(implret1); 408| 891| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 891| 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| 891| const ToType canary2{25}; 414| 891| const auto outputbuffer_first_run = outputbuffer; 415| 891| std::ranges::fill(outputbuffer, canary2); 416| 891| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 891| src.size(), outputbuffer.data()); 418| | 419| 891| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 891] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 891| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 381, False: 510] | Branch (423:29): [True: 381, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 381| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 381| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 381] ------------------ 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| 381| } 440| 891| } 441| 891| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 381, False: 510] ------------------ 442| 891| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 297| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 170, False: 127] ------------------ 447| 510| for (auto& e : results) { ------------------ | Branch (447:20): [True: 510, False: 170] ------------------ 448| 510| e.outputhash.clear(); 449| 510| } 450| 170| } 451| | 452| 297| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 297| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 297] ------------------ 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| 297| } else { 474| 297| ret.implementations_agree = true; 475| 297| } 476| 297| return ret; 477| 297| } _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 891| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 891| } else { 405| 891| return r.error == simdutf::error_code::SUCCESS; 406| 891| } 407| 891| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_1EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDsEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 594| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_37clENSt3__14spanIKcLm18446744073709551615EEE: 554| 395| +[](std::span chardata) { \ 555| 395| const auto c = \ 556| 395| Conversion{ \ 559| 395| &I::lenfunc, &I::conversionfunc, \ 560| 395| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 395|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 395| ::std::void_t(); \ | | 1180| 395| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 395| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 395| constexpr auto _size = _name.size(); \ | | 1183| 395| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 395| return _nameof; }() ------------------ 561| 395| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 395|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 395| ::std::void_t(); \ | | 1180| 395| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 395| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 395| constexpr auto _size = _name.size(); \ | | 1183| 395| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 395| return _nameof; }() ------------------ 562| 395| c.fuzz(chardata); \ 563| 395| } \ _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 395| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 395| const FromSpan from{reinterpret_cast(chardata.data()), 178| 395| chardata.size() / sizeof(FromType)}; 179| | 180| 395| static const bool do_print_testcase = 181| 395| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 395| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 395] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 395| do { 189| | // step 0 - is the input valid? 190| 395| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 395| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 395] | 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| 395| From == UtfEncodings::UTF8) { 198| 395| if (!count_the_input(from) && !allow_implementations_to_differ) ------------------ | Branch (198:13): [True: 0, False: 395] | Branch (198:39): [True: 0, Folded] ------------------ 199| 0| break; 200| 395| } 201| | 202| | // step 2 - what is the required size of the output? 203| 395| const auto [output_length, length_agree] = 204| 395| calculate_length(from, inputisvalid); 205| 395| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 395] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 395| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 222, False: 173] | Branch (208:28): [True: 0, False: 222] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 395| const auto [written, outputs_agree] = 215| 395| do_conversion(from, output_length, inputisvalid); 216| 395| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 395] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 395| return; 221| 395| } 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| 395| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 230| 395| validation_result verify_valid_input(FromSpan src) const { 231| 395| validation_result ret{}; 232| | 233| 395| auto input_validation = ValidationFunctionTrait::ValidationWithErrors; 234| 395| const auto implementations = get_supported_implementations(); 235| 395| std::vector results; 236| 395| results.reserve(implementations.size()); 237| | 238| 1.18k| for (auto impl : implementations) { ------------------ | Branch (238:20): [True: 1.18k, False: 395] ------------------ 239| 1.18k| results.push_back( 240| 1.18k| std::invoke(input_validation, impl, src.data(), src.size())); 241| | 242| | // make sure the validation variant that returns a bool agrees 243| 1.18k| const bool validation1 = results.back().error == simdutf::SUCCESS; 244| 1.18k| const bool validation2 = 245| 1.18k| std::invoke(ValidationFunctionTrait::Validation, impl, 246| 1.18k| src.data(), src.size()); 247| 1.18k| if (validation1 != validation2) { ------------------ | Branch (247:11): [True: 0, False: 1.18k] ------------------ 248| 0| std::cerr << "begin errormessage for verify_valid_input()\n"; 249| 0| std::cerr << ValidationFunctionTrait::ValidationWithErrorsName 250| 0| << " gives " << validation1 << " while " 251| 0| << ValidationFunctionTrait::ValidationName << " gave " 252| 0| << validation2 << " for implementation " << impl->name() 253| 0| << '\n'; 254| 0| std::cerr << "end errormessage\n"; 255| 0| std::abort(); 256| 0| } 257| 1.18k| } 258| | 259| 395| auto neq = [](const auto& a, const auto& b) { return a != b; }; 260| 395| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (260:9): [True: 0, False: 395] ------------------ 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| 395| } else { 273| 395| ret.implementations_agree = true; 274| 395| } 275| 395| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 395| return r.error == simdutf::SUCCESS; 277| 395| }); 278| 395| return ret; 279| 395| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE18verify_valid_inputIvQneT_LS0_4EEENSB_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clIS7_S7_EEDaSJ_SM_: 259| 790| 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| 741| ret.valid = std::ranges::all_of(results, [](const simdutf::result& r) { 276| 741| return r.error == simdutf::SUCCESS; 277| 741| }); _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEE: 289| 395| bool count_the_input(FromSpan src) const { 290| 395| const auto implementations = get_supported_implementations(); 291| 395| std::vector results; 292| 395| results.reserve(implementations.size()); 293| | 294| 1.18k| for (auto impl : implementations) { ------------------ | Branch (294:20): [True: 1.18k, False: 395] ------------------ 295| 1.18k| 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| 1.18k| } else if constexpr (From == UtfEncodings::UTF8) { 301| 1.18k| ret = impl->count_utf8(src.data(), src.size()); 302| 1.18k| } 303| 1.18k| results.push_back(ret); 304| 1.18k| } 305| 395| auto neq = [](const auto& a, const auto& b) { return a != b; }; 306| 395| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (306:9): [True: 0, False: 395] ------------------ 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| 395| return true; 321| 395| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE15count_the_inputENSt3__14spanIS3_Lm18446744073709551615EEEENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 305| 790| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 395| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 395| length_result ret{}; 344| | 345| 395| const auto implementations = get_supported_implementations(); 346| 395| std::vector results; 347| 395| results.reserve(implementations.size()); 348| | 349| 1.18k| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 1.18k, False: 395] ------------------ 350| 1.18k| const auto len = invoke_lengthcalc(impl, src); 351| 1.18k| results.push_back(len); 352| 1.18k| ret.length.push_back(len); 353| 1.18k| } 354| | 355| 395| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 395| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 395] ------------------ 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| 395| } else { 375| 395| ret.implementations_agree = true; 376| 395| } 377| 395| return ret; 378| 395| } _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSF_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 1.18k| FromSpan src) const { 331| 1.18k| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 1.18k| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSH_SK_: 355| 790| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEb: 382| 395| const bool inputisvalid) const { 383| 395| conversion_result ret{}; 384| | 385| 395| const auto implementations = get_supported_implementations(); 386| | 387| 395| std::vector> results; 388| 395| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 395| std::vector> outputbuffers; 393| 395| outputbuffers.reserve(implementations.size()); 394| 1.58k| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 1.18k, False: 395] ------------------ 395| 1.18k| auto impl = implementations[i]; 396| 1.18k| const ToType canary1{42}; 397| 1.18k| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 1.18k| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 1.18k| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 1.18k| const auto success = [](const ConversionResult& r) -> bool { 402| 1.18k| if constexpr (std::is_same_v) { 403| 1.18k| return r != 0; 404| 1.18k| } else { 405| 1.18k| return r.error == simdutf::error_code::SUCCESS; 406| 1.18k| } 407| 1.18k| }(implret1); 408| 1.18k| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 1.18k| 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| 1.18k| const ToType canary2{25}; 414| 1.18k| const auto outputbuffer_first_run = outputbuffer; 415| 1.18k| std::ranges::fill(outputbuffer, canary2); 416| 1.18k| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 1.18k| src.size(), outputbuffer.data()); 418| | 419| 1.18k| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 1.18k] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 1.18k| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 519, False: 666] | Branch (423:29): [True: 519, False: 0] ------------------ 424| | // only care about the output if the input is valid 425| 519| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 519| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 519] ------------------ 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| 519| } 440| 1.18k| } 441| 1.18k| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 519, False: 666] ------------------ 442| 1.18k| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 395| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 222, False: 173] ------------------ 447| 666| for (auto& e : results) { ------------------ | Branch (447:20): [True: 666, False: 222] ------------------ 448| 666| e.outputhash.clear(); 449| 666| } 450| 222| } 451| | 452| 395| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 395| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 395] ------------------ 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| 395| } else { 474| 395| ret.implementations_agree = true; 475| 395| } 476| 395| return ret; 477| 395| } _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKS7_E_clESM_: 401| 1.18k| const auto success = [](const ConversionResult& r) -> bool { 402| | if constexpr (std::is_same_v) { 403| | return r != 0; 404| 1.18k| } else { 405| 1.18k| return r.error == simdutf::error_code::SUCCESS; 406| 1.18k| } 407| 1.18k| }(implret1); _ZZNK10ConversionIL12UtfEncodings2ELS0_3EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFNS1_6resultES4_mPDiEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSC_6vectorImNSC_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultIS7_ESU_EEDaSN_SQ_: 452| 790| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_38clENSt3__14spanIKcLm18446744073709551615EEE: 554| 50| +[](std::span chardata) { \ 555| 50| const auto c = \ 556| 50| Conversion{ \ 559| 50| &I::lenfunc, &I::conversionfunc, \ 560| 50| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 50|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 50| ::std::void_t(); \ | | 1180| 50| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 50| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 50| constexpr auto _size = _name.size(); \ | | 1183| 50| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 50| return _nameof; }() ------------------ 561| 50| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 50|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 50| ::std::void_t(); \ | | 1180| 50| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 50| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 50| constexpr auto _size = _name.size(); \ | | 1183| 50| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 50| return _nameof; }() ------------------ 562| 50| c.fuzz(chardata); \ 563| 50| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE4fuzzENSt3__14spanIS5_Lm18446744073709551615EEE: 175| 50| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 50| const FromSpan from{reinterpret_cast(chardata.data()), 178| 50| chardata.size() / sizeof(FromType)}; 179| | 180| 50| static const bool do_print_testcase = 181| 50| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 50| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 50] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 50| do { 189| | // step 0 - is the input valid? 190| 50| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 50| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 50] | 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| 50| const auto [output_length, length_agree] = 204| 50| calculate_length(from, inputisvalid); 205| 50| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 50] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 50| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 50] | 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| 50| const auto [written, outputs_agree] = 215| 50| do_conversion(from, output_length, inputisvalid); 216| 50| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 50] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 50| return; 221| 50| } 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| 50| } _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 283| 50| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 50| return validation_result{.valid = true, .implementations_agree = true}; 287| 50| } _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 50| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 50| length_result ret{}; 344| | 345| 50| const auto implementations = get_supported_implementations(); 346| 50| std::vector results; 347| 50| results.reserve(implementations.size()); 348| | 349| 150| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 150, False: 50] ------------------ 350| 150| const auto len = invoke_lengthcalc(impl, src); 351| 150| results.push_back(len); 352| 150| ret.length.push_back(len); 353| 150| } 354| | 355| 50| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 50| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 50] ------------------ 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| 50| } else { 375| 50| ret.implementations_agree = true; 376| 50| } 377| 50| return ret; 378| 50| } _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 150| FromSpan src) const { 339| 150| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 150| } _ZZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 100| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 50| const bool inputisvalid) const { 383| 50| conversion_result ret{}; 384| | 385| 50| const auto implementations = get_supported_implementations(); 386| | 387| 50| std::vector> results; 388| 50| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 50| std::vector> outputbuffers; 393| 50| outputbuffers.reserve(implementations.size()); 394| 200| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 150, False: 50] ------------------ 395| 150| auto impl = implementations[i]; 396| 150| const ToType canary1{42}; 397| 150| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 150| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 150| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 150| const auto success = [](const ConversionResult& r) -> bool { 402| 150| if constexpr (std::is_same_v) { 403| 150| return r != 0; 404| 150| } else { 405| 150| return r.error == simdutf::error_code::SUCCESS; 406| 150| } 407| 150| }(implret1); 408| 150| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 150| 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| 150| const ToType canary2{25}; 414| 150| const auto outputbuffer_first_run = outputbuffer; 415| 150| std::ranges::fill(outputbuffer, canary2); 416| 150| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 150| src.size(), outputbuffer.data()); 418| | 419| 150| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 150] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 150| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 150, False: 0] | Branch (423:29): [True: 144, False: 6] ------------------ 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| 150| } 441| 150| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 144, False: 6] ------------------ 442| 150| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 50| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 50] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 50| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 50| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 50] ------------------ 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| 50| } else { 474| 50| ret.implementations_agree = true; 475| 50| } 476| 50| return ret; 477| 50| } _ZZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 150| const auto success = [](const ConversionResult& r) -> bool { 402| 150| if constexpr (std::is_same_v) { 403| 150| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 150| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_3EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDiEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 100| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_39clENSt3__14spanIKcLm18446744073709551615EEE: 554| 55| +[](std::span chardata) { \ 555| 55| const auto c = \ 556| 55| Conversion{ \ 559| 55| &I::lenfunc, &I::conversionfunc, \ 560| 55| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 55|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 55| ::std::void_t(); \ | | 1180| 55| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 55| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 55| constexpr auto _size = _name.size(); \ | | 1183| 55| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 55| return _nameof; }() ------------------ 561| 55| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 55|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 55| ::std::void_t(); \ | | 1180| 55| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 55| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 55| constexpr auto _size = _name.size(); \ | | 1183| 55| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 55| return _nameof; }() ------------------ 562| 55| c.fuzz(chardata); \ 563| 55| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE4fuzzENSt3__14spanIS5_Lm18446744073709551615EEE: 175| 55| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 55| const FromSpan from{reinterpret_cast(chardata.data()), 178| 55| chardata.size() / sizeof(FromType)}; 179| | 180| 55| static const bool do_print_testcase = 181| 55| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 55| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 55] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 55| do { 189| | // step 0 - is the input valid? 190| 55| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 55| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 55] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 55| const auto [output_length, length_agree] = 204| 55| calculate_length(from, inputisvalid); 205| 55| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 55] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 55| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 55] | Branch (208:28): [True: 0, False: 0] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 55| const auto [written, outputs_agree] = 215| 55| do_conversion(from, output_length, inputisvalid); 216| 55| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 55] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 55| return; 221| 55| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 55| } _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 283| 55| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 55| return validation_result{.valid = true, .implementations_agree = true}; 287| 55| } _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 55| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 55| length_result ret{}; 344| | 345| 55| const auto implementations = get_supported_implementations(); 346| 55| std::vector results; 347| 55| results.reserve(implementations.size()); 348| | 349| 165| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 165, False: 55] ------------------ 350| 165| const auto len = invoke_lengthcalc(impl, src); 351| 165| results.push_back(len); 352| 165| ret.length.push_back(len); 353| 165| } 354| | 355| 55| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 55| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 55] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 55| } else { 375| 55| ret.implementations_agree = true; 376| 55| } 377| 55| return ret; 378| 55| } _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 165| FromSpan src) const { 339| 165| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 165| } _ZZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 110| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 55| const bool inputisvalid) const { 383| 55| conversion_result ret{}; 384| | 385| 55| const auto implementations = get_supported_implementations(); 386| | 387| 55| std::vector> results; 388| 55| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 55| std::vector> outputbuffers; 393| 55| outputbuffers.reserve(implementations.size()); 394| 220| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 165, False: 55] ------------------ 395| 165| auto impl = implementations[i]; 396| 165| const ToType canary1{42}; 397| 165| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 165| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 165| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 165| const auto success = [](const ConversionResult& r) -> bool { 402| 165| if constexpr (std::is_same_v) { 403| 165| return r != 0; 404| 165| } else { 405| 165| return r.error == simdutf::error_code::SUCCESS; 406| 165| } 407| 165| }(implret1); 408| 165| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 165| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 165| const ToType canary2{25}; 414| 165| const auto outputbuffer_first_run = outputbuffer; 415| 165| std::ranges::fill(outputbuffer, canary2); 416| 165| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 165| src.size(), outputbuffer.data()); 418| | 419| 165| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 165] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 165| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 165, False: 0] | Branch (423:29): [True: 159, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 159| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 159| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 159] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 159| } 440| 165| } 441| 165| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 159, False: 6] ------------------ 442| 165| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 55| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 55] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 55| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 55| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 55] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 55| } else { 474| 55| ret.implementations_agree = true; 475| 55| } 476| 55| return ret; 477| 55| } _ZZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 165| const auto success = [](const ConversionResult& r) -> bool { 402| 165| if constexpr (std::is_same_v) { 403| 165| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 165| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_0EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 110| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_40clENSt3__14spanIKcLm18446744073709551615EEE: 554| 44| +[](std::span chardata) { \ 555| 44| const auto c = \ 556| 44| Conversion{ \ 559| 44| &I::lenfunc, &I::conversionfunc, \ 560| 44| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 44|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 44| ::std::void_t(); \ | | 1180| 44| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 44| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 44| constexpr auto _size = _name.size(); \ | | 1183| 44| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 44| return _nameof; }() ------------------ 561| 44| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 44|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 44| ::std::void_t(); \ | | 1180| 44| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 44| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 44| constexpr auto _size = _name.size(); \ | | 1183| 44| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 44| return _nameof; }() ------------------ 562| 44| c.fuzz(chardata); \ 563| 44| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE4fuzzENSt3__14spanIS5_Lm18446744073709551615EEE: 175| 44| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 44| const FromSpan from{reinterpret_cast(chardata.data()), 178| 44| chardata.size() / sizeof(FromType)}; 179| | 180| 44| static const bool do_print_testcase = 181| 44| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 44| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 44] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 44| do { 189| | // step 0 - is the input valid? 190| 44| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 44| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 44] | Branch (191:33): [True: 0, Folded] ------------------ 192| 0| break; 193| | 194| | // step 1 - count the input (only makes sense for some of the encodings) 195| | if constexpr (From == UtfEncodings::UTF16BE || 196| | From == UtfEncodings::UTF16LE || 197| | From == UtfEncodings::UTF8) { 198| | if (!count_the_input(from) && !allow_implementations_to_differ) 199| | break; 200| | } 201| | 202| | // step 2 - what is the required size of the output? 203| 44| const auto [output_length, length_agree] = 204| 44| calculate_length(from, inputisvalid); 205| 44| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 44] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 44| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 44] | Branch (208:28): [True: 0, False: 0] ------------------ 209| | // don't run the conversion step, it requires valid input 210| 0| return; 211| 0| } 212| | 213| | // step 3 - run the conversion 214| 44| const auto [written, outputs_agree] = 215| 44| do_conversion(from, output_length, inputisvalid); 216| 44| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 44] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 44| return; 221| 44| } while (0); ------------------ | Branch (221:14): [Folded, False: 0] ------------------ 222| | // if we come here, something failed 223| 0| std::cerr << "something failed, rerun with PRINT_FUZZ_CASE set to print a " 224| 0| "reproducer to stderr\n"; 225| 0| std::abort(); 226| 44| } _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS5_Lm18446744073709551615EEE: 283| 44| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 44| return validation_result{.valid = true, .implementations_agree = true}; 287| 44| } _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEb: 342| 44| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 44| length_result ret{}; 344| | 345| 44| const auto implementations = get_supported_implementations(); 346| 44| std::vector results; 347| 44| results.reserve(implementations.size()); 348| | 349| 132| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 132, False: 44] ------------------ 350| 132| const auto len = invoke_lengthcalc(impl, src); 351| 132| results.push_back(len); 352| 132| ret.length.push_back(len); 353| 132| } 354| | 355| 44| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 44| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 44] ------------------ 357| 0| std::cerr << "begin errormessage for calculate_length\n"; 358| 0| std::cerr << "in fuzz case invoking " << lengthcalcname << " with " 359| 0| << src.size() << " elements with valid input=" << inputisvalid 360| 0| << ":\n"; 361| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (361:31): [True: 0, False: 0] ------------------ 362| 0| std::cerr << "got return " << std::dec << results[i] 363| 0| << " from implementation " << implementations[i]->name() 364| 0| << '\n'; 365| 0| } 366| 0| std::cerr << "end errormessage\n"; 367| 0| if (inputisvalid) { ------------------ | Branch (367:11): [True: 0, False: 0] ------------------ 368| 0| ret.implementations_agree = false; 369| 0| } else { 370| 0| std::cerr 371| 0| << "implementations are allowed to disagree on invalid input\n"; 372| 0| ret.implementations_agree = true; 373| 0| } 374| 44| } else { 375| 44| ret.implementations_agree = true; 376| 44| } 377| 44| return ret; 378| 44| } _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_mEEEmSE_NSt3__14spanIS5_Lm18446744073709551615EEE: 338| 132| FromSpan src) const { 339| 132| return std::invoke(lengthcalc, impl, /*src.data(),*/ src.size()); 340| 132| } _ZZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE16calculate_lengthENSt3__14spanIS5_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 88| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 44| const bool inputisvalid) const { 383| 44| conversion_result ret{}; 384| | 385| 44| const auto implementations = get_supported_implementations(); 386| | 387| 44| std::vector> results; 388| 44| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 44| std::vector> outputbuffers; 393| 44| outputbuffers.reserve(implementations.size()); 394| 176| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 132, False: 44] ------------------ 395| 132| auto impl = implementations[i]; 396| 132| const ToType canary1{42}; 397| 132| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 132| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 132| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 132| const auto success = [](const ConversionResult& r) -> bool { 402| 132| if constexpr (std::is_same_v) { 403| 132| return r != 0; 404| 132| } else { 405| 132| return r.error == simdutf::error_code::SUCCESS; 406| 132| } 407| 132| }(implret1); 408| 132| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 132| if constexpr (use_canary_in_output) { 410| | // optionally convert again, this time with the buffer filled with 411| | // a different value. if the output differs, it means some of the buffer 412| | // was not written to by the conversion function. 413| 132| const ToType canary2{25}; 414| 132| const auto outputbuffer_first_run = outputbuffer; 415| 132| std::ranges::fill(outputbuffer, canary2); 416| 132| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 132| src.size(), outputbuffer.data()); 418| | 419| 132| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 132] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 132| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 132, False: 0] | Branch (423:29): [True: 126, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 126| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 126| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 126] ------------------ 427| 0| std::cerr << "different output the second time!\n"; 428| 0| std::cerr << "implementation " << impl->name() << " " << name 429| 0| << '\n'; 430| 0| std::cerr << "input is valid=" << inputisvalid << '\n'; 431| 0| std::cerr << "output length=" << outputbuffer.size() << '\n'; 432| 0| std::cerr << "conversion was a success? " << success << '\n'; 433| 0| for (std::size_t j = 0; j < outputbuffer.size(); ++j) { ------------------ | Branch (433:37): [True: 0, False: 0] ------------------ 434| 0| std::cerr << "output[" << j << "]\t" << +outputbuffer_first_run[j] 435| 0| << '\t' << +outputbuffer[j] << '\n'; 436| 0| } 437| 0| std::abort(); 438| 0| } 439| 126| } 440| 132| } 441| 132| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 126, False: 6] ------------------ 442| 132| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 44| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 44] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 44| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 44| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 44] ------------------ 454| 0| std::cerr << "begin errormessage for do_conversion\n"; 455| 0| std::cerr << "in fuzz case for " << name << " invoked with " << src.size() 456| 0| << " elements:\n"; 457| 0| std::cerr << "input data is valid ? " << inputisvalid << '\n'; 458| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (458:31): [True: 0, False: 0] ------------------ 459| 0| std::cerr << "got return " << std::dec << results[i] 460| 0| << " from implementation " << implementations[i]->name() 461| 0| << " using outlen=" << outlength.at(i) << '\n'; 462| 0| } 463| 0| for (std::size_t i = 0; i < results.size(); ++i) { ------------------ | Branch (463:31): [True: 0, False: 0] ------------------ 464| 0| std::cerr << "implementation " << implementations[i]->name() 465| 0| << " out: "; 466| 0| for (const auto e : outputbuffers.at(i)) { ------------------ | Branch (466:27): [True: 0, False: 0] ------------------ 467| 0| std::cerr << +e << ", "; 468| 0| } 469| 0| std::cerr << '\n'; 470| 0| } 471| 0| std::cerr << "end errormessage\n"; 472| 0| ret.implementations_agree = false; 473| 44| } else { 474| 44| ret.implementations_agree = true; 475| 44| } 476| 44| return ret; 477| 44| } _ZZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 132| const auto success = [](const ConversionResult& r) -> bool { 402| 132| if constexpr (std::is_same_v) { 403| 132| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 132| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_1EMN7simdutf14implementationEKDoFmmEMS2_KDoFmPKcmPDsEE13do_conversionENSt3__14spanIS5_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 88| auto neq = [](const auto& a, const auto& b) { return a != b; }; conversion.cpp:_ZZ18populate_functionsvENK4$_41clENSt3__14spanIKcLm18446744073709551615EEE: 554| 242| +[](std::span chardata) { \ 555| 242| const auto c = \ 556| 242| Conversion{ \ 559| 242| &I::lenfunc, &I::conversionfunc, \ 560| 242| std::string{NAMEOF(&I::lenfunc)}, \ ------------------ | | 1178| 242|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 242| ::std::void_t(); \ | | 1180| 242| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 242| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 242| constexpr auto _size = _name.size(); \ | | 1183| 242| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 242| return _nameof; }() ------------------ 561| 242| std::string{NAMEOF(&I::conversionfunc)}}; \ ------------------ | | 1178| 242|#define NAMEOF(...) []() constexpr noexcept { \ | | 1179| 242| ::std::void_t(); \ | | 1180| 242| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ | | 1181| 242| static_assert(!_name.empty(), "Expression does not have a name."); \ | | 1182| 242| constexpr auto _size = _name.size(); \ | | 1183| 242| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ | | 1184| 242| return _nameof; }() ------------------ 562| 242| c.fuzz(chardata); \ 563| 242| } \ _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE4fuzzENSt3__14spanIS3_Lm18446744073709551615EEE: 175| 242| void fuzz(std::span chardata) const { 176| | // assume the input is aligned to FromType 177| 242| const FromSpan from{reinterpret_cast(chardata.data()), 178| 242| chardata.size() / sizeof(FromType)}; 179| | 180| 242| static const bool do_print_testcase = 181| 242| std::getenv("PRINT_FUZZ_CASE") != nullptr; 182| | 183| 242| if (do_print_testcase) { ------------------ | Branch (183:9): [True: 0, False: 242] ------------------ 184| 0| dump_testcase(from, std::cerr); 185| 0| std::exit(EXIT_SUCCESS); 186| 0| } 187| | 188| 242| do { 189| | // step 0 - is the input valid? 190| 242| const auto [inputisvalid, valid_input_agree] = verify_valid_input(from); 191| 242| if (!valid_input_agree && !allow_implementations_to_differ) ------------------ | Branch (191:11): [True: 0, False: 242] | 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| 242| const auto [output_length, length_agree] = 204| 242| calculate_length(from, inputisvalid); 205| 242| if (!length_agree && !allow_implementations_to_differ) ------------------ | Branch (205:11): [True: 0, False: 242] | Branch (205:28): [True: 0, Folded] ------------------ 206| 0| break; 207| | 208| 242| if (!inputisvalid && name.find("valid") != std::string::npos) { ------------------ | Branch (208:11): [True: 0, False: 242] | 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| 242| const auto [written, outputs_agree] = 215| 242| do_conversion(from, output_length, inputisvalid); 216| 242| if (!outputs_agree && !allow_implementations_to_differ) ------------------ | Branch (216:11): [True: 0, False: 242] | Branch (216:29): [True: 0, Folded] ------------------ 217| 0| break; 218| | 219| | // coming this far means no problems were found 220| 242| return; 221| 242| } 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| 242| } _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE18verify_valid_inputIvQeqT_LS0_4EEENSA_17validation_resultENSt3__14spanIS3_Lm18446744073709551615EEE: 283| 242| validation_result verify_valid_input(FromSpan) const { 284| | // all latin1 input is valid. there is no simdutf validation function for 285| | // it. 286| 242| return validation_result{.valid = true, .implementations_agree = true}; 287| 242| } _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEb: 342| 242| length_result calculate_length(FromSpan src, const bool inputisvalid) const { 343| 242| length_result ret{}; 344| | 345| 242| const auto implementations = get_supported_implementations(); 346| 242| std::vector results; 347| 242| results.reserve(implementations.size()); 348| | 349| 726| for (auto impl : implementations) { ------------------ | Branch (349:20): [True: 726, False: 242] ------------------ 350| 726| const auto len = invoke_lengthcalc(impl, src); 351| 726| results.push_back(len); 352| 726| ret.length.push_back(len); 353| 726| } 354| | 355| 242| auto neq = [](const auto& a, const auto& b) { return a != b; }; 356| 242| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (356:9): [True: 0, False: 242] ------------------ 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| 242| } else { 375| 242| ret.implementations_agree = true; 376| 242| } 377| 242| return ret; 378| 242| } _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE17invoke_lengthcalcIvQsr3stdE14is_invocable_vIT1_PKS2_PKN23ValidationFunctionTraitIXT_EE7RawTypeEmEEEmSE_NSt3__14spanIS3_Lm18446744073709551615EEE: 330| 726| FromSpan src) const { 331| 726| return std::invoke(lengthcalc, impl, src.data(), src.size()); 332| 726| } _ZZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE16calculate_lengthENSt3__14spanIS3_Lm18446744073709551615EEEbENKUlRKT_RKT0_E_clImmEEDaSG_SJ_: 355| 484| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEb: 382| 242| const bool inputisvalid) const { 383| 242| conversion_result ret{}; 384| | 385| 242| const auto implementations = get_supported_implementations(); 386| | 387| 242| std::vector> results; 388| 242| results.reserve(implementations.size()); 389| | 390| | // put the output in a separate allocation to make access violations easier 391| | // to catch 392| 242| std::vector> outputbuffers; 393| 242| outputbuffers.reserve(implementations.size()); 394| 968| for (std::size_t i = 0; i < implementations.size(); ++i) { ------------------ | Branch (394:29): [True: 726, False: 242] ------------------ 395| 726| auto impl = implementations[i]; 396| 726| const ToType canary1{42}; 397| 726| auto& outputbuffer = outputbuffers.emplace_back(outlength.at(i), canary1); 398| 726| const auto implret1 = std::invoke(conversion, impl, src.data(), 399| 726| src.size(), outputbuffer.data()); 400| | // was the conversion successful? 401| 726| const auto success = [](const ConversionResult& r) -> bool { 402| 726| if constexpr (std::is_same_v) { 403| 726| return r != 0; 404| 726| } else { 405| 726| return r.error == simdutf::error_code::SUCCESS; 406| 726| } 407| 726| }(implret1); 408| 726| const auto hash1 = FNV1A_hash::as_str(outputbuffer); 409| 726| 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| 726| const ToType canary2{25}; 414| 726| const auto outputbuffer_first_run = outputbuffer; 415| 726| std::ranges::fill(outputbuffer, canary2); 416| 726| const auto implret2 = std::invoke(conversion, impl, src.data(), 417| 726| src.size(), outputbuffer.data()); 418| | 419| 726| if (implret1 != implret2) { ------------------ | Branch (419:13): [True: 0, False: 726] ------------------ 420| 0| std::cerr << "different return value the second time!\n"; 421| 0| std::abort(); 422| 0| } 423| 726| if (inputisvalid && success) { ------------------ | Branch (423:13): [True: 726, False: 0] | Branch (423:29): [True: 720, False: 6] ------------------ 424| | // only care about the output if the input is valid 425| 720| const auto hash2 = FNV1A_hash::as_str(outputbuffer); 426| 720| if (hash1 != hash2) { ------------------ | Branch (426:15): [True: 0, False: 720] ------------------ 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| 720| } 440| 726| } 441| 726| results.emplace_back(implret1, success ? hash1 : ""); ------------------ | Branch (441:38): [True: 720, False: 6] ------------------ 442| 726| } 443| | 444| | // do not require implementations to give the same output if 445| | // the input is not valid. 446| 242| if (!inputisvalid) { ------------------ | Branch (446:9): [True: 0, False: 242] ------------------ 447| 0| for (auto& e : results) { ------------------ | Branch (447:20): [True: 0, False: 0] ------------------ 448| 0| e.outputhash.clear(); 449| 0| } 450| 0| } 451| | 452| 242| auto neq = [](const auto& a, const auto& b) { return a != b; }; 453| 242| if (std::ranges::adjacent_find(results, neq) != results.end()) { ------------------ | Branch (453:9): [True: 0, False: 242] ------------------ 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| 242| } else { 474| 242| ret.implementations_agree = true; 475| 242| } 476| 242| return ret; 477| 242| } _ZZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKmE_clESL_: 401| 726| const auto success = [](const ConversionResult& r) -> bool { 402| 726| if constexpr (std::is_same_v) { 403| 726| return r != 0; 404| | } else { 405| | return r.error == simdutf::error_code::SUCCESS; 406| | } 407| 726| }(implret1); _ZZNK10ConversionIL12UtfEncodings4ELS0_2EMN7simdutf14implementationEKDoFmPKcmEMS2_KDoFmS4_mPcEE13do_conversionENSt3__14spanIS3_Lm18446744073709551615EEERKNSB_6vectorImNSB_9allocatorImEEEEbENKUlRKT_RKT0_E_clI6resultImEST_EEDaSM_SP_: 452| 484| auto neq = [](const auto& a, const auto& b) { return a != b; }; _ZN10FNV1A_hash14fnv1ahash_implENSt3__14spanIKcLm18446744073709551615EEE: 75| 17.3k| static constexpr std::uint64_t fnv1ahash_impl(std::span bytes) { 76| 17.3k| auto hash = offset; 77| | 78| 589M| for (auto byte : bytes) { ------------------ | Branch (78:20): [True: 589M, False: 17.3k] ------------------ 79| 589M| hash ^= static_cast(byte); 80| 589M| hash *= prime; 81| 589M| } 82| | 83| 17.3k| return hash; 84| 17.3k| } _ZN10FNV1A_hash14fnv1ahash_implIKDsLm18446744073709551615EQaaaa11is_hashableIT_Entsr3stdE9is_same_vIS2_cEntsr3stdE9is_same_vIS2_hEEEmNSt3__14spanIS2_XT0_EEE: 89| 14.6k| static constexpr std::uint64_t fnv1ahash_impl(std::span data) { 90| 14.6k| return fnv1ahash_impl({reinterpret_cast(data.data()), 91| 14.6k| data.size_bytes()}); 92| 14.6k| } _ZN10FNV1A_hash14fnv1ahash_implENSt3__14spanIKhLm18446744073709551615EEE: 65| 23.4k| fnv1ahash_impl(std::span bytes) { 66| 23.4k| auto hash = offset; 67| | 68| 2.13G| for (std::uint64_t byte : bytes) { ------------------ | Branch (68:29): [True: 2.13G, False: 23.4k] ------------------ 69| 2.13G| hash ^= byte; 70| 2.13G| hash *= prime; 71| 2.13G| } 72| | 73| 23.4k| return hash; 74| 23.4k| } _Z29get_supported_implementationsv: 26| 33.4k|get_supported_implementations() { 27| 33.4k| static const auto impl = []() -> auto { 28| 33.4k| std::vector ret; 29| 33.4k| for (auto e : simdutf::get_available_implementations()) { 30| 33.4k| std::cerr << "implementation " << e->name() << " is available? " 31| 33.4k| << e->supported_by_runtime_system() << '\n'; 32| 33.4k| if (e->supported_by_runtime_system()) { 33| 33.4k| ret.push_back(e); 34| 33.4k| } 35| 33.4k| } 36| 33.4k| return ret; 37| 33.4k| }(); 38| 33.4k| return {impl.data(), impl.size()}; 39| 33.4k|} _ZZ29get_supported_implementationsvENKUlvE_clEv: 27| 1| static const auto impl = []() -> auto { 28| 1| std::vector ret; 29| 4| for (auto e : simdutf::get_available_implementations()) { ------------------ | Branch (29:17): [True: 4, False: 1] ------------------ 30| 4| std::cerr << "implementation " << e->name() << " is available? " 31| 4| << e->supported_by_runtime_system() << '\n'; 32| 4| if (e->supported_by_runtime_system()) { ------------------ | Branch (32:11): [True: 3, False: 1] ------------------ 33| 3| ret.push_back(e); 34| 3| } 35| 4| } 36| 1| return ret; 37| 1| }(); _ZneRKN7simdutf6resultES2_: 42| 29.6k|inline bool operator!=(const simdutf::result& a, const simdutf::result& b) { 43| 29.6k| return a.count != b.count || a.error != b.error; ------------------ | Branch (43:10): [True: 0, False: 29.6k] | Branch (43:32): [True: 0, False: 29.6k] ------------------ 44| 29.6k|} _ZN10FNV1A_hash6as_strIJNSt3__16vectorIDiNS1_9allocatorIDiEEEEEEENS1_12basic_stringIcNS1_11char_traitsIcEENS3_IcEEEEDpRKT_: 100| 8.78k| template static std::string as_str(const Data&... data) { 101| 8.78k| static_assert(sizeof...(Data) > 0, "must hash with at least one argument"); 102| 8.78k| std::uint64_t h; 103| | if constexpr (sizeof...(Data) > 1) { 104| | const std::array hashes{fnv1ahash_impl(data)...}; 105| | const auto s = std::span(hashes); 106| | h = fnv1ahash_impl(s); 107| 8.78k| } else { 108| 8.78k| h = fnv1ahash_impl(data...); 109| 8.78k| } 110| 8.78k| constexpr std::size_t expected_chars = 16; 111| 8.78k| std::string ret(expected_chars, '0'); 112| 8.78k| auto c = std::to_chars(ret.data(), ret.data() + ret.size(), h, 16); 113| 8.78k| assert(c.ec == std::errc{}); ------------------ | Branch (113:5): [True: 8.78k, False: 0] ------------------ 114| 8.78k| auto nwritten = c.ptr - ret.data(); 115| 8.78k| assert(nwritten <= expected_chars); ------------------ | Branch (115:5): [True: 8.78k, False: 0] ------------------ 116| 8.78k| std::rotate(ret.data(), c.ptr, ret.data() + expected_chars); 117| 8.78k| return ret; 118| 8.78k| } _ZN10FNV1A_hash14fnv1ahash_implIDiQ11is_hashableIT_EEEmRKNSt3__16vectorIS1_NS2_9allocatorIS1_EEEE: 96| 8.78k| static constexpr std::uint64_t fnv1ahash_impl(const std::vector& data) { 97| 8.78k| return fnv1ahash_impl(std::span(data)); 98| 8.78k| } _ZN10FNV1A_hash14fnv1ahash_implIKDiLm18446744073709551615EQaaaa11is_hashableIT_Entsr3stdE9is_same_vIS2_cEntsr3stdE9is_same_vIS2_hEEEmNSt3__14spanIS2_XT0_EEE: 89| 8.78k| static constexpr std::uint64_t fnv1ahash_impl(std::span data) { 90| 8.78k| return fnv1ahash_impl({reinterpret_cast(data.data()), 91| 8.78k| data.size_bytes()}); 92| 8.78k| } _ZN10FNV1A_hash6as_strIJNSt3__16vectorIcNS1_9allocatorIcEEEEEEENS1_12basic_stringIcNS1_11char_traitsIcEES4_EEDpRKT_: 100| 17.3k| template static std::string as_str(const Data&... data) { 101| 17.3k| static_assert(sizeof...(Data) > 0, "must hash with at least one argument"); 102| 17.3k| std::uint64_t h; 103| | if constexpr (sizeof...(Data) > 1) { 104| | const std::array hashes{fnv1ahash_impl(data)...}; 105| | const auto s = std::span(hashes); 106| | h = fnv1ahash_impl(s); 107| 17.3k| } else { 108| 17.3k| h = fnv1ahash_impl(data...); 109| 17.3k| } 110| 17.3k| constexpr std::size_t expected_chars = 16; 111| 17.3k| std::string ret(expected_chars, '0'); 112| 17.3k| auto c = std::to_chars(ret.data(), ret.data() + ret.size(), h, 16); 113| 17.3k| assert(c.ec == std::errc{}); ------------------ | Branch (113:5): [True: 17.3k, False: 0] ------------------ 114| 17.3k| auto nwritten = c.ptr - ret.data(); 115| 17.3k| assert(nwritten <= expected_chars); ------------------ | Branch (115:5): [True: 17.3k, False: 0] ------------------ 116| 17.3k| std::rotate(ret.data(), c.ptr, ret.data() + expected_chars); 117| 17.3k| return ret; 118| 17.3k| } _ZN10FNV1A_hash14fnv1ahash_implIcQ11is_hashableIT_EEEmRKNSt3__16vectorIS1_NS2_9allocatorIS1_EEEE: 96| 17.3k| static constexpr std::uint64_t fnv1ahash_impl(const std::vector& data) { 97| 17.3k| return fnv1ahash_impl(std::span(data)); 98| 17.3k| } _ZN10FNV1A_hash6as_strIJNSt3__16vectorIDsNS1_9allocatorIDsEEEEEEENS1_12basic_stringIcNS1_11char_traitsIcEENS3_IcEEEEDpRKT_: 100| 14.6k| template static std::string as_str(const Data&... data) { 101| 14.6k| static_assert(sizeof...(Data) > 0, "must hash with at least one argument"); 102| 14.6k| std::uint64_t h; 103| | if constexpr (sizeof...(Data) > 1) { 104| | const std::array hashes{fnv1ahash_impl(data)...}; 105| | const auto s = std::span(hashes); 106| | h = fnv1ahash_impl(s); 107| 14.6k| } else { 108| 14.6k| h = fnv1ahash_impl(data...); 109| 14.6k| } 110| 14.6k| constexpr std::size_t expected_chars = 16; 111| 14.6k| std::string ret(expected_chars, '0'); 112| 14.6k| auto c = std::to_chars(ret.data(), ret.data() + ret.size(), h, 16); 113| 14.6k| assert(c.ec == std::errc{}); ------------------ | Branch (113:5): [True: 14.6k, False: 0] ------------------ 114| 14.6k| auto nwritten = c.ptr - ret.data(); 115| 14.6k| assert(nwritten <= expected_chars); ------------------ | Branch (115:5): [True: 14.6k, False: 0] ------------------ 116| 14.6k| std::rotate(ret.data(), c.ptr, ret.data() + expected_chars); 117| 14.6k| return ret; 118| 14.6k| } _ZN10FNV1A_hash14fnv1ahash_implIDsQ11is_hashableIT_EEEmRKNSt3__16vectorIS1_NS2_9allocatorIS1_EEEE: 96| 14.6k| static constexpr std::uint64_t fnv1ahash_impl(const std::vector& data) { 97| 14.6k| return fnv1ahash_impl(std::span(data)); 98| 14.6k| } _ZeqRKN7simdutf6resultES2_: 45| 8.01k|inline bool operator==(const simdutf::result& a, const simdutf::result& b) { 46| 8.01k| return a.count == b.count && a.error == b.error; ------------------ | Branch (46:10): [True: 8.01k, False: 0] | Branch (46:32): [True: 8.01k, False: 0] ------------------ 47| 8.01k|} _ZNK23ValidationFunctionTraitIL12UtfEncodings0EE24ValidationWithErrorsNameMUlvE_clEv: 1178| 2|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 2| ::std::void_t(); \ 1180| 2| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 2| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 2| constexpr auto _size = _name.size(); \ 1183| 2| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 2| return _nameof; }() _ZNK6nameof7cstringILt28EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 4| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt28EE4dataEv: 233| 4| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt28EE4sizeEv: 235| 4| [[nodiscard]] constexpr size_type size() const noexcept { return N; } _ZNK23ValidationFunctionTraitIL12UtfEncodings1EE24ValidationWithErrorsNameMUlvE_clEv: 1178| 2|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 2| ::std::void_t(); \ 1180| 2| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 2| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 2| constexpr auto _size = _name.size(); \ 1183| 2| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 2| return _nameof; }() _ZNK23ValidationFunctionTraitIL12UtfEncodings3EE24ValidationWithErrorsNameMUlvE_clEv: 1178| 2|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 2| ::std::void_t(); \ 1180| 2| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 2| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 2| constexpr auto _size = _name.size(); \ 1183| 2| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 2| return _nameof; }() _ZNK6nameof7cstringILt26EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 2| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt26EE4dataEv: 233| 2| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt26EE4sizeEv: 235| 2| [[nodiscard]] constexpr size_type size() const noexcept { return N; } _ZNK23ValidationFunctionTraitIL12UtfEncodings2EE24ValidationWithErrorsNameMUlvE_clEv: 1178| 2|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 2| ::std::void_t(); \ 1180| 2| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 2| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 2| constexpr auto _size = _name.size(); \ 1183| 2| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 2| return _nameof; }() _ZNK6nameof7cstringILt25EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 1.10k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt25EE4dataEv: 233| 1.10k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt25EE4sizeEv: 235| 1.10k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_0clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 96|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 96| ::std::void_t(); \ 1180| 96| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 96| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 96| constexpr auto _size = _name.size(); \ 1183| 96| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 96| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_0clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 96|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 96| ::std::void_t(); \ 1180| 96| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 96| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 96| constexpr auto _size = _name.size(); \ 1183| 96| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 96| return _nameof; }() _ZNK6nameof7cstringILt30EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 358| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt30EE4dataEv: 233| 358| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt30EE4sizeEv: 235| 358| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_1clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 224|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 224| ::std::void_t(); \ 1180| 224| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 224| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 224| constexpr auto _size = _name.size(); \ 1183| 224| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 224| return _nameof; }() _ZNK6nameof7cstringILt24EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 3.09k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt24EE4dataEv: 233| 3.09k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt24EE4sizeEv: 235| 3.09k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_1clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 224|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 224| ::std::void_t(); \ 1180| 224| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 224| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 224| constexpr auto _size = _name.size(); \ 1183| 224| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 224| return _nameof; }() _ZNK6nameof7cstringILt29EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 900| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt29EE4dataEv: 233| 900| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt29EE4sizeEv: 235| 900| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_2clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 65|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 65| ::std::void_t(); \ 1180| 65| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 65| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 65| constexpr auto _size = _name.size(); \ 1183| 65| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 65| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_2clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 65|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 65| ::std::void_t(); \ 1180| 65| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 65| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 65| constexpr auto _size = _name.size(); \ 1183| 65| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 65| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_3clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 205|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 205| ::std::void_t(); \ 1180| 205| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 205| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 205| constexpr auto _size = _name.size(); \ 1183| 205| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 205| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_3clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 205|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 205| ::std::void_t(); \ 1180| 205| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 205| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 205| constexpr auto _size = _name.size(); \ 1183| 205| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 205| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_4clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 61|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 61| ::std::void_t(); \ 1180| 61| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 61| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 61| constexpr auto _size = _name.size(); \ 1183| 61| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 61| return _nameof; }() _ZNK6nameof7cstringILt23EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 3.49k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt23EE4dataEv: 233| 3.49k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt23EE4sizeEv: 235| 3.49k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_4clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 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$_5clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 136|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 136| ::std::void_t(); \ 1180| 136| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 136| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 136| constexpr auto _size = _name.size(); \ 1183| 136| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 136| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_5clENSt3__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_functionsvENK3$_6clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 100|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 100| ::std::void_t(); \ 1180| 100| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 100| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 100| constexpr auto _size = _name.size(); \ 1183| 100| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 100| return _nameof; }() _ZNK6nameof7cstringILt22EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 4.40k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt22EE4dataEv: 233| 4.40k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt22EE4sizeEv: 235| 4.40k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_6clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 100|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 100| ::std::void_t(); \ 1180| 100| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 100| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 100| constexpr auto _size = _name.size(); \ 1183| 100| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 100| return _nameof; }() _ZNK6nameof7cstringILt27EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 326| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt27EE4dataEv: 233| 326| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt27EE4sizeEv: 235| 326| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK3$_7clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 240|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 240| ::std::void_t(); \ 1180| 240| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 240| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 240| constexpr auto _size = _name.size(); \ 1183| 240| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 240| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_7clENSt3__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; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_8clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 231|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 231| ::std::void_t(); \ 1180| 231| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 231| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 231| constexpr auto _size = _name.size(); \ 1183| 231| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 231| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_8clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 231|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 231| ::std::void_t(); \ 1180| 231| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 231| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 231| constexpr auto _size = _name.size(); \ 1183| 231| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 231| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_9clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 226|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 226| ::std::void_t(); \ 1180| 226| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 226| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 226| constexpr auto _size = _name.size(); \ 1183| 226| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 226| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK3$_9clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 226|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 226| ::std::void_t(); \ 1180| 226| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 226| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 226| constexpr auto _size = _name.size(); \ 1183| 226| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 226| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_10clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 56|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 56| ::std::void_t(); \ 1180| 56| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 56| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 56| constexpr auto _size = _name.size(); \ 1183| 56| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 56| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_10clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 56|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 56| ::std::void_t(); \ 1180| 56| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 56| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 56| constexpr auto _size = _name.size(); \ 1183| 56| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 56| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_11clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 141|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 141| ::std::void_t(); \ 1180| 141| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 141| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 141| constexpr auto _size = _name.size(); \ 1183| 141| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 141| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_11clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 141|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 141| ::std::void_t(); \ 1180| 141| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 141| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 141| constexpr auto _size = _name.size(); \ 1183| 141| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 141| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_12clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 179|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 179| ::std::void_t(); \ 1180| 179| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 179| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 179| constexpr auto _size = _name.size(); \ 1183| 179| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 179| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_12clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 179|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 179| ::std::void_t(); \ 1180| 179| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 179| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 179| constexpr auto _size = _name.size(); \ 1183| 179| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 179| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_13clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 75|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 75| ::std::void_t(); \ 1180| 75| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 75| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 75| constexpr auto _size = _name.size(); \ 1183| 75| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 75| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_13clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 75|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 75| ::std::void_t(); \ 1180| 75| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 75| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 75| constexpr auto _size = _name.size(); \ 1183| 75| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 75| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_14clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 154|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 154| ::std::void_t(); \ 1180| 154| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 154| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 154| constexpr auto _size = _name.size(); \ 1183| 154| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 154| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_14clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 154|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 154| ::std::void_t(); \ 1180| 154| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 154| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 154| constexpr auto _size = _name.size(); \ 1183| 154| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 154| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_15clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 246|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 246| ::std::void_t(); \ 1180| 246| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 246| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 246| constexpr auto _size = _name.size(); \ 1183| 246| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 246| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_15clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 246|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 246| ::std::void_t(); \ 1180| 246| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 246| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 246| constexpr auto _size = _name.size(); \ 1183| 246| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 246| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_16clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 103|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 103| ::std::void_t(); \ 1180| 103| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 103| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 103| constexpr auto _size = _name.size(); \ 1183| 103| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 103| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_16clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 103|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 103| ::std::void_t(); \ 1180| 103| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 103| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 103| constexpr auto _size = _name.size(); \ 1183| 103| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 103| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_17clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 213|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 213| ::std::void_t(); \ 1180| 213| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 213| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 213| constexpr auto _size = _name.size(); \ 1183| 213| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 213| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_17clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 213|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 213| ::std::void_t(); \ 1180| 213| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 213| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 213| constexpr auto _size = _name.size(); \ 1183| 213| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 213| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_18clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 240|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 240| ::std::void_t(); \ 1180| 240| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 240| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 240| constexpr auto _size = _name.size(); \ 1183| 240| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 240| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_18clENSt3__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; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_19clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 342|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 342| ::std::void_t(); \ 1180| 342| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 342| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 342| constexpr auto _size = _name.size(); \ 1183| 342| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 342| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_19clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 342|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 342| ::std::void_t(); \ 1180| 342| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 342| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 342| constexpr auto _size = _name.size(); \ 1183| 342| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 342| return _nameof; }() _ZNK6nameof7cstringILt21EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 751| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt21EE4dataEv: 233| 751| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt21EE4sizeEv: 235| 751| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_20clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 306|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 306| ::std::void_t(); \ 1180| 306| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 306| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 306| constexpr auto _size = _name.size(); \ 1183| 306| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 306| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_20clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 306|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 306| ::std::void_t(); \ 1180| 306| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 306| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 306| constexpr auto _size = _name.size(); \ 1183| 306| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 306| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_21clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 379|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 379| ::std::void_t(); \ 1180| 379| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 379| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 379| constexpr auto _size = _name.size(); \ 1183| 379| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 379| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_21clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 379|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 379| ::std::void_t(); \ 1180| 379| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 379| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 379| constexpr auto _size = _name.size(); \ 1183| 379| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 379| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_22clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 392|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 392| ::std::void_t(); \ 1180| 392| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 392| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 392| constexpr auto _size = _name.size(); \ 1183| 392| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 392| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_22clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 392|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 392| ::std::void_t(); \ 1180| 392| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 392| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 392| constexpr auto _size = _name.size(); \ 1183| 392| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 392| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_23clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 409|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 409| ::std::void_t(); \ 1180| 409| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 409| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 409| constexpr auto _size = _name.size(); \ 1183| 409| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 409| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_23clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 409|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 409| ::std::void_t(); \ 1180| 409| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 409| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 409| constexpr auto _size = _name.size(); \ 1183| 409| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 409| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_24clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 115|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 115| ::std::void_t(); \ 1180| 115| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 115| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 115| constexpr auto _size = _name.size(); \ 1183| 115| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 115| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_24clENSt3__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; }() _ZNK6nameof7cstringILt37EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 228| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt37EE4dataEv: 233| 228| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt37EE4sizeEv: 235| 228| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_25clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 211|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 211| ::std::void_t(); \ 1180| 211| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 211| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 211| constexpr auto _size = _name.size(); \ 1183| 211| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 211| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_25clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 211|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 211| ::std::void_t(); \ 1180| 211| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 211| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 211| constexpr auto _size = _name.size(); \ 1183| 211| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 211| return _nameof; }() _ZNK6nameof7cstringILt36EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 1.09k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt36EE4dataEv: 233| 1.09k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt36EE4sizeEv: 235| 1.09k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_26clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 324|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 324| ::std::void_t(); \ 1180| 324| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 324| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 324| constexpr auto _size = _name.size(); \ 1183| 324| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 324| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_26clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 324|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 324| ::std::void_t(); \ 1180| 324| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 324| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 324| constexpr auto _size = _name.size(); \ 1183| 324| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 324| return _nameof; }() _ZNK6nameof7cstringILt35EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 1.56k| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt35EE4dataEv: 233| 1.56k| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt35EE4sizeEv: 235| 1.56k| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_27clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 113|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 113| ::std::void_t(); \ 1180| 113| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 113| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 113| constexpr auto _size = _name.size(); \ 1183| 113| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 113| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_27clENSt3__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$_28clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 201|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 201| ::std::void_t(); \ 1180| 201| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 201| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 201| constexpr auto _size = _name.size(); \ 1183| 201| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 201| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_28clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 201|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 201| ::std::void_t(); \ 1180| 201| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 201| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 201| constexpr auto _size = _name.size(); \ 1183| 201| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 201| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_29clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 327|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 327| ::std::void_t(); \ 1180| 327| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 327| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 327| constexpr auto _size = _name.size(); \ 1183| 327| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 327| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_29clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 327|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 327| ::std::void_t(); \ 1180| 327| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 327| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 327| constexpr auto _size = _name.size(); \ 1183| 327| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 327| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_30clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 227|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 227| ::std::void_t(); \ 1180| 227| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 227| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 227| constexpr auto _size = _name.size(); \ 1183| 227| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 227| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_30clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 227|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 227| ::std::void_t(); \ 1180| 227| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 227| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 227| constexpr auto _size = _name.size(); \ 1183| 227| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 227| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_31clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 330|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 330| ::std::void_t(); \ 1180| 330| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 330| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 330| constexpr auto _size = _name.size(); \ 1183| 330| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 330| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_31clENSt3__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$_32clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 357|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 357| ::std::void_t(); \ 1180| 357| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 357| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 357| constexpr auto _size = _name.size(); \ 1183| 357| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 357| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_32clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 357|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 357| ::std::void_t(); \ 1180| 357| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 357| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 357| constexpr auto _size = _name.size(); \ 1183| 357| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 357| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_33clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 462|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 462| ::std::void_t(); \ 1180| 462| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 462| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 462| constexpr auto _size = _name.size(); \ 1183| 462| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 462| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_33clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 462|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 462| ::std::void_t(); \ 1180| 462| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 462| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 462| constexpr auto _size = _name.size(); \ 1183| 462| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 462| return _nameof; }() _ZNK6nameof7cstringILt33EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 857| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt33EE4dataEv: 233| 857| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt33EE4sizeEv: 235| 857| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_34clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 261|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 261| ::std::void_t(); \ 1180| 261| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 261| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 261| constexpr auto _size = _name.size(); \ 1183| 261| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 261| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_34clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 261|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 261| ::std::void_t(); \ 1180| 261| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 261| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 261| constexpr auto _size = _name.size(); \ 1183| 261| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 261| return _nameof; }() _ZNK6nameof7cstringILt34EEcvNSt3__117basic_string_viewIcNS2_11char_traitsIcEEEEEv: 269| 261| [[nodiscard]] constexpr operator string_view() const noexcept { return {data(), size()}; } _ZNK6nameof7cstringILt34EE4dataEv: 233| 261| [[nodiscard]] constexpr const_pointer data() const noexcept { return chars_; } _ZNK6nameof7cstringILt34EE4sizeEv: 235| 261| [[nodiscard]] constexpr size_type size() const noexcept { return N; } conversion.cpp:_ZZZ18populate_functionsvENK4$_35clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 386|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 386| ::std::void_t(); \ 1180| 386| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 386| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 386| constexpr auto _size = _name.size(); \ 1183| 386| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 386| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_35clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 386|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 386| ::std::void_t(); \ 1180| 386| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 386| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 386| constexpr auto _size = _name.size(); \ 1183| 386| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 386| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_36clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 297|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 297| ::std::void_t(); \ 1180| 297| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 297| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 297| constexpr auto _size = _name.size(); \ 1183| 297| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 297| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_36clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 297|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 297| ::std::void_t(); \ 1180| 297| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 297| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 297| constexpr auto _size = _name.size(); \ 1183| 297| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 297| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_37clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 395|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 395| ::std::void_t(); \ 1180| 395| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 395| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 395| constexpr auto _size = _name.size(); \ 1183| 395| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 395| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_37clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 395|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 395| ::std::void_t(); \ 1180| 395| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 395| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 395| constexpr auto _size = _name.size(); \ 1183| 395| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 395| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_38clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 50|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 50| ::std::void_t(); \ 1180| 50| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 50| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 50| constexpr auto _size = _name.size(); \ 1183| 50| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 50| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_38clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 50|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 50| ::std::void_t(); \ 1180| 50| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 50| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 50| constexpr auto _size = _name.size(); \ 1183| 50| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 50| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_39clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 55|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 55| ::std::void_t(); \ 1180| 55| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 55| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 55| constexpr auto _size = _name.size(); \ 1183| 55| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 55| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_39clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 55|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 55| ::std::void_t(); \ 1180| 55| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 55| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 55| constexpr auto _size = _name.size(); \ 1183| 55| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 55| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_40clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 44|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 44| ::std::void_t(); \ 1180| 44| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 44| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 44| constexpr auto _size = _name.size(); \ 1183| 44| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 44| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_40clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 44|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 44| ::std::void_t(); \ 1180| 44| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 44| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 44| constexpr auto _size = _name.size(); \ 1183| 44| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 44| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_41clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE1_clEv: 1178| 242|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 242| ::std::void_t(); \ 1180| 242| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 242| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 242| constexpr auto _size = _name.size(); \ 1183| 242| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 242| return _nameof; }() conversion.cpp:_ZZZ18populate_functionsvENK4$_41clENSt3__14spanIKcLm18446744073709551615EEEENKUlvE2_clEv: 1178| 242|#define NAMEOF(...) []() constexpr noexcept { \ 1179| 242| ::std::void_t(); \ 1180| 242| constexpr auto _name = ::nameof::detail::pretty_name(#__VA_ARGS__); \ 1181| 242| static_assert(!_name.empty(), "Expression does not have a name."); \ 1182| 242| constexpr auto _size = _name.size(); \ 1183| 242| constexpr auto _nameof = ::nameof::cstring<_size>{_name}; \ 1184| 242| return _nameof; }() _ZN7simdutf6resultC2ENS_10error_codeEm: 87| 78.3k| : error{err}, count{pos} {} _ZNK7simdutf6result6is_errEv: 93| 5.40k| simdutf_really_inline simdutf_constexpr23 bool is_err() const noexcept { 94| 5.40k| return error != error_code::SUCCESS; 95| 5.40k| } _ZN7simdutf11full_resultC2ENS_10error_codeEmm: 111| 5.89k| : error{err}, input_count{pos_in}, output_count{pos_out} {} _ZNK7simdutf11full_resultcvNS_6resultEEv: 117| 3.90k| simdutf_really_inline simdutf_constexpr23 operator result() const noexcept { 118| 3.90k| if (error == error_code::SUCCESS) { ------------------ | Branch (118:9): [True: 2.38k, False: 1.51k] ------------------ 119| 2.38k| return result{error, output_count}; 120| 2.38k| } else { 121| 1.51k| return result{error, input_count}; 122| 1.51k| } 123| 3.90k| } _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| 297| utf16_length_from_latin1(size_t length) const noexcept { 6363| 297| return length; 6364| 297| } _ZNK7simdutf14implementation24latin1_length_from_utf32Em: 6550| 990| latin1_length_from_utf32(size_t length) const noexcept { 6551| 990| return length; 6552| 990| } _ZNK7simdutf14implementation24latin1_length_from_utf16Em: 6587| 1.07k| latin1_length_from_utf16(size_t length) const noexcept { 6588| 1.07k| return length; 6589| 1.07k| } _ZNK7simdutf14implementation24utf32_length_from_latin1Em: 6620| 150| utf32_length_from_latin1(size_t length) const noexcept { 6621| 150| return length; 6622| 150| } _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| 484| size_t len) { 11| 484| const uint8_t *c = reinterpret_cast(buf); 12| 484| size_t answer = 0; 13| 3.82k| for (size_t i = 0; i < len; i++) { ------------------ | Branch (13:22): [True: 3.34k, False: 484] ------------------ 14| 3.34k| if ((c[i] >> 7)) { ------------------ | Branch (14:9): [True: 1.00k, False: 2.33k] ------------------ 15| 1.00k| answer++; 16| 1.00k| } 17| 3.34k| } 18| 484| return answer + len; 19| 484|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115latin1_to_utf167convertILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 14| 228| char16_t *utf16_output) { 15| 228| size_t pos = 0; 16| 228| char16_t *start{utf16_output}; 17| | 18| 2.81M| while (pos < len) { ------------------ | Branch (18:10): [True: 2.81M, False: 228] ------------------ 19| 2.81M| uint16_t word = 20| 2.81M| uint8_t(data[pos]); // extend Latin-1 char to 16-bit Unicode code point 21| 2.81M| *utf16_output++ = 22| 2.81M| char16_t(match_system(big_endian) ? word : u16_swap_bytes(word)); ------------------ | Branch (22:18): [True: 2.81M, Folded] ------------------ 23| 2.81M| pos++; 24| 2.81M| } 25| | 26| 228| return utf16_output - start; 27| 228|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115latin1_to_utf167convertILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 14| 294| char16_t *utf16_output) { 15| 294| size_t pos = 0; 16| 294| char16_t *start{utf16_output}; 17| | 18| 2.88M| while (pos < len) { ------------------ | Branch (18:10): [True: 2.88M, False: 294] ------------------ 19| 2.88M| uint16_t word = 20| 2.88M| uint8_t(data[pos]); // extend Latin-1 char to 16-bit Unicode code point 21| 2.88M| *utf16_output++ = 22| 2.88M| char16_t(match_system(big_endian) ? word : u16_swap_bytes(word)); ------------------ | Branch (22:18): [Folded, False: 2.88M] ------------------ 23| 2.88M| pos++; 24| 2.88M| } 25| | 26| 294| return utf16_output - start; 27| 294|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115latin1_to_utf327convertIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mPDi: 14| 248| char32_t *utf32_output) { 15| 248| char32_t *start{utf32_output}; 16| 2.69M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (16:22): [True: 2.69M, False: 248] ------------------ 17| 2.69M| *utf32_output++ = uint8_t(data[i]); 18| 2.69M| } 19| 248| return utf32_output - start; 20| 248|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf87convertEPKcmPc: 66| 1.37k| char *utf8_output) { 67| 1.37k| return convert(reinterpret_cast(buf), len, 68| 1.37k| utf8_output); 69| 1.37k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf87convertIPKhPcQaasr7simdutf6detailE22indexes_into_byte_likeIT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEEmS7_mS8_: 17| 1.37k| OutputPtr utf8_output) { 18| | // const unsigned char *data = reinterpret_cast(buf); 19| 1.37k| size_t pos = 0; 20| 1.37k| size_t utf8_pos = 0; 21| | 22| 4.22M| while (pos < len) { ------------------ | Branch (22:10): [True: 4.21M, False: 1.37k] ------------------ 23| |#if SIMDUTF_CPLUSPLUS23 24| | if !consteval 25| |#endif 26| 4.21M| { 27| | // try to convert the next block of 16 ASCII bytes 28| 4.21M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (28:11): [True: 4.20M, False: 12.6k] ------------------ 29| | // they are ascii 30| 4.20M| uint64_t v1; 31| 4.20M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 32| 4.20M| uint64_t v2; 33| 4.20M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 34| 4.20M| uint64_t v{v1 | 35| 4.20M| v2}; // We are only interested in these bits: 1000 1000 1000 36| | // 1000, so it makes sense to concatenate everything 37| 4.20M| if ((v & 0x8080808080808080) == ------------------ | Branch (37:13): [True: 1.98M, False: 2.21M] ------------------ 38| 4.20M| 0) { // if NONE of these are set, e.g. all of them are zero, then 39| | // everything is ASCII 40| 1.98M| size_t final_pos = pos + 16; 41| 33.8M| while (pos < final_pos) { ------------------ | Branch (41:18): [True: 31.8M, False: 1.98M] ------------------ 42| 31.8M| utf8_output[utf8_pos++] = char(data[pos]); 43| 31.8M| pos++; 44| 31.8M| } 45| 1.98M| continue; 46| 1.98M| } 47| 4.20M| } // if (pos + 16 <= len) 48| 4.21M| } // !consteval scope 49| | 50| 2.23M| unsigned char byte = data[pos]; 51| 2.23M| if ((byte & 0x80) == 0) { // if ASCII ------------------ | Branch (51:9): [True: 1.38M, False: 848k] ------------------ 52| | // will generate one UTF-8 bytes 53| 1.38M| utf8_output[utf8_pos++] = char(byte); 54| 1.38M| pos++; 55| 1.38M| } else { 56| | // will generate two UTF-8 bytes 57| 848k| utf8_output[utf8_pos++] = char((byte >> 6) | 0b11000000); 58| 848k| utf8_output[utf8_pos++] = char((byte & 0b111111) | 0b10000000); 59| 848k| pos++; 60| 848k| } 61| 2.23M| } // while 62| 1.37k| return utf8_pos; 63| 1.37k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf823utf8_length_from_latin1IPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_m: 152| 242|utf8_length_from_latin1(InputPtr input, size_t length) noexcept { 153| 242| size_t answer = length; 154| 242| size_t i = 0; 155| | 156| |#if SIMDUTF_CPLUSPLUS23 157| | if !consteval 158| |#endif 159| 242| { 160| 242| auto pop = [](uint64_t v) { 161| 242| return (size_t)(((v >> 7) & UINT64_C(0x0101010101010101)) * 162| 242| UINT64_C(0x0101010101010101) >> 163| 242| 56); 164| 242| }; 165| 531k| for (; i + 32 <= length; i += 32) { ------------------ | Branch (165:12): [True: 531k, False: 242] ------------------ 166| 531k| uint64_t v; 167| 531k| memcpy(&v, input + i, 8); 168| 531k| answer += pop(v); 169| 531k| memcpy(&v, input + i + 8, sizeof(v)); 170| 531k| answer += pop(v); 171| 531k| memcpy(&v, input + i + 16, sizeof(v)); 172| 531k| answer += pop(v); 173| 531k| memcpy(&v, input + i + 24, sizeof(v)); 174| 531k| answer += pop(v); 175| 531k| } 176| 480| for (; i + 8 <= length; i += 8) { ------------------ | Branch (176:12): [True: 238, False: 242] ------------------ 177| 238| uint64_t v; 178| 238| memcpy(&v, input + i, sizeof(v)); 179| 238| answer += pop(v); 180| 238| } 181| 242| } // !consteval scope 182| 664| for (; i + 1 <= length; i += 1) { ------------------ | Branch (182:10): [True: 422, False: 242] ------------------ 183| 422| answer += static_cast(input[i]) >> 7; 184| 422| } 185| 242| return answer; 186| 242|} simdutf.cpp:_ZZN7simdutf6scalar12_GLOBAL__N_114latin1_to_utf823utf8_length_from_latin1IPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mENKUlmE_clEm: 160| 2.12M| auto pop = [](uint64_t v) { 161| 2.12M| return (size_t)(((v >> 7) & UINT64_C(0x0101010101010101)) * 162| | UINT64_C(0x0101010101010101) >> 163| 2.12M| 56); 164| 2.12M| }; _ZN7simdutf6scalar14u16_swap_bytesEt: 8| 207M|u16_swap_bytes(const uint16_t word) { 9| 207M| return uint16_t((word >> 8) | (word << 8)); 10| 207M|} _ZN7simdutf6scalar5utf1614swap_if_neededILNS_10endiannessE1EEEtt: 27| 95.8M|template constexpr uint16_t swap_if_needed(uint16_t c) { 28| 95.8M| return !match_system(big_endian) ? scalar::u16_swap_bytes(c) : c; ------------------ | Branch (28:10): [True: 95.8M, Folded] ------------------ 29| 95.8M|} _ZN7simdutf6scalar5utf1614swap_if_neededILNS_10endiannessE0EEEtt: 27| 85.4M|template constexpr uint16_t swap_if_needed(uint16_t c) { 28| 85.4M| return !match_system(big_endian) ? scalar::u16_swap_bytes(c) : c; ------------------ | Branch (28:10): [Folded, False: 85.4M] ------------------ 29| 85.4M|} _ZN7simdutf6scalar5utf168validateILNS_10endiannessE0EEEbPKDsm: 22| 3.87k|validate(const char16_t *data, size_t len) noexcept { 23| 3.87k| uint64_t pos = 0; 24| 18.7M| while (pos < len) { ------------------ | Branch (24:10): [True: 18.7M, False: 2.64k] ------------------ 25| 18.7M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 26| 18.7M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (26:9): [True: 18.3k, False: 18.7M] ------------------ 27| 18.3k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 234, False: 18.0k] ------------------ 28| 234| return false; 29| 234| } 30| 18.0k| char16_t diff = char16_t(word - 0xD800); 31| 18.0k| if (diff > 0x3FF) { ------------------ | Branch (31:11): [True: 412, False: 17.6k] ------------------ 32| 412| return false; 33| 412| } 34| 17.6k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (34:28): [Folded, False: 17.6k] ------------------ 35| 17.6k| ? u16_swap_bytes(data[pos + 1]) 36| 17.6k| : data[pos + 1]; 37| 17.6k| char16_t diff2 = char16_t(next_word - 0xDC00); 38| 17.6k| if (diff2 > 0x3FF) { ------------------ | Branch (38:11): [True: 586, False: 17.0k] ------------------ 39| 586| return false; 40| 586| } 41| 17.0k| pos += 2; 42| 18.7M| } else { 43| 18.7M| pos++; 44| 18.7M| } 45| 18.7M| } 46| 2.64k| return true; 47| 3.87k|} _ZN7simdutf6scalar5utf168validateILNS_10endiannessE1EEEbPKDsm: 22| 3.79k|validate(const char16_t *data, size_t len) noexcept { 23| 3.79k| uint64_t pos = 0; 24| 22.2M| while (pos < len) { ------------------ | Branch (24:10): [True: 22.2M, False: 2.57k] ------------------ 25| 22.2M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 26| 22.2M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (26:9): [True: 17.5k, False: 22.2M] ------------------ 27| 17.5k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 243, False: 17.2k] ------------------ 28| 243| return false; 29| 243| } 30| 17.2k| char16_t diff = char16_t(word - 0xD800); 31| 17.2k| if (diff > 0x3FF) { ------------------ | Branch (31:11): [True: 420, False: 16.8k] ------------------ 32| 420| return false; 33| 420| } 34| 16.8k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (34:28): [True: 16.8k, Folded] ------------------ 35| 16.8k| ? u16_swap_bytes(data[pos + 1]) 36| 16.8k| : data[pos + 1]; 37| 16.8k| char16_t diff2 = char16_t(next_word - 0xDC00); 38| 16.8k| if (diff2 > 0x3FF) { ------------------ | Branch (38:11): [True: 552, False: 16.3k] ------------------ 39| 552| return false; 40| 552| } 41| 16.3k| pos += 2; 42| 22.2M| } else { 43| 22.2M| pos++; 44| 22.2M| } 45| 22.2M| } 46| 2.57k| return true; 47| 3.79k|} _ZN7simdutf6scalar5utf1620validate_with_errorsILNS_10endiannessE0EEENS_6resultEPKDsm: 51| 4.12k|validate_with_errors(const char16_t *data, size_t len) noexcept { 52| 4.12k| size_t pos = 0; 53| 18.7M| while (pos < len) { ------------------ | Branch (53:10): [True: 18.7M, False: 2.64k] ------------------ 54| 18.7M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 55| 18.7M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (55:9): [True: 18.6k, False: 18.7M] ------------------ 56| 18.6k| if (pos + 1 >= len) { ------------------ | Branch (56:11): [True: 234, False: 18.3k] ------------------ 57| 234| return result(error_code::SURROGATE, pos); 58| 234| } 59| 18.3k| char16_t diff = char16_t(word - 0xD800); 60| 18.3k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 540, False: 17.8k] ------------------ 61| 540| return result(error_code::SURROGATE, pos); 62| 540| } 63| 17.8k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (63:28): [Folded, False: 17.8k] ------------------ 64| 17.8k| ? u16_swap_bytes(data[pos + 1]) 65| 17.8k| : data[pos + 1]; 66| 17.8k| char16_t diff2 = uint16_t(next_word - 0xDC00); 67| 17.8k| if (diff2 > 0x3FF) { ------------------ | Branch (67:11): [True: 711, False: 17.1k] ------------------ 68| 711| return result(error_code::SURROGATE, pos); 69| 711| } 70| 17.1k| pos += 2; 71| 18.7M| } else { 72| 18.7M| pos++; 73| 18.7M| } 74| 18.7M| } 75| 2.64k| return result(error_code::SUCCESS, pos); 76| 4.12k|} _ZN7simdutf6scalar5utf1620validate_with_errorsILNS_10endiannessE1EEENS_6resultEPKDsm: 51| 4.00k|validate_with_errors(const char16_t *data, size_t len) noexcept { 52| 4.00k| size_t pos = 0; 53| 22.2M| while (pos < len) { ------------------ | Branch (53:10): [True: 22.2M, False: 2.57k] ------------------ 54| 22.2M| char16_t word = scalar::utf16::swap_if_needed(data[pos]); 55| 22.2M| if ((word & 0xF800) == 0xD800) { ------------------ | Branch (55:9): [True: 17.8k, False: 22.2M] ------------------ 56| 17.8k| if (pos + 1 >= len) { ------------------ | Branch (56:11): [True: 243, False: 17.5k] ------------------ 57| 243| return result(error_code::SURROGATE, pos); 58| 243| } 59| 17.5k| char16_t diff = char16_t(word - 0xD800); 60| 17.5k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 507, False: 17.0k] ------------------ 61| 507| return result(error_code::SURROGATE, pos); 62| 507| } 63| 17.0k| char16_t next_word = !match_system(big_endian) ------------------ | Branch (63:28): [True: 17.0k, Folded] ------------------ 64| 17.0k| ? u16_swap_bytes(data[pos + 1]) 65| 17.0k| : data[pos + 1]; 66| 17.0k| char16_t diff2 = uint16_t(next_word - 0xDC00); 67| 17.0k| if (diff2 > 0x3FF) { ------------------ | Branch (67:11): [True: 678, False: 16.3k] ------------------ 68| 678| return result(error_code::SURROGATE, pos); 69| 678| } 70| 16.3k| pos += 2; 71| 22.2M| } else { 72| 22.2M| pos++; 73| 22.2M| } 74| 22.2M| } 75| 2.57k| return result(error_code::SUCCESS, pos); 76| 4.00k|} _ZN7simdutf6scalar5utf1617count_code_pointsILNS_10endiannessE0EEEmPKDsm: 79| 4.99k|simdutf_constexpr23 size_t count_code_points(const char16_t *p, size_t len) { 80| | // We are not BOM aware. 81| 4.99k| size_t counter{0}; 82| 26.3M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (82:22): [True: 26.3M, False: 4.99k] ------------------ 83| 26.3M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 84| 26.3M| counter += ((word & 0xFC00) != 0xDC00); 85| 26.3M| } 86| 4.99k| return counter; 87| 4.99k|} _ZN7simdutf6scalar5utf1617count_code_pointsILNS_10endiannessE1EEEmPKDsm: 79| 4.93k|simdutf_constexpr23 size_t count_code_points(const char16_t *p, size_t len) { 80| | // We are not BOM aware. 81| 4.93k| size_t counter{0}; 82| 27.9M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (82:22): [True: 27.9M, False: 4.93k] ------------------ 83| 27.9M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 84| 27.9M| counter += ((word & 0xFC00) != 0xDC00); 85| 27.9M| } 86| 4.93k| return counter; 87| 4.93k|} _ZN7simdutf6scalar5utf1622utf8_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 91| 2.33k| size_t len) { 92| | // We are not BOM aware. 93| 2.33k| size_t counter{0}; 94| 16.1M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (94:22): [True: 16.1M, False: 2.33k] ------------------ 95| 16.1M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 96| 16.1M| counter++; // ASCII 97| 16.1M| counter += static_cast( 98| 16.1M| word > 99| 16.1M| 0x7F); // non-ASCII is at least 2 bytes, surrogates are 2*2 == 4 bytes 100| 16.1M| counter += static_cast((word > 0x7FF && word <= 0xD7FF) || ------------------ | Branch (100:37): [True: 3.21M, False: 12.9M] | Branch (100:53): [True: 2.90M, False: 303k] ------------------ 101| 13.2M| (word >= 0xE000)); // three-byte ------------------ | Branch (101:36): [True: 236k, False: 13.0M] ------------------ 102| 16.1M| } 103| 2.33k| return counter; 104| 2.33k|} _ZN7simdutf6scalar5utf1622utf8_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 91| 2.18k| size_t len) { 92| | // We are not BOM aware. 93| 2.18k| size_t counter{0}; 94| 16.5M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (94:22): [True: 16.5M, False: 2.18k] ------------------ 95| 16.5M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 96| 16.5M| counter++; // ASCII 97| 16.5M| counter += static_cast( 98| 16.5M| word > 99| 16.5M| 0x7F); // non-ASCII is at least 2 bytes, surrogates are 2*2 == 4 bytes 100| 16.5M| counter += static_cast((word > 0x7FF && word <= 0xD7FF) || ------------------ | Branch (100:37): [True: 3.64M, False: 12.9M] | Branch (100:53): [True: 3.48M, False: 166k] ------------------ 101| 13.1M| (word >= 0xE000)); // three-byte ------------------ | Branch (101:36): [True: 146k, False: 12.9M] ------------------ 102| 16.5M| } 103| 2.18k| return counter; 104| 2.18k|} _ZN7simdutf6scalar5utf1623utf32_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 108| 420| size_t len) { 109| | // We are not BOM aware. 110| 420| size_t counter{0}; 111| 4.85M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (111:22): [True: 4.85M, False: 420] ------------------ 112| 4.85M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 113| 4.85M| counter += ((word & 0xFC00) != 0xDC00); 114| 4.85M| } 115| 420| return counter; 116| 420|} _ZN7simdutf6scalar5utf1623utf32_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 108| 448| size_t len) { 109| | // We are not BOM aware. 110| 448| size_t counter{0}; 111| 6.29M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (111:22): [True: 6.29M, False: 448] ------------------ 112| 6.29M| char16_t word = scalar::utf16::swap_if_needed(p[i]); 113| 6.29M| counter += ((word & 0xFC00) != 0xDC00); 114| 6.29M| } 115| 448| return counter; 116| 448|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin17convertILNS_10endiannessE0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 304| OutputPtr latin_output) { 18| 304| if (len == 0) { ------------------ | Branch (18:7): [True: 6, False: 298] ------------------ 19| 6| return 0; 20| 6| } 21| 298| size_t pos = 0; 22| 298| const auto latin_output_start = latin_output; 23| 298| uint16_t word = 0; 24| 298| uint16_t too_large = 0; 25| | 26| 3.88M| while (pos < len) { ------------------ | Branch (26:10): [True: 3.88M, False: 298] ------------------ 27| 3.88M| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (27:12): [Folded, False: 3.88M] ------------------ 28| 3.88M| too_large |= word; 29| 3.88M| *latin_output++ = char(word & 0xFF); 30| 3.88M| pos++; 31| 3.88M| } 32| 298| if ((too_large & 0xFF00) != 0) { ------------------ | Branch (32:7): [True: 188, False: 110] ------------------ 33| 188| return 0; 34| 188| } 35| | 36| 110| return latin_output - latin_output_start; 37| 298|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin17convertILNS_10endiannessE1EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 242| OutputPtr latin_output) { 18| 242| if (len == 0) { ------------------ | Branch (18:7): [True: 6, False: 236] ------------------ 19| 6| return 0; 20| 6| } 21| 236| size_t pos = 0; 22| 236| const auto latin_output_start = latin_output; 23| 236| uint16_t word = 0; 24| 236| uint16_t too_large = 0; 25| | 26| 1.06M| while (pos < len) { ------------------ | Branch (26:10): [True: 1.06M, False: 236] ------------------ 27| 1.06M| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (27:12): [True: 1.06M, Folded] ------------------ 28| 1.06M| too_large |= word; 29| 1.06M| *latin_output++ = char(word & 0xFF); 30| 1.06M| pos++; 31| 1.06M| } 32| 236| if ((too_large & 0xFF00) != 0) { ------------------ | Branch (32:7): [True: 138, False: 98] ------------------ 33| 138| return 0; 34| 138| } 35| | 36| 98| return latin_output - latin_output_start; 37| 236|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin119convert_with_errorsILNS_10endiannessE0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEENS_6resultES8_mS9_: 45| 444| OutputPtr latin_output) { 46| 444| if (len == 0) { ------------------ | Branch (46:7): [True: 4, False: 440] ------------------ 47| 4| return result(error_code::SUCCESS, 0); 48| 4| } 49| 440| size_t pos = 0; 50| 440| auto start = latin_output; 51| 440| uint16_t word; 52| | 53| 395k| while (pos < len) { ------------------ | Branch (53:10): [True: 395k, False: 202] ------------------ 54| |#if SIMDUTF_CPLUSPLUS23 55| | if !consteval 56| |#endif 57| 395k| { 58| 395k| if (pos + 16 <= len) { // if it is safe to read 32 more bytes, check that ------------------ | Branch (58:11): [True: 393k, False: 1.87k] ------------------ 59| | // they are Latin1 60| 393k| uint64_t v1, v2, v3, v4; 61| 393k| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 62| 393k| ::memcpy(&v2, data + pos + 4, sizeof(uint64_t)); 63| 393k| ::memcpy(&v3, data + pos + 8, sizeof(uint64_t)); 64| 393k| ::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| 393k| if (((v1 | v2 | v3 | v4) & 0xFF00FF00FF00FF00) == 0) { ------------------ | Branch (79:13): [True: 393k, False: 186] ------------------ 80| 393k| size_t final_pos = pos + 16; 81| 6.68M| while (pos < final_pos) { ------------------ | Branch (81:18): [True: 6.29M, False: 393k] ------------------ 82| 6.29M| *latin_output++ = !match_system(big_endian) ------------------ | Branch (82:31): [Folded, False: 6.29M] ------------------ 83| 6.29M| ? char(u16_swap_bytes(data[pos])) 84| 6.29M| : char(data[pos]); 85| 6.29M| pos++; 86| 6.29M| } 87| 393k| continue; 88| 393k| } 89| 393k| } 90| 395k| } 91| | 92| 2.05k| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (92:12): [Folded, False: 2.05k] ------------------ 93| 2.05k| if ((word & 0xFF00) == 0) { ------------------ | Branch (93:9): [True: 1.82k, False: 238] ------------------ 94| 1.82k| *latin_output++ = char(word & 0xFF); 95| 1.82k| pos++; 96| 1.82k| } else { 97| 238| return result(error_code::TOO_LARGE, pos); 98| 238| } 99| 2.05k| } 100| 202| return result(error_code::SUCCESS, latin_output - start); 101| 440|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf16_to_latin119convert_with_errorsILNS_10endiannessE1EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEENS_6resultES8_mS9_: 45| 414| OutputPtr latin_output) { 46| 414| if (len == 0) { ------------------ | Branch (46:7): [True: 4, False: 410] ------------------ 47| 4| return result(error_code::SUCCESS, 0); 48| 4| } 49| 410| size_t pos = 0; 50| 410| auto start = latin_output; 51| 410| uint16_t word; 52| | 53| 459k| while (pos < len) { ------------------ | Branch (53:10): [True: 459k, False: 152] ------------------ 54| |#if SIMDUTF_CPLUSPLUS23 55| | if !consteval 56| |#endif 57| 459k| { 58| 459k| if (pos + 16 <= len) { // if it is safe to read 32 more bytes, check that ------------------ | Branch (58:11): [True: 457k, False: 1.36k] ------------------ 59| | // they are Latin1 60| 457k| uint64_t v1, v2, v3, v4; 61| 457k| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 62| 457k| ::memcpy(&v2, data + pos + 4, sizeof(uint64_t)); 63| 457k| ::memcpy(&v3, data + pos + 8, sizeof(uint64_t)); 64| 457k| ::memcpy(&v4, data + pos + 12, sizeof(uint64_t)); 65| | 66| 457k| if constexpr (!match_system(big_endian)) { 67| 457k| v1 = (v1 >> 8) | (v1 << (64 - 8)); 68| 457k| } 69| 457k| if constexpr (!match_system(big_endian)) { 70| 457k| v2 = (v2 >> 8) | (v2 << (64 - 8)); 71| 457k| } 72| 457k| if constexpr (!match_system(big_endian)) { 73| 457k| v3 = (v3 >> 8) | (v3 << (64 - 8)); 74| 457k| } 75| 457k| if constexpr (!match_system(big_endian)) { 76| 457k| v4 = (v4 >> 8) | (v4 << (64 - 8)); 77| 457k| } 78| | 79| 457k| if (((v1 | v2 | v3 | v4) & 0xFF00FF00FF00FF00) == 0) { ------------------ | Branch (79:13): [True: 457k, False: 252] ------------------ 80| 457k| size_t final_pos = pos + 16; 81| 7.77M| while (pos < final_pos) { ------------------ | Branch (81:18): [True: 7.31M, False: 457k] ------------------ 82| 7.31M| *latin_output++ = !match_system(big_endian) ------------------ | Branch (82:31): [True: 7.31M, Folded] ------------------ 83| 7.31M| ? char(u16_swap_bytes(data[pos])) 84| 7.31M| : char(data[pos]); 85| 7.31M| pos++; 86| 7.31M| } 87| 457k| continue; 88| 457k| } 89| 457k| } 90| 459k| } 91| | 92| 1.61k| word = !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (92:12): [True: 1.61k, Folded] ------------------ 93| 1.61k| if ((word & 0xFF00) == 0) { ------------------ | Branch (93:9): [True: 1.35k, False: 258] ------------------ 94| 1.35k| *latin_output++ = char(word & 0xFF); 95| 1.35k| pos++; 96| 1.35k| } else { 97| 258| return result(error_code::TOO_LARGE, pos); 98| 258| } 99| 1.61k| } 100| 152| return result(error_code::SUCCESS, latin_output - start); 101| 410|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf327convertILNS_10endiannessE0EEEmPKDsmPDi: 11| 786| char32_t *utf32_output) { 12| 786| size_t pos = 0; 13| 786| char32_t *start{utf32_output}; 14| 742k| while (pos < len) { ------------------ | Branch (14:10): [True: 742k, False: 450] ------------------ 15| 742k| uint16_t word = 16| 742k| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [Folded, False: 742k] ------------------ 17| 742k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 740k, False: 2.10k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 740k| *utf32_output++ = char32_t(word); 20| 740k| pos++; 21| 740k| } else { 22| | // must be a surrogate pair 23| 2.10k| uint16_t diff = uint16_t(word - 0xD800); 24| 2.10k| if (diff > 0x3FF) { ------------------ | Branch (24:11): [True: 126, False: 1.97k] ------------------ 25| 126| return 0; 26| 126| } 27| 1.97k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 84, False: 1.89k] ------------------ 28| 84| return 0; 29| 84| } // minimal bound checking 30| 1.89k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (30:28): [Folded, False: 1.89k] ------------------ 31| 1.89k| ? u16_swap_bytes(data[pos + 1]) 32| 1.89k| : data[pos + 1]; 33| 1.89k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 34| 1.89k| if (diff2 > 0x3FF) { ------------------ | Branch (34:11): [True: 126, False: 1.76k] ------------------ 35| 126| return 0; 36| 126| } 37| 1.76k| uint32_t value = (diff << 10) + diff2 + 0x10000; 38| 1.76k| *utf32_output++ = char32_t(value); 39| 1.76k| pos += 2; 40| 1.76k| } 41| 742k| } 42| 450| return utf32_output - start; 43| 786|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf327convertILNS_10endiannessE1EEEmPKDsmPDi: 11| 894| char32_t *utf32_output) { 12| 894| size_t pos = 0; 13| 894| char32_t *start{utf32_output}; 14| 1.06M| while (pos < len) { ------------------ | Branch (14:10): [True: 1.05M, False: 542] ------------------ 15| 1.05M| uint16_t word = 16| 1.05M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [True: 1.05M, Folded] ------------------ 17| 1.05M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 1.05M, False: 2.23k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 1.05M| *utf32_output++ = char32_t(word); 20| 1.05M| pos++; 21| 1.05M| } else { 22| | // must be a surrogate pair 23| 2.23k| uint16_t diff = uint16_t(word - 0xD800); 24| 2.23k| if (diff > 0x3FF) { ------------------ | Branch (24:11): [True: 114, False: 2.11k] ------------------ 25| 114| return 0; 26| 114| } 27| 2.11k| if (pos + 1 >= len) { ------------------ | Branch (27:11): [True: 120, False: 1.99k] ------------------ 28| 120| return 0; 29| 120| } // minimal bound checking 30| 1.99k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (30:28): [True: 1.99k, Folded] ------------------ 31| 1.99k| ? u16_swap_bytes(data[pos + 1]) 32| 1.99k| : data[pos + 1]; 33| 1.99k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 34| 1.99k| if (diff2 > 0x3FF) { ------------------ | Branch (34:11): [True: 118, False: 1.87k] ------------------ 35| 118| return 0; 36| 118| } 37| 1.87k| uint32_t value = (diff << 10) + diff2 + 0x10000; 38| 1.87k| *utf32_output++ = char32_t(value); 39| 1.87k| pos += 2; 40| 1.87k| } 41| 1.05M| } 42| 542| return utf32_output - start; 43| 894|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3219convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKDsmPDi: 47| 872| char32_t *utf32_output) { 48| 872| size_t pos = 0; 49| 872| char32_t *start{utf32_output}; 50| 2.51M| while (pos < len) { ------------------ | Branch (50:10): [True: 2.51M, False: 524] ------------------ 51| 2.51M| uint16_t word = 52| 2.51M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (52:9): [Folded, False: 2.51M] ------------------ 53| 2.51M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (53:9): [True: 2.50M, False: 11.1k] ------------------ 54| | // No surrogate pair, extend 16-bit word to 32-bit word 55| 2.50M| *utf32_output++ = char32_t(word); 56| 2.50M| pos++; 57| 2.50M| } else { 58| | // must be a surrogate pair 59| 11.1k| uint16_t diff = uint16_t(word - 0xD800); 60| 11.1k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 102, False: 11.0k] ------------------ 61| 102| return result(error_code::SURROGATE, pos); 62| 102| } 63| 11.0k| if (pos + 1 >= len) { ------------------ | Branch (63:11): [True: 90, False: 10.9k] ------------------ 64| 90| return result(error_code::SURROGATE, pos); 65| 90| } // minimal bound checking 66| 10.9k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (66:28): [Folded, False: 10.9k] ------------------ 67| 10.9k| ? u16_swap_bytes(data[pos + 1]) 68| 10.9k| : data[pos + 1]; 69| 10.9k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 70| 10.9k| if (diff2 > 0x3FF) { ------------------ | Branch (70:11): [True: 156, False: 10.7k] ------------------ 71| 156| return result(error_code::SURROGATE, pos); 72| 156| } 73| 10.7k| uint32_t value = (diff << 10) + diff2 + 0x10000; 74| 10.7k| *utf32_output++ = char32_t(value); 75| 10.7k| pos += 2; 76| 10.7k| } 77| 2.51M| } 78| 524| return result(error_code::SUCCESS, utf32_output - start); 79| 872|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3219convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKDsmPDi: 47| 910| char32_t *utf32_output) { 48| 910| size_t pos = 0; 49| 910| char32_t *start{utf32_output}; 50| 2.53M| while (pos < len) { ------------------ | Branch (50:10): [True: 2.53M, False: 466] ------------------ 51| 2.53M| uint16_t word = 52| 2.53M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (52:9): [True: 2.53M, Folded] ------------------ 53| 2.53M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (53:9): [True: 2.52M, False: 11.5k] ------------------ 54| | // No surrogate pair, extend 16-bit word to 32-bit word 55| 2.52M| *utf32_output++ = char32_t(word); 56| 2.52M| pos++; 57| 2.52M| } else { 58| | // must be a surrogate pair 59| 11.5k| uint16_t diff = uint16_t(word - 0xD800); 60| 11.5k| if (diff > 0x3FF) { ------------------ | Branch (60:11): [True: 178, False: 11.3k] ------------------ 61| 178| return result(error_code::SURROGATE, pos); 62| 178| } 63| 11.3k| if (pos + 1 >= len) { ------------------ | Branch (63:11): [True: 90, False: 11.2k] ------------------ 64| 90| return result(error_code::SURROGATE, pos); 65| 90| } // minimal bound checking 66| 11.2k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (66:28): [True: 11.2k, Folded] ------------------ 67| 11.2k| ? u16_swap_bytes(data[pos + 1]) 68| 11.2k| : data[pos + 1]; 69| 11.2k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 70| 11.2k| if (diff2 > 0x3FF) { ------------------ | Branch (70:11): [True: 176, False: 11.0k] ------------------ 71| 176| return result(error_code::SURROGATE, pos); 72| 176| } 73| 11.0k| uint32_t value = (diff << 10) + diff2 + 0x10000; 74| 11.0k| *utf32_output++ = char32_t(value); 75| 11.0k| pos += 2; 76| 11.0k| } 77| 2.53M| } 78| 466| return result(error_code::SUCCESS, utf32_output - start); 79| 910|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3213convert_validILNS_10endiannessE0EEEmPKDsmPDi: 11| 120| char32_t *utf32_output) { 12| 120| size_t pos = 0; 13| 120| char32_t *start{utf32_output}; 14| 516k| while (pos < len) { ------------------ | Branch (14:10): [True: 516k, False: 120] ------------------ 15| 516k| uint16_t word = 16| 516k| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [Folded, False: 516k] ------------------ 17| 516k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 512k, False: 4.00k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 512k| *utf32_output++ = char32_t(word); 20| 512k| pos++; 21| 512k| } else { 22| | // must be a surrogate pair 23| 4.00k| uint16_t diff = uint16_t(word - 0xD800); 24| 4.00k| if (pos + 1 >= len) { ------------------ | Branch (24:11): [True: 0, False: 4.00k] ------------------ 25| 0| return 0; 26| 0| } // minimal bound checking 27| 4.00k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (27:28): [Folded, False: 4.00k] ------------------ 28| 4.00k| ? u16_swap_bytes(data[pos + 1]) 29| 4.00k| : data[pos + 1]; 30| 4.00k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 31| 4.00k| uint32_t value = (diff << 10) + diff2 + 0x10000; 32| 4.00k| *utf32_output++ = char32_t(value); 33| 4.00k| pos += 2; 34| 4.00k| } 35| 516k| } 36| 120| return utf32_output - start; 37| 120|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf16_to_utf3213convert_validILNS_10endiannessE1EEEmPKDsmPDi: 11| 166| char32_t *utf32_output) { 12| 166| size_t pos = 0; 13| 166| char32_t *start{utf32_output}; 14| 5.00M| while (pos < len) { ------------------ | Branch (14:10): [True: 5.00M, False: 166] ------------------ 15| 5.00M| uint16_t word = 16| 5.00M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (16:9): [True: 5.00M, Folded] ------------------ 17| 5.00M| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (17:9): [True: 5.00M, False: 1.43k] ------------------ 18| | // No surrogate pair, extend 16-bit word to 32-bit word 19| 5.00M| *utf32_output++ = char32_t(word); 20| 5.00M| pos++; 21| 5.00M| } else { 22| | // must be a surrogate pair 23| 1.43k| uint16_t diff = uint16_t(word - 0xD800); 24| 1.43k| if (pos + 1 >= len) { ------------------ | Branch (24:11): [True: 0, False: 1.43k] ------------------ 25| 0| return 0; 26| 0| } // minimal bound checking 27| 1.43k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (27:28): [True: 1.43k, Folded] ------------------ 28| 1.43k| ? u16_swap_bytes(data[pos + 1]) 29| 1.43k| : data[pos + 1]; 30| 1.43k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 31| 1.43k| uint32_t value = (diff << 10) + diff2 + 0x10000; 32| 1.43k| *utf32_output++ = char32_t(value); 33| 1.43k| pos += 2; 34| 1.43k| } 35| 5.00M| } 36| 166| return utf32_output - start; 37| 166|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf87convertILNS_10endiannessE0EPKDsPcQsr7simdutf6detailE18indexes_into_utf16IT0_EEEmS8_mT1_: 17| 1.91k| OutputPtr utf8_output) { 18| 1.91k| size_t pos = 0; 19| 1.91k| const auto start = utf8_output; 20| 3.08M| while (pos < len) { ------------------ | Branch (20:10): [True: 3.07M, False: 1.51k] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 3.07M| { 25| | // try to convert the next block of 8 bytes 26| 3.07M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 3.07M, False: 3.32k] ------------------ 27| | // they are ascii 28| 3.07M| uint64_t v; 29| 3.07M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| | if constexpr (!match_system(big_endian)) { 31| | v = (v >> 8) | (v << (64 - 8)); 32| | } 33| 3.07M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 2.06M, False: 1.01M] ------------------ 34| 2.06M| size_t final_pos = pos + 4; 35| 10.3M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 8.25M, False: 2.06M] ------------------ 36| 8.25M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [Folded, False: 8.25M] ------------------ 37| 8.25M| ? char(u16_swap_bytes(data[pos])) 38| 8.25M| : char(data[pos]); 39| 8.25M| pos++; 40| 8.25M| } 41| 2.06M| continue; 42| 2.06M| } 43| 3.07M| } 44| 3.07M| } 45| 1.01M| uint16_t word = 46| 1.01M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (46:9): [Folded, False: 1.01M] ------------------ 47| 1.01M| if ((word & 0xFF80) == 0) { ------------------ | Branch (47:9): [True: 203k, False: 812k] ------------------ 48| | // will generate one UTF-8 bytes 49| 203k| *utf8_output++ = char(word); 50| 203k| pos++; 51| 812k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (51:16): [True: 352k, False: 459k] ------------------ 52| | // will generate two UTF-8 bytes 53| | // we have 0b110XXXXX 0b10XXXXXX 54| 352k| *utf8_output++ = char((word >> 6) | 0b11000000); 55| 352k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 56| 352k| pos++; 57| 459k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (57:16): [True: 452k, False: 6.79k] ------------------ 58| | // will generate three UTF-8 bytes 59| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 60| 452k| *utf8_output++ = char((word >> 12) | 0b11100000); 61| 452k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 62| 452k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 63| 452k| pos++; 64| 452k| } else { 65| | // must be a surrogate pair 66| 6.79k| if (pos + 1 >= len) { ------------------ | Branch (66:11): [True: 54, False: 6.73k] ------------------ 67| 54| return 0; 68| 54| } 69| 6.73k| uint16_t diff = uint16_t(word - 0xD800); 70| 6.73k| if (diff > 0x3FF) { ------------------ | Branch (70:11): [True: 136, False: 6.60k] ------------------ 71| 136| return 0; 72| 136| } 73| 6.60k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (73:28): [Folded, False: 6.60k] ------------------ 74| 6.60k| ? u16_swap_bytes(data[pos + 1]) 75| 6.60k| : data[pos + 1]; 76| 6.60k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 77| 6.60k| if (diff2 > 0x3FF) { ------------------ | Branch (77:11): [True: 208, False: 6.39k] ------------------ 78| 208| return 0; 79| 208| } 80| 6.39k| uint32_t value = (diff << 10) + diff2 + 0x10000; 81| | // will generate four UTF-8 bytes 82| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 83| 6.39k| *utf8_output++ = char((value >> 18) | 0b11110000); 84| 6.39k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 85| 6.39k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 86| 6.39k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 87| 6.39k| pos += 2; 88| 6.39k| } 89| 1.01M| } 90| 1.51k| return utf8_output - start; 91| 1.91k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf87convertILNS_10endiannessE1EPKDsPcQsr7simdutf6detailE18indexes_into_utf16IT0_EEEmS8_mT1_: 17| 1.74k| OutputPtr utf8_output) { 18| 1.74k| size_t pos = 0; 19| 1.74k| const auto start = utf8_output; 20| 4.21M| while (pos < len) { ------------------ | Branch (20:10): [True: 4.21M, False: 1.47k] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 4.21M| { 25| | // try to convert the next block of 8 bytes 26| 4.21M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 4.21M, False: 3.26k] ------------------ 27| | // they are ascii 28| 4.21M| uint64_t v; 29| 4.21M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| 4.21M| if constexpr (!match_system(big_endian)) { 31| 4.21M| v = (v >> 8) | (v << (64 - 8)); 32| 4.21M| } 33| 4.21M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.68M, False: 2.52M] ------------------ 34| 1.68M| size_t final_pos = pos + 4; 35| 8.42M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 6.74M, False: 1.68M] ------------------ 36| 6.74M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [True: 6.74M, Folded] ------------------ 37| 6.74M| ? char(u16_swap_bytes(data[pos])) 38| 6.74M| : char(data[pos]); 39| 6.74M| pos++; 40| 6.74M| } 41| 1.68M| continue; 42| 1.68M| } 43| 4.21M| } 44| 4.21M| } 45| 2.53M| uint16_t word = 46| 2.53M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (46:9): [True: 2.53M, Folded] ------------------ 47| 2.53M| if ((word & 0xFF80) == 0) { ------------------ | Branch (47:9): [True: 287k, False: 2.24M] ------------------ 48| | // will generate one UTF-8 bytes 49| 287k| *utf8_output++ = char(word); 50| 287k| pos++; 51| 2.24M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (51:16): [True: 96.7k, False: 2.14M] ------------------ 52| | // will generate two UTF-8 bytes 53| | // we have 0b110XXXXX 0b10XXXXXX 54| 96.7k| *utf8_output++ = char((word >> 6) | 0b11000000); 55| 96.7k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 56| 96.7k| pos++; 57| 2.14M| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (57:16): [True: 2.14M, False: 5.22k] ------------------ 58| | // will generate three UTF-8 bytes 59| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 60| 2.14M| *utf8_output++ = char((word >> 12) | 0b11100000); 61| 2.14M| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 62| 2.14M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 63| 2.14M| pos++; 64| 2.14M| } else { 65| | // must be a surrogate pair 66| 5.22k| if (pos + 1 >= len) { ------------------ | Branch (66:11): [True: 42, False: 5.18k] ------------------ 67| 42| return 0; 68| 42| } 69| 5.18k| uint16_t diff = uint16_t(word - 0xD800); 70| 5.18k| if (diff > 0x3FF) { ------------------ | Branch (70:11): [True: 78, False: 5.10k] ------------------ 71| 78| return 0; 72| 78| } 73| 5.10k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (73:28): [True: 5.10k, Folded] ------------------ 74| 5.10k| ? u16_swap_bytes(data[pos + 1]) 75| 5.10k| : data[pos + 1]; 76| 5.10k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 77| 5.10k| if (diff2 > 0x3FF) { ------------------ | Branch (77:11): [True: 146, False: 4.96k] ------------------ 78| 146| return 0; 79| 146| } 80| 4.96k| uint32_t value = (diff << 10) + diff2 + 0x10000; 81| | // will generate four UTF-8 bytes 82| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 83| 4.96k| *utf8_output++ = char((value >> 18) | 0b11110000); 84| 4.96k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 85| 4.96k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 86| 4.96k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 87| 4.96k| pos += 2; 88| 4.96k| } 89| 2.53M| } 90| 1.47k| return utf8_output - start; 91| 1.74k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf819convert_with_errorsILNS_10endiannessE0ELb0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT1_Esr7simdutf6detailE26index_assignable_from_charIT2_EEENS_11full_resultES8_mS9_m: 101| 1.68k| size_t utf8_len = 0) { 102| 1.68k| if (check_output && utf8_len == 0) { ------------------ | Branch (102:7): [Folded, False: 1.68k] | Branch (102:23): [True: 0, False: 0] ------------------ 103| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, 0, 0); 104| 0| } 105| | 106| 1.68k| size_t pos = 0; 107| 1.68k| auto start = utf8_output; 108| 1.68k| auto end = utf8_output + utf8_len; 109| | 110| 5.76M| while (pos < len) { ------------------ | Branch (110:10): [True: 5.76M, False: 1.24k] ------------------ 111| |#if SIMDUTF_CPLUSPLUS23 112| | if !consteval 113| |#endif 114| 5.76M| { 115| | // try to convert the next block of 8 bytes 116| 5.76M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (116:11): [True: 5.76M, False: 2.94k] ------------------ 117| | // they are ascii 118| 5.76M| uint64_t v; 119| 5.76M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 120| | if constexpr (!match_system(big_endian)) 121| | v = (v >> 8) | (v << (64 - 8)); 122| 5.76M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (122:13): [True: 1.52M, False: 4.23M] ------------------ 123| 1.52M| size_t final_pos = pos + 4; 124| 7.64M| while (pos < final_pos) { ------------------ | Branch (124:18): [True: 6.11M, False: 1.52M] ------------------ 125| 6.11M| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (125:17): [Folded, False: 6.11M] | 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| 6.11M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (129:30): [Folded, False: 6.11M] ------------------ 130| 6.11M| ? char(u16_swap_bytes(data[pos])) 131| 6.11M| : char(data[pos]); 132| 6.11M| pos++; 133| 6.11M| } 134| 1.52M| continue; 135| 1.52M| } 136| 5.76M| } 137| 5.76M| } 138| | 139| 4.23M| uint16_t word = 140| 4.23M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (140:9): [Folded, False: 4.23M] ------------------ 141| 4.23M| if ((word & 0xFF80) == 0) { ------------------ | Branch (141:9): [True: 234k, False: 4.00M] ------------------ 142| | // will generate one UTF-8 bytes 143| 234k| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (143:11): [Folded, False: 234k] | 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| 234k| *utf8_output++ = char(word); 148| 234k| pos++; 149| 4.00M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (149:16): [True: 1.15M, False: 2.85M] ------------------ 150| | // will generate two UTF-8 bytes 151| | // we have 0b110XXXXX 0b10XXXXXX 152| 1.15M| if (check_output && size_t(end - utf8_output) < 2) { ------------------ | Branch (152:11): [Folded, False: 1.15M] | 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| 1.15M| *utf8_output++ = char((word >> 6) | 0b11000000); 157| 1.15M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 158| 1.15M| pos++; 159| | 160| 2.85M| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (160:16): [True: 2.84M, False: 5.18k] ------------------ 161| | // will generate three UTF-8 bytes 162| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 163| 2.84M| if (check_output && size_t(end - utf8_output) < 3) { ------------------ | Branch (163:11): [Folded, False: 2.84M] | 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| 2.84M| *utf8_output++ = char((word >> 12) | 0b11100000); 168| 2.84M| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 169| 2.84M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 170| 2.84M| pos++; 171| 2.84M| } else { 172| | 173| 5.18k| if (check_output && size_t(end - utf8_output) < 4) { ------------------ | Branch (173:11): [Folded, False: 5.18k] | 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| 5.18k| if (pos + 1 >= len) { ------------------ | Branch (178:11): [True: 84, False: 5.09k] ------------------ 179| 84| return full_result(error_code::SURROGATE, pos, utf8_output - start); 180| 84| } 181| 5.09k| uint16_t diff = uint16_t(word - 0xD800); 182| 5.09k| if (diff > 0x3FF) { ------------------ | Branch (182:11): [True: 148, False: 4.94k] ------------------ 183| 148| return full_result(error_code::SURROGATE, pos, utf8_output - start); 184| 148| } 185| 4.94k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (185:28): [Folded, False: 4.94k] ------------------ 186| 4.94k| ? u16_swap_bytes(data[pos + 1]) 187| 4.94k| : data[pos + 1]; 188| 4.94k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 189| 4.94k| if (diff2 > 0x3FF) { ------------------ | Branch (189:11): [True: 210, False: 4.73k] ------------------ 190| 210| return full_result(error_code::SURROGATE, pos, utf8_output - start); 191| 210| } 192| 4.73k| uint32_t value = (diff << 10) + diff2 + 0x10000; 193| | // will generate four UTF-8 bytes 194| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 195| 4.73k| *utf8_output++ = char((value >> 18) | 0b11110000); 196| 4.73k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 197| 4.73k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 198| 4.73k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 199| 4.73k| pos += 2; 200| 4.73k| } 201| 4.23M| } 202| 1.24k| return full_result(error_code::SUCCESS, pos, utf8_output - start); 203| 1.68k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf819convert_with_errorsILNS_10endiannessE1ELb0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT1_Esr7simdutf6detailE26index_assignable_from_charIT2_EEENS_11full_resultES8_mS9_m: 101| 1.60k| size_t utf8_len = 0) { 102| 1.60k| if (check_output && utf8_len == 0) { ------------------ | Branch (102:7): [Folded, False: 1.60k] | Branch (102:23): [True: 0, False: 0] ------------------ 103| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, 0, 0); 104| 0| } 105| | 106| 1.60k| size_t pos = 0; 107| 1.60k| auto start = utf8_output; 108| 1.60k| auto end = utf8_output + utf8_len; 109| | 110| 3.47M| while (pos < len) { ------------------ | Branch (110:10): [True: 3.46M, False: 1.13k] ------------------ 111| |#if SIMDUTF_CPLUSPLUS23 112| | if !consteval 113| |#endif 114| 3.46M| { 115| | // try to convert the next block of 8 bytes 116| 3.46M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (116:11): [True: 3.46M, False: 2.71k] ------------------ 117| | // they are ascii 118| 3.46M| uint64_t v; 119| 3.46M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 120| | if constexpr (!match_system(big_endian)) 121| 3.46M| v = (v >> 8) | (v << (64 - 8)); 122| 3.46M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (122:13): [True: 1.91M, False: 1.55M] ------------------ 123| 1.91M| size_t final_pos = pos + 4; 124| 9.57M| while (pos < final_pos) { ------------------ | Branch (124:18): [True: 7.65M, False: 1.91M] ------------------ 125| 7.65M| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (125:17): [Folded, False: 7.65M] | 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| 7.65M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (129:30): [True: 7.65M, Folded] ------------------ 130| 7.65M| ? char(u16_swap_bytes(data[pos])) 131| 7.65M| : char(data[pos]); 132| 7.65M| pos++; 133| 7.65M| } 134| 1.91M| continue; 135| 1.91M| } 136| 3.46M| } 137| 3.46M| } 138| | 139| 1.55M| uint16_t word = 140| 1.55M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (140:9): [True: 1.55M, Folded] ------------------ 141| 1.55M| if ((word & 0xFF80) == 0) { ------------------ | Branch (141:9): [True: 271k, False: 1.28M] ------------------ 142| | // will generate one UTF-8 bytes 143| 271k| if (check_output && size_t(end - utf8_output) < 1) { ------------------ | Branch (143:11): [Folded, False: 271k] | 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| 271k| *utf8_output++ = char(word); 148| 271k| pos++; 149| 1.28M| } else if ((word & 0xF800) == 0) { ------------------ | Branch (149:16): [True: 128k, False: 1.15M] ------------------ 150| | // will generate two UTF-8 bytes 151| | // we have 0b110XXXXX 0b10XXXXXX 152| 128k| if (check_output && size_t(end - utf8_output) < 2) { ------------------ | Branch (152:11): [Folded, False: 128k] | 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| 128k| *utf8_output++ = char((word >> 6) | 0b11000000); 157| 128k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 158| 128k| pos++; 159| | 160| 1.15M| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (160:16): [True: 1.14M, False: 7.83k] ------------------ 161| | // will generate three UTF-8 bytes 162| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 163| 1.14M| if (check_output && size_t(end - utf8_output) < 3) { ------------------ | Branch (163:11): [Folded, False: 1.14M] | Branch (163:27): [True: 0, False: 0] ------------------ 164| 0| return full_result(error_code::OUTPUT_BUFFER_TOO_SMALL, pos, 165| 0| utf8_output - start); 166| 0| } 167| 1.14M| *utf8_output++ = char((word >> 12) | 0b11100000); 168| 1.14M| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 169| 1.14M| *utf8_output++ = char((word & 0b111111) | 0b10000000); 170| 1.14M| pos++; 171| 1.14M| } else { 172| | 173| 7.83k| if (check_output && size_t(end - utf8_output) < 4) { ------------------ | Branch (173:11): [Folded, False: 7.83k] | 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| 7.83k| if (pos + 1 >= len) { ------------------ | Branch (178:11): [True: 66, False: 7.76k] ------------------ 179| 66| return full_result(error_code::SURROGATE, pos, utf8_output - start); 180| 66| } 181| 7.76k| uint16_t diff = uint16_t(word - 0xD800); 182| 7.76k| if (diff > 0x3FF) { ------------------ | Branch (182:11): [True: 180, False: 7.58k] ------------------ 183| 180| return full_result(error_code::SURROGATE, pos, utf8_output - start); 184| 180| } 185| 7.58k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (185:28): [True: 7.58k, Folded] ------------------ 186| 7.58k| ? u16_swap_bytes(data[pos + 1]) 187| 7.58k| : data[pos + 1]; 188| 7.58k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 189| 7.58k| if (diff2 > 0x3FF) { ------------------ | Branch (189:11): [True: 228, False: 7.35k] ------------------ 190| 228| return full_result(error_code::SURROGATE, pos, utf8_output - start); 191| 228| } 192| 7.35k| uint32_t value = (diff << 10) + diff2 + 0x10000; 193| | // will generate four UTF-8 bytes 194| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 195| 7.35k| *utf8_output++ = char((value >> 18) | 0b11110000); 196| 7.35k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 197| 7.35k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 198| 7.35k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 199| 7.35k| pos += 2; 200| 7.35k| } 201| 1.55M| } 202| 1.13k| return full_result(error_code::SUCCESS, pos, utf8_output - start); 203| 1.60k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf813convert_validILNS_10endiannessE0EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 384| OutputPtr utf8_output) { 18| 384| size_t pos = 0; 19| 384| auto start = utf8_output; 20| 1.50M| while (pos < len) { ------------------ | Branch (20:10): [True: 1.50M, False: 384] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 1.50M| { 25| | // try to convert the next block of 4 ASCII characters 26| 1.50M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 1.50M, False: 786] ------------------ 27| | // they are ascii 28| 1.50M| uint64_t v; 29| 1.50M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| | if constexpr (!match_system(big_endian)) { 31| | v = (v >> 8) | (v << (64 - 8)); 32| | } 33| 1.50M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.17M, False: 334k] ------------------ 34| 1.17M| size_t final_pos = pos + 4; 35| 5.87M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 4.69M, False: 1.17M] ------------------ 36| 4.69M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [Folded, False: 4.69M] ------------------ 37| 4.69M| ? char(u16_swap_bytes(data[pos])) 38| 4.69M| : char(data[pos]); 39| 4.69M| pos++; 40| 4.69M| } 41| 1.17M| continue; 42| 1.17M| } 43| 1.50M| } 44| 1.50M| } 45| | 46| 334k| uint16_t word = 47| 334k| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (47:9): [Folded, False: 334k] ------------------ 48| 334k| if ((word & 0xFF80) == 0) { ------------------ | Branch (48:9): [True: 152k, False: 182k] ------------------ 49| | // will generate one UTF-8 bytes 50| 152k| *utf8_output++ = char(word); 51| 152k| pos++; 52| 182k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (52:16): [True: 50.8k, False: 131k] ------------------ 53| | // will generate two UTF-8 bytes 54| | // we have 0b110XXXXX 0b10XXXXXX 55| 50.8k| *utf8_output++ = char((word >> 6) | 0b11000000); 56| 50.8k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 57| 50.8k| pos++; 58| 131k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (58:16): [True: 129k, False: 2.62k] ------------------ 59| | // will generate three UTF-8 bytes 60| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 61| 129k| *utf8_output++ = char((word >> 12) | 0b11100000); 62| 129k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 63| 129k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 64| 129k| pos++; 65| 129k| } else { 66| | // must be a surrogate pair 67| 2.62k| uint16_t diff = uint16_t(word - 0xD800); 68| 2.62k| if (pos + 1 >= len) { ------------------ | Branch (68:11): [True: 0, False: 2.62k] ------------------ 69| 0| return 0; 70| 0| } // minimal bound checking 71| 2.62k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (71:28): [Folded, False: 2.62k] ------------------ 72| 2.62k| ? u16_swap_bytes(data[pos + 1]) 73| 2.62k| : data[pos + 1]; 74| 2.62k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 75| 2.62k| uint32_t value = (diff << 10) + diff2 + 0x10000; 76| | // will generate four UTF-8 bytes 77| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 78| 2.62k| *utf8_output++ = char((value >> 18) | 0b11110000); 79| 2.62k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 80| 2.62k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 81| 2.62k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 82| 2.62k| pos += 2; 83| 2.62k| } 84| 334k| } 85| 384| return utf8_output - start; 86| 384|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf16_to_utf813convert_validILNS_10endiannessE1EPKDsPcQaasr7simdutf6detailE18indexes_into_utf16IT0_Esr7simdutf6detailE26index_assignable_from_charIT1_EEEmS8_mS9_: 17| 414| OutputPtr utf8_output) { 18| 414| size_t pos = 0; 19| 414| auto start = utf8_output; 20| 2.91M| while (pos < len) { ------------------ | Branch (20:10): [True: 2.91M, False: 414] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 2.91M| { 25| | // try to convert the next block of 4 ASCII characters 26| 2.91M| if (pos + 4 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (26:11): [True: 2.91M, False: 858] ------------------ 27| | // they are ascii 28| 2.91M| uint64_t v; 29| 2.91M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 30| 2.91M| if constexpr (!match_system(big_endian)) { 31| 2.91M| v = (v >> 8) | (v << (64 - 8)); 32| 2.91M| } 33| 2.91M| if ((v & 0xFF80FF80FF80FF80) == 0) { ------------------ | Branch (33:13): [True: 1.77M, False: 1.13M] ------------------ 34| 1.77M| size_t final_pos = pos + 4; 35| 8.87M| while (pos < final_pos) { ------------------ | Branch (35:18): [True: 7.10M, False: 1.77M] ------------------ 36| 7.10M| *utf8_output++ = !match_system(big_endian) ------------------ | Branch (36:30): [True: 7.10M, Folded] ------------------ 37| 7.10M| ? char(u16_swap_bytes(data[pos])) 38| 7.10M| : char(data[pos]); 39| 7.10M| pos++; 40| 7.10M| } 41| 1.77M| continue; 42| 1.77M| } 43| 2.91M| } 44| 2.91M| } 45| | 46| 1.13M| uint16_t word = 47| 1.13M| !match_system(big_endian) ? u16_swap_bytes(data[pos]) : data[pos]; ------------------ | Branch (47:9): [True: 1.13M, Folded] ------------------ 48| 1.13M| if ((word & 0xFF80) == 0) { ------------------ | Branch (48:9): [True: 266k, False: 869k] ------------------ 49| | // will generate one UTF-8 bytes 50| 266k| *utf8_output++ = char(word); 51| 266k| pos++; 52| 869k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (52:16): [True: 116k, False: 753k] ------------------ 53| | // will generate two UTF-8 bytes 54| | // we have 0b110XXXXX 0b10XXXXXX 55| 116k| *utf8_output++ = char((word >> 6) | 0b11000000); 56| 116k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 57| 116k| pos++; 58| 753k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (58:16): [True: 750k, False: 2.83k] ------------------ 59| | // will generate three UTF-8 bytes 60| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 61| 750k| *utf8_output++ = char((word >> 12) | 0b11100000); 62| 750k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 63| 750k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 64| 750k| pos++; 65| 750k| } else { 66| | // must be a surrogate pair 67| 2.83k| uint16_t diff = uint16_t(word - 0xD800); 68| 2.83k| if (pos + 1 >= len) { ------------------ | Branch (68:11): [True: 0, False: 2.83k] ------------------ 69| 0| return 0; 70| 0| } // minimal bound checking 71| 2.83k| uint16_t next_word = !match_system(big_endian) ------------------ | Branch (71:28): [True: 2.83k, Folded] ------------------ 72| 2.83k| ? u16_swap_bytes(data[pos + 1]) 73| 2.83k| : data[pos + 1]; 74| 2.83k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 75| 2.83k| uint32_t value = (diff << 10) + diff2 + 0x10000; 76| | // will generate four UTF-8 bytes 77| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 78| 2.83k| *utf8_output++ = char((value >> 18) | 0b11110000); 79| 2.83k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 80| 2.83k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 81| 2.83k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 82| 2.83k| pos += 2; 83| 2.83k| } 84| 1.13M| } 85| 414| return utf8_output - start; 86| 414|} _ZN7simdutf6scalar5utf328validateEPKDim: 25| 5.57k| size_t len) noexcept { 26| 5.57k| return validate(reinterpret_cast(buf), len); 27| 5.57k|} _ZN7simdutf6scalar5utf3220validate_with_errorsEPKDim: 49| 7.66k|validate_with_errors(const char32_t *buf, size_t len) noexcept { 50| 7.66k| return validate_with_errors(reinterpret_cast(buf), len); 51| 7.66k|} _ZN7simdutf6scalar5utf3222utf8_length_from_utf32EPKDim: 54| 2.71k| size_t len) { 55| | // We are not BOM aware. 56| 2.71k| size_t counter{0}; 57| 2.46M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (57:22): [True: 2.46M, False: 2.71k] ------------------ 58| | // credit: @ttsugriy for the vectorizable approach 59| 2.46M| counter++; // ASCII 60| 2.46M| counter += static_cast(p[i] > 0x7F); // two-byte 61| 2.46M| counter += static_cast(p[i] > 0x7FF); // three-byte 62| 2.46M| counter += static_cast(p[i] > 0xFFFF); // four-bytes 63| 2.46M| } 64| 2.71k| return counter; 65| 2.71k|} _ZN7simdutf6scalar5utf3223utf16_length_from_utf32EPKDim: 68| 4.01k|utf16_length_from_utf32(const char32_t *p, size_t len) { 69| | // We are not BOM aware. 70| 4.01k| size_t counter{0}; 71| 4.24M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (71:22): [True: 4.24M, False: 4.01k] ------------------ 72| 4.24M| counter++; // non-surrogate word 73| 4.24M| counter += static_cast(p[i] > 0xFFFF); // surrogate pair 74| 4.24M| } 75| 4.01k| return counter; 76| 4.01k|} _ZN7simdutf6scalar5utf328validateIPKjQsr7simdutf6detailE19indexes_into_uint32IT_EEEbS5_m: 13| 5.57k| size_t len) noexcept { 14| 5.57k| uint64_t pos = 0; 15| 1.11M| for (; pos < len; pos++) { ------------------ | Branch (15:10): [True: 1.10M, False: 2.43k] ------------------ 16| 1.10M| uint32_t word = data[pos]; 17| 1.10M| if (word > 0x10FFFF || (word >= 0xD800 && word <= 0xDFFF)) { ------------------ | Branch (17:9): [True: 2.67k, False: 1.10M] | Branch (17:29): [True: 33.9k, False: 1.07M] | Branch (17:47): [True: 459, False: 33.4k] ------------------ 18| 3.13k| return false; 19| 3.13k| } 20| 1.10M| } 21| 2.43k| return true; 22| 5.57k|} _ZN7simdutf6scalar5utf3220validate_with_errorsIPKjQsr7simdutf6detailE19indexes_into_uint32IT_EEENS_6resultES5_m: 34| 7.66k|validate_with_errors(InputPtr data, size_t len) noexcept { 35| 7.66k| size_t pos = 0; 36| 1.11M| for (; pos < len; pos++) { ------------------ | Branch (36:10): [True: 1.11M, False: 2.43k] ------------------ 37| 1.11M| uint32_t word = data[pos]; 38| 1.11M| if (word > 0x10FFFF) { ------------------ | Branch (38:9): [True: 4.38k, False: 1.10M] ------------------ 39| 4.38k| return result(error_code::TOO_LARGE, pos); 40| 4.38k| } 41| 1.10M| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (41:9): [True: 35.3k, False: 1.07M] | Branch (41:27): [True: 849, False: 34.4k] ------------------ 42| 849| return result(error_code::SURROGATE, pos); 43| 849| } 44| 1.10M| } 45| 2.43k| return result(error_code::SUCCESS, pos); 46| 7.66k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf32_to_latin17convertEPKDimPc: 12| 486| char *latin1_output) { 13| 486| char *start = latin1_output; 14| 486| uint32_t utf32_char; 15| 486| size_t pos = 0; 16| 486| uint32_t too_large = 0; 17| | 18| 1.18M| while (pos < len) { ------------------ | Branch (18:10): [True: 1.18M, False: 486] ------------------ 19| 1.18M| utf32_char = (uint32_t)data[pos]; 20| 1.18M| too_large |= utf32_char; 21| 1.18M| *latin1_output++ = (char)(utf32_char & 0xFF); 22| 1.18M| pos++; 23| 1.18M| } 24| 486| if ((too_large & 0xFFFFFF00) != 0) { ------------------ | Branch (24:7): [True: 370, False: 116] ------------------ 25| 370| return 0; 26| 370| } 27| 116| return latin1_output - start; 28| 486|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_115utf32_to_latin119convert_with_errorsEPKDimPc: 32| 1.32k| char *latin1_output) { 33| 1.32k| char *start{latin1_output}; 34| 1.32k| size_t pos = 0; 35| 213k| while (pos < len) { ------------------ | Branch (35:10): [True: 213k, False: 108] ------------------ 36| |#if SIMDUTF_CPLUSPLUS23 37| | if !consteval 38| |#endif 39| 213k| { 40| 213k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (40:11): [True: 212k, False: 336] ------------------ 41| | // they are Latin1 42| 212k| uint64_t v; 43| 212k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 44| 212k| if ((v & 0xFFFFFF00FFFFFF00) == 0) { ------------------ | Branch (44:13): [True: 211k, False: 1.10k] ------------------ 45| 211k| *latin1_output++ = char(data[pos]); 46| 211k| *latin1_output++ = char(data[pos + 1]); 47| 211k| pos += 2; 48| 211k| continue; 49| 211k| } 50| 212k| } 51| 213k| } 52| | 53| 1.43k| uint32_t utf32_char = data[pos]; 54| 1.43k| if ((utf32_char & 0xFFFFFF00) == ------------------ | Branch (54:9): [True: 224, False: 1.21k] ------------------ 55| 1.43k| 0) { // Check if the character can be represented in Latin-1 56| 224| *latin1_output++ = (char)(utf32_char & 0xFF); 57| 224| pos++; 58| 1.21k| } else { 59| 1.21k| return result(error_code::TOO_LARGE, pos); 60| 1.21k| }; 61| 224| } 62| 108| return result(error_code::SUCCESS, latin1_output - start); 63| 1.32k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf167convertILNS_10endiannessE0EEEmPKDimPDs: 11| 1.68k| char16_t *utf16_output) { 12| 1.68k| size_t pos = 0; 13| 1.68k| char16_t *start{utf16_output}; 14| 495k| while (pos < len) { ------------------ | Branch (14:10): [True: 494k, False: 874] ------------------ 15| 494k| uint32_t word = data[pos]; 16| 494k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 479k, False: 15.2k] ------------------ 17| 479k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (17:11): [True: 1.66k, False: 478k] | Branch (17:29): [True: 176, False: 1.48k] ------------------ 18| 176| return 0; 19| 176| } 20| | // will not generate a surrogate pair 21| 479k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (21:25): [Folded, False: 479k] ------------------ 22| 479k| ? char16_t(u16_swap_bytes(uint16_t(word))) 23| 479k| : char16_t(word); 24| 479k| } else { 25| | // will generate a surrogate pair 26| 15.2k| if (word > 0x10FFFF) { ------------------ | Branch (26:11): [True: 636, False: 14.5k] ------------------ 27| 636| return 0; 28| 636| } 29| 14.5k| word -= 0x10000; 30| 14.5k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 31| 14.5k| 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| 14.5k| *utf16_output++ = char16_t(high_surrogate); 37| 14.5k| *utf16_output++ = char16_t(low_surrogate); 38| 14.5k| } 39| 494k| pos++; 40| 494k| } 41| 874| return utf16_output - start; 42| 1.68k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf167convertILNS_10endiannessE1EEEmPKDimPDs: 11| 1.32k| char16_t *utf16_output) { 12| 1.32k| size_t pos = 0; 13| 1.32k| char16_t *start{utf16_output}; 14| 39.2k| while (pos < len) { ------------------ | Branch (14:10): [True: 38.5k, False: 630] ------------------ 15| 38.5k| uint32_t word = data[pos]; 16| 38.5k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 33.4k, False: 5.12k] ------------------ 17| 33.4k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (17:11): [True: 1.02k, False: 32.4k] | Branch (17:29): [True: 94, False: 926] ------------------ 18| 94| return 0; 19| 94| } 20| | // will not generate a surrogate pair 21| 33.3k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (21:25): [True: 33.3k, Folded] ------------------ 22| 33.3k| ? char16_t(u16_swap_bytes(uint16_t(word))) 23| 33.3k| : char16_t(word); 24| 33.3k| } else { 25| | // will generate a surrogate pair 26| 5.12k| if (word > 0x10FFFF) { ------------------ | Branch (26:11): [True: 598, False: 4.52k] ------------------ 27| 598| return 0; 28| 598| } 29| 4.52k| word -= 0x10000; 30| 4.52k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 31| 4.52k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 32| 4.52k| if constexpr (!match_system(big_endian)) { 33| 4.52k| high_surrogate = u16_swap_bytes(high_surrogate); 34| 4.52k| low_surrogate = u16_swap_bytes(low_surrogate); 35| 4.52k| } 36| 4.52k| *utf16_output++ = char16_t(high_surrogate); 37| 4.52k| *utf16_output++ = char16_t(low_surrogate); 38| 4.52k| } 39| 37.8k| pos++; 40| 37.8k| } 41| 630| return utf16_output - start; 42| 1.32k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1619convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKDimPDs: 46| 2.11k| char16_t *utf16_output) { 47| 2.11k| size_t pos = 0; 48| 2.11k| char16_t *start{utf16_output}; 49| 75.6k| while (pos < len) { ------------------ | Branch (49:10): [True: 75.1k, False: 504] ------------------ 50| 75.1k| uint32_t word = data[pos]; 51| 75.1k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (51:9): [True: 66.6k, False: 8.55k] ------------------ 52| 66.6k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (52:11): [True: 1.75k, False: 64.8k] | Branch (52:29): [True: 330, False: 1.42k] ------------------ 53| 330| return result(error_code::SURROGATE, pos); 54| 330| } 55| | // will not generate a surrogate pair 56| 66.2k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (56:25): [Folded, False: 66.2k] ------------------ 57| 66.2k| ? char16_t(u16_swap_bytes(uint16_t(word))) 58| 66.2k| : char16_t(word); 59| 66.2k| } else { 60| | // will generate a surrogate pair 61| 8.55k| if (word > 0x10FFFF) { ------------------ | Branch (61:11): [True: 1.27k, False: 7.27k] ------------------ 62| 1.27k| return result(error_code::TOO_LARGE, pos); 63| 1.27k| } 64| 7.27k| word -= 0x10000; 65| 7.27k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 66| 7.27k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 67| | if constexpr (!match_system(big_endian)) { 68| | high_surrogate = u16_swap_bytes(high_surrogate); 69| | low_surrogate = u16_swap_bytes(low_surrogate); 70| | } 71| 7.27k| *utf16_output++ = char16_t(high_surrogate); 72| 7.27k| *utf16_output++ = char16_t(low_surrogate); 73| 7.27k| } 74| 73.5k| pos++; 75| 73.5k| } 76| 504| return result(error_code::SUCCESS, utf16_output - start); 77| 2.11k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1619convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKDimPDs: 46| 1.95k| char16_t *utf16_output) { 47| 1.95k| size_t pos = 0; 48| 1.95k| char16_t *start{utf16_output}; 49| 82.6k| while (pos < len) { ------------------ | Branch (49:10): [True: 82.1k, False: 520] ------------------ 50| 82.1k| uint32_t word = data[pos]; 51| 82.1k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (51:9): [True: 78.1k, False: 4.01k] ------------------ 52| 78.1k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (52:11): [True: 1.71k, False: 76.3k] | Branch (52:29): [True: 252, False: 1.46k] ------------------ 53| 252| return result(error_code::SURROGATE, pos); 54| 252| } 55| | // will not generate a surrogate pair 56| 77.8k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (56:25): [True: 77.8k, Folded] ------------------ 57| 77.8k| ? char16_t(u16_swap_bytes(uint16_t(word))) 58| 77.8k| : char16_t(word); 59| 77.8k| } else { 60| | // will generate a surrogate pair 61| 4.01k| if (word > 0x10FFFF) { ------------------ | Branch (61:11): [True: 1.18k, False: 2.82k] ------------------ 62| 1.18k| return result(error_code::TOO_LARGE, pos); 63| 1.18k| } 64| 2.82k| word -= 0x10000; 65| 2.82k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 66| 2.82k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 67| 2.82k| if constexpr (!match_system(big_endian)) { 68| 2.82k| high_surrogate = u16_swap_bytes(high_surrogate); 69| 2.82k| low_surrogate = u16_swap_bytes(low_surrogate); 70| 2.82k| } 71| 2.82k| *utf16_output++ = char16_t(high_surrogate); 72| 2.82k| *utf16_output++ = char16_t(low_surrogate); 73| 2.82k| } 74| 80.6k| pos++; 75| 80.6k| } 76| 520| return result(error_code::SUCCESS, utf16_output - start); 77| 1.95k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1613convert_validILNS_10endiannessE0EEEmPKDimPDs: 11| 252| char16_t *utf16_output) { 12| 252| size_t pos = 0; 13| 252| char16_t *start{utf16_output}; 14| 305k| while (pos < len) { ------------------ | Branch (14:10): [True: 304k, False: 252] ------------------ 15| 304k| uint32_t word = data[pos]; 16| 304k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 301k, False: 3.48k] ------------------ 17| | // will not generate a surrogate pair 18| 301k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (18:25): [Folded, False: 301k] ------------------ 19| 301k| ? char16_t(u16_swap_bytes(uint16_t(word))) 20| 301k| : char16_t(word); 21| 301k| pos++; 22| 301k| } else { 23| | // will generate a surrogate pair 24| 3.48k| word -= 0x10000; 25| 3.48k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 26| 3.48k| 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| 3.48k| *utf16_output++ = char16_t(high_surrogate); 32| 3.48k| *utf16_output++ = char16_t(low_surrogate); 33| 3.48k| pos++; 34| 3.48k| } 35| 304k| } 36| 252| return utf16_output - start; 37| 252|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf32_to_utf1613convert_validILNS_10endiannessE1EEEmPKDimPDs: 11| 114| char16_t *utf16_output) { 12| 114| size_t pos = 0; 13| 114| char16_t *start{utf16_output}; 14| 6.75k| while (pos < len) { ------------------ | Branch (14:10): [True: 6.63k, False: 114] ------------------ 15| 6.63k| uint32_t word = data[pos]; 16| 6.63k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (16:9): [True: 3.00k, False: 3.63k] ------------------ 17| | // will not generate a surrogate pair 18| 3.00k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (18:25): [True: 3.00k, Folded] ------------------ 19| 3.00k| ? char16_t(u16_swap_bytes(uint16_t(word))) 20| 3.00k| : char16_t(word); 21| 3.00k| pos++; 22| 3.63k| } else { 23| | // will generate a surrogate pair 24| 3.63k| word -= 0x10000; 25| 3.63k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 26| 3.63k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 27| 3.63k| if constexpr (!match_system(big_endian)) { 28| 3.63k| high_surrogate = u16_swap_bytes(high_surrogate); 29| 3.63k| low_surrogate = u16_swap_bytes(low_surrogate); 30| 3.63k| } 31| 3.63k| *utf16_output++ = char16_t(high_surrogate); 32| 3.63k| *utf16_output++ = char16_t(low_surrogate); 33| 3.63k| pos++; 34| 3.63k| } 35| 6.63k| } 36| 114| return utf16_output - start; 37| 114|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf32_to_utf87convertIPKDiPcQaasr7simdutf6detailE18indexes_into_utf32IT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEEmS7_mS8_: 17| 1.83k| OutputPtr utf8_output) { 18| 1.83k| size_t pos = 0; 19| 1.83k| auto start = utf8_output; 20| 81.8k| while (pos < len) { ------------------ | Branch (20:10): [True: 81.1k, False: 694] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 81.1k| { // try to convert the next block of 2 ASCII characters 25| 81.1k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (25:11): [True: 80.2k, False: 938] ------------------ 26| | // they are ascii 27| 80.2k| uint64_t v; 28| 80.2k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 29| 80.2k| if ((v & 0xFFFFFF80FFFFFF80) == 0) { ------------------ | Branch (29:13): [True: 54.1k, False: 26.0k] ------------------ 30| 54.1k| *utf8_output++ = char(data[pos]); 31| 54.1k| *utf8_output++ = char(data[pos + 1]); 32| 54.1k| pos += 2; 33| 54.1k| continue; 34| 54.1k| } 35| 80.2k| } 36| 81.1k| } 37| | 38| 26.9k| uint32_t word = data[pos]; 39| 26.9k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (39:9): [True: 3.43k, False: 23.5k] ------------------ 40| | // will generate one UTF-8 bytes 41| 3.43k| *utf8_output++ = char(word); 42| 3.43k| pos++; 43| 23.5k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (43:16): [True: 9.70k, False: 13.8k] ------------------ 44| | // will generate two UTF-8 bytes 45| | // we have 0b110XXXXX 0b10XXXXXX 46| 9.70k| *utf8_output++ = char((word >> 6) | 0b11000000); 47| 9.70k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 48| 9.70k| pos++; 49| 13.8k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (49:16): [True: 8.89k, False: 4.95k] ------------------ 50| | // will generate three UTF-8 bytes 51| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 52| 8.89k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (52:11): [True: 6.27k, False: 2.62k] | Branch (52:29): [True: 192, False: 6.07k] ------------------ 53| 192| return 0; 54| 192| } 55| 8.70k| *utf8_output++ = char((word >> 12) | 0b11100000); 56| 8.70k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 57| 8.70k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 58| 8.70k| pos++; 59| 8.70k| } else { 60| | // will generate four UTF-8 bytes 61| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 62| 4.95k| if (word > 0x10FFFF) { ------------------ | Branch (62:11): [True: 944, False: 4.01k] ------------------ 63| 944| return 0; 64| 944| } 65| 4.01k| *utf8_output++ = char((word >> 18) | 0b11110000); 66| 4.01k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 67| 4.01k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 68| 4.01k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 69| 4.01k| pos++; 70| 4.01k| } 71| 26.9k| } 72| 694| return utf8_output - start; 73| 1.83k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf32_to_utf819convert_with_errorsIPKDiPcQaasr7simdutf6detailE18indexes_into_utf32IT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEENS_6resultES7_mS8_: 81| 2.76k| OutputPtr utf8_output) { 82| 2.76k| size_t pos = 0; 83| 2.76k| auto start = utf8_output; 84| 273k| while (pos < len) { ------------------ | Branch (84:10): [True: 272k, False: 694] ------------------ 85| |#if SIMDUTF_CPLUSPLUS23 86| | if !consteval 87| |#endif 88| 272k| { // try to convert the next block of 2 ASCII characters 89| 272k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (89:11): [True: 271k, False: 1.10k] ------------------ 90| | // they are ascii 91| 271k| uint64_t v; 92| 271k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 93| 271k| if ((v & 0xFFFFFF80FFFFFF80) == 0) { ------------------ | Branch (93:13): [True: 217k, False: 54.1k] ------------------ 94| 217k| *utf8_output++ = char(data[pos]); 95| 217k| *utf8_output++ = char(data[pos + 1]); 96| 217k| pos += 2; 97| 217k| continue; 98| 217k| } 99| 271k| } 100| 272k| } 101| | 102| 55.2k| uint32_t word = data[pos]; 103| 55.2k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (103:9): [True: 5.58k, False: 49.6k] ------------------ 104| | // will generate one UTF-8 bytes 105| 5.58k| *utf8_output++ = char(word); 106| 5.58k| pos++; 107| 49.6k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (107:16): [True: 28.5k, False: 21.1k] ------------------ 108| | // will generate two UTF-8 bytes 109| | // we have 0b110XXXXX 0b10XXXXXX 110| 28.5k| *utf8_output++ = char((word >> 6) | 0b11000000); 111| 28.5k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 112| 28.5k| pos++; 113| 28.5k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (113:16): [True: 9.40k, False: 11.7k] ------------------ 114| | // will generate three UTF-8 bytes 115| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 116| 9.40k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (116:11): [True: 6.20k, False: 3.20k] | Branch (116:29): [True: 342, False: 5.85k] ------------------ 117| 342| return result(error_code::SURROGATE, pos); 118| 342| } 119| 9.05k| *utf8_output++ = char((word >> 12) | 0b11100000); 120| 9.05k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 121| 9.05k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 122| 9.05k| pos++; 123| 11.7k| } else { 124| | // will generate four UTF-8 bytes 125| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 126| 11.7k| if (word > 0x10FFFF) { ------------------ | Branch (126:11): [True: 1.72k, False: 10.0k] ------------------ 127| 1.72k| return result(error_code::TOO_LARGE, pos); 128| 1.72k| } 129| 10.0k| *utf8_output++ = char((word >> 18) | 0b11110000); 130| 10.0k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 131| 10.0k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 132| 10.0k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 133| 10.0k| pos++; 134| 10.0k| } 135| 55.2k| } 136| 694| return result(error_code::SUCCESS, utf8_output - start); 137| 2.76k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf32_to_utf813convert_validIPKDiPcQaasr7simdutf6detailE18indexes_into_utf32IT_Esr7simdutf6detailE26index_assignable_from_charIT0_EEEmS7_mS8_: 17| 188| OutputPtr utf8_output) { 18| 188| size_t pos = 0; 19| 188| auto start = utf8_output; 20| 84.6k| while (pos < len) { ------------------ | Branch (20:10): [True: 84.4k, False: 188] ------------------ 21| |#if SIMDUTF_CPLUSPLUS23 22| | if !consteval 23| |#endif 24| 84.4k| { // try to convert the next block of 2 ASCII characters 25| 84.4k| if (pos + 2 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (25:11): [True: 84.2k, False: 170] ------------------ 26| | // they are ascii 27| 84.2k| uint64_t v; 28| 84.2k| ::memcpy(&v, data + pos, sizeof(uint64_t)); 29| 84.2k| if ((v & 0xFFFFFF80FFFFFF80) == 0) { ------------------ | Branch (29:13): [True: 74.3k, False: 9.92k] ------------------ 30| 74.3k| *utf8_output++ = char(data[pos]); 31| 74.3k| *utf8_output++ = char(data[pos + 1]); 32| 74.3k| pos += 2; 33| 74.3k| continue; 34| 74.3k| } 35| 84.2k| } 36| 84.4k| } 37| | 38| 10.0k| uint32_t word = data[pos]; 39| 10.0k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (39:9): [True: 642, False: 9.45k] ------------------ 40| | // will generate one UTF-8 bytes 41| 642| *utf8_output++ = char(word); 42| 642| pos++; 43| 9.45k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (43:16): [True: 5.26k, False: 4.19k] ------------------ 44| | // will generate two UTF-8 bytes 45| | // we have 0b110XXXXX 0b10XXXXXX 46| 5.26k| *utf8_output++ = char((word >> 6) | 0b11000000); 47| 5.26k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 48| 5.26k| pos++; 49| 5.26k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (49:16): [True: 860, False: 3.33k] ------------------ 50| | // will generate three UTF-8 bytes 51| | // we have 0b1110XXXX 0b10XXXXXX 0b10XXXXXX 52| 860| *utf8_output++ = char((word >> 12) | 0b11100000); 53| 860| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 54| 860| *utf8_output++ = char((word & 0b111111) | 0b10000000); 55| 860| pos++; 56| 3.33k| } else { 57| | // will generate four UTF-8 bytes 58| | // we have 0b11110XXX 0b10XXXXXX 0b10XXXXXX 0b10XXXXXX 59| 3.33k| *utf8_output++ = char((word >> 18) | 0b11110000); 60| 3.33k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 61| 3.33k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 62| 3.33k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 63| 3.33k| pos++; 64| 3.33k| } 65| 10.0k| } 66| 188| return utf8_output - start; 67| 188|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf88validateEPKcm: 113| 3.52k| size_t len) noexcept { 114| 3.52k| return validate(reinterpret_cast(buf), len); 115| 3.52k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf88validateIPKhEEbT_m: 14| 3.52k| size_t len) noexcept { 15| 3.52k| static_assert( 16| 3.52k| std::is_same::type, uint8_t>::value, 17| 3.52k| "dereferencing the data pointer must result in a uint8_t"); 18| 3.52k| uint64_t pos = 0; 19| 3.52k| uint32_t code_point = 0; 20| 8.53M| while (pos < len) { ------------------ | Branch (20:10): [True: 8.52M, False: 614] ------------------ 21| 8.52M| uint64_t next_pos; 22| |#if SIMDUTF_CPLUSPLUS23 23| | if !consteval 24| |#endif 25| 8.52M| { // check if the next 16 bytes are ascii. 26| 8.52M| next_pos = pos + 16; 27| 8.52M| if (next_pos <= len) { // if it is safe to read 16 more bytes, check ------------------ | Branch (27:11): [True: 8.52M, False: 3.35k] ------------------ 28| | // that they are ascii 29| 8.52M| uint64_t v1{}; 30| 8.52M| std::memcpy(&v1, data + pos, sizeof(uint64_t)); 31| 8.52M| uint64_t v2{}; 32| 8.52M| std::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 33| 8.52M| uint64_t v{v1 | v2}; 34| 8.52M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (34:13): [True: 7.48M, False: 1.03M] ------------------ 35| 7.48M| pos = next_pos; 36| 7.48M| continue; 37| 7.48M| } 38| 8.52M| } 39| 8.52M| } 40| | 41| 1.04M| unsigned char byte = data[pos]; 42| | 43| 5.41M| while (byte < 0b10000000) { ------------------ | Branch (43:12): [True: 4.37M, False: 1.04M] ------------------ 44| 4.37M| if (++pos == len) { ------------------ | Branch (44:11): [True: 1.10k, False: 4.37M] ------------------ 45| 1.10k| return true; 46| 1.10k| } 47| 4.37M| byte = data[pos]; 48| 4.37M| } 49| | 50| 1.04M| if ((byte & 0b11100000) == 0b11000000) { ------------------ | Branch (50:9): [True: 692k, False: 349k] ------------------ 51| 692k| next_pos = pos + 2; 52| 692k| if (next_pos > len) { ------------------ | Branch (52:11): [True: 80, False: 692k] ------------------ 53| 80| return false; 54| 80| } 55| 692k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (55:11): [True: 242, False: 692k] ------------------ 56| 242| return false; 57| 242| } 58| | // range check 59| 692k| code_point = (byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 60| 692k| if (code_point < 0x80) { ------------------ | Branch (60:11): [True: 31, False: 692k] ------------------ 61| 31| return false; 62| 31| } 63| 692k| } else if ((byte & 0b11110000) == 0b11100000) { ------------------ | Branch (63:16): [True: 295k, False: 53.9k] ------------------ 64| 295k| next_pos = pos + 3; 65| 295k| if (next_pos > len) { ------------------ | Branch (65:11): [True: 48, False: 295k] ------------------ 66| 48| return false; 67| 48| } 68| 295k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (68:11): [True: 110, False: 295k] ------------------ 69| 110| return false; 70| 110| } 71| 295k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (71:11): [True: 60, False: 294k] ------------------ 72| 60| return false; 73| 60| } 74| | // range check 75| 294k| code_point = (byte & 0b00001111) << 12 | 76| 294k| (data[pos + 1] & 0b00111111) << 6 | 77| 294k| (data[pos + 2] & 0b00111111); 78| 294k| if ((code_point < 0x800) || ------------------ | Branch (78:11): [True: 47, False: 294k] ------------------ 79| 294k| (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (79:12): [True: 8.61k, False: 286k] | Branch (79:35): [True: 22, False: 8.59k] ------------------ 80| 69| return false; 81| 69| } 82| 294k| } else if ((byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (82:16): [True: 53.0k, False: 837] ------------------ 83| 53.0k| next_pos = pos + 4; 84| 53.0k| if (next_pos > len) { ------------------ | Branch (84:11): [True: 40, False: 53.0k] ------------------ 85| 40| return false; 86| 40| } 87| 53.0k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (87:11): [True: 102, False: 52.9k] ------------------ 88| 102| return false; 89| 102| } 90| 52.9k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (90:11): [True: 52, False: 52.8k] ------------------ 91| 52| return false; 92| 52| } 93| 52.8k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (93:11): [True: 37, False: 52.8k] ------------------ 94| 37| return false; 95| 37| } 96| | // range check 97| 52.8k| code_point = 98| 52.8k| (byte & 0b00000111) << 18 | (data[pos + 1] & 0b00111111) << 12 | 99| 52.8k| (data[pos + 2] & 0b00111111) << 6 | (data[pos + 3] & 0b00111111); 100| 52.8k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (100:11): [True: 49, False: 52.7k] | Branch (100:35): [True: 47, False: 52.7k] ------------------ 101| 96| return false; 102| 96| } 103| 52.8k| } else { 104| | // we may have a continuation 105| 837| return false; 106| 837| } 107| 1.03M| pos = next_pos; 108| 1.03M| } 109| 614| return true; 110| 3.52k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf820validate_with_errorsEPKcm: 221| 6.87k|validate_with_errors(const char *buf, size_t len) noexcept { 222| 6.87k| return validate_with_errors(reinterpret_cast(buf), len); 223| 6.87k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf820validate_with_errorsIPKhEENS_6resultET_m: 119| 6.87k|validate_with_errors(BytePtr data, size_t len) noexcept { 120| 6.87k| static_assert( 121| 6.87k| std::is_same::type, uint8_t>::value, 122| 6.87k| "dereferencing the data pointer must result in a uint8_t"); 123| 6.87k| size_t pos = 0; 124| 6.87k| uint32_t code_point = 0; 125| 8.53M| while (pos < len) { ------------------ | Branch (125:10): [True: 8.53M, False: 614] ------------------ 126| | // check of the next 16 bytes are ascii. 127| 8.53M| size_t next_pos = pos + 16; 128| 8.53M| if (next_pos <= ------------------ | Branch (128:9): [True: 8.53M, False: 5.19k] ------------------ 129| 8.53M| len) { // if it is safe to read 16 more bytes, check that they are ascii 130| 8.53M| uint64_t v1; 131| 8.53M| std::memcpy(&v1, data + pos, sizeof(uint64_t)); 132| 8.53M| uint64_t v2; 133| 8.53M| std::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 134| 8.53M| uint64_t v{v1 | v2}; 135| 8.53M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (135:11): [True: 7.48M, False: 1.04M] ------------------ 136| 7.48M| pos = next_pos; 137| 7.48M| continue; 138| 7.48M| } 139| 8.53M| } 140| 1.04M| unsigned char byte = data[pos]; 141| | 142| 5.43M| while (byte < 0b10000000) { ------------------ | Branch (142:12): [True: 4.39M, False: 1.04M] ------------------ 143| 4.39M| if (++pos == len) { ------------------ | Branch (143:11): [True: 1.10k, False: 4.38M] ------------------ 144| 1.10k| return result(error_code::SUCCESS, len); 145| 1.10k| } 146| 4.38M| byte = data[pos]; 147| 4.38M| } 148| | 149| 1.04M| if ((byte & 0b11100000) == 0b11000000) { ------------------ | Branch (149:9): [True: 695k, False: 353k] ------------------ 150| 695k| next_pos = pos + 2; 151| 695k| if (next_pos > len) { ------------------ | Branch (151:11): [True: 240, False: 695k] ------------------ 152| 240| return result(error_code::TOO_SHORT, pos); 153| 240| } 154| 695k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (154:11): [True: 726, False: 694k] ------------------ 155| 726| return result(error_code::TOO_SHORT, pos); 156| 726| } 157| | // range check 158| 694k| code_point = (byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 159| 694k| if (code_point < 0x80) { ------------------ | Branch (159:11): [True: 93, False: 694k] ------------------ 160| 93| return result(error_code::OVERLONG, pos); 161| 93| } 162| 694k| } else if ((byte & 0b11110000) == 0b11100000) { ------------------ | Branch (162:16): [True: 296k, False: 56.4k] ------------------ 163| 296k| next_pos = pos + 3; 164| 296k| if (next_pos > len) { ------------------ | Branch (164:11): [True: 144, False: 296k] ------------------ 165| 144| return result(error_code::TOO_SHORT, pos); 166| 144| } 167| 296k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (167:11): [True: 330, False: 296k] ------------------ 168| 330| return result(error_code::TOO_SHORT, pos); 169| 330| } 170| 296k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (170:11): [True: 180, False: 296k] ------------------ 171| 180| return result(error_code::TOO_SHORT, pos); 172| 180| } 173| | // range check 174| 296k| code_point = (byte & 0b00001111) << 12 | 175| 296k| (data[pos + 1] & 0b00111111) << 6 | 176| 296k| (data[pos + 2] & 0b00111111); 177| 296k| if (code_point < 0x800) { ------------------ | Branch (177:11): [True: 141, False: 295k] ------------------ 178| 141| return result(error_code::OVERLONG, pos); 179| 141| } 180| 295k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (180:11): [True: 9.19k, False: 286k] | Branch (180:34): [True: 66, False: 9.13k] ------------------ 181| 66| return result(error_code::SURROGATE, pos); 182| 66| } 183| 295k| } else if ((byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (183:16): [True: 54.1k, False: 2.25k] ------------------ 184| 54.1k| next_pos = pos + 4; 185| 54.1k| if (next_pos > len) { ------------------ | Branch (185:11): [True: 120, False: 54.0k] ------------------ 186| 120| return result(error_code::TOO_SHORT, pos); 187| 120| } 188| 54.0k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (188:11): [True: 306, False: 53.7k] ------------------ 189| 306| return result(error_code::TOO_SHORT, pos); 190| 306| } 191| 53.7k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (191:11): [True: 156, False: 53.5k] ------------------ 192| 156| return result(error_code::TOO_SHORT, pos); 193| 156| } 194| 53.5k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (194:11): [True: 111, False: 53.4k] ------------------ 195| 111| return result(error_code::TOO_SHORT, pos); 196| 111| } 197| | // range check 198| 53.4k| code_point = 199| 53.4k| (byte & 0b00000111) << 18 | (data[pos + 1] & 0b00111111) << 12 | 200| 53.4k| (data[pos + 2] & 0b00111111) << 6 | (data[pos + 3] & 0b00111111); 201| 53.4k| if (code_point <= 0xffff) { ------------------ | Branch (201:11): [True: 147, False: 53.3k] ------------------ 202| 147| return result(error_code::OVERLONG, pos); 203| 147| } 204| 53.3k| if (0x10ffff < code_point) { ------------------ | Branch (204:11): [True: 141, False: 53.1k] ------------------ 205| 141| return result(error_code::TOO_LARGE, pos); 206| 141| } 207| 53.3k| } else { 208| | // we either have too many continuation bytes or an invalid leading byte 209| 2.25k| if ((byte & 0b11000000) == 0b10000000) { ------------------ | Branch (209:11): [True: 1.07k, False: 1.18k] ------------------ 210| 1.07k| return result(error_code::TOO_LONG, pos); 211| 1.18k| } else { 212| 1.18k| return result(error_code::HEADER_BITS, pos); 213| 1.18k| } 214| 2.25k| } 215| 1.04M| pos = next_pos; 216| 1.04M| } 217| 614| return result(error_code::SUCCESS, len); 218| 6.87k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf831rewind_and_validate_with_errorsEPKcS4_m: 232| 3.60k| const char *start, const char *buf, size_t len) noexcept { 233| | // First check that we start with a leading byte 234| 3.60k| if ((*start & 0b11000000) == 0b10000000) { ------------------ | Branch (234:7): [True: 252, False: 3.35k] ------------------ 235| 252| return result(error_code::TOO_LONG, 0); 236| 252| } 237| 3.35k| size_t extra_len{0}; 238| | // A leading byte cannot be further than 4 bytes away 239| 3.55k| for (int i = 0; i < 5; i++) { ------------------ | Branch (239:19): [True: 3.55k, False: 0] ------------------ 240| 3.55k| unsigned char byte = *buf; 241| 3.55k| if ((byte & 0b11000000) != 0b10000000) { ------------------ | Branch (241:9): [True: 3.35k, False: 200] ------------------ 242| 3.35k| break; 243| 3.35k| } else { 244| 200| buf--; 245| 200| extra_len++; 246| 200| } 247| 3.55k| } 248| | 249| 3.35k| result res = validate_with_errors(buf, len + extra_len); 250| 3.35k| res.count -= extra_len; // Might underflow 251| 3.35k| return res; 252| 3.60k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf817count_code_pointsIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_m: 258| 15.3k|simdutf_constexpr23 size_t count_code_points(InputPtr data, size_t len) { 259| 15.3k| size_t counter{0}; 260| 266M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (260:22): [True: 266M, False: 15.3k] ------------------ 261| | // -65 is 0b10111111, anything larger in two-complement's should start a new 262| | // code point. 263| 266M| if (int8_t(data[i]) > -65) { ------------------ | Branch (263:9): [True: 263M, False: 3.07M] ------------------ 264| 263M| counter++; 265| 263M| } 266| 266M| } 267| 15.3k| return counter; 268| 15.3k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_14utf822utf16_length_from_utf8IPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_m: 274| 5.77k|simdutf_constexpr23 size_t utf16_length_from_utf8(InputPtr data, size_t len) { 275| 5.77k| size_t counter{0}; 276| 88.2M| for (size_t i = 0; i < len; i++) { ------------------ | Branch (276:22): [True: 88.2M, False: 5.77k] ------------------ 277| 88.2M| if (int8_t(data[i]) > -65) { ------------------ | Branch (277:9): [True: 87.0M, False: 1.22M] ------------------ 278| 87.0M| counter++; 279| 87.0M| } 280| 88.2M| if (uint8_t(data[i]) >= 240) { ------------------ | Branch (280:9): [True: 97.8k, False: 88.1M] ------------------ 281| 97.8k| counter++; 282| 97.8k| } 283| 88.2M| } 284| 5.77k| return counter; 285| 5.77k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_114utf8_to_latin130rewind_and_convert_with_errorsEmPKcmPc: 176| 1.03k| char *latin1_output) { 177| 1.03k| 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| 1.03k| 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| 1.03k| bool found_leading_bytes{false}; 185| | // important: it is i <= how_far_back and not 'i < how_far_back'. 186| 1.13k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (186:22): [True: 1.04k, False: 92] ------------------ 187| 1.04k| unsigned char byte = buf[-static_cast(i)]; 188| 1.04k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 189| 1.04k| if (found_leading_bytes) { ------------------ | Branch (189:9): [True: 944, False: 96] ------------------ 190| 944| if (i > 0 && byte < 128) { ------------------ | Branch (190:11): [True: 4, False: 940] | 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| 940| buf -= i; 196| 940| extra_len = i; 197| 940| break; 198| 944| } 199| 1.04k| } 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| 1.03k| if (!found_leading_bytes) { ------------------ | Branch (208:7): [True: 92, False: 940] ------------------ 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| 92| return result(error_code::TOO_LONG, 0 - how_far_back); 214| 92| } 215| 940| result res = convert_with_errors(buf, len + extra_len, latin1_output); 216| 940| if (res.error) { ------------------ | Branch (216:7): [True: 528, False: 412] ------------------ 217| 528| res.count -= extra_len; 218| 528| } 219| 940| return res; 220| 1.03k|} 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| 1.73M| while (pos < len) { ------------------ | Branch (96:10): [True: 1.73M, False: 622] ------------------ 97| |#if SIMDUTF_CPLUSPLUS23 98| | if !consteval 99| |#endif 100| 1.73M| { 101| | // try to convert the next block of 16 ASCII bytes 102| 1.73M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (102:11): [True: 1.72M, False: 6.43k] ------------------ 103| | // they are ascii 104| 1.72M| uint64_t v1; 105| 1.72M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 106| 1.72M| uint64_t v2; 107| 1.72M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 108| 1.72M| uint64_t v{v1 | v2}; // We are only interested in these bits: 1000 1000 109| | // 1000 1000...etc 110| 1.72M| if ((v & 0x8080808080808080) == ------------------ | Branch (110:13): [True: 1.38M, False: 344k] ------------------ 111| 1.72M| 0) { // if NONE of these are set, e.g. all of them are zero, then 112| | // everything is ASCII 113| 1.38M| size_t final_pos = pos + 16; 114| 23.4M| while (pos < final_pos) { ------------------ | Branch (114:18): [True: 22.1M, False: 1.38M] ------------------ 115| 22.1M| *latin_output++ = char(data[pos]); 116| 22.1M| pos++; 117| 22.1M| } 118| 1.38M| continue; 119| 1.38M| } 120| 1.72M| } 121| 1.73M| } 122| | // suppose it is not an all ASCII byte sequence 123| 351k| uint8_t leading_byte = data[pos]; // leading byte 124| 351k| if (leading_byte < 0b10000000) { ------------------ | Branch (124:9): [True: 307k, False: 43.6k] ------------------ 125| | // converting one ASCII byte !!! 126| 307k| *latin_output++ = char(leading_byte); 127| 307k| pos++; 128| 307k| } else if ((leading_byte & 0b11100000) == ------------------ | Branch (128:16): [True: 43.1k, False: 516] ------------------ 129| 43.6k| 0b11000000) { // the first three bits indicate: 130| | // We have a two-byte UTF-8 131| 43.1k| if (pos + 1 >= len) { ------------------ | Branch (131:11): [True: 48, False: 43.0k] ------------------ 132| 48| return result(error_code::TOO_SHORT, pos); 133| 48| } // minimal bound checking 134| 43.0k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (134:11): [True: 216, False: 42.8k] ------------------ 135| 216| return result(error_code::TOO_SHORT, pos); 136| 216| } // 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| 42.8k| uint32_t code_point = 140| 42.8k| (leading_byte & 0b00011111) << 6 | 141| 42.8k| (data[pos + 1] & 142| 42.8k| 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| 42.8k| if (code_point < 0x80) { ------------------ | Branch (147:11): [True: 12, False: 42.8k] ------------------ 148| 12| return result(error_code::OVERLONG, pos); 149| 12| } 150| 42.8k| if (0xFF < code_point) { ------------------ | Branch (150:11): [True: 48, False: 42.7k] ------------------ 151| 48| return result(error_code::TOO_LARGE, pos); 152| 48| } // We only care about the range 129-255 which is Non-ASCII latin1 153| | // characters 154| 42.7k| *latin_output++ = char(code_point); 155| 42.7k| pos += 2; 156| 42.7k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (156:16): [True: 84, False: 432] ------------------ 157| | // We have a three-byte UTF-8 158| 84| return result(error_code::TOO_LARGE, pos); 159| 432| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (159:16): [True: 72, False: 360] ------------------ 160| | // we have a 4-byte UTF-8 word. 161| 72| return result(error_code::TOO_LARGE, pos); 162| 360| } else { 163| | // we either have too many continuation bytes or an invalid leading byte 164| 360| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (164:11): [True: 144, False: 216] ------------------ 165| 144| return result(error_code::TOO_LONG, pos); 166| 144| } 167| | 168| 216| return result(error_code::HEADER_BITS, pos); 169| 360| } 170| 351k| } 171| 622| 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| 1.28k| OutputPtr latin_output) { 18| 1.28k| size_t pos = 0; 19| 1.28k| auto start = latin_output; 20| | 21| 1.59M| while (pos < len) { ------------------ | Branch (21:10): [True: 1.59M, False: 556] ------------------ 22| |#if SIMDUTF_CPLUSPLUS23 23| | if !consteval 24| |#endif 25| 1.59M| { 26| | // try to convert the next block of 16 ASCII bytes 27| 1.59M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (27:11): [True: 1.58M, False: 4.47k] ------------------ 28| | // they are ascii 29| 1.58M| uint64_t v1; 30| 1.58M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 31| 1.58M| uint64_t v2; 32| 1.58M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 33| 1.58M| uint64_t v{v1 | v2}; // We are only interested in these bits: 1000 1000 34| | // 1000 1000 .... etc 35| 1.58M| if ((v & 0x8080808080808080) == ------------------ | Branch (35:13): [True: 1.44M, False: 138k] ------------------ 36| 1.58M| 0) { // if NONE of these are set, e.g. all of them are zero, then 37| | // everything is ASCII 38| 1.44M| size_t final_pos = pos + 16; 39| 24.5M| while (pos < final_pos) { ------------------ | Branch (39:18): [True: 23.1M, False: 1.44M] ------------------ 40| 23.1M| *latin_output++ = char(data[pos]); 41| 23.1M| pos++; 42| 23.1M| } 43| 1.44M| continue; 44| 1.44M| } 45| 1.58M| } 46| 1.59M| } 47| | 48| | // suppose it is not an all ASCII byte sequence 49| 143k| uint8_t leading_byte = data[pos]; // leading byte 50| 143k| if (leading_byte < 0b10000000) { ------------------ | Branch (50:9): [True: 125k, False: 17.4k] ------------------ 51| | // converting one ASCII byte !!! 52| 125k| *latin_output++ = char(leading_byte); 53| 125k| pos++; 54| 125k| } else if ((leading_byte & 0b11100000) == ------------------ | Branch (54:16): [True: 16.9k, False: 498] ------------------ 55| 17.4k| 0b11000000) { // the first three bits indicate: 56| | // We have a two-byte UTF-8 57| 16.9k| if (pos + 1 >= len) { ------------------ | Branch (57:11): [True: 60, False: 16.8k] ------------------ 58| 60| return 0; 59| 60| } // minimal bound checking 60| 16.8k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (60:11): [True: 104, False: 16.7k] ------------------ 61| 104| return 0; 62| 104| } // 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| 16.7k| uint32_t code_point = 66| 16.7k| (leading_byte & 0b00011111) << 6 | 67| 16.7k| (data[pos + 1] & 68| 16.7k| 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| 16.7k| if (code_point < 0x80 || 0xFF < code_point) { ------------------ | Branch (73:11): [True: 14, False: 16.7k] | Branch (73:32): [True: 56, False: 16.7k] ------------------ 74| 70| 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| 70| } 78| 16.7k| *latin_output++ = char(code_point); 79| 16.7k| pos += 2; 80| 16.7k| } else { 81| 498| return 0; 82| 498| } 83| 143k| } 84| 556| return latin_output - start; 85| 1.28k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf167convertILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.70k| char16_t *utf16_output) { 17| 1.70k| size_t pos = 0; 18| 1.70k| char16_t *start{utf16_output}; 19| 2.16M| while (pos < len) { ------------------ | Branch (19:10): [True: 2.16M, False: 740] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| | // try to convert the next block of 16 ASCII bytes 24| 2.16M| { 25| 2.16M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (25:11): [True: 2.16M, False: 5.06k] ------------------ 26| | // they are ascii 27| 2.16M| uint64_t v1; 28| 2.16M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 29| 2.16M| uint64_t v2; 30| 2.16M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 31| 2.16M| uint64_t v{v1 | v2}; 32| 2.16M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (32:13): [True: 1.17M, False: 984k] ------------------ 33| 1.17M| size_t final_pos = pos + 16; 34| 20.0M| while (pos < final_pos) { ------------------ | Branch (34:18): [True: 18.8M, False: 1.17M] ------------------ 35| 18.8M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (35:31): [Folded, False: 18.8M] ------------------ 36| 18.8M| ? char16_t(u16_swap_bytes(data[pos])) 37| 18.8M| : char16_t(data[pos]); 38| 18.8M| pos++; 39| 18.8M| } 40| 1.17M| continue; 41| 1.17M| } 42| 2.16M| } 43| 2.16M| } 44| | 45| 989k| uint8_t leading_byte = data[pos]; // leading byte 46| 989k| if (leading_byte < 0b10000000) { ------------------ | Branch (46:9): [True: 805k, False: 183k] ------------------ 47| | // converting one ASCII byte !!! 48| 805k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (48:25): [Folded, False: 805k] ------------------ 49| 805k| ? char16_t(u16_swap_bytes(leading_byte)) 50| 805k| : char16_t(leading_byte); 51| 805k| pos++; 52| 805k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (52:16): [True: 110k, False: 73.0k] ------------------ 53| | // We have a two-byte UTF-8, it should become 54| | // a single UTF-16 word. 55| 110k| if (pos + 1 >= len) { ------------------ | Branch (55:11): [True: 42, False: 110k] ------------------ 56| 42| return 0; 57| 42| } // minimal bound checking 58| 110k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (58:11): [True: 92, False: 110k] ------------------ 59| 92| return 0; 60| 92| } 61| | // range check 62| 110k| uint32_t code_point = 63| 110k| (leading_byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 64| 110k| if (code_point < 0x80) { ------------------ | Branch (64:11): [True: 18, False: 110k] ------------------ 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| 110k| *utf16_output++ = char16_t(code_point); 71| 110k| pos += 2; 72| 110k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (72:16): [True: 64.4k, False: 8.57k] ------------------ 73| | // We have a three-byte UTF-8, it should become 74| | // a single UTF-16 word. 75| 64.4k| if (pos + 2 >= len) { ------------------ | Branch (75:11): [True: 36, False: 64.4k] ------------------ 76| 36| return 0; 77| 36| } // minimal bound checking 78| | 79| 64.4k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (79:11): [True: 50, False: 64.3k] ------------------ 80| 50| return 0; 81| 50| } 82| 64.3k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (82:11): [True: 52, False: 64.3k] ------------------ 83| 52| return 0; 84| 52| } 85| | // range check 86| 64.3k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 87| 64.3k| (data[pos + 1] & 0b00111111) << 6 | 88| 64.3k| (data[pos + 2] & 0b00111111); 89| 64.3k| if (code_point < 0x800 || (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (89:11): [True: 54, False: 64.2k] | Branch (89:34): [True: 740, False: 63.5k] | Branch (89:57): [True: 10, False: 730] ------------------ 90| 64| return 0; 91| 64| } 92| | if constexpr (!match_system(big_endian)) { 93| | code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 94| | } 95| 64.2k| *utf16_output++ = char16_t(code_point); 96| 64.2k| pos += 3; 97| 64.2k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (97:16): [True: 8.16k, False: 408] ------------------ 98| | // we have a 4-byte UTF-8 word. 99| 8.16k| if (pos + 3 >= len) { ------------------ | Branch (99:11): [True: 30, False: 8.13k] ------------------ 100| 30| return 0; 101| 30| } // minimal bound checking 102| 8.13k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (102:11): [True: 58, False: 8.07k] ------------------ 103| 58| return 0; 104| 58| } 105| 8.07k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (105:11): [True: 18, False: 8.05k] ------------------ 106| 18| return 0; 107| 18| } 108| 8.05k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (108:11): [True: 36, False: 8.02k] ------------------ 109| 36| return 0; 110| 36| } 111| | 112| | // range check 113| 8.02k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 114| 8.02k| (data[pos + 1] & 0b00111111) << 12 | 115| 8.02k| (data[pos + 2] & 0b00111111) << 6 | 116| 8.02k| (data[pos + 3] & 0b00111111); 117| 8.02k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (117:11): [True: 16, False: 8.00k] | Branch (117:35): [True: 44, False: 7.96k] ------------------ 118| 60| return 0; 119| 60| } 120| 7.96k| code_point -= 0x10000; 121| 7.96k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 122| 7.96k| 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| 7.96k| *utf16_output++ = char16_t(high_surrogate); 128| 7.96k| *utf16_output++ = char16_t(low_surrogate); 129| 7.96k| pos += 4; 130| 7.96k| } else { 131| 408| return 0; 132| 408| } 133| 989k| } 134| 740| return utf16_output - start; 135| 1.70k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf167convertILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.59k| char16_t *utf16_output) { 17| 1.59k| size_t pos = 0; 18| 1.59k| char16_t *start{utf16_output}; 19| 1.27M| while (pos < len) { ------------------ | Branch (19:10): [True: 1.27M, False: 608] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| | // try to convert the next block of 16 ASCII bytes 24| 1.27M| { 25| 1.27M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (25:11): [True: 1.27M, False: 4.23k] ------------------ 26| | // they are ascii 27| 1.27M| uint64_t v1; 28| 1.27M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 29| 1.27M| uint64_t v2; 30| 1.27M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 31| 1.27M| uint64_t v{v1 | v2}; 32| 1.27M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (32:13): [True: 947k, False: 324k] ------------------ 33| 947k| size_t final_pos = pos + 16; 34| 16.1M| while (pos < final_pos) { ------------------ | Branch (34:18): [True: 15.1M, False: 947k] ------------------ 35| 15.1M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (35:31): [True: 15.1M, Folded] ------------------ 36| 15.1M| ? char16_t(u16_swap_bytes(data[pos])) 37| 15.1M| : char16_t(data[pos]); 38| 15.1M| pos++; 39| 15.1M| } 40| 947k| continue; 41| 947k| } 42| 1.27M| } 43| 1.27M| } 44| | 45| 329k| uint8_t leading_byte = data[pos]; // leading byte 46| 329k| if (leading_byte < 0b10000000) { ------------------ | Branch (46:9): [True: 270k, False: 58.1k] ------------------ 47| | // converting one ASCII byte !!! 48| 270k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (48:25): [True: 270k, Folded] ------------------ 49| 270k| ? char16_t(u16_swap_bytes(leading_byte)) 50| 270k| : char16_t(leading_byte); 51| 270k| pos++; 52| 270k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (52:16): [True: 33.7k, False: 24.3k] ------------------ 53| | // We have a two-byte UTF-8, it should become 54| | // a single UTF-16 word. 55| 33.7k| if (pos + 1 >= len) { ------------------ | Branch (55:11): [True: 42, False: 33.7k] ------------------ 56| 42| return 0; 57| 42| } // minimal bound checking 58| 33.7k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (58:11): [True: 80, False: 33.6k] ------------------ 59| 80| return 0; 60| 80| } 61| | // range check 62| 33.6k| uint32_t code_point = 63| 33.6k| (leading_byte & 0b00011111) << 6 | (data[pos + 1] & 0b00111111); 64| 33.6k| if (code_point < 0x80) { ------------------ | Branch (64:11): [True: 20, False: 33.6k] ------------------ 65| 20| return 0; 66| 20| } 67| 33.6k| if constexpr (!match_system(big_endian)) { 68| 33.6k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 69| 33.6k| } 70| 33.6k| *utf16_output++ = char16_t(code_point); 71| 33.6k| pos += 2; 72| 33.6k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (72:16): [True: 21.5k, False: 2.79k] ------------------ 73| | // We have a three-byte UTF-8, it should become 74| | // a single UTF-16 word. 75| 21.5k| if (pos + 2 >= len) { ------------------ | Branch (75:11): [True: 30, False: 21.5k] ------------------ 76| 30| return 0; 77| 30| } // minimal bound checking 78| | 79| 21.5k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (79:11): [True: 78, False: 21.4k] ------------------ 80| 78| return 0; 81| 78| } 82| 21.4k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (82:11): [True: 46, False: 21.4k] ------------------ 83| 46| return 0; 84| 46| } 85| | // range check 86| 21.4k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 87| 21.4k| (data[pos + 1] & 0b00111111) << 6 | 88| 21.4k| (data[pos + 2] & 0b00111111); 89| 21.4k| if (code_point < 0x800 || (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (89:11): [True: 40, False: 21.3k] | Branch (89:34): [True: 398, False: 20.9k] | Branch (89:57): [True: 20, False: 378] ------------------ 90| 60| return 0; 91| 60| } 92| 21.3k| if constexpr (!match_system(big_endian)) { 93| 21.3k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 94| 21.3k| } 95| 21.3k| *utf16_output++ = char16_t(code_point); 96| 21.3k| pos += 3; 97| 21.3k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (97:16): [True: 2.40k, False: 386] ------------------ 98| | // we have a 4-byte UTF-8 word. 99| 2.40k| if (pos + 3 >= len) { ------------------ | Branch (99:11): [True: 30, False: 2.37k] ------------------ 100| 30| return 0; 101| 30| } // minimal bound checking 102| 2.37k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (102:11): [True: 62, False: 2.31k] ------------------ 103| 62| return 0; 104| 62| } 105| 2.31k| if ((data[pos + 2] & 0b11000000) != 0b10000000) { ------------------ | Branch (105:11): [True: 42, False: 2.27k] ------------------ 106| 42| return 0; 107| 42| } 108| 2.27k| if ((data[pos + 3] & 0b11000000) != 0b10000000) { ------------------ | Branch (108:11): [True: 30, False: 2.24k] ------------------ 109| 30| return 0; 110| 30| } 111| | 112| | // range check 113| 2.24k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 114| 2.24k| (data[pos + 1] & 0b00111111) << 12 | 115| 2.24k| (data[pos + 2] & 0b00111111) << 6 | 116| 2.24k| (data[pos + 3] & 0b00111111); 117| 2.24k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (117:11): [True: 48, False: 2.19k] | Branch (117:35): [True: 36, False: 2.15k] ------------------ 118| 84| return 0; 119| 84| } 120| 2.15k| code_point -= 0x10000; 121| 2.15k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 122| 2.15k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 123| 2.15k| if constexpr (!match_system(big_endian)) { 124| 2.15k| high_surrogate = u16_swap_bytes(high_surrogate); 125| 2.15k| low_surrogate = u16_swap_bytes(low_surrogate); 126| 2.15k| } 127| 2.15k| *utf16_output++ = char16_t(high_surrogate); 128| 2.15k| *utf16_output++ = char16_t(low_surrogate); 129| 2.15k| pos += 4; 130| 2.15k| } else { 131| 386| return 0; 132| 386| } 133| 329k| } 134| 608| return utf16_output - start; 135| 1.59k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1619convert_with_errorsILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEENS_6resultES7_mPDs: 142| 1.70k| char16_t *utf16_output) { 143| 1.70k| size_t pos = 0; 144| 1.70k| char16_t *start{utf16_output}; 145| 2.99M| while (pos < len) { ------------------ | Branch (145:10): [True: 2.99M, False: 754] ------------------ 146| |#if SIMDUTF_CPLUSPLUS23 147| | if !consteval 148| |#endif 149| 2.99M| { 150| | // try to convert the next block of 16 ASCII bytes 151| 2.99M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (151:11): [True: 2.98M, False: 6.11k] ------------------ 152| | // they are ascii 153| 2.98M| uint64_t v1; 154| 2.98M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 155| 2.98M| uint64_t v2; 156| 2.98M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 157| 2.98M| uint64_t v{v1 | v2}; 158| 2.98M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (158:13): [True: 1.37M, False: 1.61M] ------------------ 159| 1.37M| size_t final_pos = pos + 16; 160| 23.3M| while (pos < final_pos) { ------------------ | Branch (160:18): [True: 21.9M, False: 1.37M] ------------------ 161| 21.9M| const char16_t byte = uint8_t(data[pos]); 162| 21.9M| *utf16_output++ = 163| 21.9M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (163:17): [Folded, False: 21.9M] ------------------ 164| 21.9M| pos++; 165| 21.9M| } 166| 1.37M| continue; 167| 1.37M| } 168| 2.98M| } 169| 2.99M| } 170| | 171| 1.61M| auto leading_byte = uint8_t(data[pos]); // leading byte 172| 1.61M| if (leading_byte < 0b10000000) { ------------------ | Branch (172:9): [True: 1.33M, False: 287k] ------------------ 173| | // converting one ASCII byte !!! 174| 1.33M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (174:25): [Folded, False: 1.33M] ------------------ 175| 1.33M| ? char16_t(u16_swap_bytes(leading_byte)) 176| 1.33M| : char16_t(leading_byte); 177| 1.33M| pos++; 178| 1.33M| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (178:16): [True: 194k, False: 93.5k] ------------------ 179| | // We have a two-byte UTF-8, it should become 180| | // a single UTF-16 word. 181| 194k| if (pos + 1 >= len) { ------------------ | Branch (181:11): [True: 60, False: 194k] ------------------ 182| 60| return result(error_code::TOO_SHORT, pos); 183| 60| } // minimal bound checking 184| 194k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (184:11): [True: 102, False: 194k] ------------------ 185| 102| return result(error_code::TOO_SHORT, pos); 186| 102| } 187| | // range check 188| 194k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 189| 194k| (uint8_t(data[pos + 1]) & 0b00111111); 190| 194k| if (code_point < 0x80) { ------------------ | Branch (190:11): [True: 24, False: 193k] ------------------ 191| 24| return result(error_code::OVERLONG, pos); 192| 24| } 193| | if constexpr (!match_system(big_endian)) { 194| | code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 195| | } 196| 193k| *utf16_output++ = char16_t(code_point); 197| 193k| pos += 2; 198| 193k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (198:16): [True: 74.4k, False: 19.0k] ------------------ 199| | // We have a three-byte UTF-8, it should become 200| | // a single UTF-16 word. 201| 74.4k| if (pos + 2 >= len) { ------------------ | Branch (201:11): [True: 30, False: 74.4k] ------------------ 202| 30| return result(error_code::TOO_SHORT, pos); 203| 30| } // minimal bound checking 204| | 205| 74.4k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (205:11): [True: 72, False: 74.3k] ------------------ 206| 72| return result(error_code::TOO_SHORT, pos); 207| 72| } 208| 74.3k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (208:11): [True: 36, False: 74.3k] ------------------ 209| 36| return result(error_code::TOO_SHORT, pos); 210| 36| } 211| | // range check 212| 74.3k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 213| 74.3k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 214| 74.3k| (uint8_t(data[pos + 2]) & 0b00111111); 215| 74.3k| if (code_point < 0x800) { ------------------ | Branch (215:11): [True: 30, False: 74.3k] ------------------ 216| 30| return result(error_code::OVERLONG, pos); 217| 30| } 218| 74.3k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (218:11): [True: 3.05k, False: 71.2k] | Branch (218:34): [True: 18, False: 3.03k] ------------------ 219| 18| return result(error_code::SURROGATE, pos); 220| 18| } 221| | if constexpr (!match_system(big_endian)) { 222| | code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 223| | } 224| 74.3k| *utf16_output++ = char16_t(code_point); 225| 74.3k| pos += 3; 226| 74.3k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (226:16): [True: 18.6k, False: 402] ------------------ 227| | // we have a 4-byte UTF-8 word. 228| 18.6k| if (pos + 3 >= len) { ------------------ | Branch (228:11): [True: 24, False: 18.6k] ------------------ 229| 24| return result(error_code::TOO_SHORT, pos); 230| 24| } // minimal bound checking 231| 18.6k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (231:11): [True: 48, False: 18.5k] ------------------ 232| 48| return result(error_code::TOO_SHORT, pos); 233| 48| } 234| 18.5k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (234:11): [True: 24, False: 18.5k] ------------------ 235| 24| return result(error_code::TOO_SHORT, pos); 236| 24| } 237| 18.5k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (237:11): [True: 12, False: 18.5k] ------------------ 238| 12| return result(error_code::TOO_SHORT, pos); 239| 12| } 240| | 241| | // range check 242| 18.5k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 243| 18.5k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 244| 18.5k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 245| 18.5k| (uint8_t(data[pos + 3]) & 0b00111111); 246| 18.5k| if (code_point <= 0xffff) { ------------------ | Branch (246:11): [True: 36, False: 18.5k] ------------------ 247| 36| return result(error_code::OVERLONG, pos); 248| 36| } 249| 18.5k| if (0x10ffff < code_point) { ------------------ | Branch (249:11): [True: 30, False: 18.4k] ------------------ 250| 30| return result(error_code::TOO_LARGE, pos); 251| 30| } 252| 18.4k| code_point -= 0x10000; 253| 18.4k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 254| 18.4k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 255| | if constexpr (!match_system(big_endian)) { 256| | high_surrogate = u16_swap_bytes(high_surrogate); 257| | low_surrogate = u16_swap_bytes(low_surrogate); 258| | } 259| 18.4k| *utf16_output++ = char16_t(high_surrogate); 260| 18.4k| *utf16_output++ = char16_t(low_surrogate); 261| 18.4k| pos += 4; 262| 18.4k| } else { 263| | // we either have too many continuation bytes or an invalid leading byte 264| 402| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (264:11): [True: 186, False: 216] ------------------ 265| 186| return result(error_code::TOO_LONG, pos); 266| 216| } else { 267| 216| return result(error_code::HEADER_BITS, pos); 268| 216| } 269| 402| } 270| 1.61M| } 271| 754| return result(error_code::SUCCESS, utf16_output - start); 272| 1.70k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1619convert_with_errorsILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEENS_6resultES7_mPDs: 142| 2.23k| char16_t *utf16_output) { 143| 2.23k| size_t pos = 0; 144| 2.23k| char16_t *start{utf16_output}; 145| 3.31M| while (pos < len) { ------------------ | Branch (145:10): [True: 3.31M, False: 1.04k] ------------------ 146| |#if SIMDUTF_CPLUSPLUS23 147| | if !consteval 148| |#endif 149| 3.31M| { 150| | // try to convert the next block of 16 ASCII bytes 151| 3.31M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (151:11): [True: 3.30M, False: 9.74k] ------------------ 152| | // they are ascii 153| 3.30M| uint64_t v1; 154| 3.30M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 155| 3.30M| uint64_t v2; 156| 3.30M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 157| 3.30M| uint64_t v{v1 | v2}; 158| 3.30M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (158:13): [True: 1.66M, False: 1.64M] ------------------ 159| 1.66M| size_t final_pos = pos + 16; 160| 28.2M| while (pos < final_pos) { ------------------ | Branch (160:18): [True: 26.6M, False: 1.66M] ------------------ 161| 26.6M| const char16_t byte = uint8_t(data[pos]); 162| 26.6M| *utf16_output++ = 163| 26.6M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (163:17): [True: 26.6M, Folded] ------------------ 164| 26.6M| pos++; 165| 26.6M| } 166| 1.66M| continue; 167| 1.66M| } 168| 3.30M| } 169| 3.31M| } 170| | 171| 1.65M| auto leading_byte = uint8_t(data[pos]); // leading byte 172| 1.65M| if (leading_byte < 0b10000000) { ------------------ | Branch (172:9): [True: 1.35M, False: 291k] ------------------ 173| | // converting one ASCII byte !!! 174| 1.35M| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (174:25): [True: 1.35M, Folded] ------------------ 175| 1.35M| ? char16_t(u16_swap_bytes(leading_byte)) 176| 1.35M| : char16_t(leading_byte); 177| 1.35M| pos++; 178| 1.35M| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (178:16): [True: 182k, False: 109k] ------------------ 179| | // We have a two-byte UTF-8, it should become 180| | // a single UTF-16 word. 181| 182k| if (pos + 1 >= len) { ------------------ | Branch (181:11): [True: 72, False: 182k] ------------------ 182| 72| return result(error_code::TOO_SHORT, pos); 183| 72| } // minimal bound checking 184| 182k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (184:11): [True: 156, False: 182k] ------------------ 185| 156| return result(error_code::TOO_SHORT, pos); 186| 156| } 187| | // range check 188| 182k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 189| 182k| (uint8_t(data[pos + 1]) & 0b00111111); 190| 182k| if (code_point < 0x80) { ------------------ | Branch (190:11): [True: 12, False: 182k] ------------------ 191| 12| return result(error_code::OVERLONG, pos); 192| 12| } 193| 182k| if constexpr (!match_system(big_endian)) { 194| 182k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 195| 182k| } 196| 182k| *utf16_output++ = char16_t(code_point); 197| 182k| pos += 2; 198| 182k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (198:16): [True: 99.5k, False: 9.62k] ------------------ 199| | // We have a three-byte UTF-8, it should become 200| | // a single UTF-16 word. 201| 99.5k| if (pos + 2 >= len) { ------------------ | Branch (201:11): [True: 42, False: 99.4k] ------------------ 202| 42| return result(error_code::TOO_SHORT, pos); 203| 42| } // minimal bound checking 204| | 205| 99.4k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (205:11): [True: 78, False: 99.4k] ------------------ 206| 78| return result(error_code::TOO_SHORT, pos); 207| 78| } 208| 99.4k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (208:11): [True: 24, False: 99.3k] ------------------ 209| 24| return result(error_code::TOO_SHORT, pos); 210| 24| } 211| | // range check 212| 99.3k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 213| 99.3k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 214| 99.3k| (uint8_t(data[pos + 2]) & 0b00111111); 215| 99.3k| if (code_point < 0x800) { ------------------ | Branch (215:11): [True: 30, False: 99.3k] ------------------ 216| 30| return result(error_code::OVERLONG, pos); 217| 30| } 218| 99.3k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (218:11): [True: 2.53k, False: 96.8k] | Branch (218:34): [True: 24, False: 2.50k] ------------------ 219| 24| return result(error_code::SURROGATE, pos); 220| 24| } 221| 99.3k| if constexpr (!match_system(big_endian)) { 222| 99.3k| code_point = uint32_t(u16_swap_bytes(uint16_t(code_point))); 223| 99.3k| } 224| 99.3k| *utf16_output++ = char16_t(code_point); 225| 99.3k| pos += 3; 226| 99.3k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (226:16): [True: 9.16k, False: 452] ------------------ 227| | // we have a 4-byte UTF-8 word. 228| 9.16k| if (pos + 3 >= len) { ------------------ | Branch (228:11): [True: 78, False: 9.09k] ------------------ 229| 78| return result(error_code::TOO_SHORT, pos); 230| 78| } // minimal bound checking 231| 9.09k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (231:11): [True: 66, False: 9.02k] ------------------ 232| 66| return result(error_code::TOO_SHORT, pos); 233| 66| } 234| 9.02k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (234:11): [True: 66, False: 8.95k] ------------------ 235| 66| return result(error_code::TOO_SHORT, pos); 236| 66| } 237| 8.95k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (237:11): [True: 18, False: 8.94k] ------------------ 238| 18| return result(error_code::TOO_SHORT, pos); 239| 18| } 240| | 241| | // range check 242| 8.94k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 243| 8.94k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 244| 8.94k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 245| 8.94k| (uint8_t(data[pos + 3]) & 0b00111111); 246| 8.94k| if (code_point <= 0xffff) { ------------------ | Branch (246:11): [True: 48, False: 8.89k] ------------------ 247| 48| return result(error_code::OVERLONG, pos); 248| 48| } 249| 8.89k| if (0x10ffff < code_point) { ------------------ | Branch (249:11): [True: 24, False: 8.86k] ------------------ 250| 24| return result(error_code::TOO_LARGE, pos); 251| 24| } 252| 8.86k| code_point -= 0x10000; 253| 8.86k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 254| 8.86k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 255| 8.86k| if constexpr (!match_system(big_endian)) { 256| 8.86k| high_surrogate = u16_swap_bytes(high_surrogate); 257| 8.86k| low_surrogate = u16_swap_bytes(low_surrogate); 258| 8.86k| } 259| 8.86k| *utf16_output++ = char16_t(high_surrogate); 260| 8.86k| *utf16_output++ = char16_t(low_surrogate); 261| 8.86k| pos += 4; 262| 8.86k| } else { 263| | // we either have too many continuation bytes or an invalid leading byte 264| 452| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (264:11): [True: 176, False: 276] ------------------ 265| 176| return result(error_code::TOO_LONG, pos); 266| 276| } else { 267| 276| return result(error_code::HEADER_BITS, pos); 268| 276| } 269| 452| } 270| 1.65M| } 271| 1.04k| return result(error_code::SUCCESS, utf16_output - start); 272| 2.23k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1630rewind_and_convert_with_errorsILNS_10endiannessE0EEENS_6resultEmPKcmPDs: 292| 1.18k| char16_t *utf16_output) { 293| 1.18k| 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| 1.18k| 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| 1.18k| bool found_leading_bytes{false}; 301| | // important: it is i <= how_far_back and not 'i < how_far_back'. 302| 1.25k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (302:22): [True: 1.20k, False: 52] ------------------ 303| 1.20k| unsigned char byte = buf[-static_cast(i)]; 304| 1.20k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 305| 1.20k| if (found_leading_bytes) { ------------------ | Branch (305:9): [True: 1.12k, False: 72] ------------------ 306| 1.12k| if (i > 0 && byte < 128) { ------------------ | Branch (306:11): [True: 20, False: 1.10k] | Branch (306:20): [True: 20, False: 0] ------------------ 307| | // If we had to go back and the leading byte is ascii 308| | // then we can stop right away. 309| 20| return result(error_code::TOO_LONG, 0 - i + 1); 310| 20| } 311| 1.10k| buf -= i; 312| 1.10k| extra_len = i; 313| 1.10k| break; 314| 1.12k| } 315| 1.20k| } 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| 1.16k| if (!found_leading_bytes) { ------------------ | Branch (324:7): [True: 52, False: 1.10k] ------------------ 325| | // If how_far_back == 3, we may have four consecutive continuation bytes!!! 326| | // [....] [continuation] [continuation] [continuation] | [buf is 327| | // continuation] Or we possibly have a stream that does not start with a 328| | // leading byte. 329| 52| return result(error_code::TOO_LONG, 0 - how_far_back); 330| 52| } 331| 1.10k| result res = convert_with_errors(buf, len + extra_len, utf16_output); 332| 1.10k| if (res.error) { ------------------ | Branch (332:7): [True: 608, False: 500] ------------------ 333| 608| res.count -= extra_len; 334| 608| } 335| 1.10k| return res; 336| 1.16k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1630rewind_and_convert_with_errorsILNS_10endiannessE1EEENS_6resultEmPKcmPDs: 292| 1.53k| char16_t *utf16_output) { 293| 1.53k| 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| 1.53k| 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| 1.53k| bool found_leading_bytes{false}; 301| | // important: it is i <= how_far_back and not 'i < how_far_back'. 302| 1.61k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (302:22): [True: 1.54k, False: 68] ------------------ 303| 1.54k| unsigned char byte = buf[-static_cast(i)]; 304| 1.54k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 305| 1.54k| if (found_leading_bytes) { ------------------ | Branch (305:9): [True: 1.46k, False: 76] ------------------ 306| 1.46k| if (i > 0 && byte < 128) { ------------------ | Branch (306:11): [True: 8, False: 1.46k] | 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| 1.46k| buf -= i; 312| 1.46k| extra_len = i; 313| 1.46k| break; 314| 1.46k| } 315| 1.54k| } 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| 1.52k| if (!found_leading_bytes) { ------------------ | Branch (324:7): [True: 68, False: 1.46k] ------------------ 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| 68| return result(error_code::TOO_LONG, 0 - how_far_back); 330| 68| } 331| 1.46k| result res = convert_with_errors(buf, len + extra_len, utf16_output); 332| 1.46k| if (res.error) { ------------------ | Branch (332:7): [True: 768, False: 692] ------------------ 333| 768| res.count -= extra_len; 334| 768| } 335| 1.46k| return res; 336| 1.52k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE0EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.29k| char16_t *utf16_output) { 17| 1.29k| size_t pos = 0; 18| 1.29k| char16_t *start{utf16_output}; 19| 4.03M| while (pos < len) { ------------------ | Branch (19:10): [True: 4.02M, False: 1.29k] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 4.02M| { // try to convert the next block of 8 ASCII bytes 24| 4.02M| if (pos + 8 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (24:11): [True: 4.02M, False: 4.01k] ------------------ 25| | // they are ascii 26| 4.02M| uint64_t v; 27| 4.02M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 28| 4.02M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (28:13): [True: 2.75M, False: 1.27M] ------------------ 29| 2.75M| size_t final_pos = pos + 8; 30| 24.8M| while (pos < final_pos) { ------------------ | Branch (30:18): [True: 22.0M, False: 2.75M] ------------------ 31| 22.0M| const char16_t byte = uint8_t(data[pos]); 32| 22.0M| *utf16_output++ = 33| 22.0M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (33:17): [Folded, False: 22.0M] ------------------ 34| 22.0M| pos++; 35| 22.0M| } 36| 2.75M| continue; 37| 2.75M| } 38| 4.02M| } 39| 4.02M| } 40| | 41| 1.27M| auto leading_byte = uint8_t(data[pos]); // leading byte 42| 1.27M| if (leading_byte < 0b10000000) { ------------------ | Branch (42:9): [True: 865k, False: 408k] ------------------ 43| | // converting one ASCII byte !!! 44| 865k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (44:25): [Folded, False: 865k] ------------------ 45| 865k| ? char16_t(u16_swap_bytes(leading_byte)) 46| 865k| : char16_t(leading_byte); 47| 865k| pos++; 48| 865k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (48:16): [True: 233k, False: 175k] ------------------ 49| | // We have a two-byte UTF-8, it should become 50| | // a single UTF-16 word. 51| 233k| if (pos + 1 >= len) { ------------------ | Branch (51:11): [True: 0, False: 233k] ------------------ 52| 0| break; 53| 0| } // minimal bound checking 54| 233k| uint16_t code_point = uint16_t(((leading_byte & 0b00011111) << 6) | 55| 233k| (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| 233k| *utf16_output++ = char16_t(code_point); 60| 233k| pos += 2; 61| 233k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (61:16): [True: 162k, False: 12.7k] ------------------ 62| | // We have a three-byte UTF-8, it should become 63| | // a single UTF-16 word. 64| 162k| if (pos + 2 >= len) { ------------------ | Branch (64:11): [True: 0, False: 162k] ------------------ 65| 0| break; 66| 0| } // minimal bound checking 67| 162k| uint16_t code_point = 68| 162k| uint16_t(((leading_byte & 0b00001111) << 12) | 69| 162k| ((uint8_t(data[pos + 1]) & 0b00111111) << 6) | 70| 162k| (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| 162k| *utf16_output++ = char16_t(code_point); 75| 162k| pos += 3; 76| 162k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (76:16): [True: 12.7k, False: 0] ------------------ 77| | // we have a 4-byte UTF-8 word. 78| 12.7k| if (pos + 3 >= len) { ------------------ | Branch (78:11): [True: 0, False: 12.7k] ------------------ 79| 0| break; 80| 0| } // minimal bound checking 81| 12.7k| uint32_t code_point = ((leading_byte & 0b00000111) << 18) | 82| 12.7k| ((uint8_t(data[pos + 1]) & 0b00111111) << 12) | 83| 12.7k| ((uint8_t(data[pos + 2]) & 0b00111111) << 6) | 84| 12.7k| (uint8_t(data[pos + 3]) & 0b00111111); 85| 12.7k| code_point -= 0x10000; 86| 12.7k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 87| 12.7k| 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| 12.7k| *utf16_output++ = char16_t(high_surrogate); 93| 12.7k| *utf16_output++ = char16_t(low_surrogate); 94| 12.7k| pos += 4; 95| 12.7k| } else { 96| | // we may have a continuation but we do not do error checking 97| 0| return 0; 98| 0| } 99| 1.27M| } 100| 1.29k| return utf16_output - start; 101| 1.29k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE1EPKcQsr7simdutf6detailE22indexes_into_byte_likeIT0_EEEmS7_mPDs: 16| 1.38k| char16_t *utf16_output) { 17| 1.38k| size_t pos = 0; 18| 1.38k| char16_t *start{utf16_output}; 19| 2.95M| while (pos < len) { ------------------ | Branch (19:10): [True: 2.95M, False: 1.38k] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 2.95M| { // try to convert the next block of 8 ASCII bytes 24| 2.95M| if (pos + 8 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (24:11): [True: 2.95M, False: 4.49k] ------------------ 25| | // they are ascii 26| 2.95M| uint64_t v; 27| 2.95M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 28| 2.95M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (28:13): [True: 2.59M, False: 361k] ------------------ 29| 2.59M| size_t final_pos = pos + 8; 30| 23.3M| while (pos < final_pos) { ------------------ | Branch (30:18): [True: 20.7M, False: 2.59M] ------------------ 31| 20.7M| const char16_t byte = uint8_t(data[pos]); 32| 20.7M| *utf16_output++ = 33| 20.7M| !match_system(big_endian) ? u16_swap_bytes(byte) : byte; ------------------ | Branch (33:17): [True: 20.7M, Folded] ------------------ 34| 20.7M| pos++; 35| 20.7M| } 36| 2.59M| continue; 37| 2.59M| } 38| 2.95M| } 39| 2.95M| } 40| | 41| 366k| auto leading_byte = uint8_t(data[pos]); // leading byte 42| 366k| if (leading_byte < 0b10000000) { ------------------ | Branch (42:9): [True: 222k, False: 144k] ------------------ 43| | // converting one ASCII byte !!! 44| 222k| *utf16_output++ = !match_system(big_endian) ------------------ | Branch (44:25): [True: 222k, Folded] ------------------ 45| 222k| ? char16_t(u16_swap_bytes(leading_byte)) 46| 222k| : char16_t(leading_byte); 47| 222k| pos++; 48| 222k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (48:16): [True: 106k, False: 37.6k] ------------------ 49| | // We have a two-byte UTF-8, it should become 50| | // a single UTF-16 word. 51| 106k| if (pos + 1 >= len) { ------------------ | Branch (51:11): [True: 0, False: 106k] ------------------ 52| 0| break; 53| 0| } // minimal bound checking 54| 106k| uint16_t code_point = uint16_t(((leading_byte & 0b00011111) << 6) | 55| 106k| (uint8_t(data[pos + 1]) & 0b00111111)); 56| 106k| if constexpr (!match_system(big_endian)) { 57| 106k| code_point = u16_swap_bytes(uint16_t(code_point)); 58| 106k| } 59| 106k| *utf16_output++ = char16_t(code_point); 60| 106k| pos += 2; 61| 106k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (61:16): [True: 27.1k, False: 10.5k] ------------------ 62| | // We have a three-byte UTF-8, it should become 63| | // a single UTF-16 word. 64| 27.1k| if (pos + 2 >= len) { ------------------ | Branch (64:11): [True: 0, False: 27.1k] ------------------ 65| 0| break; 66| 0| } // minimal bound checking 67| 27.1k| uint16_t code_point = 68| 27.1k| uint16_t(((leading_byte & 0b00001111) << 12) | 69| 27.1k| ((uint8_t(data[pos + 1]) & 0b00111111) << 6) | 70| 27.1k| (uint8_t(data[pos + 2]) & 0b00111111)); 71| 27.1k| if constexpr (!match_system(big_endian)) { 72| 27.1k| code_point = u16_swap_bytes(uint16_t(code_point)); 73| 27.1k| } 74| 27.1k| *utf16_output++ = char16_t(code_point); 75| 27.1k| pos += 3; 76| 27.1k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (76:16): [True: 10.5k, False: 0] ------------------ 77| | // we have a 4-byte UTF-8 word. 78| 10.5k| if (pos + 3 >= len) { ------------------ | Branch (78:11): [True: 0, False: 10.5k] ------------------ 79| 0| break; 80| 0| } // minimal bound checking 81| 10.5k| uint32_t code_point = ((leading_byte & 0b00000111) << 18) | 82| 10.5k| ((uint8_t(data[pos + 1]) & 0b00111111) << 12) | 83| 10.5k| ((uint8_t(data[pos + 2]) & 0b00111111) << 6) | 84| 10.5k| (uint8_t(data[pos + 3]) & 0b00111111); 85| 10.5k| code_point -= 0x10000; 86| 10.5k| uint16_t high_surrogate = uint16_t(0xD800 + (code_point >> 10)); 87| 10.5k| uint16_t low_surrogate = uint16_t(0xDC00 + (code_point & 0x3FF)); 88| 10.5k| if constexpr (!match_system(big_endian)) { 89| 10.5k| high_surrogate = u16_swap_bytes(high_surrogate); 90| 10.5k| low_surrogate = u16_swap_bytes(low_surrogate); 91| 10.5k| } 92| 10.5k| *utf16_output++ = char16_t(high_surrogate); 93| 10.5k| *utf16_output++ = char16_t(low_surrogate); 94| 10.5k| pos += 4; 95| 10.5k| } else { 96| | // we may have a continuation but we do not do error checking 97| 0| return 0; 98| 0| } 99| 366k| } 100| 1.38k| return utf16_output - start; 101| 1.38k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf3230rewind_and_convert_with_errorsEmPKcmPDi: 249| 1.57k| char32_t *utf32_output) { 250| 1.57k| size_t extra_len{0}; 251| | // We potentially need to go back in time and find a leading byte. 252| 1.57k| size_t how_far_back = 3; // 3 bytes in the past + current position 253| 1.57k| if (how_far_back > prior_bytes) { ------------------ | Branch (253:7): [True: 852, False: 720] ------------------ 254| 852| how_far_back = prior_bytes; 255| 852| } 256| 1.57k| bool found_leading_bytes{false}; 257| | // important: it is i <= how_far_back and not 'i < how_far_back'. 258| 1.64k| for (size_t i = 0; i <= how_far_back; i++) { ------------------ | Branch (258:22): [True: 1.57k, False: 64] ------------------ 259| 1.57k| unsigned char byte = buf[-static_cast(i)]; 260| 1.57k| found_leading_bytes = ((byte & 0b11000000) != 0b10000000); 261| 1.57k| if (found_leading_bytes) { ------------------ | Branch (261:9): [True: 1.50k, False: 68] ------------------ 262| 1.50k| if (i > 0 && byte < 128) { ------------------ | Branch (262:11): [True: 4, False: 1.50k] | 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| 1.50k| buf -= i; 268| 1.50k| extra_len = i; 269| 1.50k| break; 270| 1.50k| } 271| 1.57k| } 272| | // 273| | // It is possible for this function to return a negative count in its result. 274| | // C++ Standard Section 18.1 defines size_t is in which is described 275| | // in C Standard as . C Standard Section 4.1.5 defines size_t as an 276| | // unsigned integral type of the result of the sizeof operator 277| | // 278| | // An unsigned type will simply wrap round arithmetically (well defined). 279| | // 280| 1.56k| if (!found_leading_bytes) { ------------------ | Branch (280:7): [True: 64, False: 1.50k] ------------------ 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| 64| return result(error_code::TOO_LONG, 0 - how_far_back); 286| 64| } 287| | 288| 1.50k| result res = convert_with_errors(buf, len + extra_len, utf32_output); 289| 1.50k| if (res.error) { ------------------ | Branch (289:7): [True: 820, False: 684] ------------------ 290| 820| res.count -= extra_len; 291| 820| } 292| 1.50k| return res; 293| 1.56k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf3219convert_with_errorsIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEENS_6resultES6_mPDi: 121| 2.29k| char32_t *utf32_output) { 122| 2.29k| size_t pos = 0; 123| 2.29k| char32_t *start{utf32_output}; 124| 2.75M| while (pos < len) { ------------------ | Branch (124:10): [True: 2.75M, False: 1.03k] ------------------ 125| |#if SIMDUTF_CPLUSPLUS23 126| | if !consteval 127| |#endif 128| 2.75M| { 129| | // try to convert the next block of 16 ASCII bytes 130| 2.75M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (130:11): [True: 2.74M, False: 8.72k] ------------------ 131| | // they are ascii 132| 2.74M| uint64_t v1; 133| 2.74M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 134| 2.74M| uint64_t v2; 135| 2.74M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 136| 2.74M| uint64_t v{v1 | v2}; 137| 2.74M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (137:13): [True: 1.73M, False: 1.01M] ------------------ 138| 1.73M| size_t final_pos = pos + 16; 139| 29.5M| while (pos < final_pos) { ------------------ | Branch (139:18): [True: 27.7M, False: 1.73M] ------------------ 140| 27.7M| *utf32_output++ = uint8_t(data[pos]); 141| 27.7M| pos++; 142| 27.7M| } 143| 1.73M| continue; 144| 1.73M| } 145| 2.74M| } 146| 2.75M| } 147| 1.02M| auto leading_byte = uint8_t(data[pos]); // leading byte 148| 1.02M| if (leading_byte < 0b10000000) { ------------------ | Branch (148:9): [True: 768k, False: 253k] ------------------ 149| | // converting one ASCII byte !!! 150| 768k| *utf32_output++ = char32_t(leading_byte); 151| 768k| pos++; 152| 768k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (152:16): [True: 179k, False: 73.3k] ------------------ 153| | // We have a two-byte UTF-8 154| 179k| if (pos + 1 >= len) { ------------------ | Branch (154:11): [True: 78, False: 179k] ------------------ 155| 78| return result(error_code::TOO_SHORT, pos); 156| 78| } // minimal bound checking 157| 179k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (157:11): [True: 180, False: 179k] ------------------ 158| 180| return result(error_code::TOO_SHORT, pos); 159| 180| } 160| | // range check 161| 179k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 162| 179k| (uint8_t(data[pos + 1]) & 0b00111111); 163| 179k| if (code_point < 0x80) { ------------------ | Branch (163:11): [True: 18, False: 179k] ------------------ 164| 18| return result(error_code::OVERLONG, pos); 165| 18| } 166| 179k| *utf32_output++ = char32_t(code_point); 167| 179k| pos += 2; 168| 179k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (168:16): [True: 57.9k, False: 15.4k] ------------------ 169| | // We have a three-byte UTF-8 170| 57.9k| if (pos + 2 >= len) { ------------------ | Branch (170:11): [True: 54, False: 57.9k] ------------------ 171| 54| return result(error_code::TOO_SHORT, pos); 172| 54| } // minimal bound checking 173| | 174| 57.9k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (174:11): [True: 72, False: 57.8k] ------------------ 175| 72| return result(error_code::TOO_SHORT, pos); 176| 72| } 177| 57.8k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (177:11): [True: 66, False: 57.7k] ------------------ 178| 66| return result(error_code::TOO_SHORT, pos); 179| 66| } 180| | // range check 181| 57.7k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 182| 57.7k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 183| 57.7k| (uint8_t(data[pos + 2]) & 0b00111111); 184| 57.7k| if (code_point < 0x800) { ------------------ | Branch (184:11): [True: 36, False: 57.7k] ------------------ 185| 36| return result(error_code::OVERLONG, pos); 186| 36| } 187| 57.7k| if (0xd7ff < code_point && code_point < 0xe000) { ------------------ | Branch (187:11): [True: 3.24k, False: 54.4k] | Branch (187:34): [True: 18, False: 3.22k] ------------------ 188| 18| return result(error_code::SURROGATE, pos); 189| 18| } 190| 57.7k| *utf32_output++ = char32_t(code_point); 191| 57.7k| pos += 3; 192| 57.7k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (192:16): [True: 14.9k, False: 484] ------------------ 193| | // we have a 4-byte UTF-8 word. 194| 14.9k| if (pos + 3 >= len) { ------------------ | Branch (194:11): [True: 24, False: 14.9k] ------------------ 195| 24| return result(error_code::TOO_SHORT, pos); 196| 24| } // minimal bound checking 197| 14.9k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (197:11): [True: 66, False: 14.8k] ------------------ 198| 66| return result(error_code::TOO_SHORT, pos); 199| 66| } 200| 14.8k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (200:11): [True: 42, False: 14.8k] ------------------ 201| 42| return result(error_code::TOO_SHORT, pos); 202| 42| } 203| 14.8k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (203:11): [True: 42, False: 14.7k] ------------------ 204| 42| return result(error_code::TOO_SHORT, pos); 205| 42| } 206| | 207| | // range check 208| 14.7k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 209| 14.7k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 210| 14.7k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 211| 14.7k| (uint8_t(data[pos + 3]) & 0b00111111); 212| 14.7k| if (code_point <= 0xffff) { ------------------ | Branch (212:11): [True: 42, False: 14.7k] ------------------ 213| 42| return result(error_code::OVERLONG, pos); 214| 42| } 215| 14.7k| if (0x10ffff < code_point) { ------------------ | Branch (215:11): [True: 42, False: 14.6k] ------------------ 216| 42| return result(error_code::TOO_LARGE, pos); 217| 42| } 218| 14.6k| *utf32_output++ = char32_t(code_point); 219| 14.6k| pos += 4; 220| 14.6k| } else { 221| | // we either have too many continuation bytes or an invalid leading byte 222| 484| if ((leading_byte & 0b11000000) == 0b10000000) { ------------------ | Branch (222:11): [True: 220, False: 264] ------------------ 223| 220| return result(error_code::TOO_LONG, pos); 224| 264| } else { 225| 264| return result(error_code::HEADER_BITS, pos); 226| 264| } 227| 484| } 228| 1.02M| } 229| 1.03k| return result(error_code::SUCCESS, utf32_output - start); 230| 2.29k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf327convertIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mPDi: 16| 1.79k| char32_t *utf32_output) { 17| 1.79k| size_t pos = 0; 18| 1.79k| char32_t *start{utf32_output}; 19| 2.38M| while (pos < len) { ------------------ | Branch (19:10): [True: 2.38M, False: 836] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 2.38M| { 24| | // try to convert the next block of 16 ASCII bytes 25| 2.38M| if (pos + 16 <= len) { // if it is safe to read 16 more bytes, check that ------------------ | Branch (25:11): [True: 2.38M, False: 5.52k] ------------------ 26| | // they are ascii 27| 2.38M| uint64_t v1; 28| 2.38M| ::memcpy(&v1, data + pos, sizeof(uint64_t)); 29| 2.38M| uint64_t v2; 30| 2.38M| ::memcpy(&v2, data + pos + sizeof(uint64_t), sizeof(uint64_t)); 31| 2.38M| uint64_t v{v1 | v2}; 32| 2.38M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (32:13): [True: 1.40M, False: 975k] ------------------ 33| 1.40M| size_t final_pos = pos + 16; 34| 23.9M| while (pos < final_pos) { ------------------ | Branch (34:18): [True: 22.4M, False: 1.40M] ------------------ 35| 22.4M| *utf32_output++ = uint8_t(data[pos]); 36| 22.4M| pos++; 37| 22.4M| } 38| 1.40M| continue; 39| 1.40M| } 40| 2.38M| } 41| 2.38M| } 42| 981k| auto leading_byte = uint8_t(data[pos]); // leading byte 43| 981k| if (leading_byte < 0b10000000) { ------------------ | Branch (43:9): [True: 814k, False: 166k] ------------------ 44| | // converting one ASCII byte !!! 45| 814k| *utf32_output++ = char32_t(leading_byte); 46| 814k| pos++; 47| 814k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (47:16): [True: 96.7k, False: 70.2k] ------------------ 48| | // We have a two-byte UTF-8 49| 96.7k| if (pos + 1 >= len) { ------------------ | Branch (49:11): [True: 54, False: 96.6k] ------------------ 50| 54| return 0; 51| 54| } // minimal bound checking 52| 96.6k| if ((data[pos + 1] & 0b11000000) != 0b10000000) { ------------------ | Branch (52:11): [True: 104, False: 96.5k] ------------------ 53| 104| return 0; 54| 104| } 55| | // range check 56| 96.5k| uint32_t code_point = (leading_byte & 0b00011111) << 6 | 57| 96.5k| (uint8_t(data[pos + 1]) & 0b00111111); 58| 96.5k| if (code_point < 0x80) { ------------------ | Branch (58:11): [True: 18, False: 96.5k] ------------------ 59| 18| return 0; 60| 18| } 61| 96.5k| *utf32_output++ = char32_t(code_point); 62| 96.5k| pos += 2; 63| 96.5k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (63:16): [True: 63.7k, False: 6.51k] ------------------ 64| | // We have a three-byte UTF-8 65| 63.7k| if (pos + 2 >= len) { ------------------ | Branch (65:11): [True: 36, False: 63.7k] ------------------ 66| 36| return 0; 67| 36| } // minimal bound checking 68| | 69| 63.7k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (69:11): [True: 76, False: 63.6k] ------------------ 70| 76| return 0; 71| 76| } 72| 63.6k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (72:11): [True: 30, False: 63.6k] ------------------ 73| 30| return 0; 74| 30| } 75| | // range check 76| 63.6k| uint32_t code_point = (leading_byte & 0b00001111) << 12 | 77| 63.6k| (uint8_t(data[pos + 1]) & 0b00111111) << 6 | 78| 63.6k| (uint8_t(data[pos + 2]) & 0b00111111); 79| 63.6k| if (code_point < 0x800 || (0xd7ff < code_point && code_point < 0xe000)) { ------------------ | Branch (79:11): [True: 40, False: 63.5k] | Branch (79:34): [True: 1.67k, False: 61.9k] | Branch (79:57): [True: 14, False: 1.65k] ------------------ 80| 54| return 0; 81| 54| } 82| 63.5k| *utf32_output++ = char32_t(code_point); 83| 63.5k| pos += 3; 84| 63.5k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (84:16): [True: 6.18k, False: 334] ------------------ 85| | // we have a 4-byte UTF-8 word. 86| 6.18k| if (pos + 3 >= len) { ------------------ | Branch (86:11): [True: 24, False: 6.15k] ------------------ 87| 24| return 0; 88| 24| } // minimal bound checking 89| 6.15k| if ((uint8_t(data[pos + 1]) & 0b11000000) != 0b10000000) { ------------------ | Branch (89:11): [True: 62, False: 6.09k] ------------------ 90| 62| return 0; 91| 62| } 92| 6.09k| if ((uint8_t(data[pos + 2]) & 0b11000000) != 0b10000000) { ------------------ | Branch (92:11): [True: 42, False: 6.05k] ------------------ 93| 42| return 0; 94| 42| } 95| 6.05k| if ((uint8_t(data[pos + 3]) & 0b11000000) != 0b10000000) { ------------------ | Branch (95:11): [True: 22, False: 6.03k] ------------------ 96| 22| return 0; 97| 22| } 98| | 99| | // range check 100| 6.03k| uint32_t code_point = (leading_byte & 0b00000111) << 18 | 101| 6.03k| (uint8_t(data[pos + 1]) & 0b00111111) << 12 | 102| 6.03k| (uint8_t(data[pos + 2]) & 0b00111111) << 6 | 103| 6.03k| (uint8_t(data[pos + 3]) & 0b00111111); 104| 6.03k| if (code_point <= 0xffff || 0x10ffff < code_point) { ------------------ | Branch (104:11): [True: 52, False: 5.98k] | Branch (104:35): [True: 46, False: 5.93k] ------------------ 105| 98| return 0; 106| 98| } 107| 5.93k| *utf32_output++ = char32_t(code_point); 108| 5.93k| pos += 4; 109| 5.93k| } else { 110| 334| return 0; 111| 334| } 112| 981k| } 113| 836| return utf32_output - start; 114| 1.79k|} simdutf.cpp:_ZN7simdutf6scalar12_GLOBAL__N_113utf8_to_utf3213convert_validIPKcQsr7simdutf6detailE22indexes_into_byte_likeIT_EEEmS6_mPDi: 16| 1.29k| char32_t *utf32_output) { 17| 1.29k| size_t pos = 0; 18| 1.29k| char32_t *start{utf32_output}; 19| 2.81M| while (pos < len) { ------------------ | Branch (19:10): [True: 2.81M, False: 1.29k] ------------------ 20| |#if SIMDUTF_CPLUSPLUS23 21| | if !consteval 22| |#endif 23| 2.81M| { 24| | // try to convert the next block of 8 ASCII bytes 25| 2.81M| if (pos + 8 <= len) { // if it is safe to read 8 more bytes, check that ------------------ | Branch (25:11): [True: 2.80M, False: 3.82k] ------------------ 26| | // they are ascii 27| 2.80M| uint64_t v; 28| 2.80M| ::memcpy(&v, data + pos, sizeof(uint64_t)); 29| 2.80M| if ((v & 0x8080808080808080) == 0) { ------------------ | Branch (29:13): [True: 2.32M, False: 479k] ------------------ 30| 2.32M| size_t final_pos = pos + 8; 31| 20.9M| while (pos < final_pos) { ------------------ | Branch (31:18): [True: 18.6M, False: 2.32M] ------------------ 32| 18.6M| *utf32_output++ = uint8_t(data[pos]); 33| 18.6M| pos++; 34| 18.6M| } 35| 2.32M| continue; 36| 2.32M| } 37| 2.80M| } 38| 2.81M| } 39| 483k| auto leading_byte = uint8_t(data[pos]); // leading byte 40| 483k| if (leading_byte < 0b10000000) { ------------------ | Branch (40:9): [True: 253k, False: 229k] ------------------ 41| | // converting one ASCII byte !!! 42| 253k| *utf32_output++ = char32_t(leading_byte); 43| 253k| pos++; 44| 253k| } else if ((leading_byte & 0b11100000) == 0b11000000) { ------------------ | Branch (44:16): [True: 182k, False: 47.5k] ------------------ 45| | // We have a two-byte UTF-8 46| 182k| if (pos + 1 >= len) { ------------------ | Branch (46:11): [True: 0, False: 182k] ------------------ 47| 0| break; 48| 0| } // minimal bound checking 49| 182k| *utf32_output++ = char32_t(((leading_byte & 0b00011111) << 6) | 50| 182k| (uint8_t(data[pos + 1]) & 0b00111111)); 51| 182k| pos += 2; 52| 182k| } else if ((leading_byte & 0b11110000) == 0b11100000) { ------------------ | Branch (52:16): [True: 21.7k, False: 25.8k] ------------------ 53| | // We have a three-byte UTF-8 54| 21.7k| if (pos + 2 >= len) { ------------------ | Branch (54:11): [True: 0, False: 21.7k] ------------------ 55| 0| break; 56| 0| } // minimal bound checking 57| 21.7k| *utf32_output++ = char32_t(((leading_byte & 0b00001111) << 12) | 58| 21.7k| ((uint8_t(data[pos + 1]) & 0b00111111) << 6) | 59| 21.7k| (uint8_t(data[pos + 2]) & 0b00111111)); 60| 21.7k| pos += 3; 61| 25.8k| } else if ((leading_byte & 0b11111000) == 0b11110000) { // 0b11110000 ------------------ | Branch (61:16): [True: 25.8k, False: 0] ------------------ 62| | // we have a 4-byte UTF-8 word. 63| 25.8k| if (pos + 3 >= len) { ------------------ | Branch (63:11): [True: 0, False: 25.8k] ------------------ 64| 0| break; 65| 0| } // minimal bound checking 66| 25.8k| uint32_t code_word = ((leading_byte & 0b00000111) << 18) | 67| 25.8k| ((uint8_t(data[pos + 1]) & 0b00111111) << 12) | 68| 25.8k| ((uint8_t(data[pos + 2]) & 0b00111111) << 6) | 69| 25.8k| (uint8_t(data[pos + 3]) & 0b00111111); 70| 25.8k| *utf32_output++ = char32_t(code_word); 71| 25.8k| pos += 4; 72| 25.8k| } else { 73| | // we may have a continuation but we do not do error checking 74| 0| return 0; 75| 0| } 76| 483k| } 77| 1.29k| return utf32_output - start; 78| 1.29k|} _ZNK7simdutf8fallback14implementation13validate_utf8EPKcm: 37| 3.52k|implementation::validate_utf8(const char *buf, size_t len) const noexcept { 38| 3.52k| return scalar::utf8::validate(buf, len); 39| 3.52k|} _ZNK7simdutf8fallback14implementation25validate_utf8_with_errorsEPKcm: 44| 3.52k| const char *buf, size_t len) const noexcept { 45| 3.52k| return scalar::utf8::validate_with_errors(buf, len); 46| 3.52k|} _ZNK7simdutf8fallback14implementation16validate_utf16leEPKDsm: 77| 1.38k| size_t len) const noexcept { 78| 1.38k| return scalar::utf16::validate(buf, len); 79| 1.38k|} _ZNK7simdutf8fallback14implementation16validate_utf16beEPKDsm: 85| 1.34k| size_t len) const noexcept { 86| 1.34k| return scalar::utf16::validate(buf, len); 87| 1.34k|} _ZNK7simdutf8fallback14implementation28validate_utf16le_with_errorsEPKDsm: 90| 1.38k| const char16_t *buf, size_t len) const noexcept { 91| 1.38k| return scalar::utf16::validate_with_errors(buf, len); 92| 1.38k|} _ZNK7simdutf8fallback14implementation28validate_utf16be_with_errorsEPKDsm: 95| 1.34k| const char16_t *buf, size_t len) const noexcept { 96| 1.34k| return scalar::utf16::validate_with_errors(buf, len); 97| 1.34k|} _ZNK7simdutf8fallback14implementation14validate_utf32EPKDim: 114| 2.57k|implementation::validate_utf32(const char32_t *buf, size_t len) const noexcept { 115| 2.57k| return scalar::utf32::validate(buf, len); 116| 2.57k|} _ZNK7simdutf8fallback14implementation26validate_utf32_with_errorsEPKDim: 121| 2.57k| const char32_t *buf, size_t len) const noexcept { 122| 2.57k| return scalar::utf32::validate_with_errors(buf, len); 123| 2.57k|} _ZNK7simdutf8fallback14implementation22convert_latin1_to_utf8EPKcmPc: 128| 484| const char *buf, size_t len, char *utf8_output) const noexcept { 129| 484| return scalar::latin1_to_utf8::convert(buf, len, utf8_output); 130| 484|} _ZNK7simdutf8fallback14implementation25convert_latin1_to_utf16leEPKcmPDs: 135| 88| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 136| 88| return scalar::latin1_to_utf16::convert(buf, len, 137| 88| utf16_output); 138| 88|} _ZNK7simdutf8fallback14implementation25convert_latin1_to_utf16beEPKcmPDs: 141| 110| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 142| 110| return scalar::latin1_to_utf16::convert(buf, len, 143| 110| utf16_output); 144| 110|} _ZNK7simdutf8fallback14implementation23convert_latin1_to_utf32EPKcmPDi: 149| 100| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 150| 100| return scalar::latin1_to_utf32::convert(buf, len, utf32_output); 151| 100|} _ZNK7simdutf8fallback14implementation22convert_utf8_to_latin1EPKcmPc: 156| 612| const char *buf, size_t len, char *latin1_output) const noexcept { 157| 612| return scalar::utf8_to_latin1::convert(buf, len, latin1_output); 158| 612|} _ZNK7simdutf8fallback14implementation34convert_utf8_to_latin1_with_errorsEPKcmPc: 161| 522| const char *buf, size_t len, char *latin1_output) const noexcept { 162| 522| return scalar::utf8_to_latin1::convert_with_errors(buf, len, latin1_output); 163| 522|} _ZNK7simdutf8fallback14implementation23convert_utf8_to_utf16leEPKcmPDs: 173| 784| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 174| 784| return scalar::utf8_to_utf16::convert(buf, len, 175| 784| utf16_output); 176| 784|} _ZNK7simdutf8fallback14implementation23convert_utf8_to_utf16beEPKcmPDs: 179| 758| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 180| 758| return scalar::utf8_to_utf16::convert(buf, len, 181| 758| utf16_output); 182| 758|} _ZNK7simdutf8fallback14implementation35convert_utf8_to_utf16le_with_errorsEPKcmPDs: 185| 594| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 186| 594| return scalar::utf8_to_utf16::convert_with_errors( 187| 594| buf, len, utf16_output); 188| 594|} _ZNK7simdutf8fallback14implementation35convert_utf8_to_utf16be_with_errorsEPKcmPDs: 191| 772| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 192| 772| return scalar::utf8_to_utf16::convert_with_errors( 193| 772| buf, len, utf16_output); 194| 772|} _ZNK7simdutf8fallback14implementation29convert_valid_utf8_to_utf16leEPKcmPDs: 197| 432| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 198| 432| return scalar::utf8_to_utf16::convert_valid(buf, len, 199| 432| utf16_output); 200| 432|} _ZNK7simdutf8fallback14implementation29convert_valid_utf8_to_utf16beEPKcmPDs: 203| 460| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 204| 460| return scalar::utf8_to_utf16::convert_valid(buf, len, 205| 460| utf16_output); 206| 460|} _ZNK7simdutf8fallback14implementation21convert_utf8_to_utf32EPKcmPDi: 211| 818| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 212| 818| return scalar::utf8_to_utf32::convert(buf, len, utf32_output); 213| 818|} _ZNK7simdutf8fallback14implementation33convert_utf8_to_utf32_with_errorsEPKcmPDi: 216| 790| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 217| 790| return scalar::utf8_to_utf32::convert_with_errors(buf, len, utf32_output); 218| 790|} _ZNK7simdutf8fallback14implementation27convert_valid_utf8_to_utf32EPKcmPDi: 221| 432| const char *input, size_t size, char32_t *utf32_output) const noexcept { 222| 432| return scalar::utf8_to_utf32::convert_valid(input, size, utf32_output); 223| 432|} _ZNK7simdutf8fallback14implementation25convert_utf16le_to_latin1EPKDsmPc: 228| 150| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 229| 150| return scalar::utf16_to_latin1::convert(buf, len, 230| 150| latin1_output); 231| 150|} _ZNK7simdutf8fallback14implementation25convert_utf16be_to_latin1EPKDsmPc: 234| 112| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 235| 112| return scalar::utf16_to_latin1::convert(buf, len, 236| 112| latin1_output); 237| 112|} _ZNK7simdutf8fallback14implementation37convert_utf16le_to_latin1_with_errorsEPKDsmPc: 241| 226| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 242| 226| return scalar::utf16_to_latin1::convert_with_errors( 243| 226| buf, len, latin1_output); 244| 226|} _ZNK7simdutf8fallback14implementation37convert_utf16be_to_latin1_with_errorsEPKDsmPc: 248| 230| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 249| 230| return scalar::utf16_to_latin1::convert_with_errors( 250| 230| buf, len, latin1_output); 251| 230|} _ZNK7simdutf8fallback14implementation23convert_utf16le_to_utf8EPKDsmPc: 268| 492| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 269| 492| return scalar::utf16_to_utf8::convert(buf, len, 270| 492| utf8_output); 271| 492|} _ZNK7simdutf8fallback14implementation23convert_utf16be_to_utf8EPKDsmPc: 274| 358| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 275| 358| return scalar::utf16_to_utf8::convert(buf, len, utf8_output); 276| 358|} _ZNK7simdutf8fallback14implementation35convert_utf16le_to_utf8_with_errorsEPKDsmPc: 279| 654| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 280| 654| return scalar::utf16_to_utf8::convert_with_errors( 281| 654| buf, len, utf8_output); 282| 654|} _ZNK7simdutf8fallback14implementation35convert_utf16be_to_utf8_with_errorsEPKDsmPc: 285| 648| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 286| 648| return scalar::utf16_to_utf8::convert_with_errors( 287| 648| buf, len, utf8_output); 288| 648|} _ZNK7simdutf8fallback14implementation29convert_valid_utf16le_to_utf8EPKDsmPc: 291| 384| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 292| 384| return scalar::utf16_to_utf8::convert_valid(buf, len, 293| 384| utf8_output); 294| 384|} _ZNK7simdutf8fallback14implementation29convert_valid_utf16be_to_utf8EPKDsmPc: 297| 414| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 298| 414| return scalar::utf16_to_utf8::convert_valid(buf, len, 299| 414| utf8_output); 300| 414|} _ZNK7simdutf8fallback14implementation23convert_utf32_to_latin1EPKDimPc: 305| 206| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 306| 206| return scalar::utf32_to_latin1::convert(buf, len, latin1_output); 307| 206|} _ZNK7simdutf8fallback14implementation35convert_utf32_to_latin1_with_errorsEPKDimPc: 310| 454| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 311| 454| return scalar::utf32_to_latin1::convert_with_errors(buf, len, latin1_output); 312| 454|} _ZNK7simdutf8fallback14implementation21convert_utf32_to_utf8EPKDimPc: 322| 684| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 323| 684| return scalar::utf32_to_utf8::convert(buf, len, utf8_output); 324| 684|} _ZNK7simdutf8fallback14implementation33convert_utf32_to_utf8_with_errorsEPKDimPc: 327| 924| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 328| 924| return scalar::utf32_to_utf8::convert_with_errors(buf, len, utf8_output); 329| 924|} _ZNK7simdutf8fallback14implementation27convert_valid_utf32_to_utf8EPKDimPc: 332| 188| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 333| 188| return scalar::utf32_to_utf8::convert_valid(buf, len, utf8_output); 334| 188|} _ZNK7simdutf8fallback14implementation24convert_utf32_to_utf16leEPKDimPDs: 339| 480| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 340| 480| return scalar::utf32_to_utf16::convert(buf, len, 341| 480| utf16_output); 342| 480|} _ZNK7simdutf8fallback14implementation24convert_utf32_to_utf16beEPKDimPDs: 345| 426| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 346| 426| return scalar::utf32_to_utf16::convert(buf, len, 347| 426| utf16_output); 348| 426|} _ZNK7simdutf8fallback14implementation36convert_utf32_to_utf16le_with_errorsEPKDimPDs: 351| 714| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 352| 714| return scalar::utf32_to_utf16::convert_with_errors( 353| 714| buf, len, utf16_output); 354| 714|} _ZNK7simdutf8fallback14implementation36convert_utf32_to_utf16be_with_errorsEPKDimPDs: 357| 660| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 358| 660| return scalar::utf32_to_utf16::convert_with_errors( 359| 660| buf, len, utf16_output); 360| 660|} _ZNK7simdutf8fallback14implementation30convert_valid_utf32_to_utf16leEPKDimPDs: 363| 252| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 364| 252| return scalar::utf32_to_utf16::convert_valid( 365| 252| buf, len, utf16_output); 366| 252|} _ZNK7simdutf8fallback14implementation30convert_valid_utf32_to_utf16beEPKDimPDs: 369| 114| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 370| 114| return scalar::utf32_to_utf16::convert_valid(buf, len, 371| 114| utf16_output); 372| 114|} _ZNK7simdutf8fallback14implementation24convert_utf16le_to_utf32EPKDsmPDi: 375| 308| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 376| 308| return scalar::utf16_to_utf32::convert(buf, len, 377| 308| utf32_output); 378| 308|} _ZNK7simdutf8fallback14implementation24convert_utf16be_to_utf32EPKDsmPDi: 381| 282| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 382| 282| return scalar::utf16_to_utf32::convert(buf, len, 383| 282| utf32_output); 384| 282|} _ZNK7simdutf8fallback14implementation36convert_utf16le_to_utf32_with_errorsEPKDsmPDi: 387| 402| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 388| 402| return scalar::utf16_to_utf32::convert_with_errors( 389| 402| buf, len, utf32_output); 390| 402|} _ZNK7simdutf8fallback14implementation36convert_utf16be_to_utf32_with_errorsEPKDsmPDi: 393| 422| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 394| 422| return scalar::utf16_to_utf32::convert_with_errors( 395| 422| buf, len, utf32_output); 396| 422|} _ZNK7simdutf8fallback14implementation30convert_valid_utf16le_to_utf32EPKDsmPDi: 399| 120| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 400| 120| return scalar::utf16_to_utf32::convert_valid( 401| 120| buf, len, utf32_output); 402| 120|} _ZNK7simdutf8fallback14implementation30convert_valid_utf16be_to_utf32EPKDsmPDi: 405| 166| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 406| 166| return scalar::utf16_to_utf32::convert_valid(buf, len, 407| 166| utf32_output); 408| 166|} _ZNK7simdutf8fallback14implementation13count_utf16leEPKDsm: 419| 1.38k| const char16_t *input, size_t length) const noexcept { 420| 1.38k| return scalar::utf16::count_code_points(input, length); 421| 1.38k|} _ZNK7simdutf8fallback14implementation13count_utf16beEPKDsm: 424| 1.34k| const char16_t *input, size_t length) const noexcept { 425| 1.34k| return scalar::utf16::count_code_points(input, length); 426| 1.34k|} _ZNK7simdutf8fallback14implementation10count_utf8EPKcm: 431| 3.52k|implementation::count_utf8(const char *input, size_t length) const noexcept { 432| 3.52k| return scalar::utf8::count_code_points(input, length); 433| 3.52k|} _ZNK7simdutf8fallback14implementation23latin1_length_from_utf8EPKcm: 438| 567| const char *buf, size_t len) const noexcept { 439| 567| return scalar::utf8::count_code_points(buf, len); 440| 567|} _ZNK7simdutf8fallback14implementation23utf8_length_from_latin1EPKcm: 445| 242| const char *input, size_t length) const noexcept { 446| 242| return scalar::latin1_to_utf8::utf8_length_from_latin1(input, length); 447| 242|} _ZNK7simdutf8fallback14implementation24utf8_length_from_utf16leEPKDsm: 452| 778| const char16_t *input, size_t length) const noexcept { 453| 778| return scalar::utf16::utf8_length_from_utf16(input, 454| 778| length); 455| 778|} _ZNK7simdutf8fallback14implementation24utf8_length_from_utf16beEPKDsm: 458| 727| const char16_t *input, size_t length) const noexcept { 459| 727| return scalar::utf16::utf8_length_from_utf16(input, length); 460| 727|} _ZNK7simdutf8fallback14implementation25utf32_length_from_utf16leEPKDsm: 465| 420| const char16_t *input, size_t length) const noexcept { 466| 420| return scalar::utf16::utf32_length_from_utf16(input, 467| 420| length); 468| 420|} _ZNK7simdutf8fallback14implementation25utf32_length_from_utf16beEPKDsm: 471| 448| const char16_t *input, size_t length) const noexcept { 472| 448| return scalar::utf16::utf32_length_from_utf16(input, length); 473| 448|} _ZNK7simdutf8fallback14implementation22utf16_length_from_utf8EPKcm: 478| 1.92k| const char *input, size_t length) const noexcept { 479| 1.92k| return scalar::utf8::utf16_length_from_utf8(input, length); 480| 1.92k|} _ZNK7simdutf8fallback14implementation22utf8_length_from_utf32EPKDim: 513| 904| const char32_t *input, size_t length) const noexcept { 514| 904| return scalar::utf32::utf8_length_from_utf32(input, length); 515| 904|} _ZNK7simdutf8fallback14implementation23utf16_length_from_utf32EPKDim: 520| 1.33k| const char32_t *input, size_t length) const noexcept { 521| 1.33k| return scalar::utf32::utf16_length_from_utf32(input, length); 522| 1.33k|} _ZNK7simdutf8fallback14implementation22utf32_length_from_utf8EPKcm: 527| 1.03k| const char *input, size_t length) const noexcept { 528| 1.03k| return scalar::utf8::count_code_points(input, length); 529| 1.03k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE13get_remainderEPh: 57| 6.16k|buf_block_reader::get_remainder(uint8_t *dst) const { 58| 6.16k| if (len == idx) { ------------------ | Branch (58:7): [True: 32, False: 6.12k] ------------------ 59| 32| return 0; 60| 32| } // memcpy(dst, null, 0) will trigger an error with some sanitizers 61| 6.12k| std::memset(dst, 0x20, 62| 6.12k| STEP_SIZE); // std::memset STEP_SIZE because it is more efficient 63| | // to write out 8 or 16 bytes at once. 64| 6.12k| std::memcpy(dst, buf + idx, len - idx); 65| 6.12k| return len - idx; 66| 6.16k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE13get_remainderEPh: 57| 6.16k|buf_block_reader::get_remainder(uint8_t *dst) const { 58| 6.16k| if (len == idx) { ------------------ | Branch (58:7): [True: 32, False: 6.12k] ------------------ 59| 32| return 0; 60| 32| } // memcpy(dst, null, 0) will trigger an error with some sanitizers 61| 6.12k| std::memset(dst, 0x20, 62| 6.12k| STEP_SIZE); // std::memset STEP_SIZE because it is more efficient 63| | // to write out 8 or 16 bytes at once. 64| 6.12k| std::memcpy(dst, buf + idx, len - idx); 65| 6.12k| return len - idx; 66| 6.16k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EEC2EPKhm: 36| 7.04k| : buf{_buf}, len{_len}, lenminusstep{len < STEP_SIZE ? 0 : len - STEP_SIZE}, ------------------ | Branch (36:42): [True: 2.73k, False: 4.31k] ------------------ 37| 7.04k| idx{0} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE14has_full_blockEv: 45| 4.75M|simdutf_really_inline bool buf_block_reader::has_full_block() const { 46| 4.75M| return idx < lenminusstep; 47| 4.75M|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE10full_blockEv: 51| 4.74M|buf_block_reader::full_block() const { 52| 4.74M| return &buf[idx]; 53| 4.74M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_116buf_block_readerILm64EE7advanceEv: 69| 4.74M|simdutf_really_inline void buf_block_reader::advance() { 70| 4.74M| idx += STEP_SIZE; 71| 4.74M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EEC2EPKhm: 36| 7.04k| : buf{_buf}, len{_len}, lenminusstep{len < STEP_SIZE ? 0 : len - STEP_SIZE}, ------------------ | Branch (36:42): [True: 2.73k, False: 4.31k] ------------------ 37| 7.04k| idx{0} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE14has_full_blockEv: 45| 4.75M|simdutf_really_inline bool buf_block_reader::has_full_block() const { 46| 4.75M| return idx < lenminusstep; 47| 4.75M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE10full_blockEv: 51| 4.74M|buf_block_reader::full_block() const { 52| 4.74M| return &buf[idx]; 53| 4.74M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_116buf_block_readerILm64EE7advanceEv: 69| 4.74M|simdutf_really_inline void buf_block_reader::advance() { 70| 4.74M| idx += STEP_SIZE; 71| 4.74M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE0EEEmPKDsm: 8| 1.80k| size_t size) { 9| 1.80k| size_t pos = 0; 10| 1.80k| size_t count = 0; 11| 974k| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 973k, False: 1.80k] ------------------ 12| 973k| simd16x32 input(reinterpret_cast(in + pos)); 13| | if constexpr (!match_system(big_endian)) { 14| | input.swap_bytes(); 15| | } 16| 973k| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 973k| count += count_ones(not_pair) / 2; 18| 973k| } 19| 1.80k| return count + 20| 1.80k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.80k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE1EEEmPKDsm: 8| 1.79k| size_t size) { 9| 1.79k| size_t pos = 0; 10| 1.79k| size_t count = 0; 11| 1.06M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 1.06M, False: 1.79k] ------------------ 12| 1.06M| simd16x32 input(reinterpret_cast(in + pos)); 13| 1.06M| if constexpr (!match_system(big_endian)) { 14| 1.06M| input.swap_bytes(); 15| 1.06M| } 16| 1.06M| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 1.06M| count += count_ones(not_pair) / 2; 18| 1.06M| } 19| 1.79k| return count + 20| 1.79k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.79k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 51| 420| size_t size) { 52| 420| return count_code_points(in, size); 53| 420|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 51| 448| size_t size) { 52| 448| return count_code_points(in, size); 53| 448|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE0EEEmPKDsm: 8| 1.80k| size_t size) { 9| 1.80k| size_t pos = 0; 10| 1.80k| size_t count = 0; 11| 974k| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 973k, False: 1.80k] ------------------ 12| 973k| simd16x32 input(reinterpret_cast(in + pos)); 13| | if constexpr (!match_system(big_endian)) { 14| | input.swap_bytes(); 15| | } 16| 973k| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 973k| count += count_ones(not_pair) / 2; 18| 973k| } 19| 1.80k| return count + 20| 1.80k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.80k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1617count_code_pointsILNS_10endiannessE1EEEmPKDsm: 8| 1.79k| size_t size) { 9| 1.79k| size_t pos = 0; 10| 1.79k| size_t count = 0; 11| 1.06M| for (; pos < size / 32 * 32; pos += 32) { ------------------ | Branch (11:10): [True: 1.06M, False: 1.79k] ------------------ 12| 1.06M| simd16x32 input(reinterpret_cast(in + pos)); 13| 1.06M| if constexpr (!match_system(big_endian)) { 14| 1.06M| input.swap_bytes(); 15| 1.06M| } 16| 1.06M| uint64_t not_pair = input.not_in_range(0xDC00, 0xDFFF); 17| 1.06M| count += count_ones(not_pair) / 2; 18| 1.06M| } 19| 1.79k| return count + 20| 1.79k| scalar::utf16::count_code_points(in + pos, size - pos); 21| 1.79k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE0EEEmPKDsm: 51| 420| size_t size) { 52| 420| return count_code_points(in, size); 53| 420|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1623utf32_length_from_utf16ILNS_10endiannessE1EEEmPKDsm: 51| 448| size_t size) { 52| 448| return count_code_points(in, size); 53| 448|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE0EEEmPKDsm: 10| 778| size_t size) { 11| 778| size_t pos = 0; 12| | 13| 778| using vector_u16 = simd16; 14| 778| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 778| const auto one = vector_u16::splat(1); 17| | 18| 778| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 778| 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| 778| constexpr size_t max_iterations = 65535 / 2; 26| 778| size_t iteration = max_iterations; 27| | 28| 1.00M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 1.00M, False: 778] ------------------ 29| 1.00M| 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| 1.00M| 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| 1.00M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 1.00M| 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| 1.00M| v_count += c0; 68| 1.00M| v_count += c1; 69| 1.00M| v_count += vector_u16(is_surrogate); 70| | 71| 1.00M| iteration -= 1; 72| 1.00M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 18, False: 1.00M] ------------------ 73| 18| count += v_count.sum(); 74| 18| v_count = vector_u16::zero(); 75| 18| iteration = max_iterations; 76| 18| } 77| 1.00M| } 78| | 79| 778| if (iteration > 0) { ------------------ | Branch (79:7): [True: 778, False: 0] ------------------ 80| 778| count += v_count.sum(); 81| 778| } 82| | 83| 778| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 778| size - pos); 85| 778|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE1EEEmPKDsm: 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| 1.03M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 1.03M, False: 727] ------------------ 29| 1.03M| auto input = vector_u16::load(reinterpret_cast(in + pos)); 30| 1.03M| if constexpr (!match_system(big_endian)) { 31| 1.03M| input = input.swap_bytes(); 32| 1.03M| } 33| | // 0xd800 .. 0xdbff - low surrogate 34| | // 0xdc00 .. 0xdfff - high surrogate 35| 1.03M| const auto is_surrogate = ((input & uint16_t(0xf800)) == uint16_t(0xd800)); 36| | 37| | // c0 - chars that yield 2- or 3-byte UTF-8 codes 38| 1.03M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 1.03M| const auto c1 = min(input & uint16_t(0xf800), one); 42| | 43| | /* 44| | Explanation how the counting works. 45| | 46| | In the case of a non-surrogate character we count: 47| | * always 1 -- see how `count` is initialized above; 48| | * c0 = 1 if the current char yields 2 or 3 bytes; 49| | * c1 = 1 if the current char yields 3 bytes. 50| | 51| | Thus, we always have correct count for the current char: 52| | from 1, 2 or 3 bytes. 53| | 54| | A trickier part is how we count surrogate pairs. Whether 55| | we encounter a surrogate (low or high), we count it as 56| | 3 chars and then minus 1 (`is_surrogate` is -1 or 0). 57| | Each surrogate char yields 2. A surrogate pair, that 58| | is a low surrogate followed by a high one, yields 59| | the expected 4 bytes. 60| | 61| | It also correctly handles cases when low surrogate is 62| | processed by the this loop, but high surrogate is counted 63| | by the scalar procedure. The scalar procedure uses exactly 64| | the described approach, thanks to that for valid UTF-16 65| | strings it always count correctly. 66| | */ 67| 1.03M| v_count += c0; 68| 1.03M| v_count += c1; 69| 1.03M| v_count += vector_u16(is_surrogate); 70| | 71| 1.03M| iteration -= 1; 72| 1.03M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 17, False: 1.03M] ------------------ 73| 17| count += v_count.sum(); 74| 17| v_count = vector_u16::zero(); 75| 17| iteration = max_iterations; 76| 17| } 77| 1.03M| } 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_10endiannessE0EEEmPKDsm: 10| 778| size_t size) { 11| 778| size_t pos = 0; 12| | 13| 778| using vector_u16 = simd16; 14| 778| constexpr size_t N = vector_u16::ELEMENTS; 15| | 16| 778| const auto one = vector_u16::splat(1); 17| | 18| 778| auto v_count = vector_u16::zero(); 19| | 20| | // each char16 yields at least one byte 21| 778| 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| 778| constexpr size_t max_iterations = 65535 / 2; 26| 778| size_t iteration = max_iterations; 27| | 28| 2.01M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 2.01M, False: 778] ------------------ 29| 2.01M| 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| 2.01M| 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| 2.01M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 2.01M| 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| 2.01M| v_count += c0; 68| 2.01M| v_count += c1; 69| 2.01M| v_count += vector_u16(is_surrogate); 70| | 71| 2.01M| iteration -= 1; 72| 2.01M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 47, False: 2.01M] ------------------ 73| 47| count += v_count.sum(); 74| 47| v_count = vector_u16::zero(); 75| 47| iteration = max_iterations; 76| 47| } 77| 2.01M| } 78| | 79| 778| if (iteration > 0) { ------------------ | Branch (79:7): [True: 778, False: 0] ------------------ 80| 778| count += v_count.sum(); 81| 778| } 82| | 83| 778| return count + scalar::utf16::utf8_length_from_utf16(in + pos, 84| 778| size - pos); 85| 778|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1631utf8_length_from_utf16_bytemaskILNS_10endiannessE1EEEmPKDsm: 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| 2.07M| for (; pos < size / N * N; pos += N) { ------------------ | Branch (28:10): [True: 2.07M, False: 727] ------------------ 29| 2.07M| auto input = vector_u16::load(reinterpret_cast(in + pos)); 30| 2.07M| if constexpr (!match_system(big_endian)) { 31| 2.07M| input = input.swap_bytes(); 32| 2.07M| } 33| | // 0xd800 .. 0xdbff - low surrogate 34| | // 0xdc00 .. 0xdfff - high surrogate 35| 2.07M| const auto is_surrogate = ((input & uint16_t(0xf800)) == uint16_t(0xd800)); 36| | 37| | // c0 - chars that yield 2- or 3-byte UTF-8 codes 38| 2.07M| const auto c0 = min(input & uint16_t(0xff80), one); 39| | 40| | // c1 - chars that yield 3-byte UTF-8 codes (including surrogates) 41| 2.07M| const auto c1 = min(input & uint16_t(0xf800), one); 42| | 43| | /* 44| | Explanation how the counting works. 45| | 46| | In the case of a non-surrogate character we count: 47| | * always 1 -- see how `count` is initialized above; 48| | * c0 = 1 if the current char yields 2 or 3 bytes; 49| | * c1 = 1 if the current char yields 3 bytes. 50| | 51| | Thus, we always have correct count for the current char: 52| | from 1, 2 or 3 bytes. 53| | 54| | A trickier part is how we count surrogate pairs. Whether 55| | we encounter a surrogate (low or high), we count it as 56| | 3 chars and then minus 1 (`is_surrogate` is -1 or 0). 57| | Each surrogate char yields 2. A surrogate pair, that 58| | is a low surrogate followed by a high one, yields 59| | the expected 4 bytes. 60| | 61| | It also correctly handles cases when low surrogate is 62| | processed by the this loop, but high surrogate is counted 63| | by the scalar procedure. The scalar procedure uses exactly 64| | the described approach, thanks to that for valid UTF-16 65| | strings it always count correctly. 66| | */ 67| 2.07M| v_count += c0; 68| 2.07M| v_count += c1; 69| 2.07M| v_count += vector_u16(is_surrogate); 70| | 71| 2.07M| iteration -= 1; 72| 2.07M| if (iteration == 0) { ------------------ | Branch (72:9): [True: 46, False: 2.07M] ------------------ 73| 46| count += v_count.sum(); 74| 46| v_count = vector_u16::zero(); 75| 46| iteration = max_iterations; 76| 46| } 77| 2.07M| } 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_15utf3222utf8_length_from_utf32EPKDim: 11| 904| size_t length) { 12| 904| using vector_u32 = simd32; 13| | 14| 904| 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| 904| const size_t max_increment = 3; 19| | 20| 904| 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| 904| const auto v_ffffff80 = vector_u32::splat(0xffffff80); 28| 904| const auto v_fffff800 = vector_u32::splat(0xfffff800); 29| 904| const auto v_ffff0000 = vector_u32::splat(0xffff0000); 30| 904| const auto one = vector_u32::splat(1); 31| 904|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 32| | 33| 904| size_t counter = 0; 34| | 35| | // 1. vectorized loop unrolled 4 times 36| 904| { 37| | // we use vector of uint32 counters, this is why this limit is used 38| 904| const size_t max_iterations = 39| 904| std::numeric_limits::max() / (max_increment * 4); 40| 904| size_t blocks = length / (N * 4); 41| 904| length -= blocks * (N * 4); 42| 1.10k| while (blocks != 0) { ------------------ | Branch (42:12): [True: 196, False: 904] ------------------ 43| 196| const size_t iterations = min(blocks, max_iterations); 44| 196| blocks -= iterations; 45| | 46| 196| simd32 acc = vector_u32::zero(); 47| 76.7k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (47:26): [True: 76.5k, False: 196] ------------------ 48| 76.5k| const auto in0 = vector_u32(input + 0 * N); 49| 76.5k| const auto in1 = vector_u32(input + 1 * N); 50| 76.5k| const auto in2 = vector_u32(input + 2 * N); 51| 76.5k| 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| 76.5k| acc += min(one, in0 & v_ffffff80); 70| 76.5k| acc += min(one, in1 & v_ffffff80); 71| 76.5k| acc += min(one, in2 & v_ffffff80); 72| 76.5k| acc += min(one, in3 & v_ffffff80); 73| | 74| 76.5k| acc += min(one, in0 & v_fffff800); 75| 76.5k| acc += min(one, in1 & v_fffff800); 76| 76.5k| acc += min(one, in2 & v_fffff800); 77| 76.5k| acc += min(one, in3 & v_fffff800); 78| | 79| 76.5k| acc += min(one, in0 & v_ffff0000); 80| 76.5k| acc += min(one, in1 & v_ffff0000); 81| 76.5k| acc += min(one, in2 & v_ffff0000); 82| 76.5k| acc += min(one, in3 & v_ffff0000); 83| 76.5k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 84| | 85| 76.5k| input += 4 * N; 86| 76.5k| } 87| | 88| 196| counter += acc.sum(); 89| 196| } 90| 904| } 91| | 92| | // 2. vectorized loop for tail 93| 904| { 94| 904| const size_t max_iterations = 95| 904| std::numeric_limits::max() / max_increment; 96| 904| size_t blocks = length / N; 97| 904| length -= blocks * N; 98| 1.35k| while (blocks != 0) { ------------------ | Branch (98:12): [True: 451, False: 904] ------------------ 99| 451| const size_t iterations = min(blocks, max_iterations); 100| 451| blocks -= iterations; 101| | 102| 451| auto acc = vector_u32::zero(); 103| 1.56k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (103:26): [True: 1.11k, False: 451] ------------------ 104| 1.11k| 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| 1.11k| acc += min(one, in & v_ffffff80); 112| 1.11k| acc += min(one, in & v_fffff800); 113| 1.11k| acc += min(one, in & v_ffff0000); 114| 1.11k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 115| | 116| 1.11k| input += N; 117| 1.11k| } 118| | 119| 451| counter += acc.sum(); 120| 451| } 121| 904| } 122| | 123| 904| const size_t consumed = input - start; 124| 904| if (consumed != 0) { ------------------ | Branch (124:7): [True: 499, False: 405] ------------------ 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| 499| counter += consumed; 128| 499| } 129| | 130| 904| return counter + scalar::utf32::utf8_length_from_utf32(input, length); 131| 904|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf323minImEET_S4_S4_: 8| 647|template T min(T a, T b) { return a <= b ? a : b; } ------------------ | Branch (8:48): [True: 647, False: 0] ------------------ simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf3222utf8_length_from_utf32EPKDim: 11| 904| size_t length) { 12| 904| using vector_u32 = simd32; 13| | 14| 904| 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| 904| const size_t max_increment = 3; 19| | 20| 904| 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| 904| const auto v_ffffff80 = vector_u32::splat(0xffffff80); 28| 904| const auto v_fffff800 = vector_u32::splat(0xfffff800); 29| 904| const auto v_ffff0000 = vector_u32::splat(0xffff0000); 30| 904| const auto one = vector_u32::splat(1); 31| 904|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 32| | 33| 904| size_t counter = 0; 34| | 35| | // 1. vectorized loop unrolled 4 times 36| 904| { 37| | // we use vector of uint32 counters, this is why this limit is used 38| 904| const size_t max_iterations = 39| 904| std::numeric_limits::max() / (max_increment * 4); 40| 904| size_t blocks = length / (N * 4); 41| 904| length -= blocks * (N * 4); 42| 1.37k| while (blocks != 0) { ------------------ | Branch (42:12): [True: 468, False: 904] ------------------ 43| 468| const size_t iterations = min(blocks, max_iterations); 44| 468| blocks -= iterations; 45| | 46| 468| simd32 acc = vector_u32::zero(); 47| 153k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (47:26): [True: 153k, False: 468] ------------------ 48| 153k| const auto in0 = vector_u32(input + 0 * N); 49| 153k| const auto in1 = vector_u32(input + 1 * N); 50| 153k| const auto in2 = vector_u32(input + 2 * N); 51| 153k| 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| 153k| acc += min(one, in0 & v_ffffff80); 70| 153k| acc += min(one, in1 & v_ffffff80); 71| 153k| acc += min(one, in2 & v_ffffff80); 72| 153k| acc += min(one, in3 & v_ffffff80); 73| | 74| 153k| acc += min(one, in0 & v_fffff800); 75| 153k| acc += min(one, in1 & v_fffff800); 76| 153k| acc += min(one, in2 & v_fffff800); 77| 153k| acc += min(one, in3 & v_fffff800); 78| | 79| 153k| acc += min(one, in0 & v_ffff0000); 80| 153k| acc += min(one, in1 & v_ffff0000); 81| 153k| acc += min(one, in2 & v_ffff0000); 82| 153k| acc += min(one, in3 & v_ffff0000); 83| 153k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 84| | 85| 153k| input += 4 * N; 86| 153k| } 87| | 88| 468| counter += acc.sum(); 89| 468| } 90| 904| } 91| | 92| | // 2. vectorized loop for tail 93| 904| { 94| 904| const size_t max_iterations = 95| 904| std::numeric_limits::max() / max_increment; 96| 904| size_t blocks = length / N; 97| 904| length -= blocks * N; 98| 1.39k| while (blocks != 0) { ------------------ | Branch (98:12): [True: 493, False: 904] ------------------ 99| 493| const size_t iterations = min(blocks, max_iterations); 100| 493| blocks -= iterations; 101| | 102| 493| auto acc = vector_u32::zero(); 103| 1.69k| for (size_t i = 0; i < iterations; i++) { ------------------ | Branch (103:26): [True: 1.20k, False: 493] ------------------ 104| 1.20k| const auto in = vector_u32(input); 105| | 106| |#if SIMDUTF_SIMD_HAS_UNSIGNED_CMP 107| | acc -= as_vector_u32(in > v_0000007f); 108| | acc -= as_vector_u32(in > v_000007ff); 109| | acc -= as_vector_u32(in > v_0000ffff); 110| |#else 111| 1.20k| acc += min(one, in & v_ffffff80); 112| 1.20k| acc += min(one, in & v_fffff800); 113| 1.20k| acc += min(one, in & v_ffff0000); 114| 1.20k|#endif // SIMDUTF_SIMD_HAS_UNSIGNED_CMP 115| | 116| 1.20k| input += N; 117| 1.20k| } 118| | 119| 493| counter += acc.sum(); 120| 493| } 121| 904| } 122| | 123| 904| const size_t consumed = input - start; 124| 904| if (consumed != 0) { ------------------ | Branch (124:7): [True: 537, False: 367] ------------------ 125| | // We don't count 0th bytes in the vectorized loops above, this 126| | // is why we need to count them in the end. 127| 537| counter += consumed; 128| 537| } 129| | 130| 904| return counter + scalar::utf32::utf8_length_from_utf32(input, length); 131| 904|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf323minImEET_S4_S4_: 8| 961|template T min(T a, T b) { return a <= b ? a : b; } ------------------ | Branch (8:48): [True: 961, False: 0] ------------------ simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14utf826count_code_points_bytemaskEPKcm: 21| 4.08k|count_code_points_bytemask(const char *in, size_t size) { 22| 4.08k| using vector_i8 = simd8; 23| 4.08k| using vector_u8 = simd8; 24| 4.08k| using vector_u64 = simd64; 25| | 26| 4.08k| constexpr size_t N = vector_i8::SIZE; 27| 4.08k| constexpr size_t max_iterations = 255 / 4; 28| | 29| 4.08k| size_t pos = 0; 30| 4.08k| size_t count = 0; 31| | 32| 4.08k| auto counters = vector_u64::zero(); 33| 4.08k| auto local = vector_u8::zero(); 34| 4.08k| size_t iterations = 0; 35| 1.69M| for (; pos + 4 * N <= size; pos += 4 * N) { ------------------ | Branch (35:10): [True: 1.68M, False: 4.08k] ------------------ 36| 1.68M| const auto input0 = 37| 1.68M| simd8::load(reinterpret_cast(in + pos + 0 * N)); 38| 1.68M| const auto input1 = 39| 1.68M| simd8::load(reinterpret_cast(in + pos + 1 * N)); 40| 1.68M| const auto input2 = 41| 1.68M| simd8::load(reinterpret_cast(in + pos + 2 * N)); 42| 1.68M| const auto input3 = 43| 1.68M| simd8::load(reinterpret_cast(in + pos + 3 * N)); 44| 1.68M| const auto mask0 = input0 > int8_t(-65); 45| 1.68M| const auto mask1 = input1 > int8_t(-65); 46| 1.68M| const auto mask2 = input2 > int8_t(-65); 47| 1.68M| const auto mask3 = input3 > int8_t(-65); 48| | 49| 1.68M| local -= vector_u8(mask0); 50| 1.68M| local -= vector_u8(mask1); 51| 1.68M| local -= vector_u8(mask2); 52| 1.68M| local -= vector_u8(mask3); 53| | 54| 1.68M| iterations += 1; 55| 1.68M| if (iterations == max_iterations) { ------------------ | Branch (55:9): [True: 26.5k, False: 1.66M] ------------------ 56| 26.5k| counters += sum_8bytes(local); 57| 26.5k| local = vector_u8::zero(); 58| 26.5k| iterations = 0; 59| 26.5k| } 60| 1.68M| } 61| | 62| 4.08k| if (iterations > 0) { ------------------ | Branch (62:7): [True: 1.53k, False: 2.55k] ------------------ 63| 1.53k| count += local.sum_bytes(); 64| 1.53k| } 65| | 66| 4.08k| count += counters.sum(); 67| | 68| 4.08k| return count + scalar::utf8::count_code_points(in + pos, size - pos); 69| 4.08k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14utf817count_code_pointsEPKcm: 8| 1.03k|simdutf_really_inline size_t count_code_points(const char *in, size_t size) { 9| 1.03k| size_t pos = 0; 10| 1.03k| size_t count = 0; 11| 780k| for (; pos + 64 <= size; pos += 64) { ------------------ | Branch (11:10): [True: 779k, False: 1.03k] ------------------ 12| 779k| simd8x64 input(reinterpret_cast(in + pos)); 13| 779k| uint64_t utf8_continuation_mask = input.gt(-65); 14| 779k| count += count_ones(utf8_continuation_mask); 15| 779k| } 16| 1.03k| return count + scalar::utf8::count_code_points(in + pos, size - pos); 17| 1.03k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14utf826count_code_points_bytemaskEPKcm: 21| 4.08k|count_code_points_bytemask(const char *in, size_t size) { 22| 4.08k| using vector_i8 = simd8; 23| 4.08k| using vector_u8 = simd8; 24| 4.08k| using vector_u64 = simd64; 25| | 26| 4.08k| constexpr size_t N = vector_i8::SIZE; 27| 4.08k| constexpr size_t max_iterations = 255 / 4; 28| | 29| 4.08k| size_t pos = 0; 30| 4.08k| size_t count = 0; 31| | 32| 4.08k| auto counters = vector_u64::zero(); 33| 4.08k| auto local = vector_u8::zero(); 34| 4.08k| size_t iterations = 0; 35| 3.37M| for (; pos + 4 * N <= size; pos += 4 * N) { ------------------ | Branch (35:10): [True: 3.37M, False: 4.08k] ------------------ 36| 3.37M| const auto input0 = 37| 3.37M| simd8::load(reinterpret_cast(in + pos + 0 * N)); 38| 3.37M| const auto input1 = 39| 3.37M| simd8::load(reinterpret_cast(in + pos + 1 * N)); 40| 3.37M| const auto input2 = 41| 3.37M| simd8::load(reinterpret_cast(in + pos + 2 * N)); 42| 3.37M| const auto input3 = 43| 3.37M| simd8::load(reinterpret_cast(in + pos + 3 * N)); 44| 3.37M| const auto mask0 = input0 > int8_t(-65); 45| 3.37M| const auto mask1 = input1 > int8_t(-65); 46| 3.37M| const auto mask2 = input2 > int8_t(-65); 47| 3.37M| const auto mask3 = input3 > int8_t(-65); 48| | 49| 3.37M| local -= vector_u8(mask0); 50| 3.37M| local -= vector_u8(mask1); 51| 3.37M| local -= vector_u8(mask2); 52| 3.37M| local -= vector_u8(mask3); 53| | 54| 3.37M| iterations += 1; 55| 3.37M| if (iterations == max_iterations) { ------------------ | Branch (55:9): [True: 53.2k, False: 3.32M] ------------------ 56| 53.2k| counters += sum_8bytes(local); 57| 53.2k| local = vector_u8::zero(); 58| 53.2k| iterations = 0; 59| 53.2k| } 60| 3.37M| } 61| | 62| 4.08k| if (iterations > 0) { ------------------ | Branch (62:7): [True: 2.45k, False: 1.63k] ------------------ 63| 2.45k| count += local.sum_bytes(); 64| 2.45k| } 65| | 66| 4.08k| count += counters.sum(); 67| | 68| 4.08k| return count + scalar::utf8::count_code_points(in + pos, size - pos); 69| 4.08k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14utf817count_code_pointsEPKcm: 8| 1.03k|simdutf_really_inline size_t count_code_points(const char *in, size_t size) { 9| 1.03k| size_t pos = 0; 10| 1.03k| size_t count = 0; 11| 780k| for (; pos + 64 <= size; pos += 64) { ------------------ | Branch (11:10): [True: 779k, False: 1.03k] ------------------ 12| 779k| simd8x64 input(reinterpret_cast(in + pos)); 13| 779k| uint64_t utf8_continuation_mask = input.gt(-65); 14| 779k| count += count_ones(utf8_continuation_mask); 15| 779k| } 16| 1.03k| return count + scalar::utf8::count_code_points(in + pos, size - pos); 17| 1.03k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14utf831utf16_length_from_utf8_bytemaskEPKcm: 9| 1.92k| size_t size) { 10| 1.92k| using vector_i8 = simd8; 11| 1.92k| using vector_u8 = simd8; 12| 1.92k| using vector_u64 = simd64; 13| | 14| 1.92k| constexpr size_t N = vector_i8::SIZE; 15| 1.92k| constexpr size_t max_iterations = 255 / 2; 16| | 17| 1.92k| auto counters = vector_u64::zero(); 18| 1.92k| auto local = vector_u8::zero(); 19| | 20| 1.92k| size_t iterations = 0; 21| 1.92k| size_t pos = 0; 22| 1.92k| size_t count = 0; 23| 2.75M| for (; pos + N <= size; pos += N) { ------------------ | Branch (23:10): [True: 2.75M, False: 1.92k] ------------------ 24| 2.75M| const auto input = 25| 2.75M| vector_i8::load(reinterpret_cast(in + pos)); 26| | 27| 2.75M| const auto continuation = input > int8_t(-65); 28| 2.75M| const auto utf_4bytes = vector_u8(input.value) >= uint8_t(240); 29| | 30| 2.75M| local -= vector_u8(continuation); 31| 2.75M| local -= vector_u8(utf_4bytes); 32| | 33| 2.75M| iterations += 1; 34| 2.75M| if (iterations == max_iterations) { ------------------ | Branch (34:9): [True: 21.5k, False: 2.73M] ------------------ 35| 21.5k| counters += sum_8bytes(local); 36| 21.5k| local = vector_u8::zero(); 37| 21.5k| iterations = 0; 38| 21.5k| } 39| 2.75M| } 40| | 41| 1.92k| if (iterations > 0) { ------------------ | Branch (41:7): [True: 1.23k, False: 694] ------------------ 42| 1.23k| count += local.sum_bytes(); 43| 1.23k| } 44| | 45| 1.92k| count += counters.sum(); 46| | 47| 1.92k| return count + scalar::utf8::utf16_length_from_utf8(in + pos, size - pos); 48| 1.92k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14utf831utf16_length_from_utf8_bytemaskEPKcm: 9| 1.92k| size_t size) { 10| 1.92k| using vector_i8 = simd8; 11| 1.92k| using vector_u8 = simd8; 12| 1.92k| using vector_u64 = simd64; 13| | 14| 1.92k| constexpr size_t N = vector_i8::SIZE; 15| 1.92k| constexpr size_t max_iterations = 255 / 2; 16| | 17| 1.92k| auto counters = vector_u64::zero(); 18| 1.92k| auto local = vector_u8::zero(); 19| | 20| 1.92k| size_t iterations = 0; 21| 1.92k| size_t pos = 0; 22| 1.92k| size_t count = 0; 23| 5.51M| for (; pos + N <= size; pos += N) { ------------------ | Branch (23:10): [True: 5.51M, False: 1.92k] ------------------ 24| 5.51M| const auto input = 25| 5.51M| vector_i8::load(reinterpret_cast(in + pos)); 26| | 27| 5.51M| const auto continuation = input > int8_t(-65); 28| 5.51M| const auto utf_4bytes = vector_u8(input.value) >= uint8_t(240); 29| | 30| 5.51M| local -= vector_u8(continuation); 31| 5.51M| local -= vector_u8(utf_4bytes); 32| | 33| 5.51M| iterations += 1; 34| 5.51M| if (iterations == max_iterations) { ------------------ | Branch (34:9): [True: 43.2k, False: 5.46M] ------------------ 35| 43.2k| counters += sum_8bytes(local); 36| 43.2k| local = vector_u8::zero(); 37| 43.2k| iterations = 0; 38| 43.2k| } 39| 5.51M| } 40| | 41| 1.92k| if (iterations > 0) { ------------------ | Branch (41:7): [True: 1.34k, False: 581] ------------------ 42| 1.34k| count += local.sum_bytes(); 43| 1.34k| } 44| | 45| 1.92k| count += counters.sum(); 46| | 47| 1.92k| return count + scalar::utf8::utf16_length_from_utf8(in + pos, size - pos); 48| 1.92k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoderC2Ev: 101| 1.13k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder7convertEPKcmPc: 115| 612| char *latin1_output) { 116| 612| size_t pos = 0; 117| 612| 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| 612| size_t leading_byte = 0; 125| 612| size_t margin = size; 126| 10.1k| for (; margin > 0 && leading_byte < 16; margin--) { ------------------ | Branch (126:12): [True: 10.0k, False: 144] | Branch (126:26): [True: 9.58k, False: 468] ------------------ 127| 9.58k| leading_byte += (int8_t(in[margin - 1]) > 128| 9.58k| -65); // twos complement of -65 is 1011 1111 ... 129| 9.58k| } 130| | // If the input is long enough, then we have that margin-1 is the eight last 131| | // leading byte. 132| 612| const size_t safety_margin = size - margin + 1; // to avoid overruns! 133| 692k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (133:12): [True: 691k, False: 580] ------------------ 134| 691k| simd8x64 input(reinterpret_cast(in + pos)); 135| 691k| if (input.is_ascii()) { ------------------ | Branch (135:11): [True: 635k, False: 55.6k] ------------------ 136| 635k| input.store((int8_t *)latin1_output); 137| 635k| latin1_output += 64; 138| 635k| pos += 64; 139| 635k| } else { 140| | // you might think that a for-loop would work, but under Visual Studio, 141| | // it is not good enough. 142| 55.6k| static_assert( 143| 55.6k| (simd8x64::NUM_CHUNKS == 2) || 144| 55.6k| (simd8x64::NUM_CHUNKS == 4), 145| 55.6k| "We support either two or four chunks per 64-byte block."); 146| 55.6k| auto zero = simd8{uint8_t(0)}; 147| 55.6k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 148| 55.6k| this->check_utf8_bytes(input.chunks[0], zero); 149| 55.6k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 150| | } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 151| | this->check_utf8_bytes(input.chunks[0], zero); 152| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 153| | this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 154| | this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 155| | } 156| 55.6k| uint64_t utf8_continuation_mask = 157| 55.6k| input.lt(-65 + 1); // -64 is 1100 0000 in twos complement. Note: in 158| | // this case, we also have ASCII to account for. 159| 55.6k| if (utf8_continuation_mask & 1) { ------------------ | Branch (159:13): [True: 32, False: 55.6k] ------------------ 160| 32| return 0; // error 161| 32| } 162| 55.6k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 163| 55.6k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 164| | // We process in blocks of up to 12 bytes except possibly 165| | // for fast paths which may process up to 16 bytes. For the 166| | // slow path to work, we should have at least 12 input bytes left. 167| 55.6k| size_t max_starting_point = (pos + 64) - 12; 168| | // Next loop is going to run at least five times. 169| 553k| while (pos < max_starting_point) { ------------------ | Branch (169:16): [True: 497k, False: 55.6k] ------------------ 170| | // Performance note: our ability to compute 'consumed' and 171| | // then shift and recompute is critical. If there is a 172| | // latency of, say, 4 cycles on getting 'consumed', then 173| | // the inner loop might have a total latency of about 6 cycles. 174| | // Yet we process between 6 to 12 inputs bytes, thus we get 175| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 176| | // for this section of the code. Hence, there is a limit 177| | // to how much we can further increase this latency before 178| | // it seriously harms performance. 179| 497k| size_t consumed = convert_masked_utf8_to_latin1( 180| 497k| in + pos, utf8_end_of_code_point_mask, latin1_output); 181| 497k| pos += consumed; 182| 497k| utf8_end_of_code_point_mask >>= consumed; 183| 497k| } 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| 55.6k| } 189| 691k| } 190| 580| if (errors()) { ------------------ | Branch (190:9): [True: 238, False: 342] ------------------ 191| 238| return 0; 192| 238| } 193| 342| if (pos < size) { ------------------ | Branch (193:9): [True: 338, False: 4] ------------------ 194| 338| size_t howmany = 195| 338| scalar::utf8_to_latin1::convert(in + pos, size - pos, latin1_output); 196| 338| if (howmany == 0) { ------------------ | Branch (196:11): [True: 154, False: 184] ------------------ 197| 154| return 0; 198| 154| } 199| 184| latin1_output += howmany; 200| 184| } 201| 188| return latin1_output - start; 202| 342| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 106| 164k| 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| 164k| simd8 prev1 = input.prev<1>(prev_input); 111| 164k| this->error |= check_special_cases(input, prev1); 112| 164k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin119check_special_casesENS1_4simd5simd8IhEES5_: 8| 164k|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| 164k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 20| | // 11______ 11______ 21| 164k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 22| 164k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 23| 164k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 24| 164k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 25| 164k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 26| 164k| 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| 164k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 35| | // 11110101 1000____ 36| | // 1111011_ 1000____ 37| | // 11111___ 1000____ 38| 164k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 39| 164k| constexpr const uint8_t FORBIDDEN = 0xff; 40| | 41| 164k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 42| | // 0_______ ________ 43| 164k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 44| 164k| TOO_LONG, 45| | // 10______ ________ 46| 164k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 47| | // 1100____ ________ 48| 164k| TOO_SHORT | OVERLONG_2, 49| | // 1101____ ________ 50| 164k| FORBIDDEN, 51| | // 1110____ ________ 52| 164k| FORBIDDEN, 53| | // 1111____ ________ 54| 164k| FORBIDDEN); 55| 164k| constexpr const uint8_t CARRY = 56| 164k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 57| 164k| const simd8 byte_1_low = 58| 164k| (prev1 & 0x0F) 59| 164k| .lookup_16( 60| | // ____0000 ________ 61| 164k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 62| | // ____0001 ________ 63| 164k| CARRY | OVERLONG_2, 64| | // ____001_ ________ 65| 164k| CARRY, CARRY, 66| | 67| | // ____0100 ________ 68| 164k| FORBIDDEN, 69| | // ____0101 ________ 70| 164k| FORBIDDEN, 71| | // ____011_ ________ 72| 164k| FORBIDDEN, FORBIDDEN, 73| | 74| | // ____1___ ________ 75| 164k| FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, 76| | // ____1101 ________ 77| 164k| FORBIDDEN, FORBIDDEN, FORBIDDEN); 78| 164k| const simd8 byte_2_high = input.shr<4>().lookup_16( 79| | // ________ 0_______ 80| 164k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 81| 164k| TOO_SHORT, TOO_SHORT, 82| | 83| | // ________ 1000____ 84| 164k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 85| 164k| OVERLONG_4, 86| | // ________ 1001____ 87| 164k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 88| | // ________ 101_____ 89| 164k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 90| 164k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 91| | 92| | // ________ 11______ 93| 164k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 94| 164k| return (byte_1_high & byte_1_low & byte_2_high); 95| 164k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder6errorsEv: 308| 27.6k| simdutf_really_inline bool errors() const { 309| 27.6k| return this->error.any_bits_set_anywhere(); 310| 27.6k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_114utf8_to_latin121validating_transcoder19convert_with_errorsEPKcmPc: 205| 522| char *latin1_output) { 206| 522| size_t pos = 0; 207| 522| 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| 522| size_t leading_byte = 0; 215| 522| size_t margin = size; 216| 5.08k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (216:12): [True: 4.93k, False: 144] | Branch (216:26): [True: 4.56k, False: 378] ------------------ 217| 4.56k| leading_byte += (int8_t(in[margin - 1]) > -65); 218| 4.56k| } 219| | // If the input is long enough, then we have that margin-1 is the eight last 220| | // leading byte. 221| 522| const size_t safety_margin = size - margin + 1; // to avoid overruns! 222| 355k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (222:12): [True: 354k, False: 444] ------------------ 223| 354k| simd8x64 input(reinterpret_cast(in + pos)); 224| 354k| if (input.is_ascii()) { ------------------ | Branch (224:11): [True: 327k, False: 26.6k] ------------------ 225| 327k| input.store((int8_t *)latin1_output); 226| 327k| latin1_output += 64; 227| 327k| pos += 64; 228| 327k| } else { 229| | // you might think that a for-loop would work, but under Visual Studio, 230| | // it is not good enough. 231| 26.6k| static_assert( 232| 26.6k| (simd8x64::NUM_CHUNKS == 2) || 233| 26.6k| (simd8x64::NUM_CHUNKS == 4), 234| 26.6k| "We support either two or four chunks per 64-byte block."); 235| 26.6k| auto zero = simd8{uint8_t(0)}; 236| 26.6k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 237| 26.6k| this->check_utf8_bytes(input.chunks[0], zero); 238| 26.6k| 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| 26.6k| if (errors()) { ------------------ | Branch (245:13): [True: 78, False: 26.5k] ------------------ 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| 78| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 250| 78| pos, in + pos, size - pos, latin1_output); 251| 78| res.count += pos; 252| 78| return res; 253| 78| } 254| 26.5k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 255| 26.5k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 256| 26.5k| 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| 26.5k| size_t max_starting_point = (pos + 64) - 12; 261| | // Next loop is going to run at least five times. 262| 265k| while (pos < max_starting_point) { ------------------ | Branch (262:16): [True: 238k, False: 26.5k] ------------------ 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| 238k| size_t consumed = convert_masked_utf8_to_latin1( 273| 238k| in + pos, utf8_end_of_code_point_mask, latin1_output); 274| 238k| pos += consumed; 275| 238k| utf8_end_of_code_point_mask >>= consumed; 276| 238k| } 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| 26.5k| } 282| 354k| } 283| 444| if (errors()) { ------------------ | Branch (283:9): [True: 0, False: 444] ------------------ 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| 444| if (pos < size) { ------------------ | Branch (292:9): [True: 440, 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| 440| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 297| 440| pos, in + pos, size - pos, latin1_output); 298| 440| if (res.error) { // In case of error, we want the error position ------------------ | Branch (298:11): [True: 234, False: 206] ------------------ 299| 234| res.count += pos; 300| 234| return res; 301| 234| } else { // In case of success, we want the number of word written 302| 206| latin1_output += res.count; 303| 206| } 304| 440| } 305| 210| return result(error_code::SUCCESS, latin1_output - start); 306| 444| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoderC2Ev: 101| 1.13k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder7convertEPKcmPc: 115| 612| char *latin1_output) { 116| 612| size_t pos = 0; 117| 612| 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| 612| size_t leading_byte = 0; 125| 612| size_t margin = size; 126| 10.1k| for (; margin > 0 && leading_byte < 16; margin--) { ------------------ | Branch (126:12): [True: 10.0k, False: 144] | Branch (126:26): [True: 9.58k, False: 468] ------------------ 127| 9.58k| leading_byte += (int8_t(in[margin - 1]) > 128| 9.58k| -65); // twos complement of -65 is 1011 1111 ... 129| 9.58k| } 130| | // If the input is long enough, then we have that margin-1 is the eight last 131| | // leading byte. 132| 612| const size_t safety_margin = size - margin + 1; // to avoid overruns! 133| 692k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (133:12): [True: 691k, False: 580] ------------------ 134| 691k| simd8x64 input(reinterpret_cast(in + pos)); 135| 691k| if (input.is_ascii()) { ------------------ | Branch (135:11): [True: 635k, False: 55.6k] ------------------ 136| 635k| input.store((int8_t *)latin1_output); 137| 635k| latin1_output += 64; 138| 635k| pos += 64; 139| 635k| } else { 140| | // you might think that a for-loop would work, but under Visual Studio, 141| | // it is not good enough. 142| 55.6k| static_assert( 143| 55.6k| (simd8x64::NUM_CHUNKS == 2) || 144| 55.6k| (simd8x64::NUM_CHUNKS == 4), 145| 55.6k| "We support either two or four chunks per 64-byte block."); 146| 55.6k| auto zero = simd8{uint8_t(0)}; 147| | if constexpr (simd8x64::NUM_CHUNKS == 2) { 148| | this->check_utf8_bytes(input.chunks[0], zero); 149| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 150| 55.6k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 151| 55.6k| this->check_utf8_bytes(input.chunks[0], zero); 152| 55.6k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 153| 55.6k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 154| 55.6k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 155| 55.6k| } 156| 55.6k| uint64_t utf8_continuation_mask = 157| 55.6k| input.lt(-65 + 1); // -64 is 1100 0000 in twos complement. Note: in 158| | // this case, we also have ASCII to account for. 159| 55.6k| if (utf8_continuation_mask & 1) { ------------------ | Branch (159:13): [True: 32, False: 55.6k] ------------------ 160| 32| return 0; // error 161| 32| } 162| 55.6k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 163| 55.6k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 164| | // We process in blocks of up to 12 bytes except possibly 165| | // for fast paths which may process up to 16 bytes. For the 166| | // slow path to work, we should have at least 12 input bytes left. 167| 55.6k| size_t max_starting_point = (pos + 64) - 12; 168| | // Next loop is going to run at least five times. 169| 553k| while (pos < max_starting_point) { ------------------ | Branch (169:16): [True: 497k, False: 55.6k] ------------------ 170| | // Performance note: our ability to compute 'consumed' and 171| | // then shift and recompute is critical. If there is a 172| | // latency of, say, 4 cycles on getting 'consumed', then 173| | // the inner loop might have a total latency of about 6 cycles. 174| | // Yet we process between 6 to 12 inputs bytes, thus we get 175| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 176| | // for this section of the code. Hence, there is a limit 177| | // to how much we can further increase this latency before 178| | // it seriously harms performance. 179| 497k| size_t consumed = convert_masked_utf8_to_latin1( 180| 497k| in + pos, utf8_end_of_code_point_mask, latin1_output); 181| 497k| pos += consumed; 182| 497k| utf8_end_of_code_point_mask >>= consumed; 183| 497k| } 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| 55.6k| } 189| 691k| } 190| 580| if (errors()) { ------------------ | Branch (190:9): [True: 238, False: 342] ------------------ 191| 238| return 0; 192| 238| } 193| 342| if (pos < size) { ------------------ | Branch (193:9): [True: 338, False: 4] ------------------ 194| 338| size_t howmany = 195| 338| scalar::utf8_to_latin1::convert(in + pos, size - pos, latin1_output); 196| 338| if (howmany == 0) { ------------------ | Branch (196:11): [True: 154, False: 184] ------------------ 197| 154| return 0; 198| 154| } 199| 184| latin1_output += howmany; 200| 184| } 201| 188| return latin1_output - start; 202| 342| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 106| 329k| 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| 329k| simd8 prev1 = input.prev<1>(prev_input); 111| 329k| this->error |= check_special_cases(input, prev1); 112| 329k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin119check_special_casesENS1_4simd5simd8IhEES5_: 8| 329k|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| 329k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 20| | // 11______ 11______ 21| 329k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 22| 329k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 23| 329k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 24| 329k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 25| 329k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 26| 329k| 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| 329k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 35| | // 11110101 1000____ 36| | // 1111011_ 1000____ 37| | // 11111___ 1000____ 38| 329k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 39| 329k| constexpr const uint8_t FORBIDDEN = 0xff; 40| | 41| 329k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 42| | // 0_______ ________ 43| 329k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 44| 329k| TOO_LONG, 45| | // 10______ ________ 46| 329k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 47| | // 1100____ ________ 48| 329k| TOO_SHORT | OVERLONG_2, 49| | // 1101____ ________ 50| 329k| FORBIDDEN, 51| | // 1110____ ________ 52| 329k| FORBIDDEN, 53| | // 1111____ ________ 54| 329k| FORBIDDEN); 55| 329k| constexpr const uint8_t CARRY = 56| 329k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 57| 329k| const simd8 byte_1_low = 58| 329k| (prev1 & 0x0F) 59| 329k| .lookup_16( 60| | // ____0000 ________ 61| 329k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 62| | // ____0001 ________ 63| 329k| CARRY | OVERLONG_2, 64| | // ____001_ ________ 65| 329k| CARRY, CARRY, 66| | 67| | // ____0100 ________ 68| 329k| FORBIDDEN, 69| | // ____0101 ________ 70| 329k| FORBIDDEN, 71| | // ____011_ ________ 72| 329k| FORBIDDEN, FORBIDDEN, 73| | 74| | // ____1___ ________ 75| 329k| FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, FORBIDDEN, 76| | // ____1101 ________ 77| 329k| FORBIDDEN, FORBIDDEN, FORBIDDEN); 78| 329k| const simd8 byte_2_high = input.shr<4>().lookup_16( 79| | // ________ 0_______ 80| 329k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 81| 329k| TOO_SHORT, TOO_SHORT, 82| | 83| | // ________ 1000____ 84| 329k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 85| 329k| OVERLONG_4, 86| | // ________ 1001____ 87| 329k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 88| | // ________ 101_____ 89| 329k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 90| 329k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 91| | 92| | // ________ 11______ 93| 329k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 94| 329k| return (byte_1_high & byte_1_low & byte_2_high); 95| 329k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder6errorsEv: 308| 27.6k| simdutf_really_inline bool errors() const { 309| 27.6k| return this->error.any_bits_set_anywhere(); 310| 27.6k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114utf8_to_latin121validating_transcoder19convert_with_errorsEPKcmPc: 205| 522| char *latin1_output) { 206| 522| size_t pos = 0; 207| 522| 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| 522| size_t leading_byte = 0; 215| 522| size_t margin = size; 216| 5.08k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (216:12): [True: 4.93k, False: 144] | Branch (216:26): [True: 4.56k, False: 378] ------------------ 217| 4.56k| leading_byte += (int8_t(in[margin - 1]) > -65); 218| 4.56k| } 219| | // If the input is long enough, then we have that margin-1 is the eight last 220| | // leading byte. 221| 522| const size_t safety_margin = size - margin + 1; // to avoid overruns! 222| 355k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (222:12): [True: 354k, False: 444] ------------------ 223| 354k| simd8x64 input(reinterpret_cast(in + pos)); 224| 354k| if (input.is_ascii()) { ------------------ | Branch (224:11): [True: 327k, False: 26.6k] ------------------ 225| 327k| input.store((int8_t *)latin1_output); 226| 327k| latin1_output += 64; 227| 327k| pos += 64; 228| 327k| } else { 229| | // you might think that a for-loop would work, but under Visual Studio, 230| | // it is not good enough. 231| 26.6k| static_assert( 232| 26.6k| (simd8x64::NUM_CHUNKS == 2) || 233| 26.6k| (simd8x64::NUM_CHUNKS == 4), 234| 26.6k| "We support either two or four chunks per 64-byte block."); 235| 26.6k| 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| 26.6k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 240| 26.6k| this->check_utf8_bytes(input.chunks[0], zero); 241| 26.6k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 242| 26.6k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 243| 26.6k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 244| 26.6k| } 245| 26.6k| if (errors()) { ------------------ | Branch (245:13): [True: 78, False: 26.5k] ------------------ 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| 78| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 250| 78| pos, in + pos, size - pos, latin1_output); 251| 78| res.count += pos; 252| 78| return res; 253| 78| } 254| 26.5k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 255| 26.5k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 256| 26.5k| 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| 26.5k| size_t max_starting_point = (pos + 64) - 12; 261| | // Next loop is going to run at least five times. 262| 265k| while (pos < max_starting_point) { ------------------ | Branch (262:16): [True: 238k, False: 26.5k] ------------------ 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| 238k| size_t consumed = convert_masked_utf8_to_latin1( 273| 238k| in + pos, utf8_end_of_code_point_mask, latin1_output); 274| 238k| pos += consumed; 275| 238k| utf8_end_of_code_point_mask >>= consumed; 276| 238k| } 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| 26.5k| } 282| 354k| } 283| 444| if (errors()) { ------------------ | Branch (283:9): [True: 0, False: 444] ------------------ 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| 444| if (pos < size) { ------------------ | Branch (292:9): [True: 440, 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| 440| result res = scalar::utf8_to_latin1::rewind_and_convert_with_errors( 297| 440| pos, in + pos, size - pos, latin1_output); 298| 440| if (res.error) { // In case of error, we want the error position ------------------ | Branch (298:11): [True: 234, False: 206] ------------------ 299| 234| res.count += pos; 300| 234| return res; 301| 234| } else { // In case of success, we want the number of word written 302| 206| latin1_output += res.count; 303| 206| } 304| 440| } 305| 210| return result(error_code::SUCCESS, latin1_output - start); 306| 444| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoderC2Ev: 114| 2.90k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoderC2Ev: 114| 2.90k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE0EEEmPKcmPDs: 130| 784| char16_t *utf16_output) { 131| 784| size_t pos = 0; 132| 784| 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| 784| size_t leading_byte = 0; 140| 784| size_t margin = size; 141| 7.74k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 7.49k, False: 250] | Branch (141:26): [True: 6.95k, False: 534] ------------------ 142| 6.95k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 6.95k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 784| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 577k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 576k, False: 758] ------------------ 148| 576k| simd8x64 input(reinterpret_cast(in + pos)); 149| 576k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 478k, False: 97.8k] ------------------ 150| 478k| input.store_ascii_as_utf16(utf16_output); 151| 478k| utf16_output += 64; 152| 478k| pos += 64; 153| 478k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 97.8k| static_assert( 157| 97.8k| (simd8x64::NUM_CHUNKS == 2) || 158| 97.8k| (simd8x64::NUM_CHUNKS == 4), 159| 97.8k| "We support either two or four chunks per 64-byte block."); 160| 97.8k| auto zero = simd8{uint8_t(0)}; 161| 97.8k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 162| 97.8k| this->check_utf8_bytes(input.chunks[0], zero); 163| 97.8k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 164| | } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 165| | this->check_utf8_bytes(input.chunks[0], zero); 166| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 167| | this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 168| | this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 169| | } 170| 97.8k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 97.8k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 26, False: 97.7k] ------------------ 172| 26| return 0; // error 173| 26| } 174| 97.7k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 97.7k| 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| 97.7k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 962k| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 864k, False: 97.7k] ------------------ 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| 864k| size_t consumed = convert_masked_utf8_to_utf16( 192| 864k| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 864k| pos += consumed; 194| 864k| utf8_end_of_code_point_mask >>= consumed; 195| 864k| } 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| 97.7k| } 201| 576k| } 202| 758| if (errors()) { ------------------ | Branch (202:9): [True: 296, False: 462] ------------------ 203| 296| return 0; 204| 296| } 205| 462| if (pos < size) { ------------------ | Branch (205:9): [True: 460, False: 2] ------------------ 206| 460| size_t howmany = scalar::utf8_to_utf16::convert( 207| 460| in + pos, size - pos, utf16_output); 208| 460| if (howmany == 0) { ------------------ | Branch (208:11): [True: 214, False: 246] ------------------ 209| 214| return 0; 210| 214| } 211| 246| utf16_output += howmany; 212| 246| } 213| 248| return utf16_output - start; 214| 462| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 894k| 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| 894k| simd8 prev1 = input.prev<1>(prev_input); 124| 894k| simd8 sc = check_special_cases(input, prev1); 125| 894k| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 894k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1619check_special_casesENS1_4simd5simd8IhEES5_: 8| 894k|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| 894k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 894k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 894k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 894k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 894k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 894k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 894k| 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| 894k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 894k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 894k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 894k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 894k| TOO_LONG, 40| | // 10______ ________ 41| 894k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 894k| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 894k| TOO_SHORT, 46| | // 1110____ ________ 47| 894k| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 894k| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 894k| constexpr const uint8_t CARRY = 51| 894k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 894k| const simd8 byte_1_low = 53| 894k| (prev1 & 0x0F) 54| 894k| .lookup_16( 55| | // ____0000 ________ 56| 894k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 894k| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 894k| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 894k| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 894k| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 894k| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 894k| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 894k| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 894k| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 894k| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 894k| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 894k| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 894k| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 894k| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 894k| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 894k| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 894k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 894k| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 894k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 894k| OVERLONG_4, 88| | // ________ 1001____ 89| 894k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 894k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 894k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 894k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 894k| return (byte_1_high & byte_1_low & byte_2_high); 97| 894k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1623check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 894k| const simd8 sc) { 102| 894k| simd8 prev2 = input.prev<2>(prev_input); 103| 894k| simd8 prev3 = input.prev<3>(prev_input); 104| 894k| simd8 must23 = 105| 894k| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 894k| simd8 must23_80 = must23 & uint8_t(0x80); 107| 894k| return must23_80 ^ sc; 108| 894k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder6errorsEv: 324| 198k| simdutf_really_inline bool errors() const { 325| 198k| return this->error.any_bits_set_anywhere(); 326| 198k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE1EEEmPKcmPDs: 130| 758| char16_t *utf16_output) { 131| 758| size_t pos = 0; 132| 758| 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| 758| size_t leading_byte = 0; 140| 758| size_t margin = size; 141| 6.83k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 6.59k, False: 234] | Branch (141:26): [True: 6.07k, False: 524] ------------------ 142| 6.07k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 6.07k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 758| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 574k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 573k, False: 738] ------------------ 148| 573k| simd8x64 input(reinterpret_cast(in + pos)); 149| 573k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 419k, False: 153k] ------------------ 150| 419k| input.store_ascii_as_utf16(utf16_output); 151| 419k| utf16_output += 64; 152| 419k| pos += 64; 153| 419k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 153k| static_assert( 157| 153k| (simd8x64::NUM_CHUNKS == 2) || 158| 153k| (simd8x64::NUM_CHUNKS == 4), 159| 153k| "We support either two or four chunks per 64-byte block."); 160| 153k| auto zero = simd8{uint8_t(0)}; 161| 153k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 162| 153k| this->check_utf8_bytes(input.chunks[0], zero); 163| 153k| 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| 153k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 153k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 20, False: 153k] ------------------ 172| 20| return 0; // error 173| 20| } 174| 153k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 153k| 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| 153k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 1.52M| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 1.37M, False: 153k] ------------------ 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| 1.37M| size_t consumed = convert_masked_utf8_to_utf16( 192| 1.37M| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 1.37M| pos += consumed; 194| 1.37M| utf8_end_of_code_point_mask >>= consumed; 195| 1.37M| } 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| 153k| } 201| 573k| } 202| 738| if (errors()) { ------------------ | Branch (202:9): [True: 316, False: 422] ------------------ 203| 316| return 0; 204| 316| } 205| 422| if (pos < size) { ------------------ | Branch (205:9): [True: 420, False: 2] ------------------ 206| 420| size_t howmany = scalar::utf8_to_utf16::convert( 207| 420| in + pos, size - pos, utf16_output); 208| 420| if (howmany == 0) { ------------------ | Branch (208:11): [True: 218, False: 202] ------------------ 209| 218| return 0; 210| 218| } 211| 202| utf16_output += howmany; 212| 202| } 213| 204| return utf16_output - start; 214| 422| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKcmPDs: 218| 594| char16_t *utf16_output) { 219| 594| size_t pos = 0; 220| 594| 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| 594| size_t leading_byte = 0; 228| 594| size_t margin = size; 229| 5.04k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 4.81k, False: 230] | Branch (229:26): [True: 4.44k, False: 364] ------------------ 230| 4.44k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 4.44k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 594| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 387k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 387k, False: 512] ------------------ 236| 387k| simd8x64 input(reinterpret_cast(in + pos)); 237| 387k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 294k, False: 92.8k] ------------------ 238| 294k| input.store_ascii_as_utf16(utf16_output); 239| 294k| utf16_output += 64; 240| 294k| pos += 64; 241| 294k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 92.8k| static_assert( 245| 92.8k| (simd8x64::NUM_CHUNKS == 2) || 246| 92.8k| (simd8x64::NUM_CHUNKS == 4), 247| 92.8k| "We support either two or four chunks per 64-byte block."); 248| 92.8k| auto zero = simd8{uint8_t(0)}; 249| 92.8k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 250| 92.8k| this->check_utf8_bytes(input.chunks[0], zero); 251| 92.8k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 252| | } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 253| | this->check_utf8_bytes(input.chunks[0], zero); 254| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 255| | this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 256| | this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 257| | } 258| 92.8k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 92.8k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 82, False: 92.7k] | Branch (259:25): [True: 0, False: 92.7k] ------------------ 260| | // rewind_and_convert_with_errors will seek a potential error from 261| | // in+pos onward, with the ability to go back up to pos bytes, and 262| | // read size-pos bytes forward. 263| 82| result res = 264| 82| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 82| pos, in + pos, size - pos, utf16_output); 266| 82| res.count += pos; 267| 82| return res; 268| 82| } 269| 92.7k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 92.7k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 271| | // We process in blocks of up to 12 bytes except possibly 272| | // for fast paths which may process up to 16 bytes. For the 273| | // slow path to work, we should have at least 12 input bytes left. 274| 92.7k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 911k| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 819k, False: 92.7k] ------------------ 277| | // Performance note: our ability to compute 'consumed' and 278| | // then shift and recompute is critical. If there is a 279| | // latency of, say, 4 cycles on getting 'consumed', then 280| | // the inner loop might have a total latency of about 6 cycles. 281| | // Yet we process between 6 to 12 inputs bytes, thus we get 282| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 283| | // for this section of the code. Hence, there is a limit 284| | // to how much we can further increase this latency before 285| | // it seriously harms performance. 286| 819k| size_t consumed = convert_masked_utf8_to_utf16( 287| 819k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 819k| pos += consumed; 289| 819k| utf8_end_of_code_point_mask >>= consumed; 290| 819k| } 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| 92.7k| } 296| 387k| } 297| 512| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 512] ------------------ 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| 512| if (pos < size) { ------------------ | Branch (307:9): [True: 508, 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| 508| result res = 312| 508| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 508| pos, in + pos, size - pos, utf16_output); 314| 508| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 258, False: 250] ------------------ 315| 258| res.count += pos; 316| 258| return res; 317| 258| } else { // In case of success, we want the number of word written 318| 250| utf16_output += res.count; 319| 250| } 320| 508| } 321| 254| return result(error_code::SUCCESS, utf16_output - start); 322| 512| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKcmPDs: 218| 772| char16_t *utf16_output) { 219| 772| size_t pos = 0; 220| 772| 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| 772| size_t leading_byte = 0; 228| 772| size_t margin = size; 229| 6.85k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 6.58k, False: 264] | Branch (229:26): [True: 6.08k, False: 508] ------------------ 230| 6.08k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 6.08k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 772| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 462k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 462k, False: 706] ------------------ 236| 462k| simd8x64 input(reinterpret_cast(in + pos)); 237| 462k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 358k, False: 103k] ------------------ 238| 358k| input.store_ascii_as_utf16(utf16_output); 239| 358k| utf16_output += 64; 240| 358k| pos += 64; 241| 358k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 103k| static_assert( 245| 103k| (simd8x64::NUM_CHUNKS == 2) || 246| 103k| (simd8x64::NUM_CHUNKS == 4), 247| 103k| "We support either two or four chunks per 64-byte block."); 248| 103k| auto zero = simd8{uint8_t(0)}; 249| 103k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 250| 103k| this->check_utf8_bytes(input.chunks[0], zero); 251| 103k| 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| 103k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 103k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 66, False: 103k] | Branch (259:25): [True: 0, False: 103k] ------------------ 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| 66| result res = 264| 66| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 66| pos, in + pos, size - pos, utf16_output); 266| 66| res.count += pos; 267| 66| return res; 268| 66| } 269| 103k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 103k| 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| 103k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 1.02M| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 917k, False: 103k] ------------------ 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| 917k| size_t consumed = convert_masked_utf8_to_utf16( 287| 917k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 917k| pos += consumed; 289| 917k| utf8_end_of_code_point_mask >>= consumed; 290| 917k| } 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| 103k| } 296| 462k| } 297| 706| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 706] ------------------ 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| 706| if (pos < size) { ------------------ | Branch (307:9): [True: 702, 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| 702| result res = 312| 702| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 702| pos, in + pos, size - pos, utf16_output); 314| 702| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 356, False: 346] ------------------ 315| 356| res.count += pos; 316| 356| return res; 317| 356| } else { // In case of success, we want the number of word written 318| 346| utf16_output += res.count; 319| 346| } 320| 702| } 321| 350| return result(error_code::SUCCESS, utf16_output - start); 322| 706| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE0EEEmPKcmPDs: 130| 784| char16_t *utf16_output) { 131| 784| size_t pos = 0; 132| 784| 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| 784| size_t leading_byte = 0; 140| 784| size_t margin = size; 141| 7.74k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 7.49k, False: 250] | Branch (141:26): [True: 6.95k, False: 534] ------------------ 142| 6.95k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 6.95k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 784| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 577k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 576k, False: 758] ------------------ 148| 576k| simd8x64 input(reinterpret_cast(in + pos)); 149| 576k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 478k, False: 97.8k] ------------------ 150| 478k| input.store_ascii_as_utf16(utf16_output); 151| 478k| utf16_output += 64; 152| 478k| pos += 64; 153| 478k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 97.8k| static_assert( 157| 97.8k| (simd8x64::NUM_CHUNKS == 2) || 158| 97.8k| (simd8x64::NUM_CHUNKS == 4), 159| 97.8k| "We support either two or four chunks per 64-byte block."); 160| 97.8k| auto zero = simd8{uint8_t(0)}; 161| | if constexpr (simd8x64::NUM_CHUNKS == 2) { 162| | this->check_utf8_bytes(input.chunks[0], zero); 163| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 164| 97.8k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 165| 97.8k| this->check_utf8_bytes(input.chunks[0], zero); 166| 97.8k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 167| 97.8k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 168| 97.8k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 169| 97.8k| } 170| 97.8k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 97.8k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 26, False: 97.7k] ------------------ 172| 26| return 0; // error 173| 26| } 174| 97.7k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 97.7k| 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| 97.7k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 962k| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 864k, False: 97.7k] ------------------ 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| 864k| size_t consumed = convert_masked_utf8_to_utf16( 192| 864k| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 864k| pos += consumed; 194| 864k| utf8_end_of_code_point_mask >>= consumed; 195| 864k| } 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| 97.7k| } 201| 576k| } 202| 758| if (errors()) { ------------------ | Branch (202:9): [True: 296, False: 462] ------------------ 203| 296| return 0; 204| 296| } 205| 462| if (pos < size) { ------------------ | Branch (205:9): [True: 460, False: 2] ------------------ 206| 460| size_t howmany = scalar::utf8_to_utf16::convert( 207| 460| in + pos, size - pos, utf16_output); 208| 460| if (howmany == 0) { ------------------ | Branch (208:11): [True: 214, False: 246] ------------------ 209| 214| return 0; 210| 214| } 211| 246| utf16_output += howmany; 212| 246| } 213| 248| return utf16_output - start; 214| 462| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 1.78M| 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| 1.78M| simd8 prev1 = input.prev<1>(prev_input); 124| 1.78M| simd8 sc = check_special_cases(input, prev1); 125| 1.78M| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 1.78M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1619check_special_casesENS1_4simd5simd8IhEES5_: 8| 1.78M|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| 1.78M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 1.78M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 1.78M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 1.78M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 1.78M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 1.78M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 1.78M| 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| 1.78M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 1.78M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 1.78M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 1.78M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 1.78M| TOO_LONG, 40| | // 10______ ________ 41| 1.78M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 1.78M| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 1.78M| TOO_SHORT, 46| | // 1110____ ________ 47| 1.78M| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 1.78M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 1.78M| constexpr const uint8_t CARRY = 51| 1.78M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 1.78M| const simd8 byte_1_low = 53| 1.78M| (prev1 & 0x0F) 54| 1.78M| .lookup_16( 55| | // ____0000 ________ 56| 1.78M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 1.78M| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 1.78M| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 1.78M| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 1.78M| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 1.78M| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 1.78M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 1.78M| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 1.78M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 1.78M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 1.78M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 1.78M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 1.78M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 1.78M| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 1.78M| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 1.78M| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 1.78M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 1.78M| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 1.78M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 1.78M| OVERLONG_4, 88| | // ________ 1001____ 89| 1.78M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 1.78M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 1.78M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 1.78M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 1.78M| return (byte_1_high & byte_1_low & byte_2_high); 97| 1.78M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1623check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 1.78M| const simd8 sc) { 102| 1.78M| simd8 prev2 = input.prev<2>(prev_input); 103| 1.78M| simd8 prev3 = input.prev<3>(prev_input); 104| 1.78M| simd8 must23 = 105| 1.78M| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 1.78M| simd8 must23_80 = must23 & uint8_t(0x80); 107| 1.78M| return must23_80 ^ sc; 108| 1.78M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder6errorsEv: 324| 198k| simdutf_really_inline bool errors() const { 325| 198k| return this->error.any_bits_set_anywhere(); 326| 198k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder7convertILNS_10endiannessE1EEEmPKcmPDs: 130| 758| char16_t *utf16_output) { 131| 758| size_t pos = 0; 132| 758| 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| 758| size_t leading_byte = 0; 140| 758| size_t margin = size; 141| 6.83k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (141:12): [True: 6.59k, False: 234] | Branch (141:26): [True: 6.07k, False: 524] ------------------ 142| 6.07k| leading_byte += (int8_t(in[margin - 1]) > -65); 143| 6.07k| } 144| | // If the input is long enough, then we have that margin-1 is the eight last 145| | // leading byte. 146| 758| const size_t safety_margin = size - margin + 1; // to avoid overruns! 147| 574k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (147:12): [True: 573k, False: 738] ------------------ 148| 573k| simd8x64 input(reinterpret_cast(in + pos)); 149| 573k| if (input.is_ascii()) { ------------------ | Branch (149:11): [True: 419k, False: 153k] ------------------ 150| 419k| input.store_ascii_as_utf16(utf16_output); 151| 419k| utf16_output += 64; 152| 419k| pos += 64; 153| 419k| } else { 154| | // you might think that a for-loop would work, but under Visual Studio, 155| | // it is not good enough. 156| 153k| static_assert( 157| 153k| (simd8x64::NUM_CHUNKS == 2) || 158| 153k| (simd8x64::NUM_CHUNKS == 4), 159| 153k| "We support either two or four chunks per 64-byte block."); 160| 153k| 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| 153k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 165| 153k| this->check_utf8_bytes(input.chunks[0], zero); 166| 153k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 167| 153k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 168| 153k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 169| 153k| } 170| 153k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 171| 153k| if (utf8_continuation_mask & 1) { ------------------ | Branch (171:13): [True: 20, False: 153k] ------------------ 172| 20| return 0; // error 173| 20| } 174| 153k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 175| 153k| 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| 153k| size_t max_starting_point = (pos + 64) - 12; 180| | // Next loop is going to run at least five times. 181| 1.52M| while (pos < max_starting_point) { ------------------ | Branch (181:16): [True: 1.37M, False: 153k] ------------------ 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| 1.37M| size_t consumed = convert_masked_utf8_to_utf16( 192| 1.37M| in + pos, utf8_end_of_code_point_mask, utf16_output); 193| 1.37M| pos += consumed; 194| 1.37M| utf8_end_of_code_point_mask >>= consumed; 195| 1.37M| } 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| 153k| } 201| 573k| } 202| 738| if (errors()) { ------------------ | Branch (202:9): [True: 316, False: 422] ------------------ 203| 316| return 0; 204| 316| } 205| 422| if (pos < size) { ------------------ | Branch (205:9): [True: 420, False: 2] ------------------ 206| 420| size_t howmany = scalar::utf8_to_utf16::convert( 207| 420| in + pos, size - pos, utf16_output); 208| 420| if (howmany == 0) { ------------------ | Branch (208:11): [True: 218, False: 202] ------------------ 209| 218| return 0; 210| 218| } 211| 202| utf16_output += howmany; 212| 202| } 213| 204| return utf16_output - start; 214| 422| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE0EEENS_6resultEPKcmPDs: 218| 594| char16_t *utf16_output) { 219| 594| size_t pos = 0; 220| 594| 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| 594| size_t leading_byte = 0; 228| 594| size_t margin = size; 229| 5.04k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 4.81k, False: 230] | Branch (229:26): [True: 4.44k, False: 364] ------------------ 230| 4.44k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 4.44k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 594| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 387k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 387k, False: 512] ------------------ 236| 387k| simd8x64 input(reinterpret_cast(in + pos)); 237| 387k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 294k, False: 92.8k] ------------------ 238| 294k| input.store_ascii_as_utf16(utf16_output); 239| 294k| utf16_output += 64; 240| 294k| pos += 64; 241| 294k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 92.8k| static_assert( 245| 92.8k| (simd8x64::NUM_CHUNKS == 2) || 246| 92.8k| (simd8x64::NUM_CHUNKS == 4), 247| 92.8k| "We support either two or four chunks per 64-byte block."); 248| 92.8k| auto zero = simd8{uint8_t(0)}; 249| | if constexpr (simd8x64::NUM_CHUNKS == 2) { 250| | this->check_utf8_bytes(input.chunks[0], zero); 251| | this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 252| 92.8k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 253| 92.8k| this->check_utf8_bytes(input.chunks[0], zero); 254| 92.8k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 255| 92.8k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 256| 92.8k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 257| 92.8k| } 258| 92.8k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 92.8k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 82, False: 92.7k] | Branch (259:25): [True: 0, False: 92.7k] ------------------ 260| | // rewind_and_convert_with_errors will seek a potential error from 261| | // in+pos onward, with the ability to go back up to pos bytes, and 262| | // read size-pos bytes forward. 263| 82| result res = 264| 82| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 82| pos, in + pos, size - pos, utf16_output); 266| 82| res.count += pos; 267| 82| return res; 268| 82| } 269| 92.7k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 92.7k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 271| | // We process in blocks of up to 12 bytes except possibly 272| | // for fast paths which may process up to 16 bytes. For the 273| | // slow path to work, we should have at least 12 input bytes left. 274| 92.7k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 911k| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 819k, False: 92.7k] ------------------ 277| | // Performance note: our ability to compute 'consumed' and 278| | // then shift and recompute is critical. If there is a 279| | // latency of, say, 4 cycles on getting 'consumed', then 280| | // the inner loop might have a total latency of about 6 cycles. 281| | // Yet we process between 6 to 12 inputs bytes, thus we get 282| | // a speed limit between 1 cycle/byte and 0.5 cycle/byte 283| | // for this section of the code. Hence, there is a limit 284| | // to how much we can further increase this latency before 285| | // it seriously harms performance. 286| 819k| size_t consumed = convert_masked_utf8_to_utf16( 287| 819k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 819k| pos += consumed; 289| 819k| utf8_end_of_code_point_mask >>= consumed; 290| 819k| } 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| 92.7k| } 296| 387k| } 297| 512| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 512] ------------------ 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| 512| if (pos < size) { ------------------ | Branch (307:9): [True: 508, 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| 508| result res = 312| 508| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 508| pos, in + pos, size - pos, utf16_output); 314| 508| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 258, False: 250] ------------------ 315| 258| res.count += pos; 316| 258| return res; 317| 258| } else { // In case of success, we want the number of word written 318| 250| utf16_output += res.count; 319| 250| } 320| 508| } 321| 254| return result(error_code::SUCCESS, utf16_output - start); 322| 512| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1621validating_transcoder19convert_with_errorsILNS_10endiannessE1EEENS_6resultEPKcmPDs: 218| 772| char16_t *utf16_output) { 219| 772| size_t pos = 0; 220| 772| 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| 772| size_t leading_byte = 0; 228| 772| size_t margin = size; 229| 6.85k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (229:12): [True: 6.58k, False: 264] | Branch (229:26): [True: 6.08k, False: 508] ------------------ 230| 6.08k| leading_byte += (int8_t(in[margin - 1]) > -65); 231| 6.08k| } 232| | // If the input is long enough, then we have that margin-1 is the eight last 233| | // leading byte. 234| 772| const size_t safety_margin = size - margin + 1; // to avoid overruns! 235| 462k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (235:12): [True: 462k, False: 706] ------------------ 236| 462k| simd8x64 input(reinterpret_cast(in + pos)); 237| 462k| if (input.is_ascii()) { ------------------ | Branch (237:11): [True: 358k, False: 103k] ------------------ 238| 358k| input.store_ascii_as_utf16(utf16_output); 239| 358k| utf16_output += 64; 240| 358k| pos += 64; 241| 358k| } else { 242| | // you might think that a for-loop would work, but under Visual Studio, 243| | // it is not good enough. 244| 103k| static_assert( 245| 103k| (simd8x64::NUM_CHUNKS == 2) || 246| 103k| (simd8x64::NUM_CHUNKS == 4), 247| 103k| "We support either two or four chunks per 64-byte block."); 248| 103k| 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| 103k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 253| 103k| this->check_utf8_bytes(input.chunks[0], zero); 254| 103k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 255| 103k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 256| 103k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 257| 103k| } 258| 103k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 259| 103k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (259:13): [True: 66, False: 103k] | Branch (259:25): [True: 0, False: 103k] ------------------ 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| 66| result res = 264| 66| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 265| 66| pos, in + pos, size - pos, utf16_output); 266| 66| res.count += pos; 267| 66| return res; 268| 66| } 269| 103k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 270| 103k| 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| 103k| size_t max_starting_point = (pos + 64) - 12; 275| | // Next loop is going to run at least five times. 276| 1.02M| while (pos < max_starting_point) { ------------------ | Branch (276:16): [True: 917k, False: 103k] ------------------ 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| 917k| size_t consumed = convert_masked_utf8_to_utf16( 287| 917k| in + pos, utf8_end_of_code_point_mask, utf16_output); 288| 917k| pos += consumed; 289| 917k| utf8_end_of_code_point_mask >>= consumed; 290| 917k| } 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| 103k| } 296| 462k| } 297| 706| if (errors()) { ------------------ | Branch (297:9): [True: 0, False: 706] ------------------ 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| 706| if (pos < size) { ------------------ | Branch (307:9): [True: 702, 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| 702| result res = 312| 702| scalar::utf8_to_utf16::rewind_and_convert_with_errors( 313| 702| pos, in + pos, size - pos, utf16_output); 314| 702| if (res.error) { // In case of error, we want the error position ------------------ | Branch (314:11): [True: 356, False: 346] ------------------ 315| 356| res.count += pos; 316| 356| return res; 317| 356| } else { // In case of success, we want the number of word written 318| 346| utf16_output += res.count; 319| 346| } 320| 702| } 321| 350| return result(error_code::SUCCESS, utf16_output - start); 322| 706| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE0EEEmPKcmPDs: 10| 432| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 432| size_t pos = 0; 14| 432| char16_t *start{utf16_output}; 15| 432| const size_t safety_margin = 16; // to avoid overruns! 16| 394k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 394k, False: 432] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 394k| simd8x64 in(reinterpret_cast(input + pos)); 20| 394k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 249k, False: 144k] ------------------ 21| 249k| in.store_ascii_as_utf16(utf16_output); 22| 249k| utf16_output += 64; 23| 249k| pos += 64; 24| 249k| } 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| 1.43M| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 1.29M, 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| 1.29M| size_t consumed = convert_masked_utf8_to_utf16( 56| 1.29M| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 1.29M| pos += consumed; 58| 1.29M| utf8_end_of_code_point_mask >>= consumed; 59| 1.29M| } 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| 394k| } 66| 432| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 432| input + pos, size - pos, utf16_output); 68| 432| return utf16_output - start; 69| 432|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE1EEEmPKcmPDs: 10| 460| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 460| size_t pos = 0; 14| 460| char16_t *start{utf16_output}; 15| 460| const size_t safety_margin = 16; // to avoid overruns! 16| 337k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 336k, False: 460] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 336k| simd8x64 in(reinterpret_cast(input + pos)); 20| 336k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 303k, False: 33.2k] ------------------ 21| 303k| in.store_ascii_as_utf16(utf16_output); 22| 303k| utf16_output += 64; 23| 303k| pos += 64; 24| 303k| } 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| 33.2k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 33.2k| 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| 33.2k| 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| 33.2k| 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| 319k| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 286k, False: 33.2k] ------------------ 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| 286k| size_t consumed = convert_masked_utf8_to_utf16( 56| 286k| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 286k| pos += consumed; 58| 286k| utf8_end_of_code_point_mask >>= consumed; 59| 286k| } 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| 33.2k| } 65| 336k| } 66| 460| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 460| input + pos, size - pos, utf16_output); 68| 460| return utf16_output - start; 69| 460|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE0EEEmPKcmPDs: 10| 432| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 432| size_t pos = 0; 14| 432| char16_t *start{utf16_output}; 15| 432| const size_t safety_margin = 16; // to avoid overruns! 16| 394k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 394k, False: 432] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 394k| simd8x64 in(reinterpret_cast(input + pos)); 20| 394k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 249k, False: 144k] ------------------ 21| 249k| in.store_ascii_as_utf16(utf16_output); 22| 249k| utf16_output += 64; 23| 249k| pos += 64; 24| 249k| } 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| 1.43M| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 1.29M, 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| 1.29M| size_t consumed = convert_masked_utf8_to_utf16( 56| 1.29M| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 1.29M| pos += consumed; 58| 1.29M| utf8_end_of_code_point_mask >>= consumed; 59| 1.29M| } 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| 394k| } 66| 432| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 432| input + pos, size - pos, utf16_output); 68| 432| return utf16_output - start; 69| 432|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf1613convert_validILNS_10endiannessE1EEEmPKcmPDs: 10| 460| char16_t *utf16_output) noexcept { 11| | // The implementation is not specific to haswell and should be moved to the 12| | // generic directory. 13| 460| size_t pos = 0; 14| 460| char16_t *start{utf16_output}; 15| 460| const size_t safety_margin = 16; // to avoid overruns! 16| 337k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (16:10): [True: 336k, False: 460] ------------------ 17| | // this loop could be unrolled further. For example, we could process the 18| | // mask far more than 64 bytes. 19| 336k| simd8x64 in(reinterpret_cast(input + pos)); 20| 336k| if (in.is_ascii()) { ------------------ | Branch (20:9): [True: 303k, False: 33.2k] ------------------ 21| 303k| in.store_ascii_as_utf16(utf16_output); 22| 303k| utf16_output += 64; 23| 303k| pos += 64; 24| 303k| } 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| 33.2k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 29| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 30| | // byte 31| 33.2k| 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| 33.2k| 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| 33.2k| 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| 319k| while (pos < max_starting_point) { ------------------ | Branch (41:14): [True: 286k, False: 33.2k] ------------------ 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| 286k| size_t consumed = convert_masked_utf8_to_utf16( 56| 286k| input + pos, utf8_end_of_code_point_mask, utf16_output); 57| 286k| pos += consumed; 58| 286k| utf8_end_of_code_point_mask >>= consumed; 59| 286k| } 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| 33.2k| } 65| 336k| } 66| 460| utf16_output += scalar::utf8_to_utf16::convert_valid( 67| 460| input + pos, size - pos, utf16_output); 68| 460| return utf16_output - start; 69| 460|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoderC2Ev: 114| 1.60k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder7convertEPKcmPDi: 129| 818| char32_t *utf32_output) { 130| 818| size_t pos = 0; 131| 818| 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| 818| size_t leading_byte = 0; 139| 818| size_t margin = size; 140| 7.52k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (140:12): [True: 7.26k, False: 266] | Branch (140:26): [True: 6.70k, False: 552] ------------------ 141| 6.70k| leading_byte += (int8_t(in[margin - 1]) > -65); 142| 6.70k| } 143| | // If the input is long enough, then we have that margin-1 is the fourth 144| | // last leading byte. 145| 818| const size_t safety_margin = size - margin + 1; // to avoid overruns! 146| 764k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (146:12): [True: 763k, False: 794] ------------------ 147| 763k| simd8x64 input(reinterpret_cast(in + pos)); 148| 763k| if (input.is_ascii()) { ------------------ | Branch (148:11): [True: 505k, False: 258k] ------------------ 149| 505k| input.store_ascii_as_utf32(utf32_output); 150| 505k| utf32_output += 64; 151| 505k| pos += 64; 152| 505k| } else { 153| | // you might think that a for-loop would work, but under Visual Studio, 154| | // it is not good enough. 155| 258k| static_assert( 156| 258k| (simd8x64::NUM_CHUNKS == 2) || 157| 258k| (simd8x64::NUM_CHUNKS == 4), 158| 258k| "We support either two or four chunks per 64-byte block."); 159| 258k| auto zero = simd8{uint8_t(0)}; 160| 258k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 161| 258k| this->check_utf8_bytes(input.chunks[0], zero); 162| 258k| 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| 258k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 170| 258k| if (utf8_continuation_mask & 1) { ------------------ | Branch (170:13): [True: 24, False: 258k] ------------------ 171| 24| return 0; // we have an error 172| 24| } 173| 258k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 174| 258k| 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| 258k| size_t max_starting_point = (pos + 64) - 12; 179| | // Next loop is going to run at least five times. 180| 2.57M| while (pos < max_starting_point) { ------------------ | Branch (180:16): [True: 2.31M, False: 258k] ------------------ 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| 2.31M| size_t consumed = convert_masked_utf8_to_utf32( 191| 2.31M| in + pos, utf8_end_of_code_point_mask, utf32_output); 192| 2.31M| pos += consumed; 193| 2.31M| utf8_end_of_code_point_mask >>= consumed; 194| 2.31M| } 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| 258k| } 200| 763k| } 201| 794| if (errors()) { ------------------ | Branch (201:9): [True: 306, False: 488] ------------------ 202| 306| return 0; 203| 306| } 204| 488| if (pos < size) { ------------------ | Branch (204:9): [True: 486, False: 2] ------------------ 205| 486| size_t howmany = 206| 486| scalar::utf8_to_utf32::convert(in + pos, size - pos, utf32_output); 207| 486| if (howmany == 0) { ------------------ | Branch (207:11): [True: 208, False: 278] ------------------ 208| 208| return 0; 209| 208| } 210| 278| utf32_output += howmany; 211| 278| } 212| 280| return utf32_output - start; 213| 488| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 645k| 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| 645k| simd8 prev1 = input.prev<1>(prev_input); 124| 645k| simd8 sc = check_special_cases(input, prev1); 125| 645k| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 645k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3219check_special_casesENS1_4simd5simd8IhEES5_: 8| 645k|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| 645k| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 645k| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 645k| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 645k| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 645k| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 645k| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 645k| 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| 645k| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 645k| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 645k| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 645k| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 645k| TOO_LONG, 40| | // 10______ ________ 41| 645k| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 645k| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 645k| TOO_SHORT, 46| | // 1110____ ________ 47| 645k| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 645k| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 645k| constexpr const uint8_t CARRY = 51| 645k| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 645k| const simd8 byte_1_low = 53| 645k| (prev1 & 0x0F) 54| 645k| .lookup_16( 55| | // ____0000 ________ 56| 645k| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 645k| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 645k| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 645k| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 645k| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 645k| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 645k| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 645k| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 645k| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 645k| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 645k| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 645k| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 645k| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 645k| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 645k| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 645k| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 645k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 645k| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 645k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 645k| OVERLONG_4, 88| | // ________ 1001____ 89| 645k| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 645k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 645k| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 645k| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 645k| return (byte_1_high & byte_1_low & byte_2_high); 97| 645k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3223check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 645k| const simd8 sc) { 102| 645k| simd8 prev2 = input.prev<2>(prev_input); 103| 645k| simd8 prev3 = input.prev<3>(prev_input); 104| 645k| simd8 must23 = 105| 645k| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 645k| simd8 must23_80 = must23 & uint8_t(0x80); 107| 645k| return must23_80 ^ sc; 108| 645k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder6errorsEv: 310| 65.3k| simdutf_really_inline bool errors() const { 311| 65.3k| return this->error.any_bits_set_anywhere(); 312| 65.3k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3221validating_transcoder19convert_with_errorsEPKcmPDi: 216| 790| char32_t *utf32_output) { 217| 790| size_t pos = 0; 218| 790| 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| 790| size_t leading_byte = 0; 226| 790| size_t margin = size; 227| 6.67k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (227:12): [True: 6.40k, False: 266] | Branch (227:26): [True: 5.88k, False: 524] ------------------ 228| 5.88k| leading_byte += (int8_t(in[margin - 1]) > -65); 229| 5.88k| } 230| | // If the input is long enough, then we have that margin-1 is the fourth 231| | // last leading byte. 232| 790| const size_t safety_margin = size - margin + 1; // to avoid overruns! 233| 464k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (233:12): [True: 463k, False: 702] ------------------ 234| 463k| simd8x64 input(reinterpret_cast(in + pos)); 235| 463k| if (input.is_ascii()) { ------------------ | Branch (235:11): [True: 399k, False: 63.8k] ------------------ 236| 399k| input.store_ascii_as_utf32(utf32_output); 237| 399k| utf32_output += 64; 238| 399k| pos += 64; 239| 399k| } else { 240| | // you might think that a for-loop would work, but under Visual Studio, 241| | // it is not good enough. 242| 63.8k| static_assert( 243| 63.8k| (simd8x64::NUM_CHUNKS == 2) || 244| 63.8k| (simd8x64::NUM_CHUNKS == 4), 245| 63.8k| "We support either two or four chunks per 64-byte block."); 246| 63.8k| auto zero = simd8{uint8_t(0)}; 247| 63.8k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 248| 63.8k| this->check_utf8_bytes(input.chunks[0], zero); 249| 63.8k| 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| 63.8k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 257| 63.8k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (257:13): [True: 88, False: 63.7k] | Branch (257:25): [True: 0, False: 63.7k] ------------------ 258| 88| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 259| 88| pos, in + pos, size - pos, utf32_output); 260| 88| res.count += pos; 261| 88| return res; 262| 88| } 263| 63.7k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 264| 63.7k| 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| 63.7k| size_t max_starting_point = (pos + 64) - 12; 269| | // Next loop is going to run at least five times. 270| 616k| while (pos < max_starting_point) { ------------------ | Branch (270:16): [True: 553k, False: 63.7k] ------------------ 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| 553k| size_t consumed = convert_masked_utf8_to_utf32( 281| 553k| in + pos, utf8_end_of_code_point_mask, utf32_output); 282| 553k| pos += consumed; 283| 553k| utf8_end_of_code_point_mask >>= consumed; 284| 553k| } 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| 63.7k| } 290| 463k| } 291| 702| if (errors()) { ------------------ | Branch (291:9): [True: 0, False: 702] ------------------ 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| 702| if (pos < size) { ------------------ | Branch (297:9): [True: 698, False: 4] ------------------ 298| 698| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 299| 698| pos, in + pos, size - pos, utf32_output); 300| 698| if (res.error) { // In case of error, we want the error position ------------------ | Branch (300:11): [True: 356, False: 342] ------------------ 301| 356| res.count += pos; 302| 356| return res; 303| 356| } else { // In case of success, we want the number of word written 304| 342| utf32_output += res.count; 305| 342| } 306| 698| } 307| 346| return result(error_code::SUCCESS, utf32_output - start); 308| 702| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoderC2Ev: 114| 1.60k| validating_transcoder() : error(uint8_t(0)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder7convertEPKcmPDi: 129| 818| char32_t *utf32_output) { 130| 818| size_t pos = 0; 131| 818| 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| 818| size_t leading_byte = 0; 139| 818| size_t margin = size; 140| 7.52k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (140:12): [True: 7.26k, False: 266] | Branch (140:26): [True: 6.70k, False: 552] ------------------ 141| 6.70k| leading_byte += (int8_t(in[margin - 1]) > -65); 142| 6.70k| } 143| | // If the input is long enough, then we have that margin-1 is the fourth 144| | // last leading byte. 145| 818| const size_t safety_margin = size - margin + 1; // to avoid overruns! 146| 764k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (146:12): [True: 763k, False: 794] ------------------ 147| 763k| simd8x64 input(reinterpret_cast(in + pos)); 148| 763k| if (input.is_ascii()) { ------------------ | Branch (148:11): [True: 505k, False: 258k] ------------------ 149| 505k| input.store_ascii_as_utf32(utf32_output); 150| 505k| utf32_output += 64; 151| 505k| pos += 64; 152| 505k| } else { 153| | // you might think that a for-loop would work, but under Visual Studio, 154| | // it is not good enough. 155| 258k| static_assert( 156| 258k| (simd8x64::NUM_CHUNKS == 2) || 157| 258k| (simd8x64::NUM_CHUNKS == 4), 158| 258k| "We support either two or four chunks per 64-byte block."); 159| 258k| 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| 258k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 164| 258k| this->check_utf8_bytes(input.chunks[0], zero); 165| 258k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 166| 258k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 167| 258k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 168| 258k| } 169| 258k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 170| 258k| if (utf8_continuation_mask & 1) { ------------------ | Branch (170:13): [True: 24, False: 258k] ------------------ 171| 24| return 0; // we have an error 172| 24| } 173| 258k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 174| 258k| 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| 258k| size_t max_starting_point = (pos + 64) - 12; 179| | // Next loop is going to run at least five times. 180| 2.57M| while (pos < max_starting_point) { ------------------ | Branch (180:16): [True: 2.31M, False: 258k] ------------------ 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| 2.31M| size_t consumed = convert_masked_utf8_to_utf32( 191| 2.31M| in + pos, utf8_end_of_code_point_mask, utf32_output); 192| 2.31M| pos += consumed; 193| 2.31M| utf8_end_of_code_point_mask >>= consumed; 194| 2.31M| } 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| 258k| } 200| 763k| } 201| 794| if (errors()) { ------------------ | Branch (201:9): [True: 306, False: 488] ------------------ 202| 306| return 0; 203| 306| } 204| 488| if (pos < size) { ------------------ | Branch (204:9): [True: 486, False: 2] ------------------ 205| 486| size_t howmany = 206| 486| scalar::utf8_to_utf32::convert(in + pos, size - pos, utf32_output); 207| 486| if (howmany == 0) { ------------------ | Branch (207:11): [True: 208, False: 278] ------------------ 208| 208| return 0; 209| 208| } 210| 278| utf32_output += howmany; 211| 278| } 212| 280| return utf32_output - start; 213| 488| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder16check_utf8_bytesENS1_4simd5simd8IhEES6_: 119| 1.29M| 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| 1.29M| simd8 prev1 = input.prev<1>(prev_input); 124| 1.29M| simd8 sc = check_special_cases(input, prev1); 125| 1.29M| this->error |= check_multibyte_lengths(input, prev_input, sc); 126| 1.29M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3219check_special_casesENS1_4simd5simd8IhEES5_: 8| 1.29M|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| 1.29M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 16| | // 11______ 11______ 17| 1.29M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 18| 1.29M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 19| 1.29M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 20| 1.29M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 21| 1.29M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 22| 1.29M| 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| 1.29M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 31| | // 11110101 1000____ 32| | // 1111011_ 1000____ 33| | // 11111___ 1000____ 34| 1.29M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 35| | 36| 1.29M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 37| | // 0_______ ________ 38| 1.29M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 39| 1.29M| TOO_LONG, 40| | // 10______ ________ 41| 1.29M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 42| | // 1100____ ________ 43| 1.29M| TOO_SHORT | OVERLONG_2, 44| | // 1101____ ________ 45| 1.29M| TOO_SHORT, 46| | // 1110____ ________ 47| 1.29M| TOO_SHORT | OVERLONG_3 | SURROGATE, 48| | // 1111____ ________ 49| 1.29M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 50| 1.29M| constexpr const uint8_t CARRY = 51| 1.29M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 52| 1.29M| const simd8 byte_1_low = 53| 1.29M| (prev1 & 0x0F) 54| 1.29M| .lookup_16( 55| | // ____0000 ________ 56| 1.29M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 57| | // ____0001 ________ 58| 1.29M| CARRY | OVERLONG_2, 59| | // ____001_ ________ 60| 1.29M| CARRY, CARRY, 61| | 62| | // ____0100 ________ 63| 1.29M| CARRY | TOO_LARGE, 64| | // ____0101 ________ 65| 1.29M| CARRY | TOO_LARGE | TOO_LARGE_1000, 66| | // ____011_ ________ 67| 1.29M| CARRY | TOO_LARGE | TOO_LARGE_1000, 68| 1.29M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| | 70| | // ____1___ ________ 71| 1.29M| CARRY | TOO_LARGE | TOO_LARGE_1000, 72| 1.29M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 1.29M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 1.29M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 1.29M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| | // ____1101 ________ 77| 1.29M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 78| 1.29M| CARRY | TOO_LARGE | TOO_LARGE_1000, 79| 1.29M| CARRY | TOO_LARGE | TOO_LARGE_1000); 80| 1.29M| const simd8 byte_2_high = input.shr<4>().lookup_16( 81| | // ________ 0_______ 82| 1.29M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 83| 1.29M| TOO_SHORT, TOO_SHORT, 84| | 85| | // ________ 1000____ 86| 1.29M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 87| 1.29M| OVERLONG_4, 88| | // ________ 1001____ 89| 1.29M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 90| | // ________ 101_____ 91| 1.29M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 92| 1.29M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| | 94| | // ________ 11______ 95| 1.29M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 96| 1.29M| return (byte_1_high & byte_1_low & byte_2_high); 97| 1.29M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3223check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 101| 1.29M| const simd8 sc) { 102| 1.29M| simd8 prev2 = input.prev<2>(prev_input); 103| 1.29M| simd8 prev3 = input.prev<3>(prev_input); 104| 1.29M| simd8 must23 = 105| 1.29M| simd8(must_be_2_3_continuation(prev2, prev3)); 106| 1.29M| simd8 must23_80 = must23 & uint8_t(0x80); 107| 1.29M| return must23_80 ^ sc; 108| 1.29M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder6errorsEv: 310| 65.3k| simdutf_really_inline bool errors() const { 311| 65.3k| return this->error.any_bits_set_anywhere(); 312| 65.3k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3221validating_transcoder19convert_with_errorsEPKcmPDi: 216| 790| char32_t *utf32_output) { 217| 790| size_t pos = 0; 218| 790| 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| 790| size_t leading_byte = 0; 226| 790| size_t margin = size; 227| 6.67k| for (; margin > 0 && leading_byte < 8; margin--) { ------------------ | Branch (227:12): [True: 6.40k, False: 266] | Branch (227:26): [True: 5.88k, False: 524] ------------------ 228| 5.88k| leading_byte += (int8_t(in[margin - 1]) > -65); 229| 5.88k| } 230| | // If the input is long enough, then we have that margin-1 is the fourth 231| | // last leading byte. 232| 790| const size_t safety_margin = size - margin + 1; // to avoid overruns! 233| 464k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (233:12): [True: 463k, False: 702] ------------------ 234| 463k| simd8x64 input(reinterpret_cast(in + pos)); 235| 463k| if (input.is_ascii()) { ------------------ | Branch (235:11): [True: 399k, False: 63.8k] ------------------ 236| 399k| input.store_ascii_as_utf32(utf32_output); 237| 399k| utf32_output += 64; 238| 399k| pos += 64; 239| 399k| } else { 240| | // you might think that a for-loop would work, but under Visual Studio, 241| | // it is not good enough. 242| 63.8k| static_assert( 243| 63.8k| (simd8x64::NUM_CHUNKS == 2) || 244| 63.8k| (simd8x64::NUM_CHUNKS == 4), 245| 63.8k| "We support either two or four chunks per 64-byte block."); 246| 63.8k| 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| 63.8k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 251| 63.8k| this->check_utf8_bytes(input.chunks[0], zero); 252| 63.8k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 253| 63.8k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 254| 63.8k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 255| 63.8k| } 256| 63.8k| uint64_t utf8_continuation_mask = input.lt(-65 + 1); 257| 63.8k| if (errors() || (utf8_continuation_mask & 1)) { ------------------ | Branch (257:13): [True: 88, False: 63.7k] | Branch (257:25): [True: 0, False: 63.7k] ------------------ 258| 88| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 259| 88| pos, in + pos, size - pos, utf32_output); 260| 88| res.count += pos; 261| 88| return res; 262| 88| } 263| 63.7k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 264| 63.7k| 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| 63.7k| size_t max_starting_point = (pos + 64) - 12; 269| | // Next loop is going to run at least five times. 270| 616k| while (pos < max_starting_point) { ------------------ | Branch (270:16): [True: 553k, False: 63.7k] ------------------ 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| 553k| size_t consumed = convert_masked_utf8_to_utf32( 281| 553k| in + pos, utf8_end_of_code_point_mask, utf32_output); 282| 553k| pos += consumed; 283| 553k| utf8_end_of_code_point_mask >>= consumed; 284| 553k| } 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| 63.7k| } 290| 463k| } 291| 702| if (errors()) { ------------------ | Branch (291:9): [True: 0, False: 702] ------------------ 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| 702| if (pos < size) { ------------------ | Branch (297:9): [True: 698, False: 4] ------------------ 298| 698| result res = scalar::utf8_to_utf32::rewind_and_convert_with_errors( 299| 698| pos, in + pos, size - pos, utf32_output); 300| 698| if (res.error) { // In case of error, we want the error position ------------------ | Branch (300:11): [True: 356, False: 342] ------------------ 301| 356| res.count += pos; 302| 356| return res; 303| 356| } else { // In case of success, we want the number of word written 304| 342| utf32_output += res.count; 305| 342| } 306| 698| } 307| 346| return result(error_code::SUCCESS, utf32_output - start); 308| 702| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_113utf8_to_utf3213convert_validEPKcmPDi: 9| 432| char32_t *utf32_output) noexcept { 10| 432| size_t pos = 0; 11| 432| char32_t *start{utf32_output}; 12| 432| const size_t safety_margin = 16; // to avoid overruns! 13| 310k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (13:10): [True: 310k, False: 432] ------------------ 14| 310k| simd8x64 in(reinterpret_cast(input + pos)); 15| 310k| if (in.is_ascii()) { ------------------ | Branch (15:9): [True: 256k, False: 53.2k] ------------------ 16| 256k| in.store_ascii_as_utf32(utf32_output); 17| 256k| utf32_output += 64; 18| 256k| pos += 64; 19| 256k| } else { 20| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 21| | // byte 22| 53.2k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 23| 53.2k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 24| 53.2k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 25| 53.2k| size_t max_starting_point = (pos + 64) - 12; 26| 510k| while (pos < max_starting_point) { ------------------ | Branch (26:14): [True: 457k, False: 53.2k] ------------------ 27| 457k| size_t consumed = convert_masked_utf8_to_utf32( 28| 457k| input + pos, utf8_end_of_code_point_mask, utf32_output); 29| 457k| pos += consumed; 30| 457k| utf8_end_of_code_point_mask >>= consumed; 31| 457k| } 32| 53.2k| } 33| 310k| } 34| 432| utf32_output += scalar::utf8_to_utf32::convert_valid(input + pos, size - pos, 35| 432| utf32_output); 36| 432| return utf32_output - start; 37| 432|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_113utf8_to_utf3213convert_validEPKcmPDi: 9| 432| char32_t *utf32_output) noexcept { 10| 432| size_t pos = 0; 11| 432| char32_t *start{utf32_output}; 12| 432| const size_t safety_margin = 16; // to avoid overruns! 13| 310k| while (pos + 64 + safety_margin <= size) { ------------------ | Branch (13:10): [True: 310k, False: 432] ------------------ 14| 310k| simd8x64 in(reinterpret_cast(input + pos)); 15| 310k| if (in.is_ascii()) { ------------------ | Branch (15:9): [True: 256k, False: 53.2k] ------------------ 16| 256k| in.store_ascii_as_utf32(utf32_output); 17| 256k| utf32_output += 64; 18| 256k| pos += 64; 19| 256k| } else { 20| | // -65 is 0b10111111 in two-complement's, so largest possible continuation 21| | // byte 22| 53.2k| uint64_t utf8_continuation_mask = in.lt(-65 + 1); 23| 53.2k| uint64_t utf8_leading_mask = ~utf8_continuation_mask; 24| 53.2k| uint64_t utf8_end_of_code_point_mask = utf8_leading_mask >> 1; 25| 53.2k| size_t max_starting_point = (pos + 64) - 12; 26| 510k| while (pos < max_starting_point) { ------------------ | Branch (26:14): [True: 457k, False: 53.2k] ------------------ 27| 457k| size_t consumed = convert_masked_utf8_to_utf32( 28| 457k| input + pos, utf8_end_of_code_point_mask, utf32_output); 29| 457k| pos += consumed; 30| 457k| utf8_end_of_code_point_mask >>= consumed; 31| 457k| } 32| 53.2k| } 33| 310k| } 34| 432| utf32_output += scalar::utf8_to_utf32::convert_valid(input + pos, size - pos, 35| 432| utf32_output); 36| 432| return utf32_output - start; 37| 432|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker16check_next_inputERKNS1_4simd8simd8x64IhEE: 187| 4.75M| simdutf_really_inline void check_next_input(const simd8x64 &input) { 188| 4.75M| if (simdutf_likely(is_ascii(input))) { ------------------ | | 90| 4.75M| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.90M, False: 845k] | | ------------------ ------------------ 189| 3.90M| this->error |= this->prev_incomplete; 190| 3.90M| } else { 191| | // you might think that a for-loop would work, but under Visual Studio, it 192| | // is not good enough. 193| 845k| static_assert((simd8x64::NUM_CHUNKS == 2) || 194| 845k| (simd8x64::NUM_CHUNKS == 4), 195| 845k| "We support either two or four chunks per 64-byte block."); 196| 845k| if constexpr (simd8x64::NUM_CHUNKS == 2) { 197| 845k| this->check_utf8_bytes(input.chunks[0], this->prev_input_block); 198| 845k| 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| 845k| this->prev_incomplete = 206| 845k| is_incomplete(input.chunks[simd8x64::NUM_CHUNKS - 1]); 207| 845k| this->prev_input_block = input.chunks[simd8x64::NUM_CHUNKS - 1]; 208| 845k| } 209| 4.75M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker16check_utf8_bytesENS1_4simd5simd8IhEES6_: 169| 1.69M| 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| 1.69M| simd8 prev1 = input.prev<1>(prev_input); 174| 1.69M| simd8 sc = check_special_cases(input, prev1); 175| 1.69M| this->error |= check_multibyte_lengths(input, prev_input, sc); 176| 1.69M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation19check_special_casesENS1_4simd5simd8IhEES5_: 9| 1.69M|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| 1.69M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 17| | // 11______ 11______ 18| 1.69M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 19| 1.69M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 20| 1.69M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 21| 1.69M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 22| 1.69M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 23| 1.69M| 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| 1.69M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 32| | // 11110101 1000____ 33| | // 1111011_ 1000____ 34| | // 11111___ 1000____ 35| 1.69M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 36| | 37| 1.69M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 38| | // 0_______ ________ 39| 1.69M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 40| 1.69M| TOO_LONG, 41| | // 10______ ________ 42| 1.69M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 43| | // 1100____ ________ 44| 1.69M| TOO_SHORT | OVERLONG_2, 45| | // 1101____ ________ 46| 1.69M| TOO_SHORT, 47| | // 1110____ ________ 48| 1.69M| TOO_SHORT | OVERLONG_3 | SURROGATE, 49| | // 1111____ ________ 50| 1.69M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 51| 1.69M| constexpr const uint8_t CARRY = 52| 1.69M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 53| 1.69M| const simd8 byte_1_low = 54| 1.69M| (prev1 & 0x0F) 55| 1.69M| .lookup_16( 56| | // ____0000 ________ 57| 1.69M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 58| | // ____0001 ________ 59| 1.69M| CARRY | OVERLONG_2, 60| | // ____001_ ________ 61| 1.69M| CARRY, CARRY, 62| | 63| | // ____0100 ________ 64| 1.69M| CARRY | TOO_LARGE, 65| | // ____0101 ________ 66| 1.69M| CARRY | TOO_LARGE | TOO_LARGE_1000, 67| | // ____011_ ________ 68| 1.69M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| 1.69M| CARRY | TOO_LARGE | TOO_LARGE_1000, 70| | 71| | // ____1___ ________ 72| 1.69M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 1.69M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 1.69M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 1.69M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| 1.69M| CARRY | TOO_LARGE | TOO_LARGE_1000, 77| | // ____1101 ________ 78| 1.69M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 79| 1.69M| CARRY | TOO_LARGE | TOO_LARGE_1000, 80| 1.69M| CARRY | TOO_LARGE | TOO_LARGE_1000); 81| 1.69M| const simd8 byte_2_high = input.shr<4>().lookup_16( 82| | // ________ 0_______ 83| 1.69M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 84| 1.69M| TOO_SHORT, TOO_SHORT, 85| | 86| | // ________ 1000____ 87| 1.69M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 88| 1.69M| OVERLONG_4, 89| | // ________ 1001____ 90| 1.69M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 91| | // ________ 101_____ 92| 1.69M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| 1.69M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 94| | 95| | // ________ 11______ 96| 1.69M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 97| 1.69M| return (byte_1_high & byte_1_low & byte_2_high); 98| 1.69M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation23check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 102| 1.69M| const simd8 sc) { 103| 1.69M| simd8 prev2 = input.prev<2>(prev_input); 104| 1.69M| simd8 prev3 = input.prev<3>(prev_input); 105| 1.69M| simd8 must23 = 106| 1.69M| simd8(must_be_2_3_continuation(prev2, prev3)); 107| 1.69M| simd8 must23_80 = must23 & uint8_t(0x80); 108| 1.69M| return must23_80 ^ sc; 109| 1.69M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation13is_incompleteENS1_4simd5simd8IhEE: 115| 845k|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| 845k| static const uint8_t max_array[32] = {255, 120| 845k| 255, 121| 845k| 255, 122| 845k| 255, 123| 845k| 255, 124| 845k| 255, 125| 845k| 255, 126| 845k| 255, 127| 845k| 255, 128| 845k| 255, 129| 845k| 255, 130| 845k| 255, 131| 845k| 255, 132| 845k| 255, 133| 845k| 255, 134| 845k| 255, 135| 845k| 255, 136| 845k| 255, 137| 845k| 255, 138| 845k| 255, 139| 845k| 255, 140| 845k| 255, 141| 845k| 255, 142| 845k| 255, 143| 845k| 255, 144| 845k| 255, 145| 845k| 255, 146| 845k| 255, 147| 845k| 255, 148| 845k| 0b11110000u - 1, 149| 845k| 0b11100000u - 1, 150| 845k| 0b11000000u - 1}; 151| 845k| const simd8 max_value( 152| 845k| &max_array[sizeof(max_array) - sizeof(simd8)]); 153| 845k| return input.gt_bits(max_value); 154| 845k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker9check_eofEv: 181| 6.16k| simdutf_really_inline void check_eof() { 182| | // If the previous block had incomplete UTF-8 characters at the end, an 183| | // ASCII block can't possibly finish them. 184| 6.16k| this->error |= this->prev_incomplete; 185| 6.16k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_115utf8_validation12utf8_checker6errorsEv: 212| 1.98M| simdutf_really_inline bool errors() const { 213| 1.98M| return this->error.any_bits_set_anywhere(); 214| 1.98M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker16check_next_inputERKNS1_4simd8simd8x64IhEE: 187| 4.75M| simdutf_really_inline void check_next_input(const simd8x64 &input) { 188| 4.75M| if (simdutf_likely(is_ascii(input))) { ------------------ | | 90| 4.75M| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.90M, False: 845k] | | ------------------ ------------------ 189| 3.90M| this->error |= this->prev_incomplete; 190| 3.90M| } else { 191| | // you might think that a for-loop would work, but under Visual Studio, it 192| | // is not good enough. 193| 845k| static_assert((simd8x64::NUM_CHUNKS == 2) || 194| 845k| (simd8x64::NUM_CHUNKS == 4), 195| 845k| "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| 845k| } else if constexpr (simd8x64::NUM_CHUNKS == 4) { 200| 845k| this->check_utf8_bytes(input.chunks[0], this->prev_input_block); 201| 845k| this->check_utf8_bytes(input.chunks[1], input.chunks[0]); 202| 845k| this->check_utf8_bytes(input.chunks[2], input.chunks[1]); 203| 845k| this->check_utf8_bytes(input.chunks[3], input.chunks[2]); 204| 845k| } 205| 845k| this->prev_incomplete = 206| 845k| is_incomplete(input.chunks[simd8x64::NUM_CHUNKS - 1]); 207| 845k| this->prev_input_block = input.chunks[simd8x64::NUM_CHUNKS - 1]; 208| 845k| } 209| 4.75M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker16check_utf8_bytesENS1_4simd5simd8IhEES6_: 169| 3.38M| 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| 3.38M| simd8 prev1 = input.prev<1>(prev_input); 174| 3.38M| simd8 sc = check_special_cases(input, prev1); 175| 3.38M| this->error |= check_multibyte_lengths(input, prev_input, sc); 176| 3.38M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation19check_special_casesENS1_4simd5simd8IhEES5_: 9| 3.38M|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| 3.38M| constexpr const uint8_t TOO_SHORT = 1 << 0; // 11______ 0_______ 17| | // 11______ 11______ 18| 3.38M| constexpr const uint8_t TOO_LONG = 1 << 1; // 0_______ 10______ 19| 3.38M| constexpr const uint8_t OVERLONG_3 = 1 << 2; // 11100000 100_____ 20| 3.38M| constexpr const uint8_t SURROGATE = 1 << 4; // 11101101 101_____ 21| 3.38M| constexpr const uint8_t OVERLONG_2 = 1 << 5; // 1100000_ 10______ 22| 3.38M| constexpr const uint8_t TWO_CONTS = 1 << 7; // 10______ 10______ 23| 3.38M| 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| 3.38M| constexpr const uint8_t TOO_LARGE_1000 = 1 << 6; 32| | // 11110101 1000____ 33| | // 1111011_ 1000____ 34| | // 11111___ 1000____ 35| 3.38M| constexpr const uint8_t OVERLONG_4 = 1 << 6; // 11110000 1000____ 36| | 37| 3.38M| const simd8 byte_1_high = prev1.shr<4>().lookup_16( 38| | // 0_______ ________ 39| 3.38M| TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, TOO_LONG, 40| 3.38M| TOO_LONG, 41| | // 10______ ________ 42| 3.38M| TWO_CONTS, TWO_CONTS, TWO_CONTS, TWO_CONTS, 43| | // 1100____ ________ 44| 3.38M| TOO_SHORT | OVERLONG_2, 45| | // 1101____ ________ 46| 3.38M| TOO_SHORT, 47| | // 1110____ ________ 48| 3.38M| TOO_SHORT | OVERLONG_3 | SURROGATE, 49| | // 1111____ ________ 50| 3.38M| TOO_SHORT | TOO_LARGE | TOO_LARGE_1000 | OVERLONG_4); 51| 3.38M| constexpr const uint8_t CARRY = 52| 3.38M| TOO_SHORT | TOO_LONG | TWO_CONTS; // These all have ____ in byte 1 . 53| 3.38M| const simd8 byte_1_low = 54| 3.38M| (prev1 & 0x0F) 55| 3.38M| .lookup_16( 56| | // ____0000 ________ 57| 3.38M| CARRY | OVERLONG_3 | OVERLONG_2 | OVERLONG_4, 58| | // ____0001 ________ 59| 3.38M| CARRY | OVERLONG_2, 60| | // ____001_ ________ 61| 3.38M| CARRY, CARRY, 62| | 63| | // ____0100 ________ 64| 3.38M| CARRY | TOO_LARGE, 65| | // ____0101 ________ 66| 3.38M| CARRY | TOO_LARGE | TOO_LARGE_1000, 67| | // ____011_ ________ 68| 3.38M| CARRY | TOO_LARGE | TOO_LARGE_1000, 69| 3.38M| CARRY | TOO_LARGE | TOO_LARGE_1000, 70| | 71| | // ____1___ ________ 72| 3.38M| CARRY | TOO_LARGE | TOO_LARGE_1000, 73| 3.38M| CARRY | TOO_LARGE | TOO_LARGE_1000, 74| 3.38M| CARRY | TOO_LARGE | TOO_LARGE_1000, 75| 3.38M| CARRY | TOO_LARGE | TOO_LARGE_1000, 76| 3.38M| CARRY | TOO_LARGE | TOO_LARGE_1000, 77| | // ____1101 ________ 78| 3.38M| CARRY | TOO_LARGE | TOO_LARGE_1000 | SURROGATE, 79| 3.38M| CARRY | TOO_LARGE | TOO_LARGE_1000, 80| 3.38M| CARRY | TOO_LARGE | TOO_LARGE_1000); 81| 3.38M| const simd8 byte_2_high = input.shr<4>().lookup_16( 82| | // ________ 0_______ 83| 3.38M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT, 84| 3.38M| TOO_SHORT, TOO_SHORT, 85| | 86| | // ________ 1000____ 87| 3.38M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE_1000 | 88| 3.38M| OVERLONG_4, 89| | // ________ 1001____ 90| 3.38M| TOO_LONG | OVERLONG_2 | TWO_CONTS | OVERLONG_3 | TOO_LARGE, 91| | // ________ 101_____ 92| 3.38M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 93| 3.38M| TOO_LONG | OVERLONG_2 | TWO_CONTS | SURROGATE | TOO_LARGE, 94| | 95| | // ________ 11______ 96| 3.38M| TOO_SHORT, TOO_SHORT, TOO_SHORT, TOO_SHORT); 97| 3.38M| return (byte_1_high & byte_1_low & byte_2_high); 98| 3.38M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation23check_multibyte_lengthsENS1_4simd5simd8IhEES5_S5_: 102| 3.38M| const simd8 sc) { 103| 3.38M| simd8 prev2 = input.prev<2>(prev_input); 104| 3.38M| simd8 prev3 = input.prev<3>(prev_input); 105| 3.38M| simd8 must23 = 106| 3.38M| simd8(must_be_2_3_continuation(prev2, prev3)); 107| 3.38M| simd8 must23_80 = must23 & uint8_t(0x80); 108| 3.38M| return must23_80 ^ sc; 109| 3.38M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation13is_incompleteENS1_4simd5simd8IhEE: 115| 845k|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| 845k| static const uint8_t max_array[32] = {255, 120| 845k| 255, 121| 845k| 255, 122| 845k| 255, 123| 845k| 255, 124| 845k| 255, 125| 845k| 255, 126| 845k| 255, 127| 845k| 255, 128| 845k| 255, 129| 845k| 255, 130| 845k| 255, 131| 845k| 255, 132| 845k| 255, 133| 845k| 255, 134| 845k| 255, 135| 845k| 255, 136| 845k| 255, 137| 845k| 255, 138| 845k| 255, 139| 845k| 255, 140| 845k| 255, 141| 845k| 255, 142| 845k| 255, 143| 845k| 255, 144| 845k| 255, 145| 845k| 255, 146| 845k| 255, 147| 845k| 255, 148| 845k| 0b11110000u - 1, 149| 845k| 0b11100000u - 1, 150| 845k| 0b11000000u - 1}; 151| 845k| const simd8 max_value( 152| 845k| &max_array[sizeof(max_array) - sizeof(simd8)]); 153| 845k| return input.gt_bits(max_value); 154| 845k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker9check_eofEv: 181| 6.16k| simdutf_really_inline void check_eof() { 182| | // If the previous block had incomplete UTF-8 characters at the end, an 183| | // ASCII block can't possibly finish them. 184| 6.16k| this->error |= this->prev_incomplete; 185| 6.16k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_115utf8_validation12utf8_checker6errorsEv: 212| 1.98M| simdutf_really_inline bool errors() const { 213| 1.98M| return this->error.any_bits_set_anywhere(); 214| 1.98M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation21generic_validate_utf8EPKcm: 27| 3.52k|bool generic_validate_utf8(const char *input, size_t length) { 28| 3.52k| return generic_validate_utf8( 29| 3.52k| reinterpret_cast(input), length); 30| 3.52k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation21generic_validate_utf8INS2_12utf8_checkerEEEbPKhm: 10| 3.52k|bool generic_validate_utf8(const uint8_t *input, size_t length) { 11| 3.52k| checker c{}; 12| 3.52k| buf_block_reader<64> reader(input, length); 13| 2.76M| while (reader.has_full_block()) { ------------------ | Branch (13:10): [True: 2.76M, False: 3.52k] ------------------ 14| 2.76M| simd::simd8x64 in(reader.full_block()); 15| 2.76M| c.check_next_input(in); 16| 2.76M| reader.advance(); 17| 2.76M| } 18| 3.52k| uint8_t block[64]{}; 19| 3.52k| reader.get_remainder(block); 20| 3.52k| simd::simd8x64 in(block); 21| 3.52k| c.check_next_input(in); 22| 3.52k| reader.advance(); 23| 3.52k| c.check_eof(); 24| 3.52k| return !c.errors(); 25| 3.52k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsEPKcm: 76| 3.52k|result generic_validate_utf8_with_errors(const char *input, size_t length) { 77| 3.52k| return generic_validate_utf8_with_errors( 78| 3.52k| reinterpret_cast(input), length); 79| 3.52k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsINS2_12utf8_checkerEEENS_6resultEPKhm: 36| 3.52k|result generic_validate_utf8_with_errors(const uint8_t *input, size_t length) { 37| 3.52k| checker c{}; 38| 3.52k| buf_block_reader<64> reader(input, length); 39| 3.52k| size_t count{0}; 40| 1.98M| while (reader.has_full_block()) { ------------------ | Branch (40:10): [True: 1.97M, False: 2.63k] ------------------ 41| 1.97M| simd::simd8x64 in(reader.full_block()); 42| 1.97M| c.check_next_input(in); 43| 1.97M| if (c.errors()) { ------------------ | Branch (43:9): [True: 883, False: 1.97M] ------------------ 44| 883| if (count != 0) { ------------------ | Branch (44:11): [True: 246, False: 637] ------------------ 45| 246| count--; 46| 246| } // Sometimes the error is only detected in the next chunk 47| 883| result res = scalar::utf8::rewind_and_validate_with_errors( 48| 883| reinterpret_cast(input), 49| 883| reinterpret_cast(input + count), length - count); 50| 883| res.count += count; 51| 883| return res; 52| 883| } 53| 1.97M| reader.advance(); 54| 1.97M| count += 64; 55| 1.97M| } 56| 2.63k| uint8_t block[64]{}; 57| 2.63k| reader.get_remainder(block); 58| 2.63k| simd::simd8x64 in(block); 59| 2.63k| c.check_next_input(in); 60| 2.63k| reader.advance(); 61| 2.63k| c.check_eof(); 62| 2.63k| if (c.errors()) { ------------------ | Branch (62:7): [True: 921, False: 1.71k] ------------------ 63| 921| if (count != 0) { ------------------ | Branch (63:9): [True: 229, False: 692] ------------------ 64| 229| count--; 65| 229| } // Sometimes the error is only detected in the next chunk 66| 921| result res = scalar::utf8::rewind_and_validate_with_errors( 67| 921| reinterpret_cast(input), 68| 921| reinterpret_cast(input) + count, length - count); 69| 921| res.count += count; 70| 921| return res; 71| 1.71k| } else { 72| 1.71k| return result(error_code::SUCCESS, length); 73| 1.71k| } 74| 2.63k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation21generic_validate_utf8EPKcm: 27| 3.52k|bool generic_validate_utf8(const char *input, size_t length) { 28| 3.52k| return generic_validate_utf8( 29| 3.52k| reinterpret_cast(input), length); 30| 3.52k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation21generic_validate_utf8INS2_12utf8_checkerEEEbPKhm: 10| 3.52k|bool generic_validate_utf8(const uint8_t *input, size_t length) { 11| 3.52k| checker c{}; 12| 3.52k| buf_block_reader<64> reader(input, length); 13| 2.76M| while (reader.has_full_block()) { ------------------ | Branch (13:10): [True: 2.76M, False: 3.52k] ------------------ 14| 2.76M| simd::simd8x64 in(reader.full_block()); 15| 2.76M| c.check_next_input(in); 16| 2.76M| reader.advance(); 17| 2.76M| } 18| 3.52k| uint8_t block[64]{}; 19| 3.52k| reader.get_remainder(block); 20| 3.52k| simd::simd8x64 in(block); 21| 3.52k| c.check_next_input(in); 22| 3.52k| reader.advance(); 23| 3.52k| c.check_eof(); 24| 3.52k| return !c.errors(); 25| 3.52k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsEPKcm: 76| 3.52k|result generic_validate_utf8_with_errors(const char *input, size_t length) { 77| 3.52k| return generic_validate_utf8_with_errors( 78| 3.52k| reinterpret_cast(input), length); 79| 3.52k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_115utf8_validation33generic_validate_utf8_with_errorsINS2_12utf8_checkerEEENS_6resultEPKhm: 36| 3.52k|result generic_validate_utf8_with_errors(const uint8_t *input, size_t length) { 37| 3.52k| checker c{}; 38| 3.52k| buf_block_reader<64> reader(input, length); 39| 3.52k| size_t count{0}; 40| 1.98M| while (reader.has_full_block()) { ------------------ | Branch (40:10): [True: 1.97M, False: 2.63k] ------------------ 41| 1.97M| simd::simd8x64 in(reader.full_block()); 42| 1.97M| c.check_next_input(in); 43| 1.97M| if (c.errors()) { ------------------ | Branch (43:9): [True: 883, False: 1.97M] ------------------ 44| 883| if (count != 0) { ------------------ | Branch (44:11): [True: 246, False: 637] ------------------ 45| 246| count--; 46| 246| } // Sometimes the error is only detected in the next chunk 47| 883| result res = scalar::utf8::rewind_and_validate_with_errors( 48| 883| reinterpret_cast(input), 49| 883| reinterpret_cast(input + count), length - count); 50| 883| res.count += count; 51| 883| return res; 52| 883| } 53| 1.97M| reader.advance(); 54| 1.97M| count += 64; 55| 1.97M| } 56| 2.63k| uint8_t block[64]{}; 57| 2.63k| reader.get_remainder(block); 58| 2.63k| simd::simd8x64 in(block); 59| 2.63k| c.check_next_input(in); 60| 2.63k| reader.advance(); 61| 2.63k| c.check_eof(); 62| 2.63k| if (c.errors()) { ------------------ | Branch (62:7): [True: 921, False: 1.71k] ------------------ 63| 921| if (count != 0) { ------------------ | Branch (63:9): [True: 229, False: 692] ------------------ 64| 229| count--; 65| 229| } // Sometimes the error is only detected in the next chunk 66| 921| result res = scalar::utf8::rewind_and_validate_with_errors( 67| 921| reinterpret_cast(input), 68| 921| reinterpret_cast(input) + count, length - count); 69| 921| res.count += count; 70| 921| return res; 71| 1.71k| } else { 72| 1.71k| return result(error_code::SUCCESS, length); 73| 1.71k| } 74| 2.63k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE0EEEKNS_6resultEPKDsm: 49| 2.75k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.75k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.75k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 2.74k] | | ------------------ ------------------ 51| 16| return result(error_code::SUCCESS, 0); 52| 16| } 53| | 54| 2.74k| const char16_t *start = input; 55| 2.74k| const char16_t *end = input + size; 56| | 57| 2.74k| const auto v_d8 = simd8::splat(0xd8); 58| 2.74k| const auto v_f8 = simd8::splat(0xf8); 59| 2.74k| const auto v_fc = simd8::splat(0xfc); 60| 2.74k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.33M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.33M, False: 2.53k] ------------------ 63| | // 0. Load data: since the validation takes into account only higher 64| | // byte of each word, we compress the two vectors into one which 65| | // consists only the higher bytes. 66| 2.33M| auto in0 = simd16(input); 67| 2.33M| auto in1 = 68| 2.33M| 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| 2.33M| 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| 2.33M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.33M| const uint16_t surrogates_bitmask = 77| 2.33M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.33M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.31M, False: 21.8k] ------------------ 79| 2.31M| input += 16; 80| 2.31M| } 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| 21.8k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 21.8k| const auto vH = (in & v_fc) == v_dc; 93| 21.8k| const uint16_t H = static_cast(vH.to_bitmask()); 94| | 95| | // L - word mask for low surrogates 96| | // L = not H and surrogates_wordmask 97| 21.8k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 21.8k| const uint16_t a = static_cast( 100| 21.8k| L & (H >> 1)); // A low surrogate must be followed by high one. 101| | // (A low surrogate placed in the 7th register's word 102| | // is an exception we handle.) 103| 21.8k| const uint16_t b = static_cast( 104| 21.8k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 21.8k| const uint16_t c = static_cast( 107| 21.8k| V | a | b); // Combine all the masks into the final one. 108| | 109| 21.8k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 16.8k, False: 5.06k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 16.8k| input += 16; 113| 16.8k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 4.86k, False: 204] ------------------ 114| | // The 15 lower code units of the input register contains valid UTF-16. 115| | // The 15th word may be either a low or high surrogate. It the next 116| | // iteration we 1) check if the low surrogate is followed by a high 117| | // one, 2) reject sole high surrogate. 118| 4.86k| input += 15; 119| 4.86k| } else { 120| 204| return result(error_code::SURROGATE, input - start); 121| 204| } 122| 21.8k| } 123| 2.33M| } 124| | 125| 2.53k| return result(error_code::SUCCESS, input - start); 126| 2.74k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE1EEEKNS_6resultEPKDsm: 49| 2.67k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.67k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.67k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 2.66k] | | ------------------ ------------------ 51| 16| return result(error_code::SUCCESS, 0); 52| 16| } 53| | 54| 2.66k| const char16_t *start = input; 55| 2.66k| const char16_t *end = input + size; 56| | 57| 2.66k| const auto v_d8 = simd8::splat(0xd8); 58| 2.66k| const auto v_f8 = simd8::splat(0xf8); 59| 2.66k| const auto v_fc = simd8::splat(0xfc); 60| 2.66k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.77M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.77M, False: 2.50k] ------------------ 63| | // 0. Load data: since the validation takes into account only higher 64| | // byte of each word, we compress the two vectors into one which 65| | // consists only the higher bytes. 66| 2.77M| auto in0 = simd16(input); 67| 2.77M| auto in1 = 68| 2.77M| 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| 2.77M| 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| 2.77M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.77M| const uint16_t surrogates_bitmask = 77| 2.77M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.77M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.75M, False: 18.5k] ------------------ 79| 2.75M| input += 16; 80| 2.75M| } else { 81| | // 2. We have some surrogates that have to be distinguished: 82| | // - low surrogates: 0b1101'10xx'yyyy'yyyy (0xD800..0xDBFF) 83| | // - high surrogates: 0b1101'11xx'yyyy'yyyy (0xDC00..0xDFFF) 84| | // 85| | // Fact: high surrogate has 11th bit set (3rd bit in the higher byte) 86| | 87| | // V - non-surrogate code units 88| | // V = not surrogates_wordmask 89| 18.5k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 18.5k| const auto vH = (in & v_fc) == v_dc; 93| 18.5k| const uint16_t H = static_cast(vH.to_bitmask()); 94| | 95| | // L - word mask for low surrogates 96| | // L = not H and surrogates_wordmask 97| 18.5k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 18.5k| const uint16_t a = static_cast( 100| 18.5k| L & (H >> 1)); // A low surrogate must be followed by high one. 101| | // (A low surrogate placed in the 7th register's word 102| | // is an exception we handle.) 103| 18.5k| const uint16_t b = static_cast( 104| 18.5k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 18.5k| const uint16_t c = static_cast( 107| 18.5k| V | a | b); // Combine all the masks into the final one. 108| | 109| 18.5k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 14.1k, False: 4.33k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 14.1k| input += 16; 113| 14.1k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 4.17k, False: 160] ------------------ 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| 4.17k| input += 15; 119| 4.17k| } else { 120| 160| return result(error_code::SURROGATE, input - start); 121| 160| } 122| 18.5k| } 123| 2.77M| } 124| | 125| 2.50k| return result(error_code::SUCCESS, input - start); 126| 2.66k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE0EEEKNS_6resultEPKDsm: 49| 2.75k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.75k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.75k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 2.74k] | | ------------------ ------------------ 51| 16| return result(error_code::SUCCESS, 0); 52| 16| } 53| | 54| 2.74k| const char16_t *start = input; 55| 2.74k| const char16_t *end = input + size; 56| | 57| 2.74k| const auto v_d8 = simd8::splat(0xd8); 58| 2.74k| const auto v_f8 = simd8::splat(0xf8); 59| 2.74k| const auto v_fc = simd8::splat(0xfc); 60| 2.74k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.33M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.33M, False: 2.43k] ------------------ 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| 2.33M| auto in0 = simd16(input); 67| 2.33M| auto in1 = 68| 2.33M| 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| 2.33M| 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| 2.33M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.33M| const uint16_t surrogates_bitmask = 77| 2.33M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.33M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.31M, False: 22.3k] ------------------ 79| 2.31M| input += 16; 80| 2.31M| } 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| 22.3k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 22.3k| const auto vH = (in & v_fc) == v_dc; 93| 22.3k| 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| 22.3k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 22.3k| const uint16_t a = static_cast( 100| 22.3k| 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| 22.3k| const uint16_t b = static_cast( 104| 22.3k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 22.3k| const uint16_t c = static_cast( 107| 22.3k| V | a | b); // Combine all the masks into the final one. 108| | 109| 22.3k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 17.1k, False: 5.23k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 17.1k| input += 16; 113| 17.1k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 4.93k, False: 302] ------------------ 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| 4.93k| input += 15; 119| 4.93k| } else { 120| 302| return result(error_code::SURROGATE, input - start); 121| 302| } 122| 22.3k| } 123| 2.33M| } 124| | 125| 2.43k| return result(error_code::SUCCESS, input - start); 126| 2.74k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf1626validate_utf16_with_errorsILNS_10endiannessE1EEEKNS_6resultEPKDsm: 49| 2.67k|const result validate_utf16_with_errors(const char16_t *input, size_t size) { 50| 2.67k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.67k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 2.66k] | | ------------------ ------------------ 51| 16| return result(error_code::SUCCESS, 0); 52| 16| } 53| | 54| 2.66k| const char16_t *start = input; 55| 2.66k| const char16_t *end = input + size; 56| | 57| 2.66k| const auto v_d8 = simd8::splat(0xd8); 58| 2.66k| const auto v_f8 = simd8::splat(0xf8); 59| 2.66k| const auto v_fc = simd8::splat(0xfc); 60| 2.66k| const auto v_dc = simd8::splat(0xdc); 61| | 62| 2.77M| while (input + simd16::SIZE * 2 < end) { ------------------ | Branch (62:10): [True: 2.77M, False: 2.39k] ------------------ 63| | // 0. Load data: since the validation takes into account only higher 64| | // byte of each word, we compress the two vectors into one which 65| | // consists only the higher bytes. 66| 2.77M| auto in0 = simd16(input); 67| 2.77M| auto in1 = 68| 2.77M| 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| 2.77M| 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| 2.77M| const auto surrogates_wordmask = (in & v_f8) == v_d8; 76| 2.77M| const uint16_t surrogates_bitmask = 77| 2.77M| static_cast(surrogates_wordmask.to_bitmask()); 78| 2.77M| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (78:9): [True: 2.75M, False: 19.0k] ------------------ 79| 2.75M| input += 16; 80| 2.75M| } 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| 19.0k| const uint16_t V = static_cast(~surrogates_bitmask); 90| | 91| | // H - word-mask for high surrogates: the six highest bits are 0b1101'11 92| 19.0k| const auto vH = (in & v_fc) == v_dc; 93| 19.0k| const uint16_t H = static_cast(vH.to_bitmask()); 94| | 95| | // L - word mask for low surrogates 96| | // L = not H and surrogates_wordmask 97| 19.0k| const uint16_t L = static_cast(~H & surrogates_bitmask); 98| | 99| 19.0k| const uint16_t a = static_cast( 100| 19.0k| L & (H >> 1)); // A low surrogate must be followed by high one. 101| | // (A low surrogate placed in the 7th register's word 102| | // is an exception we handle.) 103| 19.0k| const uint16_t b = static_cast( 104| 19.0k| a << 1); // Just mark that the opinput - startite fact is hold, 105| | // thanks to that we have only two masks for valid case. 106| 19.0k| const uint16_t c = static_cast( 107| 19.0k| V | a | b); // Combine all the masks into the final one. 108| | 109| 19.0k| if (c == 0xffff) { ------------------ | Branch (109:11): [True: 14.4k, False: 4.55k] ------------------ 110| | // The whole input register contains valid UTF-16, i.e., 111| | // either single code units or proper surrogate pairs. 112| 14.4k| input += 16; 113| 14.4k| } else if (c == 0x7fff) { ------------------ | Branch (113:18): [True: 4.28k, False: 266] ------------------ 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| 4.28k| input += 15; 119| 4.28k| } else { 120| 266| return result(error_code::SURROGATE, input - start); 121| 266| } 122| 19.0k| } 123| 2.77M| } 124| | 125| 2.39k| return result(error_code::SUCCESS, input - start); 126| 2.66k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf328validateEPKDim: 6| 2.57k|simdutf_really_inline bool validate(const char32_t *input, size_t size) { 7| 2.57k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.57k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 22, False: 2.54k] | | ------------------ ------------------ 8| | // empty input is valid UTF-32. protect the implementation from 9| | // handling nullptr 10| 22| return true; 11| 22| } 12| | 13| 2.54k| const char32_t *end = input + size; 14| | 15| 2.54k| using vector_u32 = simd32; 16| | 17| 2.54k| const auto standardmax = vector_u32::splat(0x10ffff); 18| 2.54k| const auto offset = vector_u32::splat(0xffff2000); 19| 2.54k| const auto standardoffsetmax = vector_u32::splat(0xfffff7ff); 20| 2.54k| auto currentmax = vector_u32::zero(); 21| 2.54k| auto currentoffsetmax = vector_u32::zero(); 22| | 23| 2.54k| constexpr size_t N = vector_u32::ELEMENTS; 24| | 25| 1.04M| while (input + N < end) { ------------------ | Branch (25:10): [True: 1.04M, False: 2.54k] ------------------ 26| 1.04M| auto in = vector_u32(input); 27| 1.04M| if constexpr (!match_system(endianness::BIG)) { 28| 1.04M| in.swap_bytes(); 29| 1.04M| } 30| | 31| 1.04M| currentmax = max(currentmax, in); 32| 1.04M| currentoffsetmax = max(currentoffsetmax, in + offset); 33| 1.04M| input += N; 34| 1.04M| } 35| | 36| 2.54k| const auto too_large = currentmax > standardmax; 37| 2.54k| if (too_large.any()) { ------------------ | Branch (37:7): [True: 942, False: 1.60k] ------------------ 38| 942| return false; 39| 942| } 40| | 41| 1.60k| const auto surrogate = currentoffsetmax > standardoffsetmax; 42| 1.60k| if (surrogate.any()) { ------------------ | Branch (42:7): [True: 27, False: 1.58k] ------------------ 43| 27| return false; 44| 27| } 45| | 46| 1.58k| return scalar::utf32::validate(input, end - input); 47| 1.60k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf3220validate_with_errorsEPKDim: 50| 2.57k| size_t size) { 51| 2.57k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.57k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 22, False: 2.54k] | | ------------------ ------------------ 52| | // empty input is valid UTF-32. protect the implementation from 53| | // handling nullptr 54| 22| return result(error_code::SUCCESS, 0); 55| 22| } 56| | 57| 2.54k| const char32_t *start = input; 58| 2.54k| const char32_t *end = input + size; 59| | 60| 2.54k| using vector_u32 = simd32; 61| | 62| 2.54k| const auto standardmax = vector_u32::splat(0x10ffff + 1); 63| 2.54k| const auto surrogate_mask = vector_u32::splat(0xfffff800); 64| 2.54k| const auto surrogate_byte = vector_u32::splat(0x0000d800); 65| | 66| 2.54k| constexpr size_t N = vector_u32::ELEMENTS; 67| | 68| 139k| while (input + N < end) { ------------------ | Branch (68:10): [True: 137k, False: 1.58k] ------------------ 69| 137k| auto in = vector_u32(input); 70| 137k| if constexpr (!match_system(endianness::BIG)) { 71| 137k| in.swap_bytes(); 72| 137k| } 73| | 74| 137k| const auto too_large = in >= standardmax; 75| 137k| const auto surrogate = (in & surrogate_mask) == surrogate_byte; 76| | 77| 137k| const auto combined = too_large | surrogate; 78| 137k| if (simdutf_unlikely(combined.any())) { ------------------ | | 93| 137k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 969, False: 136k] | | ------------------ ------------------ 79| 969| const size_t consumed = input - start; 80| 969| auto sr = scalar::utf32::validate_with_errors(input, end - input); 81| 969| sr.count += consumed; 82| | 83| 969| return sr; 84| 969| } 85| | 86| 136k| input += N; 87| 136k| } 88| | 89| 1.58k| const size_t consumed = input - start; 90| 1.58k| auto sr = scalar::utf32::validate_with_errors(input, end - input); 91| 1.58k| sr.count += consumed; 92| | 93| 1.58k| return sr; 94| 2.54k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf328validateEPKDim: 6| 2.57k|simdutf_really_inline bool validate(const char32_t *input, size_t size) { 7| 2.57k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.57k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 22, False: 2.54k] | | ------------------ ------------------ 8| | // empty input is valid UTF-32. protect the implementation from 9| | // handling nullptr 10| 22| return true; 11| 22| } 12| | 13| 2.54k| const char32_t *end = input + size; 14| | 15| 2.54k| using vector_u32 = simd32; 16| | 17| 2.54k| const auto standardmax = vector_u32::splat(0x10ffff); 18| 2.54k| const auto offset = vector_u32::splat(0xffff2000); 19| 2.54k| const auto standardoffsetmax = vector_u32::splat(0xfffff7ff); 20| 2.54k| auto currentmax = vector_u32::zero(); 21| 2.54k| auto currentoffsetmax = vector_u32::zero(); 22| | 23| 2.54k| constexpr size_t N = vector_u32::ELEMENTS; 24| | 25| 2.09M| while (input + N < end) { ------------------ | Branch (25:10): [True: 2.08M, False: 2.54k] ------------------ 26| 2.08M| auto in = vector_u32(input); 27| 2.08M| if constexpr (!match_system(endianness::BIG)) { 28| 2.08M| in.swap_bytes(); 29| 2.08M| } 30| | 31| 2.08M| currentmax = max(currentmax, in); 32| 2.08M| currentoffsetmax = max(currentoffsetmax, in + offset); 33| 2.08M| input += N; 34| 2.08M| } 35| | 36| 2.54k| const auto too_large = currentmax > standardmax; 37| 2.54k| if (too_large.any()) { ------------------ | Branch (37:7): [True: 1.08k, False: 1.46k] ------------------ 38| 1.08k| return false; 39| 1.08k| } 40| | 41| 1.46k| const auto surrogate = currentoffsetmax > standardoffsetmax; 42| 1.46k| if (surrogate.any()) { ------------------ | Branch (42:7): [True: 41, False: 1.42k] ------------------ 43| 41| return false; 44| 41| } 45| | 46| 1.42k| return scalar::utf32::validate(input, end - input); 47| 1.46k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_15utf3220validate_with_errorsEPKDim: 50| 2.57k| size_t size) { 51| 2.57k| if (simdutf_unlikely(size == 0)) { ------------------ | | 93| 2.57k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 22, False: 2.54k] | | ------------------ ------------------ 52| | // empty input is valid UTF-32. protect the implementation from 53| | // handling nullptr 54| 22| return result(error_code::SUCCESS, 0); 55| 22| } 56| | 57| 2.54k| const char32_t *start = input; 58| 2.54k| const char32_t *end = input + size; 59| | 60| 2.54k| using vector_u32 = simd32; 61| | 62| 2.54k| const auto standardmax = vector_u32::splat(0x10ffff + 1); 63| 2.54k| const auto surrogate_mask = vector_u32::splat(0xfffff800); 64| 2.54k| const auto surrogate_byte = vector_u32::splat(0x0000d800); 65| | 66| 2.54k| constexpr size_t N = vector_u32::ELEMENTS; 67| | 68| 276k| while (input + N < end) { ------------------ | Branch (68:10): [True: 275k, False: 1.42k] ------------------ 69| 275k| auto in = vector_u32(input); 70| 275k| if constexpr (!match_system(endianness::BIG)) { 71| 275k| in.swap_bytes(); 72| 275k| } 73| | 74| 275k| const auto too_large = in >= standardmax; 75| 275k| const auto surrogate = (in & surrogate_mask) == surrogate_byte; 76| | 77| 275k| const auto combined = too_large | surrogate; 78| 275k| if (simdutf_unlikely(combined.any())) { ------------------ | | 93| 275k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 1.12k, False: 274k] | | ------------------ ------------------ 79| 1.12k| const size_t consumed = input - start; 80| 1.12k| auto sr = scalar::utf32::validate_with_errors(input, end - input); 81| 1.12k| sr.count += consumed; 82| | 83| 1.12k| return sr; 84| 1.12k| } 85| | 86| 274k| input += N; 87| 274k| } 88| | 89| 1.42k| const size_t consumed = input - start; 90| 1.42k| auto sr = scalar::utf32::validate_with_errors(input, end - input); 91| 1.42k| sr.count += consumed; 92| | 93| 1.42k| return sr; 94| 2.54k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_latin1_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 88| char16_t *utf16_output) { 5| 88| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| | 7| 88| size_t i = 0; 8| 175k| for (; i < rounded_len; i += 16) { ------------------ | Branch (8:10): [True: 175k, False: 88] ------------------ 9| | // Load 16 bytes from the address (input + i) into a xmm register 10| 175k| const __m128i latin1 = 11| 175k| _mm_loadu_si128(reinterpret_cast(latin1_input + i)); 12| | 13| | // Zero extend each byte in `in` to word 14| 175k| __m256i utf16 = _mm256_cvtepu8_epi16(latin1); 15| | 16| 175k| if (big_endian) { ------------------ | Branch (16:9): [Folded, False: 175k] ------------------ 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| 175k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output + i), utf16); 25| 175k| } 26| | 27| 88| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 28| 88|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_latin1_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 110| char16_t *utf16_output) { 5| 110| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| | 7| 110| size_t i = 0; 8| 180k| for (; i < rounded_len; i += 16) { ------------------ | Branch (8:10): [True: 180k, False: 110] ------------------ 9| | // Load 16 bytes from the address (input + i) into a xmm register 10| 180k| const __m128i latin1 = 11| 180k| _mm_loadu_si128(reinterpret_cast(latin1_input + i)); 12| | 13| | // Zero extend each byte in `in` to word 14| 180k| __m256i utf16 = _mm256_cvtepu8_epi16(latin1); 15| | 16| 180k| if (big_endian) { ------------------ | Branch (16:9): [True: 180k, Folded] ------------------ 17| 180k| const __m128i swap128 = 18| 180k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 19| 180k| const __m256i swap = _mm256_set_m128i(swap128, swap128); ------------------ | | 4| 180k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 180k| _mm256_castsi128_si256(xmm2), 2) ------------------ 20| 180k| utf16 = _mm256_shuffle_epi8(utf16, swap); 21| 180k| } 22| | 23| | // Store the contents of xmm1 into the address pointed by (output + i) 24| 180k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output + i), utf16); 25| 180k| } 26| | 27| 110| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 28| 110|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_latin1_to_utf32EPKcmPDi: 3| 100| char32_t *utf32_output) { 4| 100| 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: 100] ------------------ 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| 100| return std::make_pair(buf + rounded_len, utf32_output + rounded_len); 20| 100|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_latin1_to_utf8EPKcmPc: 3| 484| char *utf8_output) { 4| 484| const char *end = latin1_input + len; 5| 484| const __m256i v_0000 = _mm256_setzero_si256(); 6| 484| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 7| 484| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 8| 484| const size_t safety_margin = 12; 9| | 10| 2.12M| while (end - latin1_input >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (10:10): [True: 2.12M, False: 484] ------------------ 11| 2.12M| __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| 2.12M| const __m128i v_80 = _mm_set1_epi8((char)0x80); 14| 2.12M| if (_mm_testz_si128(in8, v_80)) { // ASCII fast path!!!! ------------------ | Branch (14:9): [True: 1.93M, False: 191k] ------------------ 15| | // 1. store (16 bytes) 16| 1.93M| _mm_storeu_si128((__m128i *)utf8_output, in8); 17| | // 2. adjust pointers 18| 1.93M| latin1_input += 16; 19| 1.93M| utf8_output += 16; 20| 1.93M| continue; // we are done for this round! 21| 1.93M| } 22| | // We proceed only with the first 16 bytes. 23| 191k| 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| 191k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 29| 191k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 30| | 31| | // t0 = [0000|00aa|bbbb|bb00] 32| 191k| const __m256i t0 = _mm256_slli_epi16(in, 2); 33| | // t1 = [0000|00aa|0000|0000] 34| 191k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 35| | // t2 = [0000|0000|00bb|bbbb] 36| 191k| const __m256i t2 = _mm256_and_si256(in, v_003f); 37| | // t3 = [000a|aaaa|00bb|bbbb] 38| 191k| const __m256i t3 = _mm256_or_si256(t1, t2); 39| | // t4 = [1100|00aa|10bb|bbbb] 40| 191k| 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| 191k| const __m256i one_byte_bytemask = 46| 191k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 47| 191k| const uint32_t one_byte_bitmask = 48| 191k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 49| | 50| 191k| const __m256i utf8_unpacked = _mm256_blendv_epi8(t4, in, one_byte_bytemask); 51| | 52| | // 3. prepare bitmask for 8-bit lookup 53| 191k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 54| 191k| const uint32_t M1 = M0 >> 7; 55| 191k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 56| | // 4. pack the bytes 57| | 58| 191k| const uint8_t *row = 59| 191k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 60| 191k| const uint8_t *row_2 = 61| 191k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 16)] 62| 191k| [0]; 63| | 64| 191k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 65| 191k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 66| | 67| 191k| const __m256i utf8_packed = _mm256_shuffle_epi8( 68| 191k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 191k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 191k| _mm256_castsi128_si256(xmm2), 2) ------------------ 69| | // 5. store bytes 70| 191k| _mm_storeu_si128((__m128i *)utf8_output, 71| 191k| _mm256_castsi256_si128(utf8_packed)); 72| 191k| utf8_output += row[0]; 73| 191k| _mm_storeu_si128((__m128i *)utf8_output, 74| 191k| _mm256_extractf128_si256(utf8_packed, 1)); 75| 191k| utf8_output += row_2[0]; 76| | 77| | // 6. adjust pointers 78| 191k| latin1_input += 16; 79| 191k| continue; 80| | 81| 2.12M| } // while 82| 484| return std::make_pair(latin1_input, utf8_output); 83| 484|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_utf16_to_latin1ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 150| char *latin1_output) { 5| 150| const char16_t *end = buf + len; 6| 8.42k| while (end - buf >= 32) { ------------------ | Branch (6:10): [True: 8.31k, False: 108] ------------------ 7| | // Load 16 UTF-16 characters into 256-bit AVX2 register 8| 8.31k| __m256i in0 = _mm256_loadu_si256(reinterpret_cast(buf)); 9| 8.31k| __m256i in1 = 10| 8.31k| _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| 8.31k| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 21| 8.31k| if (_mm256_testz_si256(_mm256_or_si256(in0, in1), high_byte_mask)) { ------------------ | Branch (21:9): [True: 8.27k, False: 42] ------------------ 22| | // Pack 16-bit characters into 8-bit and store in latin1_output 23| 8.27k| const __m256i packed = _mm256_packus_epi16(in0, in1); 24| | 25| 8.27k| const __m256i result = _mm256_permute4x64_epi64(packed, 0b11011000); 26| | 27| 8.27k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(latin1_output), result); 28| | // Adjust pointers for the next iteration 29| 8.27k| buf += 32; 30| 8.27k| latin1_output += 32; 31| 8.27k| } else { 32| 42| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 33| 42| } 34| 8.31k| } // while 35| 108| return std::make_pair(buf, latin1_output); 36| 150|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_utf16_to_latin1ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 112| char *latin1_output) { 5| 112| const char16_t *end = buf + len; 6| 592| while (end - buf >= 32) { ------------------ | Branch (6:10): [True: 498, False: 94] ------------------ 7| | // Load 16 UTF-16 characters into 256-bit AVX2 register 8| 498| __m256i in0 = _mm256_loadu_si256(reinterpret_cast(buf)); 9| 498| __m256i in1 = 10| 498| _mm256_loadu_si256(reinterpret_cast(buf + 16)); 11| | 12| 498| if constexpr (!match_system(big_endian)) { 13| 498| const __m256i swap = _mm256_setr_epi8( 14| 498| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 15| 498| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 16| 498| in0 = _mm256_shuffle_epi8(in0, swap); 17| 498| in1 = _mm256_shuffle_epi8(in1, swap); 18| 498| } 19| | 20| 498| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 21| 498| if (_mm256_testz_si256(_mm256_or_si256(in0, in1), high_byte_mask)) { ------------------ | Branch (21:9): [True: 480, False: 18] ------------------ 22| | // Pack 16-bit characters into 8-bit and store in latin1_output 23| 480| const __m256i packed = _mm256_packus_epi16(in0, in1); 24| | 25| 480| const __m256i result = _mm256_permute4x64_epi64(packed, 0b11011000); 26| | 27| 480| _mm256_storeu_si256(reinterpret_cast<__m256i *>(latin1_output), result); 28| | // Adjust pointers for the next iteration 29| 480| buf += 32; 30| 480| latin1_output += 32; 31| 480| } else { 32| 18| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 33| 18| } 34| 498| } // while 35| 94| return std::make_pair(buf, latin1_output); 36| 112|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_140avx2_convert_utf16_to_latin1_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 41| 226| char *latin1_output) { 42| 226| const char16_t *start = buf; 43| 226| const char16_t *end = buf + len; 44| 393k| while (end - buf >= 16) { ------------------ | Branch (44:10): [True: 393k, False: 166] ------------------ 45| 393k| __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| 393k| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 55| 393k| if (_mm256_testz_si256(in, high_byte_mask)) { ------------------ | Branch (55:9): [True: 393k, False: 60] ------------------ 56| 393k| __m128i lo = _mm256_extractf128_si256(in, 0); 57| 393k| __m128i hi = _mm256_extractf128_si256(in, 1); 58| 393k| __m128i latin1_packed_lo = _mm_packus_epi16(lo, lo); 59| 393k| __m128i latin1_packed_hi = _mm_packus_epi16(hi, hi); 60| 393k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 61| 393k| latin1_packed_lo); 62| 393k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output + 8), 63| 393k| latin1_packed_hi); 64| 393k| buf += 16; 65| 393k| latin1_output += 16; 66| 393k| } else { 67| | // Fallback to scalar code for handling errors 68| 394| for (int k = 0; k < 16; k++) { ------------------ | Branch (68:23): [True: 394, False: 0] ------------------ 69| 394| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 70| 394| if (word <= 0xff) { ------------------ | Branch (70:13): [True: 334, False: 60] ------------------ 71| 334| *latin1_output++ = char(word); 72| 334| } else { 73| 60| return std::make_pair( 74| 60| result{error_code::TOO_LARGE, (size_t)(buf - start + k)}, 75| 60| latin1_output); 76| 60| } 77| 394| } 78| 0| buf += 16; 79| 0| } 80| 393k| } // while 81| 166| return std::make_pair(result{error_code::SUCCESS, (size_t)(buf - start)}, 82| 166| latin1_output); 83| 226|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_140avx2_convert_utf16_to_latin1_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 41| 230| char *latin1_output) { 42| 230| const char16_t *start = buf; 43| 230| const char16_t *end = buf + len; 44| 457k| while (end - buf >= 16) { ------------------ | Branch (44:10): [True: 457k, False: 136] ------------------ 45| 457k| __m256i in = _mm256_loadu_si256(reinterpret_cast(buf)); 46| | 47| 457k| if constexpr (!match_system(big_endian)) { 48| 457k| const __m256i swap = _mm256_setr_epi8( 49| 457k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 50| 457k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 51| 457k| in = _mm256_shuffle_epi8(in, swap); 52| 457k| } 53| | 54| 457k| __m256i high_byte_mask = _mm256_set1_epi16((int16_t)0xFF00); 55| 457k| if (_mm256_testz_si256(in, high_byte_mask)) { ------------------ | Branch (55:9): [True: 457k, False: 94] ------------------ 56| 457k| __m128i lo = _mm256_extractf128_si256(in, 0); 57| 457k| __m128i hi = _mm256_extractf128_si256(in, 1); 58| 457k| __m128i latin1_packed_lo = _mm_packus_epi16(lo, lo); 59| 457k| __m128i latin1_packed_hi = _mm_packus_epi16(hi, hi); 60| 457k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 61| 457k| latin1_packed_lo); 62| 457k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output + 8), 63| 457k| latin1_packed_hi); 64| 457k| buf += 16; 65| 457k| latin1_output += 16; 66| 457k| } else { 67| | // Fallback to scalar code for handling errors 68| 464| for (int k = 0; k < 16; k++) { ------------------ | Branch (68:23): [True: 464, False: 0] ------------------ 69| 464| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 70| 464| if (word <= 0xff) { ------------------ | Branch (70:13): [True: 370, False: 94] ------------------ 71| 370| *latin1_output++ = char(word); 72| 370| } else { 73| 94| return std::make_pair( 74| 94| result{error_code::TOO_LARGE, (size_t)(buf - start + k)}, 75| 94| latin1_output); 76| 94| } 77| 464| } 78| 0| buf += 16; 79| 0| } 80| 457k| } // while 81| 136| return std::make_pair(result{error_code::SUCCESS, (size_t)(buf - start)}, 82| 136| latin1_output); 83| 230|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf16_to_utf32ILNS_10endiannessE0EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 428| char32_t *utf32_output) { 58| 428| const char16_t *end = buf + len; 59| 428| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 428| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| | 62| 79.5k| while (end - buf >= 16) { ------------------ | Branch (62:10): [True: 79.2k, False: 318] ------------------ 63| 79.2k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 64| 79.2k| if (big_endian) { ------------------ | Branch (64:9): [Folded, False: 79.2k] ------------------ 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| 79.2k| const __m256i surrogates_bytemask = 75| 79.2k| _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| 79.2k| const uint32_t surrogates_bitmask = 80| 79.2k| 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| 79.2k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (83:9): [True: 74.4k, False: 4.75k] ------------------ 84| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 85| | // units 86| 74.4k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 87| 74.4k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 88| 74.4k| _mm256_storeu_si256( 89| 74.4k| reinterpret_cast<__m256i *>(utf32_output + 8), 90| 74.4k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 91| 74.4k| utf32_output += 16; 92| 74.4k| buf += 16; 93| | // surrogate pair(s) in a register 94| 74.4k| } else { 95| | // Let us do a scalar fallback. 96| | // It may seem wasteful to use scalar code, but being efficient with SIMD 97| | // in the presence of surrogate pairs may require non-trivial tables. 98| 4.75k| size_t forward = 15; 99| 4.75k| size_t k = 0; 100| 4.75k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (100:11): [True: 0, False: 4.75k] ------------------ 101| 0| forward = size_t(end - buf - 1); 102| 0| } 103| 69.7k| for (; k < forward; k++) { ------------------ | Branch (103:14): [True: 65.1k, False: 4.64k] ------------------ 104| 65.1k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 105| 65.1k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (105:13): [True: 59.5k, False: 5.56k] ------------------ 106| | // No surrogate pair 107| 59.5k| *utf32_output++ = char32_t(word); 108| 59.5k| } else { 109| | // must be a surrogate pair 110| 5.56k| uint16_t diff = uint16_t(word - 0xD800); 111| 5.56k| uint16_t next_word = 112| 5.56k| scalar::utf16::swap_if_needed(buf[k + 1]); 113| 5.56k| k++; 114| 5.56k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 115| 5.56k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (115:15): [True: 110, False: 5.45k] ------------------ 116| 110| return std::make_pair(nullptr, utf32_output); 117| 110| } 118| 5.45k| uint32_t value = (diff << 10) + diff2 + 0x10000; 119| 5.45k| *utf32_output++ = char32_t(value); 120| 5.45k| } 121| 65.1k| } 122| 4.64k| buf += k; 123| 4.64k| } 124| 79.2k| } // while 125| 318| return std::make_pair(buf, utf32_output); 126| 428|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf16_to_utf32ILNS_10endiannessE1EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 448| char32_t *utf32_output) { 58| 448| const char16_t *end = buf + len; 59| 448| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 448| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| | 62| 379k| while (end - buf >= 16) { ------------------ | Branch (62:10): [True: 379k, False: 374] ------------------ 63| 379k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 64| 379k| if (big_endian) { ------------------ | Branch (64:9): [True: 379k, Folded] ------------------ 65| 379k| const __m256i swap = _mm256_setr_epi8( 66| 379k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 67| 379k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 68| 379k| in = _mm256_shuffle_epi8(in, swap); 69| 379k| } 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| 379k| const __m256i surrogates_bytemask = 75| 379k| _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| 379k| const uint32_t surrogates_bitmask = 80| 379k| 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| 379k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (83:9): [True: 377k, False: 1.86k] ------------------ 84| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 85| | // units 86| 377k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 87| 377k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 88| 377k| _mm256_storeu_si256( 89| 377k| reinterpret_cast<__m256i *>(utf32_output + 8), 90| 377k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 91| 377k| utf32_output += 16; 92| 377k| buf += 16; 93| | // surrogate pair(s) in a register 94| 377k| } 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| 1.86k| size_t forward = 15; 99| 1.86k| size_t k = 0; 100| 1.86k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (100:11): [True: 0, False: 1.86k] ------------------ 101| 0| forward = size_t(end - buf - 1); 102| 0| } 103| 26.2k| for (; k < forward; k++) { ------------------ | Branch (103:14): [True: 24.4k, False: 1.79k] ------------------ 104| 24.4k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 105| 24.4k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (105:13): [True: 21.4k, False: 3.06k] ------------------ 106| | // No surrogate pair 107| 21.4k| *utf32_output++ = char32_t(word); 108| 21.4k| } else { 109| | // must be a surrogate pair 110| 3.06k| uint16_t diff = uint16_t(word - 0xD800); 111| 3.06k| uint16_t next_word = 112| 3.06k| scalar::utf16::swap_if_needed(buf[k + 1]); 113| 3.06k| k++; 114| 3.06k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 115| 3.06k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (115:15): [True: 74, False: 2.99k] ------------------ 116| 74| return std::make_pair(nullptr, utf32_output); 117| 74| } 118| 2.99k| uint32_t value = (diff << 10) + diff2 + 0x10000; 119| 2.99k| *utf32_output++ = char32_t(value); 120| 2.99k| } 121| 24.4k| } 122| 1.79k| buf += k; 123| 1.79k| } 124| 379k| } // while 125| 374| return std::make_pair(buf, utf32_output); 126| 448|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf16_to_utf32_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 138| 402| char32_t *utf32_output) { 139| 402| const char16_t *start = buf; 140| 402| const char16_t *end = buf + len; 141| 402| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 142| 402| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 143| | 144| 158k| while (end - buf >= 16) { ------------------ | Branch (144:10): [True: 157k, False: 284] ------------------ 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: 151k, False: 6.37k] ------------------ 166| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 167| | // units 168| 151k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 169| 151k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 170| 151k| _mm256_storeu_si256( 171| 151k| reinterpret_cast<__m256i *>(utf32_output + 8), 172| 151k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 173| 151k| utf32_output += 16; 174| 151k| buf += 16; 175| | // surrogate pair(s) in a register 176| 151k| } 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| 6.37k| size_t forward = 15; 181| 6.37k| size_t k = 0; 182| 6.37k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (182:11): [True: 0, False: 6.37k] ------------------ 183| 0| forward = size_t(end - buf - 1); 184| 0| } 185| 90.9k| for (; k < forward; k++) { ------------------ | Branch (185:14): [True: 84.6k, False: 6.25k] ------------------ 186| 84.6k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 187| 84.6k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (187:13): [True: 74.0k, False: 10.6k] ------------------ 188| | // No surrogate pair 189| 74.0k| *utf32_output++ = char32_t(word); 190| 74.0k| } else { 191| | // must be a surrogate pair 192| 10.6k| uint16_t diff = uint16_t(word - 0xD800); 193| 10.6k| uint16_t next_word = 194| 10.6k| scalar::utf16::swap_if_needed(buf[k + 1]); 195| 10.6k| k++; 196| 10.6k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 197| 10.6k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (197:15): [True: 118, False: 10.4k] ------------------ 198| 118| return std::make_pair( 199| 118| result(error_code::SURROGATE, buf - start + k - 1), 200| 118| utf32_output); 201| 118| } 202| 10.4k| uint32_t value = (diff << 10) + diff2 + 0x10000; 203| 10.4k| *utf32_output++ = char32_t(value); 204| 10.4k| } 205| 84.6k| } 206| 6.25k| buf += k; 207| 6.25k| } 208| 157k| } // while 209| 284| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 210| 402|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf16_to_utf32_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 138| 422| char32_t *utf32_output) { 139| 422| const char16_t *start = buf; 140| 422| const char16_t *end = buf + len; 141| 422| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 142| 422| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 143| | 144| 159k| while (end - buf >= 16) { ------------------ | Branch (144:10): [True: 159k, False: 294] ------------------ 145| 159k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 146| 159k| if (big_endian) { ------------------ | Branch (146:9): [True: 159k, Folded] ------------------ 147| 159k| const __m256i swap = _mm256_setr_epi8( 148| 159k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 149| 159k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 150| 159k| in = _mm256_shuffle_epi8(in, swap); 151| 159k| } 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| 159k| const __m256i surrogates_bytemask = 157| 159k| _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| 159k| const uint32_t surrogates_bitmask = 162| 159k| 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| 159k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (165:9): [True: 154k, False: 5.14k] ------------------ 166| | // case: we extend all sixteen 16-bit code units to sixteen 32-bit code 167| | // units 168| 154k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 169| 154k| _mm256_cvtepu16_epi32(_mm256_castsi256_si128(in))); 170| 154k| _mm256_storeu_si256( 171| 154k| reinterpret_cast<__m256i *>(utf32_output + 8), 172| 154k| _mm256_cvtepu16_epi32(_mm256_extractf128_si256(in, 1))); 173| 154k| utf32_output += 16; 174| 154k| buf += 16; 175| | // surrogate pair(s) in a register 176| 154k| } else { 177| | // Let us do a scalar fallback. 178| | // It may seem wasteful to use scalar code, but being efficient with SIMD 179| | // in the presence of surrogate pairs may require non-trivial tables. 180| 5.14k| size_t forward = 15; 181| 5.14k| size_t k = 0; 182| 5.14k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (182:11): [True: 0, False: 5.14k] ------------------ 183| 0| forward = size_t(end - buf - 1); 184| 0| } 185| 71.1k| for (; k < forward; k++) { ------------------ | Branch (185:14): [True: 66.1k, False: 5.02k] ------------------ 186| 66.1k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 187| 66.1k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (187:13): [True: 55.1k, False: 10.9k] ------------------ 188| | // No surrogate pair 189| 55.1k| *utf32_output++ = char32_t(word); 190| 55.1k| } else { 191| | // must be a surrogate pair 192| 10.9k| uint16_t diff = uint16_t(word - 0xD800); 193| 10.9k| uint16_t next_word = 194| 10.9k| scalar::utf16::swap_if_needed(buf[k + 1]); 195| 10.9k| k++; 196| 10.9k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 197| 10.9k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (197:15): [True: 128, False: 10.8k] ------------------ 198| 128| return std::make_pair( 199| 128| result(error_code::SURROGATE, buf - start + k - 1), 200| 128| utf32_output); 201| 128| } 202| 10.8k| uint32_t value = (diff << 10) + diff2 + 0x10000; 203| 10.8k| *utf32_output++ = char32_t(value); 204| 10.8k| } 205| 66.1k| } 206| 5.02k| buf += k; 207| 5.02k| } 208| 159k| } // while 209| 294| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 210| 422|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_126avx2_convert_utf16_to_utf8ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 876|avx2_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| 876| const char16_t *end = buf + len; 58| 876| const __m256i v_0000 = _mm256_setzero_si256(); 59| 876| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 876| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| 876| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 62| 876| const size_t safety_margin = 63| 876| 12; // to avoid overruns, see issue 64| | // https://github.com/simdutf/simdutf/issues/92 65| | 66| 894k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (66:10): [True: 893k, False: 708] ------------------ 67| 893k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 68| 893k| if (big_endian) { ------------------ | Branch (68:9): [Folded, False: 893k] ------------------ 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| 893k| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 76| 893k| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 758k, False: 134k] ------------------ 77| | // 1. pack the bytes 78| 758k| const __m128i utf8_packed = _mm_packus_epi16( 79| 758k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 80| | // 2. store (16 bytes) 81| 758k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 82| | // 3. adjust pointers 83| 758k| buf += 16; 84| 758k| utf8_output += 16; 85| 758k| continue; // we are done for this round! 86| 758k| } 87| | // no bits set above 7th bit 88| 134k| const __m256i one_byte_bytemask = 89| 134k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 90| 134k| const uint32_t one_byte_bitmask = 91| 134k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 92| | 93| | // no bits set above 11th bit 94| 134k| const __m256i one_or_two_bytes_bytemask = 95| 134k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 96| 134k| const uint32_t one_or_two_bytes_bitmask = 97| 134k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 98| 134k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (98:9): [True: 48.8k, False: 85.7k] ------------------ 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| 48.8k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 104| 48.8k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 105| | 106| | // t0 = [000a|aaaa|bbbb|bb00] 107| 48.8k| const __m256i t0 = _mm256_slli_epi16(in, 2); 108| | // t1 = [000a|aaaa|0000|0000] 109| 48.8k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 110| | // t2 = [0000|0000|00bb|bbbb] 111| 48.8k| const __m256i t2 = _mm256_and_si256(in, v_003f); 112| | // t3 = [000a|aaaa|00bb|bbbb] 113| 48.8k| const __m256i t3 = _mm256_or_si256(t1, t2); 114| | // t4 = [110a|aaaa|10bb|bbbb] 115| 48.8k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 116| | 117| | // 2. merge ASCII and 2-byte codewords 118| 48.8k| const __m256i utf8_unpacked = 119| 48.8k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 120| | 121| | // 3. prepare bitmask for 8-bit lookup 122| 48.8k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 123| 48.8k| const uint32_t M1 = M0 >> 7; 124| 48.8k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 125| | // 4. pack the bytes 126| | 127| 48.8k| const uint8_t *row = 128| 48.8k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 129| 48.8k| const uint8_t *row_2 = 130| 48.8k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 131| 48.8k| 16)][0]; 132| | 133| 48.8k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 134| 48.8k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 135| | 136| 48.8k| const __m256i utf8_packed = _mm256_shuffle_epi8( 137| 48.8k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 48.8k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 48.8k| _mm256_castsi128_si256(xmm2), 2) ------------------ 138| | // 5. store bytes 139| 48.8k| _mm_storeu_si128((__m128i *)utf8_output, 140| 48.8k| _mm256_castsi256_si128(utf8_packed)); 141| 48.8k| utf8_output += row[0]; 142| 48.8k| _mm_storeu_si128((__m128i *)utf8_output, 143| 48.8k| _mm256_extractf128_si256(utf8_packed, 1)); 144| 48.8k| utf8_output += row_2[0]; 145| | 146| | // 6. adjust pointers 147| 48.8k| buf += 16; 148| 48.8k| continue; 149| 48.8k| } 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| 85.7k| const __m256i surrogates_bytemask = 154| 85.7k| _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| 85.7k| const uint32_t surrogates_bitmask = 159| 85.7k| 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| 85.7k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (162:9): [True: 78.7k, False: 7.05k] ------------------ 163| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 164| 78.7k| const __m256i dup_even = _mm256_setr_epi16( 165| 78.7k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 166| 78.7k| 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| 78.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 196| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 197| 78.7k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 198| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 199| 78.7k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 195| 78.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 200| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 201| 78.7k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 195| 78.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 202| | 203| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 204| 78.7k| const __m256i s0 = _mm256_srli_epi16(in, 4); 205| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 206| 78.7k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 195| 78.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 207| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 208| 78.7k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 195| 78.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 209| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 210| 78.7k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 195| 78.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 211| 78.7k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 212| 78.7k| simdutf_vec(0b0100000000000000)); ------------------ | | 195| 78.7k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 213| 78.7k| const __m256i s4 = _mm256_xor_si256(s3, m0); 214| 78.7k|#undef simdutf_vec 215| | 216| | // 4. expand code units 16-bit => 32-bit 217| 78.7k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 218| 78.7k| 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| 78.7k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 222| 78.7k| (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| 78.7k| const uint8_t mask0 = uint8_t(mask); 243| 78.7k| const uint8_t *row0 = 244| 78.7k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 245| 78.7k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 246| 78.7k| const __m128i utf8_0 = 247| 78.7k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 248| | 249| 78.7k| const uint8_t mask1 = static_cast(mask >> 8); 250| 78.7k| const uint8_t *row1 = 251| 78.7k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 252| 78.7k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 253| 78.7k| const __m128i utf8_1 = 254| 78.7k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 255| | 256| 78.7k| const uint8_t mask2 = static_cast(mask >> 16); 257| 78.7k| const uint8_t *row2 = 258| 78.7k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 259| 78.7k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 260| 78.7k| const __m128i utf8_2 = 261| 78.7k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 262| | 263| 78.7k| const uint8_t mask3 = static_cast(mask >> 24); 264| 78.7k| const uint8_t *row3 = 265| 78.7k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 266| 78.7k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 267| 78.7k| const __m128i utf8_3 = 268| 78.7k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 269| | 270| 78.7k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 271| 78.7k| utf8_output += row0[0]; 272| 78.7k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 273| 78.7k| utf8_output += row1[0]; 274| 78.7k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 275| 78.7k| utf8_output += row2[0]; 276| 78.7k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 277| 78.7k| utf8_output += row3[0]; 278| 78.7k| buf += 16; 279| | // surrogate pair(s) in a register 280| 78.7k| } 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| 7.05k| size_t forward = 15; 285| 7.05k| size_t k = 0; 286| 7.05k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (286:11): [True: 0, False: 7.05k] ------------------ 287| 0| forward = size_t(end - buf - 1); 288| 0| } 289| 103k| for (; k < forward; k++) { ------------------ | Branch (289:14): [True: 96.4k, False: 6.88k] ------------------ 290| 96.4k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 291| 96.4k| if ((word & 0xFF80) == 0) { ------------------ | Branch (291:13): [True: 13.8k, False: 82.6k] ------------------ 292| 13.8k| *utf8_output++ = char(word); 293| 82.6k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (293:20): [True: 5.32k, False: 77.3k] ------------------ 294| 5.32k| *utf8_output++ = char((word >> 6) | 0b11000000); 295| 5.32k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 296| 77.3k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (296:20): [True: 68.8k, False: 8.47k] ------------------ 297| 68.8k| *utf8_output++ = char((word >> 12) | 0b11100000); 298| 68.8k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 299| 68.8k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 300| 68.8k| } else { 301| | // must be a surrogate pair 302| 8.47k| uint16_t diff = uint16_t(word - 0xD800); 303| 8.47k| uint16_t next_word = 304| 8.47k| scalar::utf16::swap_if_needed(buf[k + 1]); 305| 8.47k| k++; 306| 8.47k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 307| 8.47k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (307:15): [True: 168, False: 8.30k] ------------------ 308| 168| return std::make_pair(nullptr, utf8_output); 309| 168| } 310| 8.30k| uint32_t value = (diff << 10) + diff2 + 0x10000; 311| 8.30k| *utf8_output++ = char((value >> 18) | 0b11110000); 312| 8.30k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 313| 8.30k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 314| 8.30k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 315| 8.30k| } 316| 96.4k| } 317| 6.88k| buf += k; 318| 6.88k| } 319| 85.7k| } // while 320| 708| return std::make_pair(buf, utf8_output); 321| 876|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_126avx2_convert_utf16_to_utf8ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 772|avx2_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| 772| const char16_t *end = buf + len; 58| 772| const __m256i v_0000 = _mm256_setzero_si256(); 59| 772| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 60| 772| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 61| 772| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 62| 772| const size_t safety_margin = 63| 772| 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: 694] ------------------ 67| 1.09M| __m256i in = _mm256_loadu_si256((__m256i *)buf); 68| 1.09M| if (big_endian) { ------------------ | Branch (68:9): [True: 1.09M, Folded] ------------------ 69| 1.09M| const __m256i swap = _mm256_setr_epi8( 70| 1.09M| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 71| 1.09M| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 72| 1.09M| in = _mm256_shuffle_epi8(in, swap); 73| 1.09M| } 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: 777k, False: 316k] ------------------ 77| | // 1. pack the bytes 78| 777k| const __m128i utf8_packed = _mm_packus_epi16( 79| 777k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 80| | // 2. store (16 bytes) 81| 777k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 82| | // 3. adjust pointers 83| 777k| buf += 16; 84| 777k| utf8_output += 16; 85| 777k| continue; // we are done for this round! 86| 777k| } 87| | // no bits set above 7th bit 88| 316k| const __m256i one_byte_bytemask = 89| 316k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 90| 316k| const uint32_t one_byte_bitmask = 91| 316k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 92| | 93| | // no bits set above 11th bit 94| 316k| const __m256i one_or_two_bytes_bytemask = 95| 316k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 96| 316k| const uint32_t one_or_two_bytes_bitmask = 97| 316k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 98| 316k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (98:9): [True: 30.3k, False: 286k] ------------------ 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| 30.3k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 104| 30.3k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 105| | 106| | // t0 = [000a|aaaa|bbbb|bb00] 107| 30.3k| const __m256i t0 = _mm256_slli_epi16(in, 2); 108| | // t1 = [000a|aaaa|0000|0000] 109| 30.3k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 110| | // t2 = [0000|0000|00bb|bbbb] 111| 30.3k| const __m256i t2 = _mm256_and_si256(in, v_003f); 112| | // t3 = [000a|aaaa|00bb|bbbb] 113| 30.3k| const __m256i t3 = _mm256_or_si256(t1, t2); 114| | // t4 = [110a|aaaa|10bb|bbbb] 115| 30.3k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 116| | 117| | // 2. merge ASCII and 2-byte codewords 118| 30.3k| const __m256i utf8_unpacked = 119| 30.3k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 120| | 121| | // 3. prepare bitmask for 8-bit lookup 122| 30.3k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 123| 30.3k| const uint32_t M1 = M0 >> 7; 124| 30.3k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 125| | // 4. pack the bytes 126| | 127| 30.3k| const uint8_t *row = 128| 30.3k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 129| 30.3k| const uint8_t *row_2 = 130| 30.3k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 131| 30.3k| 16)][0]; 132| | 133| 30.3k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 134| 30.3k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 135| | 136| 30.3k| const __m256i utf8_packed = _mm256_shuffle_epi8( 137| 30.3k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 30.3k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 30.3k| _mm256_castsi128_si256(xmm2), 2) ------------------ 138| | // 5. store bytes 139| 30.3k| _mm_storeu_si128((__m128i *)utf8_output, 140| 30.3k| _mm256_castsi256_si128(utf8_packed)); 141| 30.3k| utf8_output += row[0]; 142| 30.3k| _mm_storeu_si128((__m128i *)utf8_output, 143| 30.3k| _mm256_extractf128_si256(utf8_packed, 1)); 144| 30.3k| utf8_output += row_2[0]; 145| | 146| | // 6. adjust pointers 147| 30.3k| buf += 16; 148| 30.3k| continue; 149| 30.3k| } 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| 286k| const __m256i surrogates_bytemask = 154| 286k| _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| 286k| const uint32_t surrogates_bitmask = 159| 286k| 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| 286k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (162:9): [True: 279k, False: 6.23k] ------------------ 163| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 164| 279k| const __m256i dup_even = _mm256_setr_epi16( 165| 279k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 166| 279k| 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| 279k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 196| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 197| 279k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 198| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 199| 279k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 195| 279k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 200| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 201| 279k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 195| 279k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 202| | 203| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 204| 279k| const __m256i s0 = _mm256_srli_epi16(in, 4); 205| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 206| 279k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 195| 279k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 207| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 208| 279k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 195| 279k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 209| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 210| 279k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 195| 279k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 211| 279k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 212| 279k| simdutf_vec(0b0100000000000000)); ------------------ | | 195| 279k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 213| 279k| const __m256i s4 = _mm256_xor_si256(s3, m0); 214| 279k|#undef simdutf_vec 215| | 216| | // 4. expand code units 16-bit => 32-bit 217| 279k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 218| 279k| 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| 279k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 222| 279k| (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| 279k| const uint8_t mask0 = uint8_t(mask); 243| 279k| const uint8_t *row0 = 244| 279k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 245| 279k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 246| 279k| const __m128i utf8_0 = 247| 279k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 248| | 249| 279k| const uint8_t mask1 = static_cast(mask >> 8); 250| 279k| const uint8_t *row1 = 251| 279k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 252| 279k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 253| 279k| const __m128i utf8_1 = 254| 279k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 255| | 256| 279k| const uint8_t mask2 = static_cast(mask >> 16); 257| 279k| const uint8_t *row2 = 258| 279k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 259| 279k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 260| 279k| const __m128i utf8_2 = 261| 279k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 262| | 263| 279k| const uint8_t mask3 = static_cast(mask >> 24); 264| 279k| const uint8_t *row3 = 265| 279k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 266| 279k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 267| 279k| const __m128i utf8_3 = 268| 279k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 269| | 270| 279k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 271| 279k| utf8_output += row0[0]; 272| 279k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 273| 279k| utf8_output += row1[0]; 274| 279k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 275| 279k| utf8_output += row2[0]; 276| 279k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 277| 279k| utf8_output += row3[0]; 278| 279k| buf += 16; 279| | // surrogate pair(s) in a register 280| 279k| } 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| 6.23k| size_t forward = 15; 285| 6.23k| size_t k = 0; 286| 6.23k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (286:11): [True: 0, False: 6.23k] ------------------ 287| 0| forward = size_t(end - buf - 1); 288| 0| } 289| 92.0k| for (; k < forward; k++) { ------------------ | Branch (289:14): [True: 85.8k, False: 6.15k] ------------------ 290| 85.8k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 291| 85.8k| if ((word & 0xFF80) == 0) { ------------------ | Branch (291:13): [True: 14.3k, False: 71.4k] ------------------ 292| 14.3k| *utf8_output++ = char(word); 293| 71.4k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (293:20): [True: 2.82k, False: 68.6k] ------------------ 294| 2.82k| *utf8_output++ = char((word >> 6) | 0b11000000); 295| 2.82k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 296| 68.6k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (296:20): [True: 61.4k, False: 7.23k] ------------------ 297| 61.4k| *utf8_output++ = char((word >> 12) | 0b11100000); 298| 61.4k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 299| 61.4k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 300| 61.4k| } else { 301| | // must be a surrogate pair 302| 7.23k| uint16_t diff = uint16_t(word - 0xD800); 303| 7.23k| uint16_t next_word = 304| 7.23k| scalar::utf16::swap_if_needed(buf[k + 1]); 305| 7.23k| k++; 306| 7.23k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 307| 7.23k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (307:15): [True: 78, False: 7.15k] ------------------ 308| 78| return std::make_pair(nullptr, utf8_output); 309| 78| } 310| 7.15k| uint32_t value = (diff << 10) + diff2 + 0x10000; 311| 7.15k| *utf8_output++ = char((value >> 18) | 0b11110000); 312| 7.15k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 313| 7.15k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 314| 7.15k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 315| 7.15k| } 316| 85.8k| } 317| 6.15k| buf += k; 318| 6.15k| } 319| 286k| } // while 320| 694| return std::make_pair(buf, utf8_output); 321| 772|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_138avx2_convert_utf16_to_utf8_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 333| 654| char *utf8_output) { 334| 654| const char16_t *start = buf; 335| 654| const char16_t *end = buf + len; 336| | 337| 654| const __m256i v_0000 = _mm256_setzero_si256(); 338| 654| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 339| 654| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 340| 654| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 341| 654| const size_t safety_margin = 342| 654| 12; // to avoid overruns, see issue 343| | // https://github.com/simdutf/simdutf/issues/92 344| | 345| 647k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (345:10): [True: 646k, False: 504] ------------------ 346| 646k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 347| 646k| if (big_endian) { ------------------ | Branch (347:9): [Folded, False: 646k] ------------------ 348| 0| const __m256i swap = _mm256_setr_epi8( 349| 0| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 350| 0| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 351| 0| in = _mm256_shuffle_epi8(in, swap); 352| 0| } 353| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 354| 646k| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 355| 646k| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (355:9): [True: 353k, False: 292k] ------------------ 356| | // 1. pack the bytes 357| 353k| const __m128i utf8_packed = _mm_packus_epi16( 358| 353k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 359| | // 2. store (16 bytes) 360| 353k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 361| | // 3. adjust pointers 362| 353k| buf += 16; 363| 353k| utf8_output += 16; 364| 353k| continue; // we are done for this round! 365| 353k| } 366| | // no bits set above 7th bit 367| 292k| const __m256i one_byte_bytemask = 368| 292k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 369| 292k| const uint32_t one_byte_bitmask = 370| 292k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 371| | 372| | // no bits set above 11th bit 373| 292k| const __m256i one_or_two_bytes_bytemask = 374| 292k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 375| 292k| const uint32_t one_or_two_bytes_bitmask = 376| 292k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 377| 292k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (377:9): [True: 74.3k, False: 218k] ------------------ 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| 74.3k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 383| 74.3k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 384| | 385| | // t0 = [000a|aaaa|bbbb|bb00] 386| 74.3k| const __m256i t0 = _mm256_slli_epi16(in, 2); 387| | // t1 = [000a|aaaa|0000|0000] 388| 74.3k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 389| | // t2 = [0000|0000|00bb|bbbb] 390| 74.3k| const __m256i t2 = _mm256_and_si256(in, v_003f); 391| | // t3 = [000a|aaaa|00bb|bbbb] 392| 74.3k| const __m256i t3 = _mm256_or_si256(t1, t2); 393| | // t4 = [110a|aaaa|10bb|bbbb] 394| 74.3k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 395| | 396| | // 2. merge ASCII and 2-byte codewords 397| 74.3k| const __m256i utf8_unpacked = 398| 74.3k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 399| | 400| | // 3. prepare bitmask for 8-bit lookup 401| 74.3k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 402| 74.3k| const uint32_t M1 = M0 >> 7; 403| 74.3k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 404| | // 4. pack the bytes 405| | 406| 74.3k| const uint8_t *row = 407| 74.3k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 408| 74.3k| const uint8_t *row_2 = 409| 74.3k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 410| 74.3k| 16)][0]; 411| | 412| 74.3k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 413| 74.3k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 414| | 415| 74.3k| const __m256i utf8_packed = _mm256_shuffle_epi8( 416| 74.3k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 74.3k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 74.3k| _mm256_castsi128_si256(xmm2), 2) ------------------ 417| | // 5. store bytes 418| 74.3k| _mm_storeu_si128((__m128i *)utf8_output, 419| 74.3k| _mm256_castsi256_si128(utf8_packed)); 420| 74.3k| utf8_output += row[0]; 421| 74.3k| _mm_storeu_si128((__m128i *)utf8_output, 422| 74.3k| _mm256_extractf128_si256(utf8_packed, 1)); 423| 74.3k| utf8_output += row_2[0]; 424| | 425| | // 6. adjust pointers 426| 74.3k| buf += 16; 427| 74.3k| continue; 428| 74.3k| } 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| 218k| const __m256i surrogates_bytemask = 433| 218k| _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| 218k| const uint32_t surrogates_bitmask = 438| 218k| 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| 218k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (441:9): [True: 214k, False: 3.99k] ------------------ 442| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 443| 214k| const __m256i dup_even = _mm256_setr_epi16( 444| 214k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 445| 214k| 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| 214k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 475| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 476| 214k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 477| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 478| 214k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 474| 214k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 479| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 480| 214k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 474| 214k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 481| | 482| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 483| 214k| const __m256i s0 = _mm256_srli_epi16(in, 4); 484| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 485| 214k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 474| 214k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 486| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 487| 214k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 474| 214k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 488| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 489| 214k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 474| 214k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 490| 214k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 491| 214k| simdutf_vec(0b0100000000000000)); ------------------ | | 474| 214k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 492| 214k| const __m256i s4 = _mm256_xor_si256(s3, m0); 493| 214k|#undef simdutf_vec 494| | 495| | // 4. expand code units 16-bit => 32-bit 496| 214k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 497| 214k| 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| 214k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 501| 214k| (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| 214k| const uint8_t mask0 = uint8_t(mask); 522| 214k| const uint8_t *row0 = 523| 214k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 524| 214k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 525| 214k| const __m128i utf8_0 = 526| 214k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 527| | 528| 214k| const uint8_t mask1 = static_cast(mask >> 8); 529| 214k| const uint8_t *row1 = 530| 214k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 531| 214k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 532| 214k| const __m128i utf8_1 = 533| 214k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 534| | 535| 214k| const uint8_t mask2 = static_cast(mask >> 16); 536| 214k| const uint8_t *row2 = 537| 214k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 538| 214k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 539| 214k| const __m128i utf8_2 = 540| 214k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 541| | 542| 214k| const uint8_t mask3 = static_cast(mask >> 24); 543| 214k| const uint8_t *row3 = 544| 214k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 545| 214k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 546| 214k| const __m128i utf8_3 = 547| 214k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 548| | 549| 214k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 550| 214k| utf8_output += row0[0]; 551| 214k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 552| 214k| utf8_output += row1[0]; 553| 214k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 554| 214k| utf8_output += row2[0]; 555| 214k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 556| 214k| utf8_output += row3[0]; 557| 214k| buf += 16; 558| | // surrogate pair(s) in a register 559| 214k| } 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| 3.99k| size_t forward = 15; 564| 3.99k| size_t k = 0; 565| 3.99k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (565:11): [True: 0, False: 3.99k] ------------------ 566| 0| forward = size_t(end - buf - 1); 567| 0| } 568| 58.2k| for (; k < forward; k++) { ------------------ | Branch (568:14): [True: 54.4k, False: 3.84k] ------------------ 569| 54.4k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 570| 54.4k| if ((word & 0xFF80) == 0) { ------------------ | Branch (570:13): [True: 10.2k, False: 44.2k] ------------------ 571| 10.2k| *utf8_output++ = char(word); 572| 44.2k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (572:20): [True: 2.13k, False: 42.1k] ------------------ 573| 2.13k| *utf8_output++ = char((word >> 6) | 0b11000000); 574| 2.13k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 575| 42.1k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (575:20): [True: 37.5k, False: 4.51k] ------------------ 576| 37.5k| *utf8_output++ = char((word >> 12) | 0b11100000); 577| 37.5k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 578| 37.5k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 579| 37.5k| } else { 580| | // must be a surrogate pair 581| 4.51k| uint16_t diff = uint16_t(word - 0xD800); 582| 4.51k| uint16_t next_word = 583| 4.51k| scalar::utf16::swap_if_needed(buf[k + 1]); 584| 4.51k| k++; 585| 4.51k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 586| 4.51k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (586:15): [True: 150, False: 4.36k] ------------------ 587| 150| return std::make_pair( 588| 150| result(error_code::SURROGATE, buf - start + k - 1), 589| 150| utf8_output); 590| 150| } 591| 4.36k| uint32_t value = (diff << 10) + diff2 + 0x10000; 592| 4.36k| *utf8_output++ = char((value >> 18) | 0b11110000); 593| 4.36k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 594| 4.36k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 595| 4.36k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 596| 4.36k| } 597| 54.4k| } 598| 3.84k| buf += k; 599| 3.84k| } 600| 218k| } // while 601| 504| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 602| 654|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_138avx2_convert_utf16_to_utf8_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 333| 648| char *utf8_output) { 334| 648| const char16_t *start = buf; 335| 648| const char16_t *end = buf + len; 336| | 337| 648| const __m256i v_0000 = _mm256_setzero_si256(); 338| 648| const __m256i v_f800 = _mm256_set1_epi16((int16_t)0xf800); 339| 648| const __m256i v_d800 = _mm256_set1_epi16((int16_t)0xd800); 340| 648| const __m256i v_c080 = _mm256_set1_epi16((int16_t)0xc080); 341| 648| const size_t safety_margin = 342| 648| 12; // to avoid overruns, see issue 343| | // https://github.com/simdutf/simdutf/issues/92 344| | 345| 576k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (345:10): [True: 575k, False: 472] ------------------ 346| 575k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 347| 575k| if (big_endian) { ------------------ | Branch (347:9): [True: 575k, Folded] ------------------ 348| 575k| const __m256i swap = _mm256_setr_epi8( 349| 575k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 350| 575k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 351| 575k| in = _mm256_shuffle_epi8(in, swap); 352| 575k| } 353| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 354| 575k| const __m256i v_ff80 = _mm256_set1_epi16((int16_t)0xff80); 355| 575k| if (_mm256_testz_si256(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (355:9): [True: 431k, False: 144k] ------------------ 356| | // 1. pack the bytes 357| 431k| const __m128i utf8_packed = _mm_packus_epi16( 358| 431k| _mm256_castsi256_si128(in), _mm256_extractf128_si256(in, 1)); 359| | // 2. store (16 bytes) 360| 431k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 361| | // 3. adjust pointers 362| 431k| buf += 16; 363| 431k| utf8_output += 16; 364| 431k| continue; // we are done for this round! 365| 431k| } 366| | // no bits set above 7th bit 367| 144k| const __m256i one_byte_bytemask = 368| 144k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_ff80), v_0000); 369| 144k| const uint32_t one_byte_bitmask = 370| 144k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 371| | 372| | // no bits set above 11th bit 373| 144k| const __m256i one_or_two_bytes_bytemask = 374| 144k| _mm256_cmpeq_epi16(_mm256_and_si256(in, v_f800), v_0000); 375| 144k| const uint32_t one_or_two_bytes_bitmask = 376| 144k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 377| 144k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (377:9): [True: 27.4k, False: 116k] ------------------ 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| 27.4k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 383| 27.4k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 384| | 385| | // t0 = [000a|aaaa|bbbb|bb00] 386| 27.4k| const __m256i t0 = _mm256_slli_epi16(in, 2); 387| | // t1 = [000a|aaaa|0000|0000] 388| 27.4k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 389| | // t2 = [0000|0000|00bb|bbbb] 390| 27.4k| const __m256i t2 = _mm256_and_si256(in, v_003f); 391| | // t3 = [000a|aaaa|00bb|bbbb] 392| 27.4k| const __m256i t3 = _mm256_or_si256(t1, t2); 393| | // t4 = [110a|aaaa|10bb|bbbb] 394| 27.4k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 395| | 396| | // 2. merge ASCII and 2-byte codewords 397| 27.4k| const __m256i utf8_unpacked = 398| 27.4k| _mm256_blendv_epi8(t4, in, one_byte_bytemask); 399| | 400| | // 3. prepare bitmask for 8-bit lookup 401| 27.4k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 402| 27.4k| const uint32_t M1 = M0 >> 7; 403| 27.4k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 404| | // 4. pack the bytes 405| | 406| 27.4k| const uint8_t *row = 407| 27.4k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 408| 27.4k| const uint8_t *row_2 = 409| 27.4k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 410| 27.4k| 16)][0]; 411| | 412| 27.4k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 413| 27.4k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 414| | 415| 27.4k| const __m256i utf8_packed = _mm256_shuffle_epi8( 416| 27.4k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 27.4k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 27.4k| _mm256_castsi128_si256(xmm2), 2) ------------------ 417| | // 5. store bytes 418| 27.4k| _mm_storeu_si128((__m128i *)utf8_output, 419| 27.4k| _mm256_castsi256_si128(utf8_packed)); 420| 27.4k| utf8_output += row[0]; 421| 27.4k| _mm_storeu_si128((__m128i *)utf8_output, 422| 27.4k| _mm256_extractf128_si256(utf8_packed, 1)); 423| 27.4k| utf8_output += row_2[0]; 424| | 425| | // 6. adjust pointers 426| 27.4k| buf += 16; 427| 27.4k| continue; 428| 27.4k| } 429| | // 1. Check if there are any surrogate word in the input chunk. 430| | // We have also deal with situation when there is a surrogate word 431| | // at the end of a chunk. 432| 116k| const __m256i surrogates_bytemask = 433| 116k| _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| 116k| const uint32_t surrogates_bitmask = 438| 116k| 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| 116k| if (surrogates_bitmask == 0x00000000) { ------------------ | Branch (441:9): [True: 110k, False: 6.42k] ------------------ 442| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 443| 110k| const __m256i dup_even = _mm256_setr_epi16( 444| 110k| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 445| 110k| 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| 110k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 475| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 476| 110k| const __m256i t0 = _mm256_shuffle_epi8(in, dup_even); 477| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 478| 110k| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 474| 110k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 479| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 480| 110k| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 474| 110k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 481| | 482| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 483| 110k| const __m256i s0 = _mm256_srli_epi16(in, 4); 484| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 485| 110k| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 474| 110k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 486| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 487| 110k| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 474| 110k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 488| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 489| 110k| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 474| 110k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 490| 110k| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 491| 110k| simdutf_vec(0b0100000000000000)); ------------------ | | 474| 110k|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 492| 110k| const __m256i s4 = _mm256_xor_si256(s3, m0); 493| 110k|#undef simdutf_vec 494| | 495| | // 4. expand code units 16-bit => 32-bit 496| 110k| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 497| 110k| 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| 110k| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 501| 110k| (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| 110k| const uint8_t mask0 = uint8_t(mask); 522| 110k| const uint8_t *row0 = 523| 110k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 524| 110k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 525| 110k| const __m128i utf8_0 = 526| 110k| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 527| | 528| 110k| const uint8_t mask1 = static_cast(mask >> 8); 529| 110k| const uint8_t *row1 = 530| 110k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 531| 110k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 532| 110k| const __m128i utf8_1 = 533| 110k| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 534| | 535| 110k| const uint8_t mask2 = static_cast(mask >> 16); 536| 110k| const uint8_t *row2 = 537| 110k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 538| 110k| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 539| 110k| const __m128i utf8_2 = 540| 110k| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 541| | 542| 110k| const uint8_t mask3 = static_cast(mask >> 24); 543| 110k| const uint8_t *row3 = 544| 110k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 545| 110k| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 546| 110k| const __m128i utf8_3 = 547| 110k| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 548| | 549| 110k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 550| 110k| utf8_output += row0[0]; 551| 110k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 552| 110k| utf8_output += row1[0]; 553| 110k| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 554| 110k| utf8_output += row2[0]; 555| 110k| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 556| 110k| utf8_output += row3[0]; 557| 110k| buf += 16; 558| | // surrogate pair(s) in a register 559| 110k| } 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| 6.42k| size_t forward = 15; 564| 6.42k| size_t k = 0; 565| 6.42k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (565:11): [True: 0, False: 6.42k] ------------------ 566| 0| forward = size_t(end - buf - 1); 567| 0| } 568| 94.3k| for (; k < forward; k++) { ------------------ | Branch (568:14): [True: 88.0k, False: 6.24k] ------------------ 569| 88.0k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 570| 88.0k| if ((word & 0xFF80) == 0) { ------------------ | Branch (570:13): [True: 24.9k, False: 63.1k] ------------------ 571| 24.9k| *utf8_output++ = char(word); 572| 63.1k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (572:20): [True: 4.46k, False: 58.6k] ------------------ 573| 4.46k| *utf8_output++ = char((word >> 6) | 0b11000000); 574| 4.46k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 575| 58.6k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (575:20): [True: 51.4k, False: 7.17k] ------------------ 576| 51.4k| *utf8_output++ = char((word >> 12) | 0b11100000); 577| 51.4k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 578| 51.4k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 579| 51.4k| } else { 580| | // must be a surrogate pair 581| 7.17k| uint16_t diff = uint16_t(word - 0xD800); 582| 7.17k| uint16_t next_word = 583| 7.17k| scalar::utf16::swap_if_needed(buf[k + 1]); 584| 7.17k| k++; 585| 7.17k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 586| 7.17k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (586:15): [True: 176, False: 7.00k] ------------------ 587| 176| return std::make_pair( 588| 176| result(error_code::SURROGATE, buf - start + k - 1), 589| 176| utf8_output); 590| 176| } 591| 7.00k| uint32_t value = (diff << 10) + diff2 + 0x10000; 592| 7.00k| *utf8_output++ = char((value >> 18) | 0b11110000); 593| 7.00k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 594| 7.00k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 595| 7.00k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 596| 7.00k| } 597| 88.0k| } 598| 6.24k| buf += k; 599| 6.24k| } 600| 116k| } // while 601| 472| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 602| 648|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128avx2_convert_utf32_to_latin1EPKDimPc: 3| 206| char *latin1_output) { 4| 206| const size_t rounded_len = 5| 206| len & ~0x1F; // Round down to nearest multiple of 32 6| | 7| 206| const __m256i high_bytes_mask = _mm256_set1_epi32(0xFFFFFF00); 8| | 9| 808| for (size_t i = 0; i < rounded_len; i += 4 * 8) { ------------------ | Branch (9:22): [True: 638, False: 170] ------------------ 10| 638| __m256i a = _mm256_loadu_si256((__m256i *)(buf + 0 * 8)); 11| 638| __m256i b = _mm256_loadu_si256((__m256i *)(buf + 1 * 8)); 12| 638| __m256i c = _mm256_loadu_si256((__m256i *)(buf + 2 * 8)); 13| 638| __m256i d = _mm256_loadu_si256((__m256i *)(buf + 3 * 8)); 14| | 15| 638| const __m256i check_combined = 16| 638| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 17| | 18| 638| if (!_mm256_testz_si256(check_combined, high_bytes_mask)) { ------------------ | Branch (18:9): [True: 36, False: 602] ------------------ 19| 36| return std::make_pair(nullptr, latin1_output); 20| 36| } 21| | 22| 602| b = _mm256_slli_epi32(b, 1 * 8); 23| 602| c = _mm256_slli_epi32(c, 2 * 8); 24| 602| 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| 602| const __m256i t0 = 35| 602| _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| 602| const __m256i shuffle_bytes = 40| 602| _mm256_setr_epi8(0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15, 41| 602| 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15); 42| | 43| 602| 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| 602| const __m256i shuffle_dwords = _mm256_setr_epi32(0, 4, 1, 5, 2, 6, 3, 7); 48| 602| const __m256i t2 = _mm256_permutevar8x32_epi32(t1, shuffle_dwords); 49| |// clang format on 50| | 51| 602| _mm256_storeu_si256((__m256i *)latin1_output, t2); 52| | 53| 602| latin1_output += 32; 54| 602| buf += 32; 55| 602| } 56| | 57| 170| return std::make_pair(buf, latin1_output); 58| 206|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_140avx2_convert_utf32_to_latin1_with_errorsEPKDimPc: 62| 454| char *latin1_output) { 63| 454| const size_t rounded_len = 64| 454| len & ~0x1F; // Round down to nearest multiple of 32 65| | 66| 454| const char32_t *start = buf; 67| | 68| 454| const __m256i high_bytes_mask = _mm256_set1_epi32(0xFFFFFF00); 69| | 70| 13.6k| for (size_t i = 0; i < rounded_len; i += 4 * 8) { ------------------ | Branch (70:22): [True: 13.3k, False: 332] ------------------ 71| 13.3k| __m256i a = _mm256_loadu_si256((__m256i *)(buf + 0 * 8)); 72| 13.3k| __m256i b = _mm256_loadu_si256((__m256i *)(buf + 1 * 8)); 73| 13.3k| __m256i c = _mm256_loadu_si256((__m256i *)(buf + 2 * 8)); 74| 13.3k| __m256i d = _mm256_loadu_si256((__m256i *)(buf + 3 * 8)); 75| | 76| 13.3k| const __m256i check_combined = 77| 13.3k| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 78| | 79| 13.3k| if (!_mm256_testz_si256(check_combined, high_bytes_mask)) { ------------------ | Branch (79:9): [True: 122, False: 13.2k] ------------------ 80| | // Fallback to scalar code for handling errors 81| 780| for (int k = 0; k < 4 * 8; k++) { ------------------ | Branch (81:23): [True: 780, False: 0] ------------------ 82| 780| char32_t codepoint = buf[k]; 83| 780| if (codepoint <= 0xFF) { ------------------ | Branch (83:13): [True: 658, False: 122] ------------------ 84| 658| *latin1_output++ = static_cast(codepoint); 85| 658| } else { 86| 122| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 87| 122| latin1_output); 88| 122| } 89| 780| } 90| 122| } 91| | 92| 13.2k| b = _mm256_slli_epi32(b, 1 * 8); 93| 13.2k| c = _mm256_slli_epi32(c, 2 * 8); 94| 13.2k| d = _mm256_slli_epi32(d, 3 * 8); 95| | 96| 13.2k| const __m256i t0 = 97| 13.2k| _mm256_or_si256(_mm256_or_si256(a, b), _mm256_or_si256(c, d)); 98| | 99| 13.2k| const __m256i shuffle_bytes = 100| 13.2k| _mm256_setr_epi8(0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15, 101| 13.2k| 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15); 102| | 103| 13.2k| const __m256i t1 = _mm256_shuffle_epi8(t0, shuffle_bytes); 104| | 105| 13.2k| const __m256i shuffle_dwords = _mm256_setr_epi32(0, 4, 1, 5, 2, 6, 3, 7); 106| 13.2k| const __m256i t2 = _mm256_permutevar8x32_epi32(t1, shuffle_dwords); 107| | 108| 13.2k| _mm256_storeu_si256((__m256i *)latin1_output, t2); 109| | 110| 13.2k| latin1_output += 32; 111| 13.2k| buf += 32; 112| 13.2k| } 113| | 114| 332| return std::make_pair(result(error_code::SUCCESS, buf - start), 115| 332| latin1_output); 116| 454|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf32_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKDiPDsEES7_mS8_: 4| 732| char16_t *utf16_output) { 5| 732| const char32_t *end = buf + len; 6| | 7| 732| const size_t safety_margin = 8| 732| 12; // to avoid overruns, see issue 9| | // https://github.com/simdutf/simdutf/issues/92 10| 732| __m256i forbidden_bytemask = _mm256_setzero_si256(); 11| | 12| 732| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 13| 732| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 14| 732| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 15| | 16| 99.9k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (16:10): [True: 99.3k, False: 602] ------------------ 17| 99.3k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 18| | 19| 99.3k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 99.3k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 96.3k, False: 3.04k] | | ------------------ ------------------ 20| | // no bits set above 16th bit <=> can pack to UTF16 21| | // without surrogate pairs 22| 96.3k| forbidden_bytemask = _mm256_or_si256( 23| 96.3k| forbidden_bytemask, 24| 96.3k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800)); 25| | 26| 96.3k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 27| 96.3k| _mm256_extractf128_si256(in, 1)); 28| 96.3k| if (big_endian) { ------------------ | Branch (28:11): [Folded, False: 96.3k] ------------------ 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| 96.3k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 34| 96.3k| utf16_output += 8; 35| 96.3k| buf += 8; 36| 96.3k| } else { 37| 3.04k| size_t forward = 7; 38| 3.04k| size_t k = 0; 39| 3.04k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (39:11): [True: 0, False: 3.04k] ------------------ 40| 0| forward = size_t(end - buf - 1); 41| 0| } 42| 23.6k| for (; k < forward; k++) { ------------------ | Branch (42:14): [True: 20.7k, False: 2.91k] ------------------ 43| 20.7k| uint32_t word = buf[k]; 44| 20.7k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (44:13): [True: 5.02k, False: 15.7k] ------------------ 45| | // will not generate a surrogate pair 46| 5.02k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (46:15): [True: 762, False: 4.26k] | Branch (46:33): [True: 6, False: 756] ------------------ 47| 6| return std::make_pair(nullptr, utf16_output); 48| 6| } 49| 5.02k| *utf16_output++ = 50| 5.02k| big_endian ------------------ | Branch (50:15): [Folded, False: 5.02k] ------------------ 51| 5.02k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 52| 5.02k| : char16_t(word); 53| 15.7k| } else { 54| | // will generate a surrogate pair 55| 15.7k| if (word > 0x10FFFF) { ------------------ | Branch (55:15): [True: 124, False: 15.6k] ------------------ 56| 124| return std::make_pair(nullptr, utf16_output); 57| 124| } 58| 15.6k| word -= 0x10000; 59| 15.6k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 60| 15.6k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 61| 15.6k| if (big_endian) { ------------------ | Branch (61:15): [Folded, False: 15.6k] ------------------ 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| 15.6k| *utf16_output++ = char16_t(high_surrogate); 68| 15.6k| *utf16_output++ = char16_t(low_surrogate); 69| 15.6k| } 70| 20.7k| } 71| 2.91k| buf += k; 72| 2.91k| } 73| 99.3k| } 74| | 75| | // check for invalid input 76| 602| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (76:7): [True: 42, False: 560] ------------------ 77| 42| return std::make_pair(nullptr, utf16_output); 78| 42| } 79| | 80| 560| return std::make_pair(buf, utf16_output); 81| 602|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_127avx2_convert_utf32_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKDiPDsEES7_mS8_: 4| 540| char16_t *utf16_output) { 5| 540| const char32_t *end = buf + len; 6| | 7| 540| const size_t safety_margin = 8| 540| 12; // to avoid overruns, see issue 9| | // https://github.com/simdutf/simdutf/issues/92 10| 540| __m256i forbidden_bytemask = _mm256_setzero_si256(); 11| | 12| 540| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 13| 540| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 14| 540| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 15| | 16| 5.61k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (16:10): [True: 5.18k, False: 432] ------------------ 17| 5.18k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 18| | 19| 5.18k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 5.18k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 3.92k, False: 1.25k] | | ------------------ ------------------ 20| | // no bits set above 16th bit <=> can pack to UTF16 21| | // without surrogate pairs 22| 3.92k| forbidden_bytemask = _mm256_or_si256( 23| 3.92k| forbidden_bytemask, 24| 3.92k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800)); 25| | 26| 3.92k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 27| 3.92k| _mm256_extractf128_si256(in, 1)); 28| 3.92k| if (big_endian) { ------------------ | Branch (28:11): [True: 3.92k, Folded] ------------------ 29| 3.92k| const __m128i swap = 30| 3.92k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 31| 3.92k| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 32| 3.92k| } 33| 3.92k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 34| 3.92k| utf16_output += 8; 35| 3.92k| buf += 8; 36| 3.92k| } else { 37| 1.25k| size_t forward = 7; 38| 1.25k| size_t k = 0; 39| 1.25k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (39:11): [True: 0, False: 1.25k] ------------------ 40| 0| forward = size_t(end - buf - 1); 41| 0| } 42| 9.53k| for (; k < forward; k++) { ------------------ | Branch (42:14): [True: 8.38k, False: 1.14k] ------------------ 43| 8.38k| uint32_t word = buf[k]; 44| 8.38k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (44:13): [True: 1.90k, False: 6.48k] ------------------ 45| | // will not generate a surrogate pair 46| 1.90k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (46:15): [True: 298, False: 1.60k] | Branch (46:33): [True: 6, False: 292] ------------------ 47| 6| return std::make_pair(nullptr, utf16_output); 48| 6| } 49| 1.89k| *utf16_output++ = 50| 1.89k| big_endian ------------------ | Branch (50:15): [True: 1.89k, Folded] ------------------ 51| 1.89k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 52| 1.89k| : char16_t(word); 53| 6.48k| } else { 54| | // will generate a surrogate pair 55| 6.48k| if (word > 0x10FFFF) { ------------------ | Branch (55:15): [True: 102, False: 6.38k] ------------------ 56| 102| return std::make_pair(nullptr, utf16_output); 57| 102| } 58| 6.38k| word -= 0x10000; 59| 6.38k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 60| 6.38k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 61| 6.38k| if (big_endian) { ------------------ | Branch (61:15): [True: 6.38k, Folded] ------------------ 62| 6.38k| high_surrogate = 63| 6.38k| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 64| 6.38k| low_surrogate = 65| 6.38k| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 66| 6.38k| } 67| 6.38k| *utf16_output++ = char16_t(high_surrogate); 68| 6.38k| *utf16_output++ = char16_t(low_surrogate); 69| 6.38k| } 70| 8.38k| } 71| 1.14k| buf += k; 72| 1.14k| } 73| 5.18k| } 74| | 75| | // check for invalid input 76| 432| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (76:7): [True: 22, False: 410] ------------------ 77| 22| return std::make_pair(nullptr, utf16_output); 78| 22| } 79| | 80| 410| return std::make_pair(buf, utf16_output); 81| 432|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf32_to_utf16_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 86| 714| char16_t *utf16_output) { 87| 714| const char32_t *start = buf; 88| 714| const char32_t *end = buf + len; 89| | 90| 714| const size_t safety_margin = 91| 714| 12; // to avoid overruns, see issue 92| | // https://github.com/simdutf/simdutf/issues/92 93| | 94| 714| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 95| 714| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 96| 714| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 97| | 98| 9.59k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (98:10): [True: 9.06k, False: 526] ------------------ 99| 9.06k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 100| | 101| 9.06k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 9.06k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 7.23k, False: 1.83k] | | ------------------ ------------------ 102| | // no bits set above 16th bit <=> can pack to UTF16 without surrogate 103| | // pairs 104| 7.23k| const __m256i forbidden_bytemask = 105| 7.23k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800); 106| 7.23k| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != ------------------ | Branch (106:11): [True: 36, False: 7.20k] ------------------ 107| 7.23k| 0x0) { 108| 36| return std::make_pair(result(error_code::SURROGATE, buf - start), 109| 36| utf16_output); 110| 36| } 111| | 112| 7.20k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 113| 7.20k| _mm256_extractf128_si256(in, 1)); 114| 7.20k| if (big_endian) { ------------------ | Branch (114:11): [Folded, False: 7.20k] ------------------ 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| 7.20k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 120| 7.20k| utf16_output += 8; 121| 7.20k| buf += 8; 122| 7.20k| } else { 123| 1.83k| size_t forward = 7; 124| 1.83k| size_t k = 0; 125| 1.83k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (125:11): [True: 0, False: 1.83k] ------------------ 126| 0| forward = size_t(end - buf - 1); 127| 0| } 128| 13.8k| for (; k < forward; k++) { ------------------ | Branch (128:14): [True: 12.1k, False: 1.67k] ------------------ 129| 12.1k| uint32_t word = buf[k]; 130| 12.1k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (130:13): [True: 5.68k, False: 6.50k] ------------------ 131| | // will not generate a surrogate pair 132| 5.68k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (132:15): [True: 626, False: 5.05k] | Branch (132:33): [True: 6, False: 620] ------------------ 133| 6| return std::make_pair( 134| 6| result(error_code::SURROGATE, buf - start + k), utf16_output); 135| 6| } 136| 5.67k| *utf16_output++ = 137| 5.67k| big_endian ------------------ | Branch (137:15): [Folded, False: 5.67k] ------------------ 138| 5.67k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 139| 5.67k| : char16_t(word); 140| 6.50k| } else { 141| | // will generate a surrogate pair 142| 6.50k| if (word > 0x10FFFF) { ------------------ | Branch (142:15): [True: 146, False: 6.35k] ------------------ 143| 146| return std::make_pair( 144| 146| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 145| 146| } 146| 6.35k| word -= 0x10000; 147| 6.35k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 148| 6.35k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 149| 6.35k| if (big_endian) { ------------------ | Branch (149:15): [Folded, False: 6.35k] ------------------ 150| 0| high_surrogate = 151| 0| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 152| 0| low_surrogate = 153| 0| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 154| 0| } 155| 6.35k| *utf16_output++ = char16_t(high_surrogate); 156| 6.35k| *utf16_output++ = char16_t(low_surrogate); 157| 6.35k| } 158| 12.1k| } 159| 1.67k| buf += k; 160| 1.67k| } 161| 9.06k| } 162| | 163| 526| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 164| 714|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_139avx2_convert_utf32_to_utf16_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 86| 660| char16_t *utf16_output) { 87| 660| const char32_t *start = buf; 88| 660| const char32_t *end = buf + len; 89| | 90| 660| const size_t safety_margin = 91| 660| 12; // to avoid overruns, see issue 92| | // https://github.com/simdutf/simdutf/issues/92 93| | 94| 660| const __m256i v_ffff0000 = _mm256_set1_epi32((int32_t)0xffff0000); 95| 660| const __m256i v_f800 = _mm256_set1_epi32((uint32_t)0xf800); 96| 660| const __m256i v_d800 = _mm256_set1_epi32((uint32_t)0xd800); 97| | 98| 10.3k| while (end - buf >= std::ptrdiff_t(8 + safety_margin)) { ------------------ | Branch (98:10): [True: 9.89k, False: 476] ------------------ 99| 9.89k| const __m256i in = _mm256_loadu_si256((__m256i *)buf); 100| | 101| 9.89k| if (simdutf_likely(_mm256_testz_si256(in, v_ffff0000))) { ------------------ | | 90| 9.89k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 9.12k, False: 762] | | ------------------ ------------------ 102| | // no bits set above 16th bit <=> can pack to UTF16 without surrogate 103| | // pairs 104| 9.12k| const __m256i forbidden_bytemask = 105| 9.12k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_f800), v_d800); 106| 9.12k| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != ------------------ | Branch (106:11): [True: 12, False: 9.11k] ------------------ 107| 9.12k| 0x0) { 108| 12| return std::make_pair(result(error_code::SURROGATE, buf - start), 109| 12| utf16_output); 110| 12| } 111| | 112| 9.11k| __m128i utf16_packed = _mm_packus_epi32(_mm256_castsi256_si128(in), 113| 9.11k| _mm256_extractf128_si256(in, 1)); 114| 9.11k| if (big_endian) { ------------------ | Branch (114:11): [True: 9.11k, Folded] ------------------ 115| 9.11k| const __m128i swap = 116| 9.11k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 117| 9.11k| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 118| 9.11k| } 119| 9.11k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 120| 9.11k| utf16_output += 8; 121| 9.11k| buf += 8; 122| 9.11k| } else { 123| 762| size_t forward = 7; 124| 762| size_t k = 0; 125| 762| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (125:11): [True: 0, False: 762] ------------------ 126| 0| forward = size_t(end - buf - 1); 127| 0| } 128| 5.25k| for (; k < forward; k++) { ------------------ | Branch (128:14): [True: 4.66k, False: 590] ------------------ 129| 4.66k| uint32_t word = buf[k]; 130| 4.66k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (130:13): [True: 2.51k, False: 2.15k] ------------------ 131| | // will not generate a surrogate pair 132| 2.51k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (132:15): [True: 762, False: 1.75k] | Branch (132:33): [True: 10, False: 752] ------------------ 133| 10| return std::make_pair( 134| 10| result(error_code::SURROGATE, buf - start + k), utf16_output); 135| 10| } 136| 2.50k| *utf16_output++ = 137| 2.50k| big_endian ------------------ | Branch (137:15): [True: 2.50k, Folded] ------------------ 138| 2.50k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 139| 2.50k| : char16_t(word); 140| 2.50k| } else { 141| | // will generate a surrogate pair 142| 2.15k| if (word > 0x10FFFF) { ------------------ | Branch (142:15): [True: 162, False: 1.98k] ------------------ 143| 162| return std::make_pair( 144| 162| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 145| 162| } 146| 1.98k| word -= 0x10000; 147| 1.98k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 148| 1.98k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 149| 1.98k| if (big_endian) { ------------------ | Branch (149:15): [True: 1.98k, Folded] ------------------ 150| 1.98k| high_surrogate = 151| 1.98k| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 152| 1.98k| low_surrogate = 153| 1.98k| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 154| 1.98k| } 155| 1.98k| *utf16_output++ = char16_t(high_surrogate); 156| 1.98k| *utf16_output++ = char16_t(low_surrogate); 157| 1.98k| } 158| 4.66k| } 159| 590| buf += k; 160| 590| } 161| 9.89k| } 162| | 163| 476| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 164| 660|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_126avx2_convert_utf32_to_utf8EPKDimPc: 2| 872|avx2_convert_utf32_to_utf8(const char32_t *buf, size_t len, char *utf8_output) { 3| 872| const char32_t *end = buf + len; 4| 872| const __m256i v_0000 = _mm256_setzero_si256(); 5| 872| const __m256i v_ffff0000 = _mm256_set1_epi32((uint32_t)0xffff0000); 6| 872| const __m256i v_ff80 = _mm256_set1_epi16((uint16_t)0xff80); 7| 872| const __m256i v_f800 = _mm256_set1_epi16((uint16_t)0xf800); 8| 872| const __m256i v_c080 = _mm256_set1_epi16((uint16_t)0xc080); 9| 872| const __m256i v_7fffffff = _mm256_set1_epi32((uint32_t)0x7fffffff); 10| 872| __m256i running_max = _mm256_setzero_si256(); 11| 872| __m256i forbidden_bytemask = _mm256_setzero_si256(); 12| | 13| 872| const size_t safety_margin = 14| 872| 12; // to avoid overruns, see issue 15| | // https://github.com/simdutf/simdutf/issues/92 16| | 17| 18.8k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (17:10): [True: 18.2k, False: 652] ------------------ 18| 18.2k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 19| 18.2k| __m256i nextin = _mm256_loadu_si256((__m256i *)buf + 1); 20| 18.2k| 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| 18.2k| __m256i in_16 = _mm256_packus_epi32(_mm256_and_si256(in, v_7fffffff), 25| 18.2k| _mm256_and_si256(nextin, v_7fffffff)); 26| 18.2k| 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| 18.2k| if (_mm256_testz_si256(in_16, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (31:9): [True: 14.3k, False: 3.88k] ------------------ 32| | // 1. pack the bytes 33| 14.3k| const __m128i utf8_packed = _mm_packus_epi16( 34| 14.3k| _mm256_castsi256_si128(in_16), _mm256_extractf128_si256(in_16, 1)); 35| | // 2. store (16 bytes) 36| 14.3k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 37| | // 3. adjust pointers 38| 14.3k| buf += 16; 39| 14.3k| utf8_output += 16; 40| 14.3k| continue; // we are done for this round! 41| 14.3k| } 42| | // no bits set above 7th bit 43| 3.88k| const __m256i one_byte_bytemask = 44| 3.88k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_ff80), v_0000); 45| 3.88k| const uint32_t one_byte_bitmask = 46| 3.88k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 47| | 48| | // no bits set above 11th bit 49| 3.88k| const __m256i one_or_two_bytes_bytemask = 50| 3.88k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_0000); 51| 3.88k| const uint32_t one_or_two_bytes_bitmask = 52| 3.88k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 53| 3.88k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (53:9): [True: 1.61k, False: 2.27k] ------------------ 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| 1.61k| const __m256i v_1f00 = _mm256_set1_epi16((int16_t)0x1f00); 58| 1.61k| const __m256i v_003f = _mm256_set1_epi16((int16_t)0x003f); 59| | 60| | // t0 = [000a|aaaa|bbbb|bb00] 61| 1.61k| const __m256i t0 = _mm256_slli_epi16(in_16, 2); 62| | // t1 = [000a|aaaa|0000|0000] 63| 1.61k| const __m256i t1 = _mm256_and_si256(t0, v_1f00); 64| | // t2 = [0000|0000|00bb|bbbb] 65| 1.61k| const __m256i t2 = _mm256_and_si256(in_16, v_003f); 66| | // t3 = [000a|aaaa|00bb|bbbb] 67| 1.61k| const __m256i t3 = _mm256_or_si256(t1, t2); 68| | // t4 = [110a|aaaa|10bb|bbbb] 69| 1.61k| const __m256i t4 = _mm256_or_si256(t3, v_c080); 70| | 71| | // 2. merge ASCII and 2-byte codewords 72| 1.61k| const __m256i utf8_unpacked = 73| 1.61k| _mm256_blendv_epi8(t4, in_16, one_byte_bytemask); 74| | 75| | // 3. prepare bitmask for 8-bit lookup 76| 1.61k| const uint32_t M0 = one_byte_bitmask & 0x55555555; 77| 1.61k| const uint32_t M1 = M0 >> 7; 78| 1.61k| const uint32_t M2 = (M1 | M0) & 0x00ff00ff; 79| | // 4. pack the bytes 80| | 81| 1.61k| const uint8_t *row = 82| 1.61k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2)][0]; 83| 1.61k| const uint8_t *row_2 = 84| 1.61k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[uint8_t(M2 >> 85| 1.61k| 16)][0]; 86| | 87| 1.61k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 88| 1.61k| const __m128i shuffle_2 = _mm_loadu_si128((__m128i *)(row_2 + 1)); 89| | 90| 1.61k| const __m256i utf8_packed = _mm256_shuffle_epi8( 91| 1.61k| utf8_unpacked, _mm256_setr_m128i(shuffle, shuffle_2)); ------------------ | | 7| 1.61k| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 8| 1.61k| _mm256_castsi128_si256(xmm2), 2) ------------------ 92| | // 5. store bytes 93| 1.61k| _mm_storeu_si128((__m128i *)utf8_output, 94| 1.61k| _mm256_castsi256_si128(utf8_packed)); 95| 1.61k| utf8_output += row[0]; 96| 1.61k| _mm_storeu_si128((__m128i *)utf8_output, 97| 1.61k| _mm256_extractf128_si256(utf8_packed, 1)); 98| 1.61k| utf8_output += row_2[0]; 99| | 100| | // 6. adjust pointers 101| 1.61k| buf += 16; 102| 1.61k| continue; 103| 1.61k| } 104| | // Must check for overflow in packing 105| 2.27k| const __m256i saturation_bytemask = _mm256_cmpeq_epi32( 106| 2.27k| _mm256_and_si256(_mm256_or_si256(in, nextin), v_ffff0000), v_0000); 107| 2.27k| const uint32_t saturation_bitmask = 108| 2.27k| static_cast(_mm256_movemask_epi8(saturation_bytemask)); 109| 2.27k| if (saturation_bitmask == 0xffffffff) { ------------------ | Branch (109:9): [True: 814, False: 1.45k] ------------------ 110| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 111| 814| const __m256i v_d800 = _mm256_set1_epi16((uint16_t)0xd800); 112| 814| forbidden_bytemask = _mm256_or_si256( 113| 814| forbidden_bytemask, 114| 814| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_d800)); 115| | 116| 814| const __m256i dup_even = _mm256_setr_epi16( 117| 814| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 118| 814| 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| 814|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 148| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 149| 814| const __m256i t0 = _mm256_shuffle_epi8(in_16, dup_even); 150| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 151| 814| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 147| 814|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 152| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 153| 814| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 147| 814|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 154| | 155| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 156| 814| const __m256i s0 = _mm256_srli_epi16(in_16, 4); 157| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 158| 814| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 147| 814|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 159| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 160| 814| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 147| 814|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 161| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 162| 814| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 147| 814|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 163| 814| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 164| 814| simdutf_vec(0b0100000000000000)); ------------------ | | 147| 814|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 165| 814| const __m256i s4 = _mm256_xor_si256(s3, m0); 166| 814|#undef simdutf_vec 167| | 168| | // 4. expand code units 16-bit => 32-bit 169| 814| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 170| 814| 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| 814| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 174| 814| (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| 814| const uint8_t mask0 = uint8_t(mask); 195| 814| const uint8_t *row0 = 196| 814| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 197| 814| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 198| 814| const __m128i utf8_0 = 199| 814| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 200| | 201| 814| const uint8_t mask1 = static_cast(mask >> 8); 202| 814| const uint8_t *row1 = 203| 814| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 204| 814| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 205| 814| const __m128i utf8_1 = 206| 814| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 207| | 208| 814| const uint8_t mask2 = static_cast(mask >> 16); 209| 814| const uint8_t *row2 = 210| 814| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 211| 814| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 212| 814| const __m128i utf8_2 = 213| 814| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 214| | 215| 814| const uint8_t mask3 = static_cast(mask >> 24); 216| 814| const uint8_t *row3 = 217| 814| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 218| 814| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 219| 814| const __m128i utf8_3 = 220| 814| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 221| | 222| 814| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 223| 814| utf8_output += row0[0]; 224| 814| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 225| 814| utf8_output += row1[0]; 226| 814| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 227| 814| utf8_output += row2[0]; 228| 814| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 229| 814| utf8_output += row3[0]; 230| 814| buf += 16; 231| 1.45k| } 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| 1.45k| size_t forward = 15; 237| 1.45k| size_t k = 0; 238| 1.45k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (238:11): [True: 0, False: 1.45k] ------------------ 239| 0| forward = size_t(end - buf - 1); 240| 0| } 241| 20.9k| for (; k < forward; k++) { ------------------ | Branch (241:14): [True: 19.6k, False: 1.23k] ------------------ 242| 19.6k| uint32_t word = buf[k]; 243| 19.6k| if ((word & 0xFFFFFF80) == 0) { // 1-byte (ASCII) ------------------ | Branch (243:13): [True: 9.54k, False: 10.1k] ------------------ 244| 9.54k| *utf8_output++ = char(word); 245| 10.1k| } else if ((word & 0xFFFFF800) == 0) { // 2-byte ------------------ | Branch (245:20): [True: 1.34k, False: 8.77k] ------------------ 246| 1.34k| *utf8_output++ = char((word >> 6) | 0b11000000); 247| 1.34k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 248| 8.77k| } else if ((word & 0xFFFF0000) == 0) { // 3-byte ------------------ | Branch (248:20): [True: 2.86k, False: 5.90k] ------------------ 249| 2.86k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (249:15): [True: 1.10k, False: 1.76k] | Branch (249:33): [True: 30, False: 1.07k] ------------------ 250| 30| return std::make_pair(nullptr, utf8_output); 251| 30| } 252| 2.83k| *utf8_output++ = char((word >> 12) | 0b11100000); 253| 2.83k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 254| 2.83k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 255| 5.90k| } else { // 4-byte 256| 5.90k| if (word > 0x10FFFF) { ------------------ | Branch (256:15): [True: 190, False: 5.71k] ------------------ 257| 190| return std::make_pair(nullptr, utf8_output); 258| 190| } 259| 5.71k| *utf8_output++ = char((word >> 18) | 0b11110000); 260| 5.71k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 261| 5.71k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 262| 5.71k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 263| 5.71k| } 264| 19.6k| } 265| 1.23k| buf += k; 266| 1.23k| } 267| 2.27k| } // while 268| | 269| | // check for invalid input 270| 652| const __m256i v_10ffff = _mm256_set1_epi32((uint32_t)0x10ffff); 271| 652| if (static_cast(_mm256_movemask_epi8(_mm256_cmpeq_epi32( ------------------ | Branch (271:7): [True: 34, False: 618] ------------------ 272| 652| _mm256_max_epu32(running_max, v_10ffff), v_10ffff))) != 0xffffffff) { 273| 34| return std::make_pair(nullptr, utf8_output); 274| 34| } 275| | 276| 618| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (276:7): [True: 54, False: 564] ------------------ 277| 54| return std::make_pair(nullptr, utf8_output); 278| 54| } 279| | 280| 564| return std::make_pair(buf, utf8_output); 281| 618|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_138avx2_convert_utf32_to_utf8_with_errorsEPKDimPc: 285| 924| char *utf8_output) { 286| 924| const char32_t *end = buf + len; 287| 924| const char32_t *start = buf; 288| | 289| 924| const __m256i v_0000 = _mm256_setzero_si256(); 290| 924| const __m256i v_ffff0000 = _mm256_set1_epi32((uint32_t)0xffff0000); 291| 924| const __m256i v_ff80 = _mm256_set1_epi16((uint16_t)0xff80); 292| 924| const __m256i v_f800 = _mm256_set1_epi16((uint16_t)0xf800); 293| 924| const __m256i v_c080 = _mm256_set1_epi16((uint16_t)0xc080); 294| 924| const __m256i v_7fffffff = _mm256_set1_epi32((uint32_t)0x7fffffff); 295| 924| const __m256i v_10ffff = _mm256_set1_epi32((uint32_t)0x10ffff); 296| | 297| 924| const size_t safety_margin = 298| 924| 12; // to avoid overruns, see issue 299| | // https://github.com/simdutf/simdutf/issues/92 300| | 301| 30.8k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (301:10): [True: 30.1k, False: 664] ------------------ 302| 30.1k| __m256i in = _mm256_loadu_si256((__m256i *)buf); 303| 30.1k| __m256i nextin = _mm256_loadu_si256((__m256i *)buf + 1); 304| | // Check for too large input 305| 30.1k| const __m256i max_input = 306| 30.1k| _mm256_max_epu32(_mm256_max_epu32(in, nextin), v_10ffff); 307| 30.1k| if (static_cast(_mm256_movemask_epi8( ------------------ | Branch (307:9): [True: 172, False: 29.9k] ------------------ 308| 30.1k| _mm256_cmpeq_epi32(max_input, v_10ffff))) != 0xffffffff) { 309| 172| return std::make_pair(result(error_code::TOO_LARGE, buf - start), 310| 172| utf8_output); 311| 172| } 312| | 313| | // Pack 32-bit UTF-32 code units to 16-bit UTF-16 code units with unsigned 314| | // saturation 315| 29.9k| __m256i in_16 = _mm256_packus_epi32(_mm256_and_si256(in, v_7fffffff), 316| 29.9k| _mm256_and_si256(nextin, v_7fffffff)); 317| 29.9k| 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| 29.9k| if (_mm256_testz_si256(in_16, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (322:9): [True: 25.4k, False: 4.49k] ------------------ 323| | // 1. pack the bytes 324| 25.4k| const __m128i utf8_packed = _mm_packus_epi16( 325| 25.4k| _mm256_castsi256_si128(in_16), _mm256_extractf128_si256(in_16, 1)); 326| | // 2. store (16 bytes) 327| 25.4k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 328| | // 3. adjust pointers 329| 25.4k| buf += 16; 330| 25.4k| utf8_output += 16; 331| 25.4k| continue; // we are done for this round! 332| 25.4k| } 333| | // no bits set above 7th bit 334| 4.49k| const __m256i one_byte_bytemask = 335| 4.49k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_ff80), v_0000); 336| 4.49k| const uint32_t one_byte_bitmask = 337| 4.49k| static_cast(_mm256_movemask_epi8(one_byte_bytemask)); 338| | 339| | // no bits set above 11th bit 340| 4.49k| const __m256i one_or_two_bytes_bytemask = 341| 4.49k| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_0000); 342| 4.49k| const uint32_t one_or_two_bytes_bitmask = 343| 4.49k| static_cast(_mm256_movemask_epi8(one_or_two_bytes_bytemask)); 344| 4.49k| if (one_or_two_bytes_bitmask == 0xffffffff) { ------------------ | Branch (344:9): [True: 2.51k, False: 1.98k] ------------------ 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| 1.98k| const __m256i saturation_bytemask = _mm256_cmpeq_epi32( 397| 1.98k| _mm256_and_si256(_mm256_or_si256(in, nextin), v_ffff0000), v_0000); 398| 1.98k| const uint32_t saturation_bitmask = 399| 1.98k| static_cast(_mm256_movemask_epi8(saturation_bytemask)); 400| 1.98k| if (saturation_bitmask == 0xffffffff) { ------------------ | Branch (400:9): [True: 752, False: 1.23k] ------------------ 401| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 402| | 403| | // Check for illegal surrogate code units 404| 752| const __m256i v_d800 = _mm256_set1_epi16((uint16_t)0xd800); 405| 752| const __m256i forbidden_bytemask = 406| 752| _mm256_cmpeq_epi16(_mm256_and_si256(in_16, v_f800), v_d800); 407| 752| if (static_cast(_mm256_movemask_epi8(forbidden_bytemask)) != ------------------ | Branch (407:11): [True: 52, False: 700] ------------------ 408| 752| 0x0) { 409| 52| return std::make_pair(result(error_code::SURROGATE, buf - start), 410| 52| utf8_output); 411| 52| } 412| | 413| 700| const __m256i dup_even = _mm256_setr_epi16( 414| 700| 0x0000, 0x0202, 0x0404, 0x0606, 0x0808, 0x0a0a, 0x0c0c, 0x0e0e, 415| 700| 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| 700|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) 445| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 446| 700| const __m256i t0 = _mm256_shuffle_epi8(in_16, dup_even); 447| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 448| 700| const __m256i t1 = _mm256_and_si256(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 444| 700|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 449| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 450| 700| const __m256i t2 = _mm256_or_si256(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 444| 700|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 451| | 452| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 453| 700| const __m256i s0 = _mm256_srli_epi16(in_16, 4); 454| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 455| 700| const __m256i s1 = _mm256_and_si256(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 444| 700|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 456| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 457| 700| const __m256i s2 = _mm256_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 444| 700|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 458| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 459| 700| const __m256i s3 = _mm256_or_si256(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 444| 700|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 460| 700| const __m256i m0 = _mm256_andnot_si256(one_or_two_bytes_bytemask, 461| 700| simdutf_vec(0b0100000000000000)); ------------------ | | 444| 700|#define simdutf_vec(x) _mm256_set1_epi16(static_cast(x)) ------------------ 462| 700| const __m256i s4 = _mm256_xor_si256(s3, m0); 463| 700|#undef simdutf_vec 464| | 465| | // 4. expand code units 16-bit => 32-bit 466| 700| const __m256i out0 = _mm256_unpacklo_epi16(t2, s4); 467| 700| 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| 700| const uint32_t mask = (one_byte_bitmask & 0x55555555) | 471| 700| (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| 700| const uint8_t mask0 = uint8_t(mask); 492| 700| const uint8_t *row0 = 493| 700| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 494| 700| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 495| 700| const __m128i utf8_0 = 496| 700| _mm_shuffle_epi8(_mm256_castsi256_si128(out0), shuffle0); 497| | 498| 700| const uint8_t mask1 = static_cast(mask >> 8); 499| 700| const uint8_t *row1 = 500| 700| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 501| 700| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 502| 700| const __m128i utf8_1 = 503| 700| _mm_shuffle_epi8(_mm256_castsi256_si128(out1), shuffle1); 504| | 505| 700| const uint8_t mask2 = static_cast(mask >> 16); 506| 700| const uint8_t *row2 = 507| 700| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask2][0]; 508| 700| const __m128i shuffle2 = _mm_loadu_si128((__m128i *)(row2 + 1)); 509| 700| const __m128i utf8_2 = 510| 700| _mm_shuffle_epi8(_mm256_extractf128_si256(out0, 1), shuffle2); 511| | 512| 700| const uint8_t mask3 = static_cast(mask >> 24); 513| 700| const uint8_t *row3 = 514| 700| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask3][0]; 515| 700| const __m128i shuffle3 = _mm_loadu_si128((__m128i *)(row3 + 1)); 516| 700| const __m128i utf8_3 = 517| 700| _mm_shuffle_epi8(_mm256_extractf128_si256(out1, 1), shuffle3); 518| | 519| 700| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 520| 700| utf8_output += row0[0]; 521| 700| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 522| 700| utf8_output += row1[0]; 523| 700| _mm_storeu_si128((__m128i *)utf8_output, utf8_2); 524| 700| utf8_output += row2[0]; 525| 700| _mm_storeu_si128((__m128i *)utf8_output, utf8_3); 526| 700| utf8_output += row3[0]; 527| 700| buf += 16; 528| 1.23k| } 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| 1.23k| size_t forward = 15; 534| 1.23k| size_t k = 0; 535| 1.23k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (535:11): [True: 0, False: 1.23k] ------------------ 536| 0| forward = size_t(end - buf - 1); 537| 0| } 538| 19.3k| for (; k < forward; k++) { ------------------ | Branch (538:14): [True: 18.1k, False: 1.19k] ------------------ 539| 18.1k| uint32_t word = buf[k]; 540| 18.1k| if ((word & 0xFFFFFF80) == 0) { // 1-byte (ASCII) ------------------ | Branch (540:13): [True: 5.56k, False: 12.6k] ------------------ 541| 5.56k| *utf8_output++ = char(word); 542| 12.6k| } else if ((word & 0xFFFFF800) == 0) { // 2-byte ------------------ | Branch (542:20): [True: 1.65k, False: 10.9k] ------------------ 543| 1.65k| *utf8_output++ = char((word >> 6) | 0b11000000); 544| 1.65k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 545| 10.9k| } else if ((word & 0xFFFF0000) == 0) { // 3-byte ------------------ | Branch (545:20): [True: 2.50k, False: 8.47k] ------------------ 546| 2.50k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (546:15): [True: 1.23k, False: 1.26k] | Branch (546:33): [True: 36, False: 1.19k] ------------------ 547| 36| return std::make_pair( 548| 36| result(error_code::SURROGATE, buf - start + k), utf8_output); 549| 36| } 550| 2.46k| *utf8_output++ = char((word >> 12) | 0b11100000); 551| 2.46k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 552| 2.46k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 553| 8.47k| } else { // 4-byte 554| 8.47k| if (word > 0x10FFFF) { ------------------ | Branch (554:15): [True: 0, False: 8.47k] ------------------ 555| 0| return std::make_pair( 556| 0| result(error_code::TOO_LARGE, buf - start + k), utf8_output); 557| 0| } 558| 8.47k| *utf8_output++ = char((word >> 18) | 0b11110000); 559| 8.47k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 560| 8.47k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 561| 8.47k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 562| 8.47k| } 563| 18.1k| } 564| 1.19k| buf += k; 565| 1.19k| } 566| 1.98k| } // while 567| | 568| 664| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 569| 924|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_129convert_masked_utf8_to_latin1EPKcmRPc: 9| 736k| 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| 736k| const __m128i in = _mm_loadu_si128((__m128i *)input); 21| | 22| 736k| const uint16_t input_utf8_end_of_code_point_mask = 23| 736k| utf8_end_of_code_point_mask & 24| 736k| 0xfff; // we are only processing 12 bytes in case it is not all ASCII 25| | 26| 736k| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (26:7): [True: 4.70k, False: 731k] ------------------ 27| | // We process the data in chunks of 12 bytes. 28| 4.70k| _mm_storeu_si128(reinterpret_cast<__m128i *>(latin1_output), in); 29| 4.70k| latin1_output += 12; // We wrote 12 characters. 30| 4.70k| return 12; // We consumed 1 bytes. 31| 4.70k| } 32| | /// We do not have a fast path available, so we fallback. 33| 731k| const uint8_t idx = 34| 731k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 35| 731k| const uint8_t consumed = 36| 731k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 37| | // this indicates an invalid input: 38| 731k| if (idx >= 64) { ------------------ | Branch (38:7): [True: 24.9k, False: 706k] ------------------ 39| 24.9k| return consumed; 40| 24.9k| } 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| 706k| const __m128i sh = 48| 706k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 49| 706k| const __m128i perm = _mm_shuffle_epi8(in, sh); 50| 706k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 51| 706k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 52| 706k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 53| 706k| 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| 706k| _mm_storel_epi64((__m128i *)latin1_output, latin1_packed); 58| 706k| latin1_output += 6; // We wrote 6 bytes. 59| 706k| return consumed; 60| 731k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE0EEEmPKcmRPDs: 10| 2.97M| 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| 2.97M| const __m128i swap = 23| 2.97M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 2.97M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 2.97M| const uint16_t input_utf8_end_of_code_point_mask = 26| 2.97M| utf8_end_of_code_point_mask & 0xfff; 27| 2.97M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 10.4k, False: 2.96M] ------------------ 28| | // We process the data in chunks of 12 bytes. 29| 10.4k| __m256i ascii = _mm256_cvtepu8_epi16(in); 30| 10.4k| if (big_endian) { ------------------ | Branch (30:9): [Folded, False: 10.4k] ------------------ 31| 0| const __m256i swap256 = _mm256_setr_epi8( 32| 0| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 33| 0| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 34| 0| ascii = _mm256_shuffle_epi8(ascii, swap256); 35| 0| } 36| 10.4k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output), ascii); 37| 10.4k| utf16_output += 12; // We wrote 12 16-bit characters. 38| 10.4k| return 12; // We consumed 12 bytes. 39| 10.4k| } 40| 2.96M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (40:7): [True: 11.5k, False: 2.95M] ------------------ 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| 11.5k| const __m128i sh = 45| 11.5k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 46| 11.5k| const __m128i perm = _mm_shuffle_epi8(in, sh); 47| 11.5k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 48| 11.5k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 49| 11.5k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 50| 11.5k| if (big_endian) ------------------ | Branch (50:9): [Folded, False: 11.5k] ------------------ 51| 0| composed = _mm_shuffle_epi8(composed, swap); 52| 11.5k| _mm_storeu_si128((__m128i *)utf16_output, composed); 53| 11.5k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 54| 11.5k| return 16; 55| 11.5k| } 56| 2.95M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (56:7): [True: 4.17k, False: 2.94M] ------------------ 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| 4.17k| const __m128i sh = 61| 4.17k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 62| 4.17k| const __m128i perm = _mm_shuffle_epi8(in, sh); 63| 4.17k| const __m128i ascii = 64| 4.17k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 65| 4.17k| const __m128i middlebyte = 66| 4.17k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 67| 4.17k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 68| 4.17k| const __m128i highbyte = 69| 4.17k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 70| 4.17k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 71| 4.17k| const __m128i composed = 72| 4.17k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 73| 4.17k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 74| 4.17k| if (big_endian) ------------------ | Branch (74:9): [Folded, False: 4.17k] ------------------ 75| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 76| 4.17k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 77| 4.17k| utf16_output += 4; 78| 4.17k| return 12; 79| 4.17k| } 80| | 81| 2.94M| const uint8_t idx = simdutf::tables::utf8_to_utf16::utf8bigindex 82| 2.94M| [input_utf8_end_of_code_point_mask][0]; 83| 2.94M| const uint8_t consumed = simdutf::tables::utf8_to_utf16::utf8bigindex 84| 2.94M| [input_utf8_end_of_code_point_mask][1]; 85| 2.94M| if (idx < 64) { ------------------ | Branch (85:7): [True: 2.50M, False: 441k] ------------------ 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| 2.50M| const __m128i sh = _mm_loadu_si128( 93| 2.50M| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 94| 2.50M| const __m128i perm = _mm_shuffle_epi8(in, sh); 95| 2.50M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 96| 2.50M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 97| 2.50M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 98| 2.50M| if (big_endian) ------------------ | Branch (98:9): [Folded, False: 2.50M] ------------------ 99| 0| composed = _mm_shuffle_epi8(composed, swap); 100| 2.50M| _mm_storeu_si128((__m128i *)utf16_output, composed); 101| 2.50M| utf16_output += 6; // We wrote 12 bytes, 6 code points. There is a potential 102| | // overflow of 4 bytes. 103| 2.50M| } else if (idx < 145) { ------------------ | Branch (103:14): [True: 388k, False: 53.1k] ------------------ 104| | // FOUR (4) input code-code units 105| 388k| const __m128i sh = _mm_loadu_si128( 106| 388k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 107| 388k| const __m128i perm = _mm_shuffle_epi8(in, sh); 108| 388k| const __m128i ascii = 109| 388k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 110| 388k| const __m128i middlebyte = 111| 388k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 112| 388k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 113| 388k| const __m128i highbyte = 114| 388k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 115| 388k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 116| 388k| const __m128i composed = 117| 388k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 118| 388k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 119| 388k| if (big_endian) ------------------ | Branch (119:9): [Folded, False: 388k] ------------------ 120| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 121| 388k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 122| 388k| utf16_output += 4; // Here we overflow by 8 bytes. 123| 388k| } else if (idx < 209) { ------------------ | Branch (123:14): [True: 52.0k, False: 1.13k] ------------------ 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| 52.0k| const __m128i sh = _mm_loadu_si128( 135| 52.0k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 136| 52.0k| const __m128i perm = _mm_shuffle_epi8(in, sh); 137| 52.0k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 138| 52.0k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 139| 52.0k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 140| 52.0k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 141| | // correct for spurious high bit 142| 52.0k| const __m128i correct = 143| 52.0k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 144| 52.0k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 145| 52.0k| 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| 52.0k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 149| 52.0k| 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| 52.0k| const __m128i composed = 155| 52.0k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 156| 52.0k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 157| 52.0k| const __m128i composedminus = 158| 52.0k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 159| 52.0k| const __m128i lowtenbits = 160| 52.0k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 161| | // Notice the 0x3ff mask: 162| 52.0k| const __m128i hightenbits = 163| 52.0k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 164| 52.0k| const __m128i lowtenbitsadd = 165| 52.0k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 166| 52.0k| const __m128i hightenbitsadd = 167| 52.0k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 168| 52.0k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 169| 52.0k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 170| 52.0k| uint32_t basic_buffer[4]; 171| 52.0k| uint32_t basic_buffer_swap[4]; 172| 52.0k| if (big_endian) { ------------------ | Branch (172:9): [Folded, False: 52.0k] ------------------ 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| 52.0k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 178| 52.0k| uint32_t surrogate_buffer[4]; 179| 52.0k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 180| 208k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (180:24): [True: 156k, False: 52.0k] ------------------ 181| 156k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (181:11): [True: 43.7k, False: 112k] ------------------ 182| 43.7k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 183| 43.7k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 184| 43.7k| utf16_output += 2; 185| 112k| } else { 186| 112k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (186:27): [Folded, False: 112k] ------------------ 187| 112k| : uint16_t(basic_buffer[i]); 188| 112k| utf16_output++; 189| 112k| } 190| 156k| } 191| 52.0k| } else { 192| | // here we know that there is an error but we do not handle errors 193| 1.13k| } 194| 2.94M| return consumed; 195| 2.95M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE1EEEmPKcmRPDs: 10| 2.57M| 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| 2.57M| const __m128i swap = 23| 2.57M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 2.57M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 2.57M| const uint16_t input_utf8_end_of_code_point_mask = 26| 2.57M| utf8_end_of_code_point_mask & 0xfff; 27| 2.57M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 7.29k, False: 2.56M] ------------------ 28| | // We process the data in chunks of 12 bytes. 29| 7.29k| __m256i ascii = _mm256_cvtepu8_epi16(in); 30| 7.29k| if (big_endian) { ------------------ | Branch (30:9): [True: 7.29k, Folded] ------------------ 31| 7.29k| const __m256i swap256 = _mm256_setr_epi8( 32| 7.29k| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 33| 7.29k| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 34| 7.29k| ascii = _mm256_shuffle_epi8(ascii, swap256); 35| 7.29k| } 36| 7.29k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf16_output), ascii); 37| 7.29k| utf16_output += 12; // We wrote 12 16-bit characters. 38| 7.29k| return 12; // We consumed 12 bytes. 39| 7.29k| } 40| 2.56M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (40:7): [True: 10.5k, False: 2.55M] ------------------ 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| 10.5k| const __m128i sh = 45| 10.5k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 46| 10.5k| const __m128i perm = _mm_shuffle_epi8(in, sh); 47| 10.5k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 48| 10.5k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 49| 10.5k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 50| 10.5k| if (big_endian) ------------------ | Branch (50:9): [True: 10.5k, Folded] ------------------ 51| 10.5k| composed = _mm_shuffle_epi8(composed, swap); 52| 10.5k| _mm_storeu_si128((__m128i *)utf16_output, composed); 53| 10.5k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 54| 10.5k| return 16; 55| 10.5k| } 56| 2.55M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (56:7): [True: 3.38k, False: 2.55M] ------------------ 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| 3.38k| const __m128i sh = 61| 3.38k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 62| 3.38k| const __m128i perm = _mm_shuffle_epi8(in, sh); 63| 3.38k| const __m128i ascii = 64| 3.38k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 65| 3.38k| const __m128i middlebyte = 66| 3.38k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 67| 3.38k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 68| 3.38k| const __m128i highbyte = 69| 3.38k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 70| 3.38k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 71| 3.38k| const __m128i composed = 72| 3.38k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 73| 3.38k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 74| 3.38k| if (big_endian) ------------------ | Branch (74:9): [True: 3.38k, Folded] ------------------ 75| 3.38k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 76| 3.38k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 77| 3.38k| utf16_output += 4; 78| 3.38k| return 12; 79| 3.38k| } 80| | 81| 2.55M| const uint8_t idx = simdutf::tables::utf8_to_utf16::utf8bigindex 82| 2.55M| [input_utf8_end_of_code_point_mask][0]; 83| 2.55M| const uint8_t consumed = simdutf::tables::utf8_to_utf16::utf8bigindex 84| 2.55M| [input_utf8_end_of_code_point_mask][1]; 85| 2.55M| if (idx < 64) { ------------------ | Branch (85:7): [True: 2.32M, False: 227k] ------------------ 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| 2.32M| const __m128i sh = _mm_loadu_si128( 93| 2.32M| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 94| 2.32M| const __m128i perm = _mm_shuffle_epi8(in, sh); 95| 2.32M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 96| 2.32M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 97| 2.32M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 98| 2.32M| if (big_endian) ------------------ | Branch (98:9): [True: 2.32M, Folded] ------------------ 99| 2.32M| composed = _mm_shuffle_epi8(composed, swap); 100| 2.32M| _mm_storeu_si128((__m128i *)utf16_output, composed); 101| 2.32M| utf16_output += 6; // We wrote 12 bytes, 6 code points. There is a potential 102| | // overflow of 4 bytes. 103| 2.32M| } else if (idx < 145) { ------------------ | Branch (103:14): [True: 197k, False: 30.0k] ------------------ 104| | // FOUR (4) input code-code units 105| 197k| const __m128i sh = _mm_loadu_si128( 106| 197k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 107| 197k| const __m128i perm = _mm_shuffle_epi8(in, sh); 108| 197k| const __m128i ascii = 109| 197k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 110| 197k| const __m128i middlebyte = 111| 197k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 112| 197k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 113| 197k| const __m128i highbyte = 114| 197k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 115| 197k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 116| 197k| const __m128i composed = 117| 197k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 118| 197k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 119| 197k| if (big_endian) ------------------ | Branch (119:9): [True: 197k, Folded] ------------------ 120| 197k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 121| 197k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 122| 197k| utf16_output += 4; // Here we overflow by 8 bytes. 123| 197k| } else if (idx < 209) { ------------------ | Branch (123:14): [True: 29.0k, False: 978] ------------------ 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| 29.0k| const __m128i sh = _mm_loadu_si128( 135| 29.0k| (const __m128i *)simdutf::tables::utf8_to_utf16::shufutf8[idx]); 136| 29.0k| const __m128i perm = _mm_shuffle_epi8(in, sh); 137| 29.0k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 138| 29.0k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 139| 29.0k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 140| 29.0k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 141| | // correct for spurious high bit 142| 29.0k| const __m128i correct = 143| 29.0k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 144| 29.0k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 145| 29.0k| 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| 29.0k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 149| 29.0k| 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| 29.0k| const __m128i composed = 155| 29.0k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 156| 29.0k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 157| 29.0k| const __m128i composedminus = 158| 29.0k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 159| 29.0k| const __m128i lowtenbits = 160| 29.0k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 161| | // Notice the 0x3ff mask: 162| 29.0k| const __m128i hightenbits = 163| 29.0k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 164| 29.0k| const __m128i lowtenbitsadd = 165| 29.0k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 166| 29.0k| const __m128i hightenbitsadd = 167| 29.0k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 168| 29.0k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 169| 29.0k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 170| 29.0k| uint32_t basic_buffer[4]; 171| 29.0k| uint32_t basic_buffer_swap[4]; 172| 29.0k| if (big_endian) { ------------------ | Branch (172:9): [True: 29.0k, Folded] ------------------ 173| 29.0k| _mm_storeu_si128((__m128i *)basic_buffer_swap, 174| 29.0k| _mm_shuffle_epi8(composed, swap)); 175| 29.0k| surrogates = _mm_shuffle_epi8(surrogates, swap); 176| 29.0k| } 177| 29.0k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 178| 29.0k| uint32_t surrogate_buffer[4]; 179| 29.0k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 180| 116k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (180:24): [True: 87.1k, False: 29.0k] ------------------ 181| 87.1k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (181:11): [True: 24.2k, False: 62.9k] ------------------ 182| 24.2k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 183| 24.2k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 184| 24.2k| utf16_output += 2; 185| 62.9k| } else { 186| 62.9k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (186:27): [True: 62.9k, Folded] ------------------ 187| 62.9k| : uint16_t(basic_buffer[i]); 188| 62.9k| utf16_output++; 189| 62.9k| } 190| 87.1k| } 191| 29.0k| } else { 192| | // here we know that there is an error but we do not handle errors 193| 978| } 194| 2.55M| return consumed; 195| 2.55M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_128convert_masked_utf8_to_utf32EPKcmRPDi: 9| 3.32M| char32_t *&utf32_output) { 10| | // we use an approach where we try to process up to 12 input bytes. 11| | // Why 12 input bytes and not 16? Because we are concerned with the size of 12| | // the lookup tables. Also 12 is nicely divisible by two and three. 13| | // 14| | // 15| | // Optimization note: our main path below is load-latency dependent. Thus it 16| | // is maybe beneficial to have fast paths that depend on branch prediction but 17| | // have less latency. This results in more instructions but, potentially, also 18| | // higher speeds. 19| | // 20| | // We first try a few fast paths. 21| 3.32M| const __m128i in = _mm_loadu_si128((__m128i *)input); 22| 3.32M| const uint16_t input_utf8_end_of_code_point_mask = 23| 3.32M| utf8_end_of_code_point_mask & 0xfff; 24| 3.32M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (24:7): [True: 9.53k, False: 3.31M] ------------------ 25| | // We process the data in chunks of 12 bytes. 26| 9.53k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output), 27| 9.53k| _mm256_cvtepu8_epi32(in)); 28| 9.53k| _mm256_storeu_si256(reinterpret_cast<__m256i *>(utf32_output + 8), 29| 9.53k| _mm256_cvtepu8_epi32(_mm_srli_si128(in, 8))); 30| 9.53k| utf32_output += 12; // We wrote 12 32-bit characters. 31| 9.53k| return 12; // We consumed 12 bytes. 32| 9.53k| } 33| 3.31M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (33:7): [True: 24.4k, False: 3.29M] ------------------ 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| 24.4k| const __m128i sh = 38| 24.4k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 39| 24.4k| const __m128i perm = _mm_shuffle_epi8(in, sh); 40| 24.4k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 41| 24.4k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 42| 24.4k| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 43| 24.4k| _mm256_storeu_si256((__m256i *)utf32_output, 44| 24.4k| _mm256_cvtepu16_epi32(composed)); 45| 24.4k| utf32_output += 8; // We wrote 16 bytes, 8 code points. 46| 24.4k| return 16; 47| 24.4k| } 48| 3.29M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (48:7): [True: 3.32k, False: 3.28M] ------------------ 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| 3.32k| const __m128i sh = 53| 3.32k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 54| 3.32k| const __m128i perm = _mm_shuffle_epi8(in, sh); 55| 3.32k| const __m128i ascii = 56| 3.32k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 57| 3.32k| const __m128i middlebyte = 58| 3.32k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 59| 3.32k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 60| 3.32k| const __m128i highbyte = 61| 3.32k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 62| 3.32k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 63| 3.32k| const __m128i composed = 64| 3.32k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 65| 3.32k| _mm_storeu_si128((__m128i *)utf32_output, composed); 66| 3.32k| utf32_output += 4; 67| 3.32k| return 12; 68| 3.32k| } 69| | /// We do not have a fast path available, so we fallback. 70| | 71| 3.28M| const uint8_t idx = 72| 3.28M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 73| 3.28M| const uint8_t consumed = 74| 3.28M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 75| 3.28M| if (idx < 64) { ------------------ | Branch (75:7): [True: 2.99M, False: 294k] ------------------ 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| 2.99M| const __m128i sh = 83| 2.99M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 84| 2.99M| const __m128i perm = _mm_shuffle_epi8(in, sh); 85| 2.99M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 86| 2.99M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 87| 2.99M| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 88| 2.99M| _mm256_storeu_si256((__m256i *)utf32_output, 89| 2.99M| _mm256_cvtepu16_epi32(composed)); 90| 2.99M| utf32_output += 6; // We wrote 24 bytes, 6 code points. There is a potential 91| | // overflow of 32 - 24 = 8 bytes. 92| 2.99M| } else if (idx < 145) { ------------------ | Branch (92:14): [True: 227k, False: 67.0k] ------------------ 93| | // FOUR (4) input code-code units 94| 227k| const __m128i sh = 95| 227k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 96| 227k| const __m128i perm = _mm_shuffle_epi8(in, sh); 97| 227k| const __m128i ascii = 98| 227k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 99| 227k| const __m128i middlebyte = 100| 227k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 101| 227k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 102| 227k| const __m128i highbyte = 103| 227k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 104| 227k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 105| 227k| const __m128i composed = 106| 227k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 107| 227k| _mm_storeu_si128((__m128i *)utf32_output, composed); 108| 227k| utf32_output += 4; 109| 227k| } else if (idx < 209) { ------------------ | Branch (109:14): [True: 65.7k, False: 1.32k] ------------------ 110| | // TWO (2) input code-code units 111| 65.7k| const __m128i sh = 112| 65.7k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 113| 65.7k| const __m128i perm = _mm_shuffle_epi8(in, sh); 114| 65.7k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 115| 65.7k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 116| 65.7k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 117| 65.7k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 118| | // correct for spurious high bit 119| 65.7k| const __m128i correct = 120| 65.7k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 121| 65.7k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 122| 65.7k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 123| 65.7k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0x07000000)); 124| 65.7k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 125| 65.7k| const __m128i composed = 126| 65.7k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 127| 65.7k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 128| 65.7k| _mm_storeu_si128((__m128i *)utf32_output, composed); 129| 65.7k| utf32_output += 130| 65.7k| 3; // We wrote 3 * 4 bytes, there is a potential overflow of 4 bytes. 131| 65.7k| } else { 132| | // here we know that there is an error but we do not handle errors 133| 1.32k| } 134| 3.28M| return consumed; 135| 3.29M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf16_gather_high_bytesILNS_10endiannessE0EEENS1_4simd5simd8IhEERKNS5_6simd16ItEESB_: 3| 2.33M| const simd16 &in1) { 4| 2.33M| if (big_endian) { ------------------ | Branch (4:7): [Folded, False: 2.33M] ------------------ 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| 2.33M| } else { 12| 2.33M| const auto t0 = in0.shr<8>(); 13| 2.33M| const auto t1 = in1.shr<8>(); 14| | 15| 2.33M| return simd16::pack(t0, t1); 16| 2.33M| } 17| 2.33M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_15utf1623utf16_gather_high_bytesILNS_10endiannessE1EEENS1_4simd5simd8IhEERKNS5_6simd16ItEESB_: 3| 2.77M| const simd16 &in1) { 4| 2.77M| if (big_endian) { ------------------ | Branch (4:7): [True: 2.77M, Folded] ------------------ 5| | // we want lower bytes 6| 2.77M| const auto mask = simd16(0x00ff); 7| 2.77M| const auto t0 = in0 & mask; 8| 2.77M| const auto t1 = in1 & mask; 9| | 10| 2.77M| return simd16::pack(t0, t1); 11| 2.77M| } 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| 2.77M|} _ZNK7simdutf7haswell14implementation13validate_utf8EPKcm: 266| 3.52k|implementation::validate_utf8(const char *buf, size_t len) const noexcept { 267| 3.52k| return haswell::utf8_validation::generic_validate_utf8(buf, len); 268| 3.52k|} _ZNK7simdutf7haswell14implementation25validate_utf8_with_errorsEPKcm: 273| 3.52k| const char *buf, size_t len) const noexcept { 274| 3.52k| return haswell::utf8_validation::generic_validate_utf8_with_errors(buf, len); 275| 3.52k|} _ZNK7simdutf7haswell14implementation16validate_utf16leEPKDsm: 312| 1.38k| size_t len) const noexcept { 313| 1.38k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.38k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 1.37k] | | ------------------ ------------------ 314| | // empty input is valid UTF-16. protect the implementation from 315| | // handling nullptr 316| 16| return true; 317| 16| } 318| 1.37k| const auto res = 319| 1.37k| haswell::utf16::validate_utf16_with_errors(buf, len); 320| 1.37k| if (res.is_err()) { ------------------ | Branch (320:7): [True: 102, False: 1.26k] ------------------ 321| 102| return false; 322| 102| } 323| | 324| 1.26k| if (res.count == len) { ------------------ | Branch (324:7): [True: 0, False: 1.26k] ------------------ 325| 0| return true; 326| 0| } 327| | 328| 1.26k| return scalar::utf16::validate(buf + res.count, 329| 1.26k| len - res.count); 330| 1.26k|} _ZNK7simdutf7haswell14implementation16validate_utf16beEPKDsm: 336| 1.34k| size_t len) const noexcept { 337| 1.34k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.34k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 1.33k] | | ------------------ ------------------ 338| | // empty input is valid UTF-16. protect the implementation from 339| | // handling nullptr 340| 16| return true; 341| 16| } 342| 1.33k| const auto res = 343| 1.33k| haswell::utf16::validate_utf16_with_errors(buf, len); 344| 1.33k| if (res.is_err()) { ------------------ | Branch (344:7): [True: 80, False: 1.25k] ------------------ 345| 80| return false; 346| 80| } 347| | 348| 1.25k| if (res.count == len) { ------------------ | Branch (348:7): [True: 0, False: 1.25k] ------------------ 349| 0| return true; 350| 0| } 351| | 352| 1.25k| return scalar::utf16::validate(buf + res.count, 353| 1.25k| len - res.count); 354| 1.25k|} _ZNK7simdutf7haswell14implementation28validate_utf16le_with_errorsEPKDsm: 357| 1.38k| const char16_t *buf, size_t len) const noexcept { 358| | 359| 1.38k| const result res = 360| 1.38k| haswell::utf16::validate_utf16_with_errors(buf, len); 361| 1.38k| if (res.count != len) { ------------------ | Branch (361:7): [True: 1.37k, False: 16] ------------------ 362| 1.37k| const result scalar_res = 363| 1.37k| scalar::utf16::validate_with_errors( 364| 1.37k| buf + res.count, len - res.count); 365| 1.37k| return result(scalar_res.error, res.count + scalar_res.count); 366| 1.37k| } else { 367| 16| return res; 368| 16| } 369| 1.38k|} _ZNK7simdutf7haswell14implementation28validate_utf16be_with_errorsEPKDsm: 372| 1.34k| const char16_t *buf, size_t len) const noexcept { 373| 1.34k| const result res = 374| 1.34k| haswell::utf16::validate_utf16_with_errors(buf, len); 375| 1.34k| if (res.count != len) { ------------------ | Branch (375:7): [True: 1.33k, False: 16] ------------------ 376| 1.33k| const result scalar_res = 377| 1.33k| scalar::utf16::validate_with_errors(buf + res.count, 378| 1.33k| len - res.count); 379| 1.33k| return result(scalar_res.error, res.count + scalar_res.count); 380| 1.33k| } else { 381| 16| return res; 382| 16| } 383| 1.34k|} _ZNK7simdutf7haswell14implementation14validate_utf32EPKDim: 398| 2.57k|implementation::validate_utf32(const char32_t *buf, size_t len) const noexcept { 399| 2.57k| return utf32::validate(buf, len); 400| 2.57k|} _ZNK7simdutf7haswell14implementation26validate_utf32_with_errorsEPKDim: 405| 2.57k| const char32_t *buf, size_t len) const noexcept { 406| 2.57k| return utf32::validate_with_errors(buf, len); 407| 2.57k|} _ZNK7simdutf7haswell14implementation22convert_latin1_to_utf8EPKcmPc: 412| 484| const char *buf, size_t len, char *utf8_output) const noexcept { 413| 484| std::pair ret = 414| 484| avx2_convert_latin1_to_utf8(buf, len, utf8_output); 415| 484| size_t converted_chars = ret.second - utf8_output; 416| | 417| 484| if (ret.first != buf + len) { ------------------ | Branch (417:7): [True: 480, False: 4] ------------------ 418| 480| const size_t scalar_converted_chars = scalar::latin1_to_utf8::convert( 419| 480| ret.first, len - (ret.first - buf), ret.second); 420| 480| converted_chars += scalar_converted_chars; 421| 480| } 422| | 423| 484| return converted_chars; 424| 484|} _ZNK7simdutf7haswell14implementation25convert_latin1_to_utf16leEPKcmPDs: 429| 88| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 430| 88| std::pair ret = 431| 88| avx2_convert_latin1_to_utf16(buf, len, utf16_output); 432| 88| if (ret.first == nullptr) { ------------------ | Branch (432:7): [True: 4, False: 84] ------------------ 433| 4| return 0; 434| 4| } 435| 84| size_t converted_chars = ret.second - utf16_output; 436| 84| if (ret.first != buf + len) { ------------------ | Branch (436:7): [True: 70, False: 14] ------------------ 437| 70| const size_t scalar_converted_chars = 438| 70| scalar::latin1_to_utf16::convert( 439| 70| ret.first, len - (ret.first - buf), ret.second); 440| 70| if (scalar_converted_chars == 0) { ------------------ | Branch (440:9): [True: 0, False: 70] ------------------ 441| 0| return 0; 442| 0| } 443| 70| converted_chars += scalar_converted_chars; 444| 70| } 445| 84| return converted_chars; 446| 84|} _ZNK7simdutf7haswell14implementation25convert_latin1_to_utf16beEPKcmPDs: 449| 110| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 450| 110| std::pair ret = 451| 110| avx2_convert_latin1_to_utf16(buf, len, utf16_output); 452| 110| if (ret.first == nullptr) { ------------------ | Branch (452:7): [True: 4, False: 106] ------------------ 453| 4| return 0; 454| 4| } 455| 106| size_t converted_chars = ret.second - utf16_output; 456| 106| if (ret.first != buf + len) { ------------------ | Branch (456:7): [True: 92, False: 14] ------------------ 457| 92| const size_t scalar_converted_chars = 458| 92| scalar::latin1_to_utf16::convert( 459| 92| ret.first, len - (ret.first - buf), ret.second); 460| 92| if (scalar_converted_chars == 0) { ------------------ | Branch (460:9): [True: 0, False: 92] ------------------ 461| 0| return 0; 462| 0| } 463| 92| converted_chars += scalar_converted_chars; 464| 92| } 465| 106| return converted_chars; 466| 106|} _ZNK7simdutf7haswell14implementation23convert_latin1_to_utf32EPKcmPDi: 471| 100| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 472| 100| std::pair ret = 473| 100| avx2_convert_latin1_to_utf32(buf, len, utf32_output); 474| 100| if (ret.first == nullptr) { ------------------ | Branch (474:7): [True: 4, False: 96] ------------------ 475| 4| return 0; 476| 4| } 477| 96| size_t converted_chars = ret.second - utf32_output; 478| 96| if (ret.first != buf + len) { ------------------ | Branch (478:7): [True: 68, False: 28] ------------------ 479| 68| const size_t scalar_converted_chars = scalar::latin1_to_utf32::convert( 480| 68| ret.first, len - (ret.first - buf), ret.second); 481| 68| if (scalar_converted_chars == 0) { ------------------ | Branch (481:9): [True: 0, False: 68] ------------------ 482| 0| return 0; 483| 0| } 484| 68| converted_chars += scalar_converted_chars; 485| 68| } 486| 96| return converted_chars; 487| 96|} _ZNK7simdutf7haswell14implementation22convert_utf8_to_latin1EPKcmPc: 492| 612| const char *buf, size_t len, char *latin1_output) const noexcept { 493| 612| utf8_to_latin1::validating_transcoder converter; 494| 612| return converter.convert(buf, len, latin1_output); 495| 612|} _ZNK7simdutf7haswell14implementation34convert_utf8_to_latin1_with_errorsEPKcmPc: 498| 522| const char *buf, size_t len, char *latin1_output) const noexcept { 499| 522| utf8_to_latin1::validating_transcoder converter; 500| 522| return converter.convert_with_errors(buf, len, latin1_output); 501| 522|} _ZNK7simdutf7haswell14implementation23convert_utf8_to_utf16leEPKcmPDs: 511| 784| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 512| 784| utf8_to_utf16::validating_transcoder converter; 513| 784| return converter.convert(buf, len, utf16_output); 514| 784|} _ZNK7simdutf7haswell14implementation23convert_utf8_to_utf16beEPKcmPDs: 517| 758| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 518| 758| utf8_to_utf16::validating_transcoder converter; 519| 758| return converter.convert(buf, len, utf16_output); 520| 758|} _ZNK7simdutf7haswell14implementation35convert_utf8_to_utf16le_with_errorsEPKcmPDs: 523| 594| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 524| 594| utf8_to_utf16::validating_transcoder converter; 525| 594| return converter.convert_with_errors(buf, len, 526| 594| utf16_output); 527| 594|} _ZNK7simdutf7haswell14implementation35convert_utf8_to_utf16be_with_errorsEPKcmPDs: 530| 772| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 531| 772| utf8_to_utf16::validating_transcoder converter; 532| 772| return converter.convert_with_errors(buf, len, utf16_output); 533| 772|} _ZNK7simdutf7haswell14implementation29convert_valid_utf8_to_utf16leEPKcmPDs: 536| 432| const char *input, size_t size, char16_t *utf16_output) const noexcept { 537| 432| return utf8_to_utf16::convert_valid(input, size, 538| 432| utf16_output); 539| 432|} _ZNK7simdutf7haswell14implementation29convert_valid_utf8_to_utf16beEPKcmPDs: 542| 460| const char *input, size_t size, char16_t *utf16_output) const noexcept { 543| 460| return utf8_to_utf16::convert_valid(input, size, 544| 460| utf16_output); 545| 460|} _ZNK7simdutf7haswell14implementation21convert_utf8_to_utf32EPKcmPDi: 550| 818| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 551| 818| utf8_to_utf32::validating_transcoder converter; 552| 818| return converter.convert(buf, len, utf32_output); 553| 818|} _ZNK7simdutf7haswell14implementation33convert_utf8_to_utf32_with_errorsEPKcmPDi: 556| 790| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 557| 790| utf8_to_utf32::validating_transcoder converter; 558| 790| return converter.convert_with_errors(buf, len, utf32_output); 559| 790|} _ZNK7simdutf7haswell14implementation27convert_valid_utf8_to_utf32EPKcmPDi: 562| 432| const char *input, size_t size, char32_t *utf32_output) const noexcept { 563| 432| return utf8_to_utf32::convert_valid(input, size, utf32_output); 564| 432|} _ZNK7simdutf7haswell14implementation25convert_utf16le_to_latin1EPKDsmPc: 569| 150| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 570| 150| std::pair ret = 571| 150| haswell::avx2_convert_utf16_to_latin1(buf, len, 572| 150| latin1_output); 573| 150| if (ret.first == nullptr) { ------------------ | Branch (573:7): [True: 46, False: 104] ------------------ 574| 46| return 0; 575| 46| } 576| 104| size_t saved_bytes = ret.second - latin1_output; 577| 104| if (ret.first != buf + len) { ------------------ | Branch (577:7): [True: 90, False: 14] ------------------ 578| 90| const size_t scalar_saved_bytes = 579| 90| scalar::utf16_to_latin1::convert( 580| 90| ret.first, len - (ret.first - buf), ret.second); 581| 90| if (scalar_saved_bytes == 0) { ------------------ | Branch (581:9): [True: 56, False: 34] ------------------ 582| 56| return 0; 583| 56| } 584| 34| saved_bytes += scalar_saved_bytes; 585| 34| } 586| 48| return saved_bytes; 587| 104|} _ZNK7simdutf7haswell14implementation25convert_utf16be_to_latin1EPKDsmPc: 590| 112| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 591| 112| std::pair ret = 592| 112| haswell::avx2_convert_utf16_to_latin1(buf, len, 593| 112| latin1_output); 594| 112| if (ret.first == nullptr) { ------------------ | Branch (594:7): [True: 22, False: 90] ------------------ 595| 22| return 0; 596| 22| } 597| 90| size_t saved_bytes = ret.second - latin1_output; 598| 90| if (ret.first != buf + len) { ------------------ | Branch (598:7): [True: 76, False: 14] ------------------ 599| 76| const size_t scalar_saved_bytes = 600| 76| scalar::utf16_to_latin1::convert( 601| 76| ret.first, len - (ret.first - buf), ret.second); 602| 76| if (scalar_saved_bytes == 0) { ------------------ | Branch (602:9): [True: 46, False: 30] ------------------ 603| 46| return 0; 604| 46| } 605| 30| saved_bytes += scalar_saved_bytes; 606| 30| } 607| 44| return saved_bytes; 608| 90|} _ZNK7simdutf7haswell14implementation37convert_utf16le_to_latin1_with_errorsEPKDsmPc: 612| 226| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 613| 226| std::pair ret = 614| 226| avx2_convert_utf16_to_latin1_with_errors( 615| 226| buf, len, latin1_output); 616| 226| if (ret.first.error) { ------------------ | Branch (616:7): [True: 60, False: 166] ------------------ 617| 60| return ret.first; 618| 60| } // Can return directly since scalar fallback already found correct 619| | // ret.first.count 620| 166| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (620:7): [True: 130, False: 36] ------------------ 621| 130| result scalar_res = 622| 130| scalar::utf16_to_latin1::convert_with_errors( 623| 130| buf + ret.first.count, len - ret.first.count, ret.second); 624| 130| if (scalar_res.error) { ------------------ | Branch (624:9): [True: 72, False: 58] ------------------ 625| 72| scalar_res.count += ret.first.count; 626| 72| return scalar_res; 627| 72| } else { 628| 58| ret.second += scalar_res.count; 629| 58| } 630| 130| } 631| 94| ret.first.count = 632| 94| ret.second - 633| 94| latin1_output; // Set count to the number of 8-bit code units written 634| 94| return ret.first; 635| 166|} _ZNK7simdutf7haswell14implementation37convert_utf16be_to_latin1_with_errorsEPKDsmPc: 639| 230| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 640| 230| std::pair ret = 641| 230| avx2_convert_utf16_to_latin1_with_errors(buf, len, 642| 230| latin1_output); 643| 230| if (ret.first.error) { ------------------ | Branch (643:7): [True: 94, False: 136] ------------------ 644| 94| return ret.first; 645| 94| } // Can return directly since scalar fallback already found correct 646| | // ret.first.count 647| 136| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (647:7): [True: 104, False: 32] ------------------ 648| 104| result scalar_res = 649| 104| scalar::utf16_to_latin1::convert_with_errors( 650| 104| buf + ret.first.count, len - ret.first.count, ret.second); 651| 104| if (scalar_res.error) { ------------------ | Branch (651:9): [True: 62, False: 42] ------------------ 652| 62| scalar_res.count += ret.first.count; 653| 62| return scalar_res; 654| 62| } else { 655| 42| ret.second += scalar_res.count; 656| 42| } 657| 104| } 658| 74| ret.first.count = 659| 74| ret.second - 660| 74| latin1_output; // Set count to the number of 8-bit code units written 661| 74| return ret.first; 662| 136|} _ZNK7simdutf7haswell14implementation23convert_utf16le_to_utf8EPKDsmPc: 679| 876| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 680| 876| std::pair ret = 681| 876| haswell::avx2_convert_utf16_to_utf8(buf, len, 682| 876| utf8_output); 683| 876| if (ret.first == nullptr) { ------------------ | Branch (683:7): [True: 172, False: 704] ------------------ 684| 172| return 0; 685| 172| } 686| 704| size_t saved_bytes = ret.second - utf8_output; 687| 704| if (ret.first != buf + len) { ------------------ | Branch (687:7): [True: 702, False: 2] ------------------ 688| 702| const size_t scalar_saved_bytes = 689| 702| scalar::utf16_to_utf8::convert( 690| 702| ret.first, len - (ret.first - buf), ret.second); 691| 702| if (scalar_saved_bytes == 0) { ------------------ | Branch (691:9): [True: 70, False: 632] ------------------ 692| 70| return 0; 693| 70| } 694| 632| saved_bytes += scalar_saved_bytes; 695| 632| } 696| 634| return saved_bytes; 697| 704|} _ZNK7simdutf7haswell14implementation23convert_utf16be_to_utf8EPKDsmPc: 700| 772| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 701| 772| std::pair ret = 702| 772| haswell::avx2_convert_utf16_to_utf8(buf, len, 703| 772| utf8_output); 704| 772| if (ret.first == nullptr) { ------------------ | Branch (704:7): [True: 82, False: 690] ------------------ 705| 82| return 0; 706| 82| } 707| 690| size_t saved_bytes = ret.second - utf8_output; 708| 690| if (ret.first != buf + len) { ------------------ | Branch (708:7): [True: 688, False: 2] ------------------ 709| 688| const size_t scalar_saved_bytes = 710| 688| scalar::utf16_to_utf8::convert( 711| 688| ret.first, len - (ret.first - buf), ret.second); 712| 688| if (scalar_saved_bytes == 0) { ------------------ | Branch (712:9): [True: 60, False: 628] ------------------ 713| 60| return 0; 714| 60| } 715| 628| saved_bytes += scalar_saved_bytes; 716| 628| } 717| 630| return saved_bytes; 718| 690|} _ZNK7simdutf7haswell14implementation35convert_utf16le_to_utf8_with_errorsEPKDsmPc: 742| 654| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 743| 654| return convert_utf16_to_utf8_with_details(buf, len, 744| 654| utf8_output); 745| 654|} _ZNK7simdutf7haswell14implementation35convert_utf16be_to_utf8_with_errorsEPKDsmPc: 748| 648| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 749| 648| return convert_utf16_to_utf8_with_details(buf, len, 750| 648| utf8_output); 751| 648|} _ZNK7simdutf7haswell14implementation29convert_valid_utf16le_to_utf8EPKDsmPc: 754| 384| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 755| 384| return convert_utf16le_to_utf8(buf, len, utf8_output); 756| 384|} _ZNK7simdutf7haswell14implementation29convert_valid_utf16be_to_utf8EPKDsmPc: 759| 414| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 760| 414| return convert_utf16be_to_utf8(buf, len, utf8_output); 761| 414|} _ZNK7simdutf7haswell14implementation21convert_utf32_to_utf8EPKDimPc: 766| 872| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 767| 872| std::pair ret = 768| 872| avx2_convert_utf32_to_utf8(buf, len, utf8_output); 769| 872| if (ret.first == nullptr) { ------------------ | Branch (769:7): [True: 312, False: 560] ------------------ 770| 312| return 0; 771| 312| } 772| 560| size_t saved_bytes = ret.second - utf8_output; 773| 560| if (ret.first != buf + len) { ------------------ | Branch (773:7): [True: 558, False: 2] ------------------ 774| 558| const size_t scalar_saved_bytes = scalar::utf32_to_utf8::convert( 775| 558| ret.first, len - (ret.first - buf), ret.second); 776| 558| if (scalar_saved_bytes == 0) { ------------------ | Branch (776:9): [True: 266, False: 292] ------------------ 777| 266| return 0; 778| 266| } 779| 292| saved_bytes += scalar_saved_bytes; 780| 292| } 781| 294| return saved_bytes; 782| 560|} _ZNK7simdutf7haswell14implementation23convert_utf32_to_latin1EPKDimPc: 787| 206| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 788| 206| std::pair ret = 789| 206| avx2_convert_utf32_to_latin1(buf, len, latin1_output); 790| 206| if (ret.first == nullptr) { ------------------ | Branch (790:7): [True: 40, False: 166] ------------------ 791| 40| return 0; 792| 40| } 793| 166| size_t saved_bytes = ret.second - latin1_output; 794| 166| if (ret.first != buf + len) { ------------------ | Branch (794:7): [True: 148, False: 18] ------------------ 795| 148| const size_t scalar_saved_bytes = scalar::utf32_to_latin1::convert( 796| 148| ret.first, len - (ret.first - buf), ret.second); 797| 148| if (scalar_saved_bytes == 0) { ------------------ | Branch (797:9): [True: 116, False: 32] ------------------ 798| 116| return 0; 799| 116| } 800| 32| saved_bytes += scalar_saved_bytes; 801| 32| } 802| 50| return saved_bytes; 803| 166|} _ZNK7simdutf7haswell14implementation35convert_utf32_to_latin1_with_errorsEPKDimPc: 806| 454| 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| 454| std::pair ret = 810| 454| avx2_convert_utf32_to_latin1_with_errors(buf, len, latin1_output); 811| 454| if (ret.first.count != len) { ------------------ | Branch (811:7): [True: 436, False: 18] ------------------ 812| 436| result scalar_res = scalar::utf32_to_latin1::convert_with_errors( 813| 436| buf + ret.first.count, len - ret.first.count, ret.second); 814| 436| if (scalar_res.error) { ------------------ | Branch (814:9): [True: 404, False: 32] ------------------ 815| 404| scalar_res.count += ret.first.count; 816| 404| return scalar_res; 817| 404| } else { 818| 32| ret.second += scalar_res.count; 819| 32| } 820| 436| } 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| 454|} _ZNK7simdutf7haswell14implementation33convert_utf32_to_utf8_with_errorsEPKDimPc: 835| 924| 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| 924| std::pair ret = 839| 924| haswell::avx2_convert_utf32_to_utf8_with_errors(buf, len, utf8_output); 840| 924| if (ret.first.count != len) { ------------------ | Branch (840:7): [True: 920, False: 4] ------------------ 841| 920| result scalar_res = scalar::utf32_to_utf8::convert_with_errors( 842| 920| buf + ret.first.count, len - ret.first.count, ret.second); 843| 920| if (scalar_res.error) { ------------------ | Branch (843:9): [True: 690, False: 230] ------------------ 844| 690| scalar_res.count += ret.first.count; 845| 690| return scalar_res; 846| 690| } else { 847| 230| ret.second += scalar_res.count; 848| 230| } 849| 920| } 850| 234| ret.first.count = 851| 234| ret.second - 852| 234| utf8_output; // Set count to the number of 8-bit code units written 853| 234| return ret.first; 854| 924|} _ZNK7simdutf7haswell14implementation24convert_utf16le_to_utf32EPKDsmPDi: 859| 428| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 860| 428| std::pair ret = 861| 428| haswell::avx2_convert_utf16_to_utf32(buf, len, 862| 428| utf32_output); 863| 428| if (ret.first == nullptr) { ------------------ | Branch (863:7): [True: 116, False: 312] ------------------ 864| 116| return 0; 865| 116| } 866| 312| size_t saved_bytes = ret.second - utf32_output; 867| 312| if (ret.first != buf + len) { ------------------ | Branch (867:7): [True: 262, False: 50] ------------------ 868| 262| const size_t scalar_saved_bytes = 869| 262| scalar::utf16_to_utf32::convert( 870| 262| ret.first, len - (ret.first - buf), ret.second); 871| 262| if (scalar_saved_bytes == 0) { ------------------ | Branch (871:9): [True: 82, False: 180] ------------------ 872| 82| return 0; 873| 82| } 874| 180| saved_bytes += scalar_saved_bytes; 875| 180| } 876| 230| return saved_bytes; 877| 312|} _ZNK7simdutf7haswell14implementation24convert_utf16be_to_utf32EPKDsmPDi: 880| 448| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 881| 448| std::pair ret = 882| 448| haswell::avx2_convert_utf16_to_utf32(buf, len, 883| 448| utf32_output); 884| 448| if (ret.first == nullptr) { ------------------ | Branch (884:7): [True: 80, False: 368] ------------------ 885| 80| return 0; 886| 80| } 887| 368| size_t saved_bytes = ret.second - utf32_output; 888| 368| if (ret.first != buf + len) { ------------------ | Branch (888:7): [True: 332, False: 36] ------------------ 889| 332| const size_t scalar_saved_bytes = 890| 332| scalar::utf16_to_utf32::convert( 891| 332| ret.first, len - (ret.first - buf), ret.second); 892| 332| if (scalar_saved_bytes == 0) { ------------------ | Branch (892:9): [True: 102, False: 230] ------------------ 893| 102| return 0; 894| 102| } 895| 230| saved_bytes += scalar_saved_bytes; 896| 230| } 897| 266| return saved_bytes; 898| 368|} _ZNK7simdutf7haswell14implementation36convert_utf16le_to_utf32_with_errorsEPKDsmPDi: 901| 402| 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| 402| std::pair ret = 905| 402| haswell::avx2_convert_utf16_to_utf32_with_errors( 906| 402| buf, len, utf32_output); 907| 402| if (ret.first.error) { ------------------ | Branch (907:7): [True: 118, False: 284] ------------------ 908| 118| return ret.first; 909| 118| } // Can return directly since scalar fallback already found correct 910| | // ret.first.count 911| 284| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (911:7): [True: 254, False: 30] ------------------ 912| 254| result scalar_res = 913| 254| scalar::utf16_to_utf32::convert_with_errors( 914| 254| buf + ret.first.count, len - ret.first.count, ret.second); 915| 254| if (scalar_res.error) { ------------------ | Branch (915:9): [True: 84, False: 170] ------------------ 916| 84| scalar_res.count += ret.first.count; 917| 84| return scalar_res; 918| 170| } else { 919| 170| ret.second += scalar_res.count; 920| 170| } 921| 254| } 922| 200| ret.first.count = 923| 200| ret.second - 924| 200| utf32_output; // Set count to the number of 8-bit code units written 925| 200| return ret.first; 926| 284|} _ZNK7simdutf7haswell14implementation36convert_utf16be_to_utf32_with_errorsEPKDsmPDi: 929| 422| 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| 422| std::pair ret = 933| 422| haswell::avx2_convert_utf16_to_utf32_with_errors( 934| 422| buf, len, utf32_output); 935| 422| if (ret.first.error) { ------------------ | Branch (935:7): [True: 128, False: 294] ------------------ 936| 128| return ret.first; 937| 128| } // Can return directly since scalar fallback already found correct 938| | // ret.first.count 939| 294| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (939:7): [True: 266, False: 28] ------------------ 940| 266| result scalar_res = 941| 266| scalar::utf16_to_utf32::convert_with_errors( 942| 266| buf + ret.first.count, len - ret.first.count, ret.second); 943| 266| if (scalar_res.error) { ------------------ | Branch (943:9): [True: 118, False: 148] ------------------ 944| 118| scalar_res.count += ret.first.count; 945| 118| return scalar_res; 946| 148| } else { 947| 148| ret.second += scalar_res.count; 948| 148| } 949| 266| } 950| 176| ret.first.count = 951| 176| ret.second - 952| 176| utf32_output; // Set count to the number of 8-bit code units written 953| 176| return ret.first; 954| 294|} _ZNK7simdutf7haswell14implementation27convert_valid_utf32_to_utf8EPKDimPc: 959| 188| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 960| 188| return convert_utf32_to_utf8(buf, len, utf8_output); 961| 188|} _ZNK7simdutf7haswell14implementation24convert_utf32_to_utf16leEPKDimPDs: 966| 732| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 967| 732| std::pair ret = 968| 732| avx2_convert_utf32_to_utf16(buf, len, utf16_output); 969| 732| if (ret.first == nullptr) { ------------------ | Branch (969:7): [True: 178, False: 554] ------------------ 970| 178| return 0; 971| 178| } 972| 554| size_t saved_bytes = ret.second - utf16_output; 973| 554| if (ret.first != buf + len) { ------------------ | Branch (973:7): [True: 550, False: 4] ------------------ 974| 550| const size_t scalar_saved_bytes = 975| 550| scalar::utf32_to_utf16::convert( 976| 550| ret.first, len - (ret.first - buf), ret.second); 977| 550| if (scalar_saved_bytes == 0) { ------------------ | Branch (977:9): [True: 178, False: 372] ------------------ 978| 178| return 0; 979| 178| } 980| 372| saved_bytes += scalar_saved_bytes; 981| 372| } 982| 376| return saved_bytes; 983| 554|} _ZNK7simdutf7haswell14implementation24convert_utf32_to_utf16beEPKDimPDs: 986| 540| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 987| 540| std::pair ret = 988| 540| avx2_convert_utf32_to_utf16(buf, len, utf16_output); 989| 540| if (ret.first == nullptr) { ------------------ | Branch (989:7): [True: 134, False: 406] ------------------ 990| 134| return 0; 991| 134| } 992| 406| size_t saved_bytes = ret.second - utf16_output; 993| 406| if (ret.first != buf + len) { ------------------ | Branch (993:7): [True: 404, False: 2] ------------------ 994| 404| const size_t scalar_saved_bytes = 995| 404| scalar::utf32_to_utf16::convert( 996| 404| ret.first, len - (ret.first - buf), ret.second); 997| 404| if (scalar_saved_bytes == 0) { ------------------ | Branch (997:9): [True: 158, False: 246] ------------------ 998| 158| return 0; 999| 158| } 1000| 246| saved_bytes += scalar_saved_bytes; 1001| 246| } 1002| 248| return saved_bytes; 1003| 406|} _ZNK7simdutf7haswell14implementation36convert_utf32_to_utf16le_with_errorsEPKDimPDs: 1006| 714| 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| 714| std::pair ret = 1010| 714| haswell::avx2_convert_utf32_to_utf16_with_errors( 1011| 714| buf, len, utf16_output); 1012| 714| if (ret.first.count != len) { ------------------ | Branch (1012:7): [True: 710, False: 4] ------------------ 1013| 710| result scalar_res = 1014| 710| scalar::utf32_to_utf16::convert_with_errors( 1015| 710| buf + ret.first.count, len - ret.first.count, ret.second); 1016| 710| if (scalar_res.error) { ------------------ | Branch (1016:9): [True: 536, False: 174] ------------------ 1017| 536| scalar_res.count += ret.first.count; 1018| 536| return scalar_res; 1019| 536| } else { 1020| 174| ret.second += scalar_res.count; 1021| 174| } 1022| 710| } 1023| 178| ret.first.count = 1024| 178| ret.second - 1025| 178| utf16_output; // Set count to the number of 8-bit code units written 1026| 178| return ret.first; 1027| 714|} _ZNK7simdutf7haswell14implementation36convert_utf32_to_utf16be_with_errorsEPKDimPDs: 1030| 660| 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| 660| std::pair ret = 1034| 660| haswell::avx2_convert_utf32_to_utf16_with_errors( 1035| 660| buf, len, utf16_output); 1036| 660| if (ret.first.count != len) { ------------------ | Branch (1036:7): [True: 656, False: 4] ------------------ 1037| 656| result scalar_res = 1038| 656| scalar::utf32_to_utf16::convert_with_errors( 1039| 656| buf + ret.first.count, len - ret.first.count, ret.second); 1040| 656| if (scalar_res.error) { ------------------ | Branch (1040:9): [True: 478, False: 178] ------------------ 1041| 478| scalar_res.count += ret.first.count; 1042| 478| return scalar_res; 1043| 478| } else { 1044| 178| ret.second += scalar_res.count; 1045| 178| } 1046| 656| } 1047| 182| ret.first.count = 1048| 182| ret.second - 1049| 182| utf16_output; // Set count to the number of 8-bit code units written 1050| 182| return ret.first; 1051| 660|} _ZNK7simdutf7haswell14implementation30convert_valid_utf32_to_utf16leEPKDimPDs: 1054| 252| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1055| 252| return convert_utf32_to_utf16le(buf, len, utf16_output); 1056| 252|} _ZNK7simdutf7haswell14implementation30convert_valid_utf32_to_utf16beEPKDimPDs: 1059| 114| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1060| 114| return convert_utf32_to_utf16be(buf, len, utf16_output); 1061| 114|} _ZNK7simdutf7haswell14implementation30convert_valid_utf16le_to_utf32EPKDsmPDi: 1064| 120| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1065| 120| return convert_utf16le_to_utf32(buf, len, utf32_output); 1066| 120|} _ZNK7simdutf7haswell14implementation30convert_valid_utf16be_to_utf32EPKDsmPDi: 1069| 166| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1070| 166| return convert_utf16be_to_utf32(buf, len, utf32_output); 1071| 166|} _ZNK7simdutf7haswell14implementation13count_utf16leEPKDsm: 1082| 1.38k| const char16_t *input, size_t length) const noexcept { 1083| 1.38k| return utf16::count_code_points(input, length); 1084| 1.38k|} _ZNK7simdutf7haswell14implementation13count_utf16beEPKDsm: 1087| 1.34k| const char16_t *input, size_t length) const noexcept { 1088| 1.34k| return utf16::count_code_points(input, length); 1089| 1.34k|} _ZNK7simdutf7haswell14implementation10count_utf8EPKcm: 1094| 4.08k|implementation::count_utf8(const char *in, size_t size) const noexcept { 1095| 4.08k| return utf8::count_code_points_bytemask(in, size); 1096| 4.08k|} _ZNK7simdutf7haswell14implementation23latin1_length_from_utf8EPKcm: 1101| 567| const char *buf, size_t len) const noexcept { 1102| 567| return count_utf8(buf, len); 1103| 567|} _ZNK7simdutf7haswell14implementation24utf8_length_from_utf16leEPKDsm: 1108| 778| const char16_t *input, size_t length) const noexcept { 1109| 778| return utf16::utf8_length_from_utf16_bytemask(input, 1110| 778| length); 1111| 778|} _ZNK7simdutf7haswell14implementation24utf8_length_from_utf16beEPKDsm: 1114| 727| const char16_t *input, size_t length) const noexcept { 1115| 727| return utf16::utf8_length_from_utf16_bytemask(input, length); 1116| 727|} _ZNK7simdutf7haswell14implementation25utf32_length_from_utf16leEPKDsm: 1121| 420| const char16_t *input, size_t length) const noexcept { 1122| 420| return utf16::utf32_length_from_utf16(input, length); 1123| 420|} _ZNK7simdutf7haswell14implementation25utf32_length_from_utf16beEPKDsm: 1126| 448| const char16_t *input, size_t length) const noexcept { 1127| 448| return utf16::utf32_length_from_utf16(input, length); 1128| 448|} _ZNK7simdutf7haswell14implementation22utf16_length_from_utf8EPKcm: 1133| 1.92k| const char *input, size_t length) const noexcept { 1134| 1.92k| return utf8::utf16_length_from_utf8_bytemask(input, length); 1135| 1.92k|} _ZNK7simdutf7haswell14implementation23utf8_length_from_latin1EPKcm: 1174| 242| const char *input, size_t len) const noexcept { 1175| 242| const uint8_t *data = reinterpret_cast(input); 1176| 242| size_t answer = len / sizeof(__m256i) * sizeof(__m256i); 1177| 242| size_t i = 0; 1178| 242| if (answer >= 2048) { // long strings optimization ------------------ | Branch (1178:7): [True: 101, False: 141] ------------------ 1179| 101| __m256i four_64bits = _mm256_setzero_si256(); 1180| 2.25k| while (i + sizeof(__m256i) <= len) { ------------------ | Branch (1180:12): [True: 2.15k, False: 101] ------------------ 1181| 2.15k| __m256i runner = _mm256_setzero_si256(); 1182| | // We can do up to 255 loops without overflow. 1183| 2.15k| size_t iterations = (len - i) / sizeof(__m256i); 1184| 2.15k| if (iterations > 255) { ------------------ | Branch (1184:11): [True: 2.05k, False: 101] ------------------ 1185| 2.05k| iterations = 255; 1186| 2.05k| } 1187| 2.15k| size_t max_i = i + iterations * sizeof(__m256i) - sizeof(__m256i); 1188| 133k| for (; i + 4 * sizeof(__m256i) <= max_i; i += 4 * sizeof(__m256i)) { ------------------ | Branch (1188:14): [True: 131k, False: 2.15k] ------------------ 1189| 131k| __m256i input1 = _mm256_loadu_si256((const __m256i *)(data + i)); 1190| 131k| __m256i input2 = 1191| 131k| _mm256_loadu_si256((const __m256i *)(data + i + sizeof(__m256i))); 1192| 131k| __m256i input3 = _mm256_loadu_si256( 1193| 131k| (const __m256i *)(data + i + 2 * sizeof(__m256i))); 1194| 131k| __m256i input4 = _mm256_loadu_si256( 1195| 131k| (const __m256i *)(data + i + 3 * sizeof(__m256i))); 1196| 131k| __m256i input12 = 1197| 131k| _mm256_add_epi8(_mm256_cmpgt_epi8(_mm256_setzero_si256(), input1), 1198| 131k| _mm256_cmpgt_epi8(_mm256_setzero_si256(), input2)); 1199| 131k| __m256i input23 = 1200| 131k| _mm256_add_epi8(_mm256_cmpgt_epi8(_mm256_setzero_si256(), input3), 1201| 131k| _mm256_cmpgt_epi8(_mm256_setzero_si256(), input4)); 1202| 131k| __m256i input1234 = _mm256_add_epi8(input12, input23); 1203| 131k| runner = _mm256_sub_epi8(runner, input1234); 1204| 131k| } 1205| 8.53k| for (; i <= max_i; i += sizeof(__m256i)) { ------------------ | Branch (1205:14): [True: 6.38k, False: 2.15k] ------------------ 1206| 6.38k| __m256i input_256_chunk = 1207| 6.38k| _mm256_loadu_si256((const __m256i *)(data + i)); 1208| 6.38k| runner = _mm256_sub_epi8( 1209| 6.38k| runner, _mm256_cmpgt_epi8(_mm256_setzero_si256(), input_256_chunk)); 1210| 6.38k| } 1211| 2.15k| four_64bits = _mm256_add_epi64( 1212| 2.15k| four_64bits, _mm256_sad_epu8(runner, _mm256_setzero_si256())); 1213| 2.15k| } 1214| 101| answer += _mm256_extract_epi64(four_64bits, 0) + 1215| 101| _mm256_extract_epi64(four_64bits, 1) + 1216| 101| _mm256_extract_epi64(four_64bits, 2) + 1217| 101| _mm256_extract_epi64(four_64bits, 3); 1218| 141| } else if (answer > 0) { ------------------ | Branch (1218:14): [True: 69, False: 72] ------------------ 1219| 1.00k| for (; i + sizeof(__m256i) <= len; i += sizeof(__m256i)) { ------------------ | Branch (1219:12): [True: 932, False: 69] ------------------ 1220| 932| __m256i latin = _mm256_loadu_si256((const __m256i *)(data + i)); 1221| 932| uint32_t non_ascii = _mm256_movemask_epi8(latin); 1222| 932| answer += count_ones(non_ascii); 1223| 932| } 1224| 69| } 1225| 242| return answer + scalar::latin1::utf8_length_from_latin1( 1226| 242| reinterpret_cast(data + i), len - i); 1227| 242|} _ZNK7simdutf7haswell14implementation22utf8_length_from_utf32EPKDim: 1232| 904| const char32_t *input, size_t length) const noexcept { 1233| 904| return utf32::utf8_length_from_utf32(input, length); 1234| 904|} _ZNK7simdutf7haswell14implementation23utf16_length_from_utf32EPKDim: 1239| 1.33k| const char32_t *input, size_t length) const noexcept { 1240| 1.33k| const __m256i v_00000000 = _mm256_setzero_si256(); 1241| 1.33k| const __m256i v_ffff0000 = _mm256_set1_epi32((uint32_t)0xffff0000); 1242| 1.33k| size_t pos = 0; 1243| 1.33k| size_t count = 0; 1244| 530k| for (; pos + 8 <= length; pos += 8) { ------------------ | Branch (1244:10): [True: 529k, False: 1.33k] ------------------ 1245| 529k| __m256i in = _mm256_loadu_si256((__m256i *)(input + pos)); 1246| 529k| const __m256i surrogate_bytemask = 1247| 529k| _mm256_cmpeq_epi32(_mm256_and_si256(in, v_ffff0000), v_00000000); 1248| 529k| const uint32_t surrogate_bitmask = 1249| 529k| static_cast(_mm256_movemask_epi8(surrogate_bytemask)); 1250| 529k| size_t surrogate_count = (32 - count_ones(surrogate_bitmask)) / 4; 1251| 529k| count += 8 + surrogate_count; 1252| 529k| } 1253| 1.33k| return count + 1254| 1.33k| scalar::utf32::utf16_length_from_utf32(input + pos, length - pos); 1255| 1.33k|} _ZNK7simdutf7haswell14implementation22utf32_length_from_utf8EPKcm: 1260| 1.03k| const char *input, size_t length) const noexcept { 1261| 1.03k| return utf8::count_code_points(input, length); 1262| 1.03k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_18is_asciiERKNS1_4simd8simd8x64IhEE: 13| 4.75M|simdutf_really_inline bool is_ascii(const simd8x64 &input) { 14| 4.75M| return input.reduce_or().is_ascii(); 15| 4.75M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_124must_be_2_3_continuationENS1_4simd5simd8IhEES4_: 22| 3.23M| const simd8 prev3) { 23| 3.23M| simd8 is_third_byte = 24| 3.23M| prev2.saturating_sub(0xe0u - 0x80); // Only 111_____ will be > 0x80 25| 3.23M| simd8 is_fourth_byte = 26| 3.23M| prev3.saturating_sub(0xf0u - 0x80); // Only 1111____ will be > 0x80 27| 3.23M| return simd8(is_third_byte | is_fourth_byte); 28| 3.23M|} _ZN7simdutf7haswell34convert_utf16_to_utf8_with_detailsILNS_10endiannessE0EEENS_11full_resultEPKDsmPc: 722| 654| const char16_t *buf, size_t len, char *utf8_output) { 723| 654| std::pair ret = 724| 654| haswell::avx2_convert_utf16_to_utf8_with_errors(buf, len, 725| 654| utf8_output); 726| 654| if (ret.first.error) { ------------------ | Branch (726:7): [True: 150, False: 504] ------------------ 727| 150| return full_result(ret.first.error, ret.first.count, 728| 150| size_t(ret.second - utf8_output)); 729| 150| } 730| 504| if (ret.first.count != len) { ------------------ | Branch (730:7): [True: 500, False: 4] ------------------ 731| 500| full_result sres = 732| 500| scalar::utf16_to_utf8::convert_with_errors( 733| 500| buf + ret.first.count, len - ret.first.count, ret.second, 0); 734| 500| return full_result(sres.error, ret.first.count + sres.input_count, 735| 500| size_t(ret.second - utf8_output) + sres.output_count); 736| 500| } 737| 4| return full_result(error_code::SUCCESS, len, 738| 4| size_t(ret.second - utf8_output)); 739| 504|} _ZN7simdutf7haswell34convert_utf16_to_utf8_with_detailsILNS_10endiannessE1EEENS_11full_resultEPKDsmPc: 722| 648| const char16_t *buf, size_t len, char *utf8_output) { 723| 648| std::pair ret = 724| 648| haswell::avx2_convert_utf16_to_utf8_with_errors(buf, len, 725| 648| utf8_output); 726| 648| if (ret.first.error) { ------------------ | Branch (726:7): [True: 176, False: 472] ------------------ 727| 176| return full_result(ret.first.error, ret.first.count, 728| 176| size_t(ret.second - utf8_output)); 729| 176| } 730| 472| if (ret.first.count != len) { ------------------ | Branch (730:7): [True: 468, False: 4] ------------------ 731| 468| full_result sres = 732| 468| scalar::utf16_to_utf8::convert_with_errors( 733| 468| buf + ret.first.count, len - ret.first.count, ret.second, 0); 734| 468| return full_result(sres.error, ret.first.count + sres.input_count, 735| 468| size_t(ret.second - utf8_output) + sres.output_count); 736| 468| } 737| 4| return full_result(error_code::SUCCESS, len, 738| 4| size_t(ret.second - utf8_output)); 739| 472|} _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| 3.35M|simdutf_really_inline long long int count_ones(uint64_t input_num) { 15| | return _popcnt64(input_num); 16| 3.35M|} _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| 21.1k| simdutf_really_inline simd8() : base8_numeric() {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEC2Ev: 124| 21.1k| simdutf_really_inline base8_numeric() : base8() {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2Ev: 71| 21.1k| simdutf_really_inline base8() : base>() {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2Ev: 14| 21.1k| simdutf_really_inline base() : value{__m256i()} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IhE9reduce_orEv: 294| 4.75M| simdutf_really_inline simd8 reduce_or() const { 295| 4.75M| return this->chunks[0] | this->chunks[1]; 296| 4.75M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEorES5_: 50| 15.2M| simdutf_really_inline Child operator|(const Child other) const { 51| 15.2M| return _mm256_or_si256(*this, other); 52| 15.2M| } 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| 4.75M| simdutf_really_inline bool is_ascii() const { 235| 4.75M| return _mm256_movemask_epi8(*this) == 0; 236| 4.75M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEoRES5_: 59| 7.30M| simdutf_really_inline Child &operator|=(const Child other) { 60| 7.30M| auto this_cast = static_cast(this); 61| 7.30M| *this_cast = *this_cast | other; 62| 7.30M| return *this_cast; 63| 7.30M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi1EEENS4_IhEES8_: 83| 3.39M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 84| | return _mm256_alignr_epi8( 85| 3.39M| *this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N); 86| 3.39M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE3shrILi4EEES4_v: 245| 6.79M| template simdutf_really_inline simd8 shr() const { 246| 6.79M| return simd8(_mm256_srli_epi16(*this, N)) & uint8_t(0xFFu >> N); 247| 6.79M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_: 157| 10.1M| L replace15) const { 158| 10.1M| return lookup_16(simd8::repeat_16( 159| 10.1M| replace0, replace1, replace2, replace3, replace4, replace5, replace6, 160| 10.1M| replace7, replace8, replace9, replace10, replace11, replace12, 161| 10.1M| replace13, replace14, replace15)); 162| 10.1M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES8_: 148| 10.1M| simdutf_really_inline simd8 lookup_16(simd8 lookup_table) const { 149| 10.1M| return _mm256_shuffle_epi8(lookup_table, *this); 150| 10.1M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE9repeat_16Ehhhhhhhhhhhhhhhh: 118| 10.1M| T v14, T v15) { 119| 10.1M| return simd8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, 120| 10.1M| v14, v15, v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, 121| 10.1M| v12, v13, v14, v15); 122| 10.1M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2Ehhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh: 207| 10.1M| : simd8(_mm256_setr_epi8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, 208| 10.1M| v12, v13, v14, v15, v16, v17, v18, v19, v20, v21, 209| 10.1M| v22, v23, v24, v25, v26, v27, v28, v29, v30, 210| 10.1M| v31)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2Eh: 195| 23.4M| simdutf_really_inline simd8(uint8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi2EEENS4_IhEES8_: 83| 3.23M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 84| | return _mm256_alignr_epi8( 85| 3.23M| *this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N); 86| 3.23M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi3EEENS4_IhEES8_: 83| 3.23M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 84| | return _mm256_alignr_epi8( 85| 3.23M| *this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N); 86| 3.23M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE14saturating_subES4_: 214| 7.30M| saturating_sub(const simd8 other) const { 215| 7.30M| return _mm256_subs_epu8(*this, other); 216| 7.30M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IbEC2EDv4_x: 95| 24.3M| simdutf_really_inline simd8(const __m256i _value) : base8(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IbNS2_5simd8IbEEEC2EDv4_x: 73| 24.3M| simdutf_really_inline base8(const __m256i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEC2EDv4_x: 17| 24.3M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEcvRKDv4_xEv: 19| 15.4M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEeoES5_: 56| 3.23M| simdutf_really_inline Child operator^(const Child other) const { 57| 3.23M| return _mm256_xor_si256(*this, other); 58| 3.23M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEC2EPKh: 197| 845k| simdutf_really_inline simd8(const uint8_t values[32]) : simd8(load(values)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE4loadEPKh: 111| 10.3M| static simdutf_really_inline simd8 load(const T values[32]) { 112| 10.3M| return _mm256_loadu_si256(reinterpret_cast(values)); 113| 10.3M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE7gt_bitsES4_: 225| 845k| gt_bits(const simd8 other) const { 226| 845k| return this->saturating_sub(other); 227| 845k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEC2EDv4_x: 17| 71.3M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEcvRKDv4_xEv: 19| 100M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IbE10to_bitmaskEv: 99| 8.87M| simdutf_really_inline uint32_t to_bitmask() const { 100| 8.87M| return uint32_t(_mm256_movemask_epi8(value)); 101| 8.87M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE21any_bits_set_anywhereEv: 241| 2.27M| simdutf_really_inline bool any_bits_set_anywhere() const { 242| 2.27M| return !bits_not_set_anywhere(); 243| 2.27M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE21bits_not_set_anywhereEv: 237| 2.27M| simdutf_really_inline bool bits_not_set_anywhere() const { 238| 2.27M| return _mm256_testz_si256(*this, *this); 239| 2.27M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaEC2EPKa: 274| 6.09M| : chunks{simd8::load(ptr), 275| 6.09M| simd8::load(ptr + sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaE4loadEPKa: 111| 21.6M| static simdutf_really_inline simd8 load(const T values[32]) { 112| 21.6M| return _mm256_loadu_si256(reinterpret_cast(values)); 113| 21.6M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEC2EDv4_x: 169| 38.3M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaEC2EDv4_x: 126| 38.3M| : base8(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5base8IaNS2_5simd8IbEEEC2EDv4_x: 73| 38.3M| simdutf_really_inline base8(const __m256i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEC2EDv4_x: 17| 38.3M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE8is_asciiEv: 298| 5.31M| simdutf_really_inline bool is_ascii() const { 299| 5.31M| return this->reduce_or().is_ascii(); 300| 5.31M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE9reduce_orEv: 294| 5.31M| simdutf_really_inline simd8 reduce_or() const { 295| 5.31M| return this->chunks[0] | this->chunks[1]; 296| 5.31M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEorES5_: 50| 5.31M| simdutf_really_inline Child operator|(const Child other) const { 51| 5.31M| return _mm256_or_si256(*this, other); 52| 5.31M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEcvRKDv4_xEv: 19| 62.0M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaE8is_asciiEv: 177| 5.31M| simdutf_really_inline bool is_ascii() const { 178| 5.31M| return _mm256_movemask_epi8(*this) == 0; 179| 5.31M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE5storeEPa: 277| 963k| simdutf_really_inline void store(T *ptr) const { 278| 963k| this->chunks[0].store(ptr + sizeof(simd8) * 0 / sizeof(T)); 279| 963k| this->chunks[1].store(ptr + sizeof(simd8) * 1 / sizeof(T)); 280| 963k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIaE5storeEPa: 129| 1.92M| simdutf_really_inline void store(T dst[32]) const { 130| 1.92M| return _mm256_storeu_si256(reinterpret_cast<__m256i *>(dst), *this); 131| 1.92M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEcvNS3_IhEEEv: 256| 2.55M|simdutf_really_inline simd8::operator simd8() const { 257| 2.55M| return this->value; 258| 2.55M|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE2ltEa: 325| 1.08M| simdutf_really_inline uint64_t lt(const T m) const { 326| 1.08M| const simd8 mask = simd8::splat(m); 327| 1.08M| return simd8x64(this->chunks[0] < mask, this->chunks[1] < mask) 328| 1.08M| .to_bitmask(); 329| 1.08M| } 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| 2.16M| simdutf_really_inline simd8 operator<(const simd8 other) const { 185| 2.16M| return _mm256_cmpgt_epi8(other, *this); 186| 2.16M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IbEC2ENS2_5simd8IbEES6_: 272| 1.86M| : chunks{chunk0, chunk1} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IbE10to_bitmaskEv: 282| 1.86M| simdutf_really_inline uint64_t to_bitmask() const { 283| 1.86M| uint64_t r_lo = uint32_t(this->chunks[0].to_bitmask()); 284| 1.86M| uint64_t r_hi = this->chunks[1].to_bitmask(); 285| 1.86M| return r_lo | (r_hi << 32); 286| 1.86M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf32EPDi: 310| 1.16M| simdutf_really_inline void store_ascii_as_utf32(char32_t *ptr) const { 311| 1.16M| this->chunks[0].store_ascii_as_utf32(ptr + sizeof(simd8) * 0); 312| 1.16M| this->chunks[1].store_ascii_as_utf32(ptr + sizeof(simd8) * 1); 313| 1.16M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf32EPDi: 36| 2.32M| simdutf_really_inline void store_ascii_as_utf32(char32_t *ptr) const { 37| 2.32M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), 38| 2.32M| _mm256_cvtepu8_epi32(_mm256_castsi256_si128(*this))); 39| 2.32M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 8), 40| 2.32M| _mm256_cvtepu8_epi32(_mm256_castsi256_si128( 41| 2.32M| _mm256_srli_si256(*this, 8)))); 42| 2.32M| _mm256_storeu_si256( 43| 2.32M| reinterpret_cast<__m256i *>(ptr + 16), 44| 2.32M| _mm256_cvtepu8_epi32(_mm256_extractf128_si256(*this, 1))); 45| 2.32M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 24), 46| | _mm256_cvtepu8_epi32(_mm_srli_si128( 47| 2.32M| _mm256_extractf128_si256(*this, 1), 8))); 48| 2.32M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE4zeroEv: 108| 102k| static simdutf_really_inline simd8 zero() { 109| 102k| return _mm256_setzero_si256(); 110| 102k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEgtES4_: 181| 11.0M| simdutf_really_inline simd8 operator>(const simd8 other) const { 182| 11.0M| return _mm256_cmpgt_epi8(*this, other); 183| 11.0M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd5simd8IaEC2Ea: 172| 9.50M| simdutf_really_inline simd8(int8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEmIENS2_5simd8IhEE: 137| 12.2M| simdutf_really_inline simd8 &operator-=(const simd8 other) { 138| 12.2M| *this = *this - other; 139| 12.2M| return *static_cast *>(this); 140| 12.2M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhEmiENS2_5simd8IhEE: 134| 12.2M| simdutf_really_inline simd8 operator-(const simd8 other) const { 135| 12.2M| return _mm256_sub_epi8(*this, other); 136| 12.2M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd10sum_8bytesENS2_5simd8IhEE: 353| 48.0k|simdutf_really_inline simd64 sum_8bytes(const simd8 v) { 354| 48.0k| return _mm256_sad_epu8(v.value, simd8::zero()); 355| 48.0k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE9sum_bytesEv: 249| 2.76k| simdutf_really_inline uint64_t sum_bytes() const { 250| 2.76k| const auto tmp = _mm256_sad_epu8(value, _mm256_setzero_si256()); 251| | 252| 2.76k| return _mm256_extract_epi64(tmp, 0) + _mm256_extract_epi64(tmp, 1) + 253| 2.76k| _mm256_extract_epi64(tmp, 2) + _mm256_extract_epi64(tmp, 3); 254| 2.76k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhEgeES4_: 229| 2.75M| operator>=(const simd8 other) const { 230| 2.75M| return other.min_val(*this) == other; 231| 2.75M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd5simd8IhE7min_valES4_: 220| 2.75M| min_val(const simd8 other) const { 221| 2.75M| return _mm256_min_epu8(other, *this); 222| 2.75M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE2gtEa: 331| 779k| simdutf_really_inline uint64_t gt(const T m) const { 332| 779k| const simd8 mask = simd8::splat(m); 333| 779k| return simd8x64(this->chunks[0] > mask, this->chunks[1] > mask) 334| 779k| .to_bitmask(); 335| 779k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEC2EDv4_x: 17| 14.2M| simdutf_really_inline base(const __m256i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEcvRKDv4_xEv: 19| 14.2M| simdutf_really_inline operator const __m256i &() const { return this->value; } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd13base8_numericIhE5splatEh: 105| 23.5M| static simdutf_really_inline simd8 splat(T _value) { 106| 23.5M| return _mm256_set1_epi8(_value); 107| 23.5M| } 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| 7.90M| const simd8 rhs) { 77| 7.90M| return _mm256_cmpeq_epi8(lhs, rhs); 78| 7.90M| } 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| 4.75M| : chunks{simd8::load(ptr), 275| 4.75M| simd8::load(ptr + sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEanES5_: 53| 11.6M| simdutf_really_inline Child operator&(const Child other) const { 54| 11.6M| return _mm256_and_si256(*this, other); 55| 11.6M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 303| 1.02M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 304| 1.02M| this->chunks[0].template store_ascii_as_utf16(ptr + 305| 1.02M| sizeof(simd8) * 0); 306| 1.02M| this->chunks[1].template store_ascii_as_utf16(ptr + 307| 1.02M| sizeof(simd8) * 1); 308| 1.02M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 22| 2.04M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 23| 2.04M| __m256i first = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(*this)); 24| 2.04M| __m256i second = _mm256_cvtepu8_epi16(_mm256_extractf128_si256(*this, 1)); 25| 2.04M| if (big_endian) { ------------------ | Branch (25:9): [Folded, False: 2.04M] ------------------ 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| 2.04M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), first); 33| 2.04M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 16), second); 34| 2.04M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 303| 1.08M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 304| 1.08M| this->chunks[0].template store_ascii_as_utf16(ptr + 305| 1.08M| sizeof(simd8) * 0); 306| 1.08M| this->chunks[1].template store_ascii_as_utf16(ptr + 307| 1.08M| sizeof(simd8) * 1); 308| 1.08M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 22| 2.16M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 23| 2.16M| __m256i first = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(*this)); 24| 2.16M| __m256i second = _mm256_cvtepu8_epi16(_mm256_extractf128_si256(*this, 1)); 25| 2.16M| if (big_endian) { ------------------ | Branch (25:9): [True: 2.16M, Folded] ------------------ 26| 2.16M| const __m256i swap = _mm256_setr_epi8( 27| 2.16M| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 28| 2.16M| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 29| 2.16M| first = _mm256_shuffle_epi8(first, swap); 30| 2.16M| second = _mm256_shuffle_epi8(second, swap); 31| 2.16M| } 32| 2.16M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr), first); 33| 2.16M| _mm256_storeu_si256(reinterpret_cast<__m256i *>(ptr + 16), second); 34| 2.16M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEorES5_: 50| 4.08M| simdutf_really_inline Child operator|(const Child other) const { 51| 4.08M| return _mm256_or_si256(*this, other); 52| 4.08M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEC2EDv4_x: 97| 71.3M| : base16_numeric(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItEC2EDv4_x: 73| 71.3M| : base16(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6base16ItNS2_6simd16IbEEEC2EDv4_x: 20| 71.3M| : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE4packERKS4_S6_: 149| 5.10M| 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| 5.10M| const __m128i lo_0 = _mm256_extracti128_si256(v0, 0); 156| 5.10M| const __m128i lo_1 = _mm256_extracti128_si256(v1, 0); 157| | 158| | // get the 1st lanes 159| 5.10M| const __m128i hi_0 = _mm256_extracti128_si256(v0, 1); 160| 5.10M| const __m128i hi_1 = _mm256_extracti128_si256(v1, 1); 161| | 162| | // build new vectors (shuffle lanes) 163| 5.10M| const __m256i t0 = _mm256_set_m128i(lo_1, lo_0); ------------------ | | 4| 5.10M| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 5.10M| _mm256_castsi128_si256(xmm2), 2) ------------------ 164| 5.10M| const __m256i t1 = _mm256_set_m128i(hi_1, hi_0); ------------------ | | 4| 5.10M| _mm256_permute2f128_si256(_mm256_castsi128_si256(xmm1), \ | | 5| 5.10M| _mm256_castsi128_si256(xmm2), 2) ------------------ 165| | 166| | // pack code units in linear order from v0 and v1 167| 5.10M| return _mm256_packus_epi16(t0, t1); 168| 5.10M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd9simd16x32ItEC2EPKt: 202| 2.04M| : chunks{simd16::load(ptr), 203| 2.04M| simd16::load(ptr + sizeof(simd16) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItE4loadEPKt: 66| 16.3M| static simdutf_really_inline simd16 load(const T values[8]) { 67| 16.3M| return _mm256_loadu_si256(reinterpret_cast(values)); 68| 16.3M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd9simd16x32ItE10swap_bytesEv: 229| 1.06M| simdutf_really_inline void swap_bytes() { 230| 1.06M| this->chunks[0] = this->chunks[0].swap_bytes(); 231| 1.06M| this->chunks[1] = this->chunks[1].swap_bytes(); 232| 1.06M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE10swap_bytesEv: 139| 3.17M| simdutf_really_inline simd16 swap_bytes() const { 140| 3.17M| const __m256i swap = _mm256_setr_epi8( 141| 3.17M| 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14, 17, 16, 19, 18, 142| 3.17M| 21, 20, 23, 22, 25, 24, 27, 26, 29, 28, 31, 30); 143| 3.17M| return _mm256_shuffle_epi8(*this, swap); 144| 3.17M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItE5splatEt: 58| 15.0M| static simdutf_really_inline simd16 splat(T _value) { 59| 15.0M| return _mm256_set1_epi16(_value); 60| 15.0M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdeqENS2_6simd16ItEES4_: 26| 10.2M| const simd16 rhs) { 27| 10.2M| return _mm256_cmpeq_epi16(lhs, rhs); 28| 10.2M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16IbEC2EDv4_x: 45| 14.2M| simdutf_really_inline simd16(const __m256i _value) : base16(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6base16IbNS2_6simd16IbEEEC2EDv4_x: 20| 14.2M| : base>(_value) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd9simd16x32IbEC2ENS2_6simd16IbEES6_: 200| 2.04M| : chunks{chunk0, chunk1} {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd9simd16x32IbE10to_bitmaskEv: 210| 2.04M| simdutf_really_inline uint64_t to_bitmask() const { 211| 2.04M| uint64_t r_lo = uint32_t(this->chunks[0].to_bitmask()); 212| 2.04M| uint64_t r_hi = this->chunks[1].to_bitmask(); 213| 2.04M| return r_lo | (r_hi << 32); 214| 2.04M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16IbE10to_bitmaskEv: 50| 4.08M| simdutf_really_inline bitmask_type to_bitmask() const { 51| 4.08M| return _mm256_movemask_epi8(*this); 52| 4.08M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE3shrILi8EEES4_v: 134| 4.66M| template simdutf_really_inline simd16 shr() const { 135| 4.66M| return simd16(_mm256_srli_epi16(*this, N)); 136| 4.66M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEleES4_: 117| 4.08M| operator<=(const simd16 other) const { 118| 4.08M| return other.max_val(*this) == other; 119| 4.08M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE7max_valES4_: 108| 4.08M| max_val(const simd16 other) const { 109| 4.08M| return _mm256_max_epu16(*this, other); 110| 4.08M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEC2EPKDs: 104| 10.2M| : simd16(load(reinterpret_cast(values))) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEC2Et: 100| 10.9M| simdutf_really_inline simd16(uint16_t _value) : simd16(splat(_value)) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd9simd16x32ItE12not_in_rangeEtt: 249| 2.04M| simdutf_really_inline uint64_t not_in_range(const T low, const T high) const { 250| 2.04M| const simd16 mask_low = simd16::splat(static_cast(low - 1)); 251| 2.04M| const simd16 mask_high = simd16::splat(static_cast(high + 1)); 252| 2.04M| return simd16x32( 253| 2.04M| (this->chunks[0] >= mask_high) | (this->chunks[0] <= mask_low), 254| 2.04M| (this->chunks[1] >= mask_high) | (this->chunks[1] <= mask_low)) 255| 2.04M| .to_bitmask(); 256| 2.04M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItEgeES4_: 121| 4.08M| operator>=(const simd16 other) const { 122| 4.08M| return other.min_val(*this) == other; 123| 4.08M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE7min_valES4_: 112| 4.08M| min_val(const simd16 other) const { 113| 4.08M| return _mm256_min_epu16(*this, other); 114| 4.08M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItE4zeroEv: 62| 1.54k| static simdutf_really_inline simd16 zero() { 63| 1.54k| return _mm256_setzero_si256(); 64| 1.54k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd3minENS2_6simd16ItEES4_: 259| 4.09M|simd16 min(const simd16 a, simd16 b) { 260| 4.09M| return _mm256_min_epu16(a.value, b.value); 261| 4.09M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItEpLENS2_6simd16ItEE: 87| 6.13M| simdutf_really_inline simd16 &operator+=(const simd16 other) { 88| 6.13M| *this = *this + other; 89| 6.13M| return *static_cast *>(this); 90| 6.13M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd14base16_numericItEplENS2_6simd16ItEE: 84| 6.13M| simdutf_really_inline simd16 operator+(const simd16 other) const { 85| 6.13M| return _mm256_add_epi16(*this, other); 86| 6.13M| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd16ItE3sumEv: 170| 1.54k| simdutf_really_inline uint64_t sum() const { 171| 1.54k| const auto lo_u16 = _mm256_and_si256(value, _mm256_set1_epi32(0x0000ffff)); 172| 1.54k| const auto hi_u16 = _mm256_srli_epi32(value, 16); 173| 1.54k| const auto sum_u32 = _mm256_add_epi32(lo_u16, hi_u16); 174| | 175| 1.54k| const auto lo_u32 = 176| 1.54k| _mm256_and_si256(sum_u32, _mm256_set1_epi64x(0xffffffff)); 177| 1.54k| const auto hi_u32 = _mm256_srli_epi64(sum_u32, 32); 178| 1.54k| const auto sum_u64 = _mm256_add_epi64(lo_u32, hi_u32); 179| | 180| 1.54k| return uint64_t(_mm256_extract_epi64(sum_u64, 0)) + 181| 1.54k| uint64_t(_mm256_extract_epi64(sum_u64, 1)) + 182| 1.54k| uint64_t(_mm256_extract_epi64(sum_u64, 2)) + 183| | uint64_t(_mm256_extract_epi64(sum_u64, 3)); 184| 1.54k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE5splatEj: 46| 18.9k| simdutf_really_inline static simd32 splat(uint32_t v) { 47| 18.9k| return _mm256_set1_epi32(v); 48| 18.9k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjEC2EDv4_x: 9| 6.32M| simdutf_really_inline simd32(const __m256i v) : value(v) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE4zeroEv: 42| 5.74k| simdutf_really_inline static simd32 zero() { 43| 5.74k| return _mm256_setzero_si256(); 44| 5.74k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjEC2IDiEEPKT_: 13| 1.48M| : value(_mm256_loadu_si256(reinterpret_cast(ptr))) {} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE10swap_bytesEv: 27| 1.18M| simdutf_really_inline simd32 swap_bytes() const { 28| 1.18M| const __m256i shuffle = 29| 1.18M| _mm256_setr_epi8(3, 2, 1, 0, 7, 6, 5, 4, 8, 9, 10, 11, 15, 14, 13, 12, 30| 1.18M| 3, 2, 1, 0, 7, 6, 5, 4, 8, 9, 10, 11, 15, 14, 13, 12); 31| | 32| 1.18M| return _mm256_shuffle_epi8(value, shuffle); 33| 1.18M| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd3maxENS2_6simd32IjEES4_: 80| 2.08M| const simd32 b) { 81| 2.08M| return _mm256_max_epu32(a.value, b.value); 82| 2.08M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdplENS2_6simd32IjEES4_: 90| 1.04M| const simd32 b) { 91| 1.04M| return _mm256_add_epi32(a.value, b.value); 92| 1.04M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdgtENS2_6simd32IjEES4_: 109| 4.15k| const simd32 b) { 110| 4.15k| return !(b >= a); 111| 4.15k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdntENS2_6simd32IbEE: 104| 4.15k|simdutf_really_inline simd32 operator!(const simd32 v) { 105| 4.15k| return _mm256_xor_si256(v.value, _mm256_set1_epi8(-1)); 106| 4.15k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IbEC2EDv4_x: 59| 421k| simdutf_really_inline simd32(const __m256i v) : value(v) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdgeENS2_6simd32IjEES4_: 100| 141k| const simd32 b) { 101| 141k| return _mm256_cmpeq_epi32(_mm256_max_epu32(a.value, b.value), a.value); 102| 141k|} simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd32IbE3anyEv: 61| 141k| simdutf_really_inline bool any() const { 62| 141k| return _mm256_movemask_epi8(value) != 0; 63| 141k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdeqENS2_6simd32IjEES4_: 95| 137k| const simd32 b) { 96| 137k| return _mm256_cmpeq_epi32(a.value, b.value); 97| 137k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdanENS2_6simd32IjEES4_: 85| 1.06M| const simd32 a) { 86| 1.06M| return _mm256_and_si256(a.value, b.value); 87| 1.06M|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simdorIbEENS2_6simd32IT_EES6_S6_: 70| 137k| const simd32 b) { 71| 137k| return _mm256_or_si256(a.value, b.value); 72| 137k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd3minENS2_6simd32IjEES4_: 75| 922k| const simd32 a) { 76| 922k| return _mm256_min_epu32(a.value, b.value); 77| 922k|} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd32IjEpLES4_: 36| 922k| simdutf_really_inline simd32 &operator+=(const simd32 other) { 37| 922k| value = _mm256_add_epi32(value, other.value); 38| 922k| return *this; 39| 922k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd32IjE3sumEv: 15| 647| simdutf_really_inline uint64_t sum() const { 16| 647| const __m256i mask = _mm256_set1_epi64x(0xffffffff); 17| 647| const __m256i t0 = _mm256_and_si256(value, mask); 18| 647| const __m256i t1 = _mm256_srli_epi64(value, 32); 19| 647| const __m256i t2 = _mm256_add_epi64(t0, t1); 20| | 21| 647| return uint64_t(_mm256_extract_epi64(t2, 0)) + 22| 647| uint64_t(_mm256_extract_epi64(t2, 1)) + 23| 647| uint64_t(_mm256_extract_epi64(t2, 2)) + 24| | uint64_t(_mm256_extract_epi64(t2, 3)); 25| 647| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd64ImE4zeroEv: 27| 6.01k| simdutf_really_inline static simd64 zero() { 28| 6.01k| return _mm256_setzero_si256(); 29| 6.01k| } simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd64ImEC2EDv4_x: 9| 54.0k| simdutf_really_inline simd64(const __m256i v) : value(v) {} simdutf.cpp:_ZN7simdutf7haswell12_GLOBAL__N_14simd6simd64ImEpLES4_: 21| 48.0k| simdutf_really_inline simd64 &operator+=(const simd64 other) { 22| 48.0k| value = _mm256_add_epi64(value, other.value); 23| 48.0k| return *this; 24| 48.0k| } simdutf.cpp:_ZNK7simdutf7haswell12_GLOBAL__N_14simd6simd64ImE3sumEv: 15| 6.01k| simdutf_really_inline uint64_t sum() const { 16| 6.01k| return _mm256_extract_epi64(value, 0) + _mm256_extract_epi64(value, 1) + 17| 6.01k| _mm256_extract_epi64(value, 2) + _mm256_extract_epi64(value, 3); 18| 6.01k| } _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| 3.88M|simdutf_really_inline long long int count_ones(uint64_t input_num) { 15| | return _popcnt64(input_num); 16| 3.88M|} _ZN7simdutf8westmere14implementationC2Ev: 18| 1| : simdutf::implementation("westmere", "Intel/AMD SSE4.2", 19| 1| internal::instruction_set::SSE42) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2Ev: 183| 21.1k| simdutf_really_inline simd8() : base8_numeric() {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEC2Ev: 118| 21.1k| simdutf_really_inline base8_numeric() : base8() {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2Ev: 68| 21.1k| simdutf_really_inline base8() : base>() {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2Ev: 13| 21.1k| simdutf_really_inline base() : value{__m128i()} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IhE9reduce_orEv: 290| 4.75M| simdutf_really_inline simd8 reduce_or() const { 291| 4.75M| return (this->chunks[0] | this->chunks[1]) | 292| 4.75M| (this->chunks[2] | this->chunks[3]); 293| 4.75M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEorES5_: 42| 31.4M| simdutf_really_inline Child operator|(const Child other) const { 43| 31.4M| return _mm_or_si128(*this, other); 44| 31.4M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2EDv2_x: 185| 315M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEC2EDv2_x: 120| 315M| : base8(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEEC2EDv2_x: 69| 315M| simdutf_really_inline base8(const __m128i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEC2EDv2_x: 16| 315M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE8is_asciiEv: 228| 4.75M| simdutf_really_inline bool is_ascii() const { 229| 4.75M| return _mm_movemask_epi8(*this) == 0; 230| 4.75M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEoRES5_: 51| 10.7M| simdutf_really_inline Child &operator|=(const Child other) { 52| 10.7M| auto this_cast = static_cast(this); 53| 10.7M| *this_cast = *this_cast | other; 54| 10.7M| return *this_cast; 55| 10.7M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi1EEENS4_IhEES8_: 79| 6.79M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 80| | return _mm_alignr_epi8(*this, prev_chunk, 16 - N); 81| 6.79M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE3shrILi4EEES4_v: 238| 13.5M| template simdutf_really_inline simd8 shr() const { 239| 13.5M| return simd8(_mm_srli_epi16(*this, N)) & uint8_t(0xFFu >> N); 240| 13.5M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_S7_: 151| 20.3M| L replace15) const { 152| 20.3M| return lookup_16(simd8::repeat_16( 153| 20.3M| replace0, replace1, replace2, replace3, replace4, replace5, replace6, 154| 20.3M| replace7, replace8, replace9, replace10, replace11, replace12, 155| 20.3M| replace13, replace14, replace15)); 156| 20.3M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE9lookup_16IhEENS2_5simd8IT_EES8_: 142| 20.3M| simdutf_really_inline simd8 lookup_16(simd8 lookup_table) const { 143| 20.3M| return _mm_shuffle_epi8(lookup_table, *this); 144| 20.3M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE9repeat_16Ehhhhhhhhhhhhhhhh: 113| 20.3M| T v14, T v15) { 114| 20.3M| return simd8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12, v13, 115| 20.3M| v14, v15); 116| 20.3M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2Ehhhhhhhhhhhhhhhh: 196| 20.3M| : simd8(_mm_setr_epi8(v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, 197| 20.3M| v12, v13, v14, v15)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2Eh: 188| 46.1M| simdutf_really_inline simd8(uint8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi2EEENS4_IhEES8_: 79| 6.46M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 80| | return _mm_alignr_epi8(*this, prev_chunk, 16 - N); 81| 6.46M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5base8IhNS2_5simd8IbEEE4prevILi3EEENS4_IhEES8_: 79| 6.46M| simdutf_really_inline simd8 prev(const simd8 prev_chunk) const { 80| | return _mm_alignr_epi8(*this, prev_chunk, 16 - N); 81| 6.46M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE14saturating_subES4_: 201| 13.7M| saturating_sub(const simd8 other) const { 202| 13.7M| return _mm_subs_epu8(*this, other); 203| 13.7M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IbEC2EDv2_x: 91| 43.5M| simdutf_really_inline simd8(const __m128i _value) : base8(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IbNS2_5simd8IbEEEC2EDv2_x: 69| 43.5M| simdutf_really_inline base8(const __m128i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEC2EDv2_x: 16| 43.5M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IbEEEcvRKDv2_xEv: 18| 43.5M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEeoES5_: 48| 6.46M| simdutf_really_inline Child operator^(const Child other) const { 49| 6.46M| return _mm_xor_si128(*this, other); 50| 6.46M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEC2EPKh: 190| 845k| simdutf_really_inline simd8(const uint8_t *values) : simd8(load(values)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE4loadEPKh: 106| 19.8M| static simdutf_really_inline simd8 load(const T values[16]) { 107| 19.8M| return _mm_loadu_si128(reinterpret_cast(values)); 108| 19.8M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE7gt_bitsES4_: 212| 845k| gt_bits(const simd8 other) const { 213| 845k| return this->saturating_sub(other); 214| 845k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEC2EDv2_x: 16| 115M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEcvRKDv2_xEv: 18| 154M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IbE10to_bitmaskEv: 95| 12.6M| simdutf_really_inline int to_bitmask() const { 96| 12.6M| return _mm_movemask_epi8(*this); 97| 12.6M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE21any_bits_set_anywhereEv: 235| 2.27M| simdutf_really_inline bool any_bits_set_anywhere() const { 236| 2.27M| return !bits_not_set_anywhere(); 237| 2.27M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE21bits_not_set_anywhereEv: 232| 2.27M| simdutf_really_inline bool bits_not_set_anywhere() const { 233| 2.27M| return _mm_testz_si128(*this, *this); 234| 2.27M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaEC2EPKa: 270| 6.09M| : chunks{simd8::load(ptr), 271| 6.09M| simd8::load(ptr + sizeof(simd8) / sizeof(T)), 272| 6.09M| simd8::load(ptr + 2 * sizeof(simd8) / sizeof(T)), 273| 6.09M| simd8::load(ptr + 3 * sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaE4loadEPKa: 106| 43.3M| static simdutf_really_inline simd8 load(const T values[16]) { 107| 43.3M| return _mm_loadu_si128(reinterpret_cast(values)); 108| 43.3M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEC2EDv2_x: 163| 80.2M| : base8_numeric(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaEC2EDv2_x: 120| 80.2M| : base8(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5base8IaNS2_5simd8IbEEEC2EDv2_x: 69| 80.2M| simdutf_really_inline base8(const __m128i _value) : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEC2EDv2_x: 16| 80.2M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE8is_asciiEv: 295| 5.31M| simdutf_really_inline bool is_ascii() const { 296| 5.31M| return this->reduce_or().is_ascii(); 297| 5.31M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE9reduce_orEv: 290| 5.31M| simdutf_really_inline simd8 reduce_or() const { 291| 5.31M| return (this->chunks[0] | this->chunks[1]) | 292| 5.31M| (this->chunks[2] | this->chunks[3]); 293| 5.31M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEorES5_: 42| 15.9M| simdutf_really_inline Child operator|(const Child other) const { 43| 15.9M| return _mm_or_si128(*this, other); 44| 15.9M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEEcvRKDv2_xEv: 18| 129M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaE8is_asciiEv: 168| 5.31M| simdutf_really_inline bool is_ascii() const { 169| 5.31M| return _mm_movemask_epi8(*this) == 0; 170| 5.31M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE5storeEPa: 275| 963k| simdutf_really_inline void store(T *ptr) const { 276| 963k| this->chunks[0].store(ptr + sizeof(simd8) * 0 / sizeof(T)); 277| 963k| this->chunks[1].store(ptr + sizeof(simd8) * 1 / sizeof(T)); 278| 963k| this->chunks[2].store(ptr + sizeof(simd8) * 2 / sizeof(T)); 279| 963k| this->chunks[3].store(ptr + sizeof(simd8) * 3 / sizeof(T)); 280| 963k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaE5storeEPa: 123| 3.85M| simdutf_really_inline void store(T dst[16]) const { 124| 3.85M| return _mm_storeu_si128(reinterpret_cast<__m128i *>(dst), *this); 125| 3.85M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEcvNS3_IhEEEv: 251| 5.96M|simdutf_really_inline simd8::operator simd8() const { 252| 5.96M| return this->value; 253| 5.96M|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE2ltEa: 326| 1.08M| simdutf_really_inline uint64_t lt(const T m) const { 327| 1.08M| const simd8 mask = simd8::splat(m); 328| 1.08M| return simd8x64(this->chunks[0] < mask, this->chunks[1] < mask, 329| 1.08M| this->chunks[2] < mask, this->chunks[3] < mask) 330| 1.08M| .to_bitmask(); 331| 1.08M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIaE5splatEa: 102| 20.8M| static simdutf_really_inline simd8 splat(T _value) { 103| 20.8M| return _mm_set1_epi8(_value); 104| 20.8M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEltES4_: 176| 4.33M| simdutf_really_inline simd8 operator<(const simd8 other) const { 177| 4.33M| return _mm_cmpgt_epi8(other, *this); 178| 4.33M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IbEC2ENS2_5simd8IbEES6_S6_S6_: 268| 1.86M| : chunks{chunk0, chunk1, chunk2, chunk3} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IbE10to_bitmaskEv: 318| 1.86M| simdutf_really_inline uint64_t to_bitmask() const { 319| 1.86M| uint64_t r0 = uint32_t(this->chunks[0].to_bitmask()); 320| 1.86M| uint64_t r1 = this->chunks[1].to_bitmask(); 321| 1.86M| uint64_t r2 = this->chunks[2].to_bitmask(); 322| 1.86M| uint64_t r3 = this->chunks[3].to_bitmask(); 323| 1.86M| return r0 | (r1 << 16) | (r2 << 32) | (r3 << 48); 324| 1.86M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf32EPDi: 311| 1.16M| simdutf_really_inline void store_ascii_as_utf32(char32_t *ptr) const { 312| 1.16M| this->chunks[0].store_ascii_as_utf32(ptr + sizeof(simd8) * 0); 313| 1.16M| this->chunks[1].store_ascii_as_utf32(ptr + sizeof(simd8) * 1); 314| 1.16M| this->chunks[2].store_ascii_as_utf32(ptr + sizeof(simd8) * 2); 315| 1.16M| this->chunks[3].store_ascii_as_utf32(ptr + sizeof(simd8) * 3); 316| 1.16M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf32EPDi: 32| 4.64M| simdutf_really_inline void store_ascii_as_utf32(char32_t *p) const { 33| 4.64M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p), _mm_cvtepu8_epi32(*this)); 34| 4.64M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 4), 35| 4.64M| _mm_cvtepu8_epi32(_mm_srli_si128(*this, 4))); 36| 4.64M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 8), 37| 4.64M| _mm_cvtepu8_epi32(_mm_srli_si128(*this, 8))); 38| 4.64M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 12), 39| | _mm_cvtepu8_epi32(_mm_srli_si128(*this, 12))); 40| 4.64M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE4zeroEv: 105| 198k| static simdutf_really_inline simd8 zero() { return _mm_setzero_si128(); } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEgtES4_: 173| 22.1M| simdutf_really_inline simd8 operator>(const simd8 other) const { 174| 22.1M| return _mm_cmpgt_epi8(*this, other); 175| 22.1M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd5simd8IaEC2Ea: 165| 19.0M| simdutf_really_inline simd8(int8_t _value) : simd8(splat(_value)) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEmIENS2_5simd8IhEE: 134| 24.5M| simdutf_really_inline simd8 &operator-=(const simd8 other) { 135| 24.5M| *this = *this - other; 136| 24.5M| return *static_cast *>(this); 137| 24.5M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhEmiENS2_5simd8IhEE: 131| 24.5M| simdutf_really_inline simd8 operator-(const simd8 other) const { 132| 24.5M| return _mm_sub_epi8(*this, other); 133| 24.5M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd10sum_8bytesENS2_5simd8IhEE: 366| 96.4k|simdutf_really_inline simd64 sum_8bytes(const simd8 v) { 367| 96.4k| return _mm_sad_epu8(v.value, simd8::zero()); 368| 96.4k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE9sum_bytesEv: 245| 3.80k| simdutf_really_inline uint64_t sum_bytes() const { 246| 3.80k| const auto tmp = _mm_sad_epu8(value, _mm_setzero_si128()); 247| 3.80k| return _mm_extract_epi64(tmp, 0) + _mm_extract_epi64(tmp, 1); 248| 3.80k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhEgeES4_: 217| 5.51M| operator>=(const simd8 other) const { 218| 5.51M| return other.min_val(*this) == other; 219| 5.51M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd5simd8IhE7min_valES4_: 207| 5.51M| min_val(const simd8 other) const { 208| 5.51M| return _mm_min_epu8(*this, other); 209| 5.51M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE2gtEa: 333| 779k| simdutf_really_inline uint64_t gt(const T m) const { 334| 779k| const simd8 mask = simd8::splat(m); 335| 779k| return simd8x64(this->chunks[0] > mask, this->chunks[1] > mask, 336| 779k| this->chunks[2] > mask, this->chunks[3] > mask) 337| 779k| .to_bitmask(); 338| 779k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEC2EDv2_x: 16| 28.5M| simdutf_really_inline base(const __m128i _value) : value(_value) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEcvRKDv2_xEv: 18| 28.5M| simdutf_really_inline operator const __m128i &() const { return this->value; } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd13base8_numericIhE5splatEh: 102| 46.1M| static simdutf_really_inline simd8 splat(T _value) { 103| 46.1M| return _mm_set1_epi8(_value); 104| 46.1M| } 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| 10.6M| const simd8 rhs) { 73| 10.6M| return _mm_cmpeq_epi8(lhs, rhs); 74| 10.6M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IhEEEanES5_: 45| 45.5M| simdutf_really_inline Child operator&(const Child other) const { 46| 45.5M| return _mm_and_si128(*this, other); 47| 45.5M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IhEC2EPKh: 270| 4.75M| : chunks{simd8::load(ptr), 271| 4.75M| simd8::load(ptr + sizeof(simd8) / sizeof(T)), 272| 4.75M| simd8::load(ptr + 2 * sizeof(simd8) / sizeof(T)), 273| 4.75M| simd8::load(ptr + 3 * sizeof(simd8) / sizeof(T))} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16ItEEEanES5_: 45| 17.8M| simdutf_really_inline Child operator&(const Child other) const { 46| 17.8M| return _mm_and_si128(*this, other); 47| 17.8M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 300| 1.02M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 301| 1.02M| this->chunks[0].template store_ascii_as_utf16(ptr + 302| 1.02M| sizeof(simd8) * 0); 303| 1.02M| this->chunks[1].template store_ascii_as_utf16(ptr + 304| 1.02M| sizeof(simd8) * 1); 305| 1.02M| this->chunks[2].template store_ascii_as_utf16(ptr + 306| 1.02M| sizeof(simd8) * 2); 307| 1.02M| this->chunks[3].template store_ascii_as_utf16(ptr + 308| 1.02M| sizeof(simd8) * 3); 309| 1.02M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE0EEEvPDs: 20| 4.09M| simdutf_really_inline void store_ascii_as_utf16(char16_t *p) const { 21| 4.09M| __m128i first = _mm_cvtepu8_epi16(*this); 22| 4.09M| __m128i second = _mm_cvtepu8_epi16(_mm_srli_si128(*this, 8)); 23| 4.09M| if (big_endian) { ------------------ | Branch (23:9): [Folded, False: 4.09M] ------------------ 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| 4.09M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p), first); 30| 4.09M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 8), second); 31| 4.09M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd8simd8x64IaE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 300| 1.08M| simdutf_really_inline void store_ascii_as_utf16(char16_t *ptr) const { 301| 1.08M| this->chunks[0].template store_ascii_as_utf16(ptr + 302| 1.08M| sizeof(simd8) * 0); 303| 1.08M| this->chunks[1].template store_ascii_as_utf16(ptr + 304| 1.08M| sizeof(simd8) * 1); 305| 1.08M| this->chunks[2].template store_ascii_as_utf16(ptr + 306| 1.08M| sizeof(simd8) * 2); 307| 1.08M| this->chunks[3].template store_ascii_as_utf16(ptr + 308| 1.08M| sizeof(simd8) * 3); 309| 1.08M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_5simd8IaEEE20store_ascii_as_utf16ILNS_10endiannessE1EEEvPDs: 20| 4.32M| simdutf_really_inline void store_ascii_as_utf16(char16_t *p) const { 21| 4.32M| __m128i first = _mm_cvtepu8_epi16(*this); 22| 4.32M| __m128i second = _mm_cvtepu8_epi16(_mm_srli_si128(*this, 8)); 23| 4.32M| if (big_endian) { ------------------ | Branch (23:9): [True: 4.32M, Folded] ------------------ 24| 4.32M| const __m128i swap = 25| 4.32M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 26| 4.32M| first = _mm_shuffle_epi8(first, swap); 27| 4.32M| second = _mm_shuffle_epi8(second, swap); 28| 4.32M| } 29| 4.32M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p), first); 30| 4.32M| _mm_storeu_si128(reinterpret_cast<__m128i *>(p + 8), second); 31| 4.32M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd12as_vector_u8ENS2_6simd32IjEE: 370| 7.51k|simdutf_really_inline simd8 as_vector_u8(const simd32 v) { 371| 7.51k| return simd8(v.value); 372| 7.51k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd4baseINS2_6simd16IbEEEorES5_: 42| 8.16M| simdutf_really_inline Child operator|(const Child other) const { 43| 8.16M| return _mm_or_si128(*this, other); 44| 8.16M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEC2EDv2_x: 79| 115M| : base16_numeric(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItEC2EDv2_x: 54| 115M| : base16(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6base16ItNS2_6simd16IbEEEC2EDv2_x: 8| 115M| : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE4packERKS4_S6_: 130| 5.10M| const simd16 &v1) { 131| 5.10M| return _mm_packus_epi16(v0, v1); 132| 5.10M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd9simd16x32ItEC2EPKt: 164| 2.04M| : chunks{simd16::load(ptr), 165| 2.04M| simd16::load(ptr + sizeof(simd16) / sizeof(T)), 166| 2.04M| simd16::load(ptr + 2 * sizeof(simd16) / sizeof(T)), 167| 2.04M| simd16::load(ptr + 3 * sizeof(simd16) / sizeof(T))} {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItE4loadEPKt: 47| 22.4M| static simdutf_really_inline simd16 load(const T values[8]) { 48| 22.4M| return _mm_loadu_si128(reinterpret_cast(values)); 49| 22.4M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd9simd16x32ItE10swap_bytesEv: 200| 1.06M| simdutf_really_inline void swap_bytes() { 201| 1.06M| this->chunks[0] = this->chunks[0].swap_bytes(); 202| 1.06M| this->chunks[1] = this->chunks[1].swap_bytes(); 203| 1.06M| this->chunks[2] = this->chunks[2].swap_bytes(); 204| 1.06M| this->chunks[3] = this->chunks[3].swap_bytes(); 205| 1.06M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE10swap_bytesEv: 121| 6.34M| simdutf_really_inline simd16 swap_bytes() const { 122| 6.34M| const __m128i swap = 123| 6.34M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 124| 6.34M| return _mm_shuffle_epi8(*this, swap); 125| 6.34M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItE5splatEt: 41| 23.2M| static simdutf_really_inline simd16 splat(T _value) { 42| 23.2M| return _mm_set1_epi16(_value); 43| 23.2M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_6simd16ItEES4_: 11| 20.4M| const simd16 rhs) { 12| 20.4M| return _mm_cmpeq_epi16(lhs, rhs); 13| 20.4M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16IbEC2EDv2_x: 28| 28.5M| simdutf_really_inline simd16(const __m128i _value) : base16(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6base16IbNS2_6simd16IbEEEC2EDv2_x: 8| 28.5M| : base>(_value) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd9simd16x32IbEC2ENS2_6simd16IbEES6_S6_S6_: 162| 2.04M| : chunks{chunk0, chunk1, chunk2, chunk3} {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd9simd16x32IbE10to_bitmaskEv: 192| 2.04M| simdutf_really_inline uint64_t to_bitmask() const { 193| 2.04M| uint64_t r0 = uint32_t(this->chunks[0].to_bitmask()); 194| 2.04M| uint64_t r1 = this->chunks[1].to_bitmask(); 195| 2.04M| uint64_t r2 = this->chunks[2].to_bitmask(); 196| 2.04M| uint64_t r3 = this->chunks[3].to_bitmask(); 197| 2.04M| return r0 | (r1 << 16) | (r2 << 32) | (r3 << 48); 198| 2.04M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16IbE10to_bitmaskEv: 33| 8.16M| simdutf_really_inline int to_bitmask() const { 34| 8.16M| return _mm_movemask_epi8(*this); 35| 8.16M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEgeES4_: 104| 8.16M| operator>=(const simd16 other) const { 105| 8.16M| return other.min_val(*this) == other; 106| 8.16M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE7min_valES4_: 95| 8.16M| min_val(const simd16 other) const { 96| 8.16M| return _mm_min_epu16(*this, other); 97| 8.16M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE3shrILi8EEES4_v: 116| 4.66M| template simdutf_really_inline simd16 shr() const { 117| 4.66M| return simd16(_mm_srli_epi16(*this, N)); 118| 4.66M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEleES4_: 100| 8.16M| operator<=(const simd16 other) const { 101| 8.16M| return other.max_val(*this) == other; 102| 8.16M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE7max_valES4_: 90| 8.16M| max_val(const simd16 other) const { 91| 8.16M| return _mm_max_epu16(*this, other); 92| 8.16M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEC2EPKDs: 86| 10.2M| : simd16(load(reinterpret_cast(values))) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd16ItEC2Et: 82| 19.1M| simdutf_really_inline simd16(uint16_t _value) : simd16(splat(_value)) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd9simd16x32ItE12not_in_rangeEtt: 228| 2.04M| simdutf_really_inline uint64_t not_in_range(const T low, const T high) const { 229| 2.04M| const simd16 mask_low = simd16::splat(static_cast(low - 1)); 230| 2.04M| const simd16 mask_high = simd16::splat(static_cast(high + 1)); 231| 2.04M| return simd16x32( 232| 2.04M| (this->chunks[0] >= mask_high) | (this->chunks[0] <= mask_low), 233| 2.04M| (this->chunks[1] >= mask_high) | (this->chunks[1] <= mask_low), 234| 2.04M| (this->chunks[2] >= mask_high) | (this->chunks[2] <= mask_low), 235| 2.04M| (this->chunks[3] >= mask_high) | (this->chunks[3] <= mask_low)) 236| 2.04M| .to_bitmask(); 237| 2.04M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItE4zeroEv: 45| 1.59k| static simdutf_really_inline simd16 zero() { return _mm_setzero_si128(); } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd3minENS2_6simd16ItEES4_: 240| 8.18M|simd16 min(const simd16 a, simd16 b) { 241| 8.18M| return _mm_min_epu16(a.value, b.value); 242| 8.18M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItEpLENS2_6simd16ItEE: 68| 12.2M| simdutf_really_inline simd16 &operator+=(const simd16 other) { 69| 12.2M| *this = *this + other; 70| 12.2M| return *static_cast *>(this); 71| 12.2M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd14base16_numericItEplENS2_6simd16ItEE: 65| 12.2M| simdutf_really_inline simd16 operator+(const simd16 other) const { 66| 12.2M| return _mm_add_epi16(*this, other); 67| 12.2M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd16ItE3sumEv: 134| 1.59k| simdutf_really_inline uint64_t sum() const { 135| 1.59k| const auto lo_u16 = _mm_and_si128(value, _mm_set1_epi32(0x0000ffff)); 136| 1.59k| const auto hi_u16 = _mm_srli_epi32(value, 16); 137| 1.59k| const auto sum_u32 = _mm_add_epi32(lo_u16, hi_u16); 138| | 139| 1.59k| const auto lo_u32 = _mm_and_si128(sum_u32, _mm_set1_epi64x(0xffffffff)); 140| 1.59k| const auto hi_u32 = _mm_srli_epi64(sum_u32, 32); 141| 1.59k| const auto sum_u64 = _mm_add_epi64(lo_u32, hi_u32); 142| | 143| 1.59k| return uint64_t(_mm_extract_epi64(sum_u64, 0)) + 144| | uint64_t(_mm_extract_epi64(sum_u64, 1)); 145| 1.59k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE5splatEj: 56| 30.5k| simdutf_really_inline static simd32 splat(uint32_t v) { 57| 30.5k| return _mm_set1_epi32(v); 58| 30.5k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjEC2EDv2_x: 9| 12.7M| simdutf_really_inline simd32(const __m128i v) : value(v) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE4zeroEv: 52| 6.05k| simdutf_really_inline static simd32 zero() { 53| 6.05k| return _mm_setzero_si128(); 54| 6.05k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjEC2IDiEEPKT_: 13| 2.97M| : value(_mm_loadu_si128(reinterpret_cast(ptr))) {} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE10swap_bytesEv: 22| 2.36M| simdutf_really_inline simd32 swap_bytes() const { 23| 2.36M| const __m128i shuffle = 24| 2.36M| _mm_setr_epi8(3, 2, 1, 0, 7, 6, 5, 4, 8, 9, 10, 11, 15, 14, 13, 12); 25| | 26| 2.36M| return _mm_shuffle_epi8(value, shuffle); 27| 2.36M| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd3maxENS2_6simd32IjEES4_: 94| 4.18M| const simd32 b) { 95| 4.18M| return _mm_max_epu32(a.value, b.value); 96| 4.18M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdplENS2_6simd32IjEES4_: 119| 2.09M| const simd32 b) { 120| 2.09M| return _mm_add_epi32(a.value, b.value); 121| 2.09M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdgtENS2_6simd32IjEES4_: 143| 15.6k| const simd32 b) { 144| 15.6k| return !(b >= a); 145| 15.6k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdntENS2_6simd32IbEE: 138| 15.6k|simdutf_really_inline simd32 operator!(const simd32 v) { 139| 15.6k| return _mm_xor_si128(v.value, _mm_set1_epi8(-1)); 140| 15.6k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IbEC2EDv2_x: 69| 872k| simdutf_really_inline simd32(const __m128i v) : value(v) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdgeENS2_6simd32IjEES4_: 134| 290k| const simd32 b) { 135| 290k| return _mm_cmpeq_epi32(_mm_max_epu32(a.value, b.value), a.value); 136| 290k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IbE3anyEv: 71| 283k| simdutf_really_inline bool any() const { 72| 283k| return _mm_movemask_epi8(value) != 0; 73| 283k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_6simd32IjEES4_: 129| 275k| const simd32 b) { 130| 275k| return _mm_cmpeq_epi32(a.value, b.value); 131| 275k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdanENS2_6simd32IjEES4_: 104| 2.12M| const simd32 b) { 105| 2.12M| return _mm_and_si128(a.value, b.value); 106| 2.12M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdorIbEENS2_6simd32IT_EES6_S6_: 84| 283k| const simd32 b) { 85| 283k| return _mm_or_si128(a.value, b.value); 86| 283k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd3minENS2_6simd32IjEES4_: 89| 1.84M| const simd32 b) { 90| 1.84M| return _mm_min_epu32(a.value, b.value); 91| 1.84M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd32IjEpLES4_: 46| 1.84M| simdutf_really_inline simd32 &operator+=(const simd32 other) { 47| 1.84M| value = _mm_add_epi32(value, other.value); 48| 1.84M| return *this; 49| 1.84M| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE3sumEv: 15| 961| simdutf_really_inline uint64_t sum() const { 16| 961| return uint64_t(_mm_extract_epi32(value, 0)) + 17| 961| uint64_t(_mm_extract_epi32(value, 1)) + 18| 961| uint64_t(_mm_extract_epi32(value, 2)) + 19| | uint64_t(_mm_extract_epi32(value, 3)); 20| 961| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdeqENS2_6simd32IjEEj: 99| 7.51k| uint32_t b) { 100| 7.51k| return _mm_cmpeq_epi32(a.value, _mm_set1_epi32(b)); 101| 7.51k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdanENS2_6simd32IjEEj: 109| 22.5k| uint32_t b) { 110| 22.5k| return _mm_and_si128(a.value, _mm_set1_epi32(b)); 111| 22.5k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IbE15to_4bit_bitmaskEv: 75| 7.51k| simdutf_really_inline uint8_t to_4bit_bitmask() const { 76| 7.51k| return uint8_t(_mm_movemask_ps(_mm_castsi128_ps(value))); 77| 7.51k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdmiENS2_6simd32IjEEj: 124| 7.51k| uint32_t b) { 125| 7.51k| return _mm_sub_epi32(a.value, _mm_set1_epi32(b)); 126| 7.51k|} simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE3shrILi10EEES4_v: 29| 7.51k| template simdutf_really_inline simd32 shr() const { 30| 7.51k| return _mm_srli_epi32(value, N); 31| 7.51k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd32IjE3shlILi16EEES4_v: 33| 7.51k| template simdutf_really_inline simd32 shl() const { 34| 7.51k| return _mm_slli_epi32(value, N); 35| 7.51k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdorENS2_6simd32IjEEj: 114| 7.51k| uint32_t b) { 115| 7.51k| return _mm_or_si128(a.value, _mm_set1_epi32(b)); 116| 7.51k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simdorIjEENS2_6simd32IT_EES6_S6_: 84| 7.51k| const simd32 b) { 85| 7.51k| return _mm_or_si128(a.value, b.value); 86| 7.51k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6selectENS2_6simd32IbEENS3_IjEES5_: 149| 7.51k| const simd32 v_false) { 150| 7.51k| return _mm_blendv_epi8(v_false.value, v_true.value, cond.value); 151| 7.51k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd64ImE4zeroEv: 26| 6.01k| simdutf_really_inline static simd64 zero() { 27| 6.01k| return _mm_setzero_si128(); 28| 6.01k| } simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd64ImEC2EDv2_x: 9| 102k| simdutf_really_inline simd64(const __m128i v) : value(v) {} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_14simd6simd64ImEpLES4_: 20| 96.4k| simdutf_really_inline simd64 &operator+=(const simd64 other) { 21| 96.4k| value = _mm_add_epi64(value, other.value); 22| 96.4k| return *this; 23| 96.4k| } simdutf.cpp:_ZNK7simdutf8westmere12_GLOBAL__N_14simd6simd64ImE3sumEv: 15| 6.01k| simdutf_really_inline uint64_t sum() const { 16| 6.01k| return _mm_extract_epi64(value, 0) + _mm_extract_epi64(value, 1); 17| 6.01k| } _ZNK7simdutf8westmere14implementation13validate_utf8EPKcm: 299| 3.52k|implementation::validate_utf8(const char *buf, size_t len) const noexcept { 300| 3.52k| return westmere::utf8_validation::generic_validate_utf8(buf, len); 301| 3.52k|} _ZNK7simdutf8westmere14implementation25validate_utf8_with_errorsEPKcm: 306| 3.52k| const char *buf, size_t len) const noexcept { 307| 3.52k| return westmere::utf8_validation::generic_validate_utf8_with_errors(buf, len); 308| 3.52k|} _ZNK7simdutf8westmere14implementation16validate_utf16leEPKDsm: 345| 1.38k| size_t len) const noexcept { 346| 1.38k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.38k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 1.37k] | | ------------------ ------------------ 347| | // empty input is valid UTF-16. protect the implementation from 348| | // handling nullptr 349| 16| return true; 350| 16| } 351| 1.37k| const auto res = 352| 1.37k| westmere::utf16::validate_utf16_with_errors(buf, len); 353| 1.37k| if (res.is_err()) { ------------------ | Branch (353:7): [True: 151, False: 1.21k] ------------------ 354| 151| return false; 355| 151| } 356| | 357| 1.21k| if (res.count == len) ------------------ | Branch (357:7): [True: 0, False: 1.21k] ------------------ 358| 0| return true; 359| | 360| 1.21k| return scalar::utf16::validate(buf + res.count, 361| 1.21k| len - res.count); 362| 1.21k|} _ZNK7simdutf8westmere14implementation16validate_utf16beEPKDsm: 368| 1.34k| size_t len) const noexcept { 369| 1.34k| if (simdutf_unlikely(len == 0)) { ------------------ | | 93| 1.34k| #define simdutf_unlikely(x) __builtin_expect(!!(x), 0) | | ------------------ | | | Branch (93:33): [True: 16, False: 1.33k] | | ------------------ ------------------ 370| | // empty input is valid UTF-16. protect the implementation from 371| | // handling nullptr 372| 16| return true; 373| 16| } 374| 1.33k| const auto res = 375| 1.33k| westmere::utf16::validate_utf16_with_errors(buf, len); 376| 1.33k| if (res.is_err()) { ------------------ | Branch (376:7): [True: 133, False: 1.19k] ------------------ 377| 133| return false; 378| 133| } 379| | 380| 1.19k| if (res.count == len) ------------------ | Branch (380:7): [True: 0, False: 1.19k] ------------------ 381| 0| return true; 382| | 383| 1.19k| return scalar::utf16::validate(buf + res.count, 384| 1.19k| len - res.count); 385| 1.19k|} _ZNK7simdutf8westmere14implementation28validate_utf16le_with_errorsEPKDsm: 388| 1.38k| const char16_t *buf, size_t len) const noexcept { 389| 1.38k| const result res = 390| 1.38k| westmere::utf16::validate_utf16_with_errors(buf, len); 391| 1.38k| if (res.count != len) { ------------------ | Branch (391:7): [True: 1.37k, False: 16] ------------------ 392| 1.37k| const result scalar_res = 393| 1.37k| scalar::utf16::validate_with_errors( 394| 1.37k| buf + res.count, len - res.count); 395| 1.37k| return result(scalar_res.error, res.count + scalar_res.count); 396| 1.37k| } else { 397| 16| return res; 398| 16| } 399| 1.38k|} _ZNK7simdutf8westmere14implementation28validate_utf16be_with_errorsEPKDsm: 402| 1.34k| const char16_t *buf, size_t len) const noexcept { 403| 1.34k| const result res = 404| 1.34k| westmere::utf16::validate_utf16_with_errors(buf, len); 405| 1.34k| if (res.count != len) { ------------------ | Branch (405:7): [True: 1.33k, False: 16] ------------------ 406| 1.33k| result scalar_res = scalar::utf16::validate_with_errors( 407| 1.33k| buf + res.count, len - res.count); 408| 1.33k| return result(scalar_res.error, res.count + scalar_res.count); 409| 1.33k| } else { 410| 16| return res; 411| 16| } 412| 1.34k|} _ZNK7simdutf8westmere14implementation14validate_utf32EPKDim: 427| 2.57k|implementation::validate_utf32(const char32_t *buf, size_t len) const noexcept { 428| 2.57k| return utf32::validate(buf, len); 429| 2.57k|} _ZNK7simdutf8westmere14implementation26validate_utf32_with_errorsEPKDim: 434| 2.57k| const char32_t *buf, size_t len) const noexcept { 435| 2.57k| return utf32::validate_with_errors(buf, len); 436| 2.57k|} _ZNK7simdutf8westmere14implementation22convert_latin1_to_utf8EPKcmPc: 441| 484| const char *buf, size_t len, char *utf8_output) const noexcept { 442| | 443| 484| std::pair ret = 444| 484| sse_convert_latin1_to_utf8(buf, len, utf8_output); 445| 484| size_t converted_chars = ret.second - utf8_output; 446| | 447| 484| if (ret.first != buf + len) { ------------------ | Branch (447:7): [True: 412, False: 72] ------------------ 448| 412| const size_t scalar_converted_chars = scalar::latin1_to_utf8::convert( 449| 412| ret.first, len - (ret.first - buf), ret.second); 450| 412| converted_chars += scalar_converted_chars; 451| 412| } 452| | 453| 484| return converted_chars; 454| 484|} _ZNK7simdutf8westmere14implementation25convert_latin1_to_utf16leEPKcmPDs: 459| 88| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 460| 88| std::pair ret = 461| 88| sse_convert_latin1_to_utf16(buf, len, utf16_output); 462| 88| if (ret.first == nullptr) { ------------------ | Branch (462:7): [True: 4, False: 84] ------------------ 463| 4| return 0; 464| 4| } 465| 84| size_t converted_chars = ret.second - utf16_output; 466| 84| if (ret.first != buf + len) { ------------------ | Branch (466:7): [True: 70, False: 14] ------------------ 467| 70| const size_t scalar_converted_chars = 468| 70| scalar::latin1_to_utf16::convert( 469| 70| ret.first, len - (ret.first - buf), ret.second); 470| 70| if (scalar_converted_chars == 0) { ------------------ | Branch (470:9): [True: 0, False: 70] ------------------ 471| 0| return 0; 472| 0| } 473| 70| converted_chars += scalar_converted_chars; 474| 70| } 475| 84| return converted_chars; 476| 84|} _ZNK7simdutf8westmere14implementation25convert_latin1_to_utf16beEPKcmPDs: 479| 110| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 480| 110| std::pair ret = 481| 110| sse_convert_latin1_to_utf16(buf, len, utf16_output); 482| 110| if (ret.first == nullptr) { ------------------ | Branch (482:7): [True: 4, False: 106] ------------------ 483| 4| return 0; 484| 4| } 485| 106| size_t converted_chars = ret.second - utf16_output; 486| 106| if (ret.first != buf + len) { ------------------ | Branch (486:7): [True: 92, False: 14] ------------------ 487| 92| const size_t scalar_converted_chars = 488| 92| scalar::latin1_to_utf16::convert( 489| 92| ret.first, len - (ret.first - buf), ret.second); 490| 92| if (scalar_converted_chars == 0) { ------------------ | Branch (490:9): [True: 0, False: 92] ------------------ 491| 0| return 0; 492| 0| } 493| 92| converted_chars += scalar_converted_chars; 494| 92| } 495| 106| return converted_chars; 496| 106|} _ZNK7simdutf8westmere14implementation23convert_latin1_to_utf32EPKcmPDi: 501| 100| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 502| 100| std::pair ret = 503| 100| sse_convert_latin1_to_utf32(buf, len, utf32_output); 504| 100| if (ret.first == nullptr) { ------------------ | Branch (504:7): [True: 4, False: 96] ------------------ 505| 4| return 0; 506| 4| } 507| 96| size_t converted_chars = ret.second - utf32_output; 508| 96| if (ret.first != buf + len) { ------------------ | Branch (508:7): [True: 80, False: 16] ------------------ 509| 80| const size_t scalar_converted_chars = scalar::latin1_to_utf32::convert( 510| 80| ret.first, len - (ret.first - buf), ret.second); 511| 80| if (scalar_converted_chars == 0) { ------------------ | Branch (511:9): [True: 0, False: 80] ------------------ 512| 0| return 0; 513| 0| } 514| 80| converted_chars += scalar_converted_chars; 515| 80| } 516| 96| return converted_chars; 517| 96|} _ZNK7simdutf8westmere14implementation22convert_utf8_to_latin1EPKcmPc: 522| 612| const char *buf, size_t len, char *latin1_output) const noexcept { 523| 612| utf8_to_latin1::validating_transcoder converter; 524| 612| return converter.convert(buf, len, latin1_output); 525| 612|} _ZNK7simdutf8westmere14implementation34convert_utf8_to_latin1_with_errorsEPKcmPc: 528| 522| const char *buf, size_t len, char *latin1_output) const noexcept { 529| 522| utf8_to_latin1::validating_transcoder converter; 530| 522| return converter.convert_with_errors(buf, len, latin1_output); 531| 522|} _ZNK7simdutf8westmere14implementation23convert_utf8_to_utf16leEPKcmPDs: 541| 784| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 542| 784| utf8_to_utf16::validating_transcoder converter; 543| 784| return converter.convert(buf, len, utf16_output); 544| 784|} _ZNK7simdutf8westmere14implementation23convert_utf8_to_utf16beEPKcmPDs: 547| 758| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 548| 758| utf8_to_utf16::validating_transcoder converter; 549| 758| return converter.convert(buf, len, utf16_output); 550| 758|} _ZNK7simdutf8westmere14implementation35convert_utf8_to_utf16le_with_errorsEPKcmPDs: 553| 594| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 554| 594| utf8_to_utf16::validating_transcoder converter; 555| 594| return converter.convert_with_errors(buf, len, 556| 594| utf16_output); 557| 594|} _ZNK7simdutf8westmere14implementation35convert_utf8_to_utf16be_with_errorsEPKcmPDs: 560| 772| const char *buf, size_t len, char16_t *utf16_output) const noexcept { 561| 772| utf8_to_utf16::validating_transcoder converter; 562| 772| return converter.convert_with_errors(buf, len, utf16_output); 563| 772|} _ZNK7simdutf8westmere14implementation29convert_valid_utf8_to_utf16leEPKcmPDs: 566| 432| const char *input, size_t size, char16_t *utf16_output) const noexcept { 567| 432| return utf8_to_utf16::convert_valid(input, size, 568| 432| utf16_output); 569| 432|} _ZNK7simdutf8westmere14implementation29convert_valid_utf8_to_utf16beEPKcmPDs: 572| 460| const char *input, size_t size, char16_t *utf16_output) const noexcept { 573| 460| return utf8_to_utf16::convert_valid(input, size, 574| 460| utf16_output); 575| 460|} _ZNK7simdutf8westmere14implementation21convert_utf8_to_utf32EPKcmPDi: 580| 818| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 581| 818| utf8_to_utf32::validating_transcoder converter; 582| 818| return converter.convert(buf, len, utf32_output); 583| 818|} _ZNK7simdutf8westmere14implementation33convert_utf8_to_utf32_with_errorsEPKcmPDi: 586| 790| const char *buf, size_t len, char32_t *utf32_output) const noexcept { 587| 790| utf8_to_utf32::validating_transcoder converter; 588| 790| return converter.convert_with_errors(buf, len, utf32_output); 589| 790|} _ZNK7simdutf8westmere14implementation27convert_valid_utf8_to_utf32EPKcmPDi: 592| 432| const char *input, size_t size, char32_t *utf32_output) const noexcept { 593| 432| return utf8_to_utf32::convert_valid(input, size, utf32_output); 594| 432|} _ZNK7simdutf8westmere14implementation25convert_utf16le_to_latin1EPKDsmPc: 599| 150| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 600| 150| std::pair ret = 601| 150| sse_convert_utf16_to_latin1(buf, len, latin1_output); 602| 150| if (ret.first == nullptr) { ------------------ | Branch (602:7): [True: 68, False: 82] ------------------ 603| 68| return 0; 604| 68| } 605| 82| size_t saved_bytes = ret.second - latin1_output; 606| | 607| 82| if (ret.first != buf + len) { ------------------ | Branch (607:7): [True: 64, False: 18] ------------------ 608| 64| const size_t scalar_saved_bytes = 609| 64| scalar::utf16_to_latin1::convert( 610| 64| ret.first, len - (ret.first - buf), ret.second); 611| 64| if (scalar_saved_bytes == 0) { ------------------ | Branch (611:9): [True: 34, False: 30] ------------------ 612| 34| return 0; 613| 34| } 614| 30| saved_bytes += scalar_saved_bytes; 615| 30| } 616| 48| return saved_bytes; 617| 82|} _ZNK7simdutf8westmere14implementation25convert_utf16be_to_latin1EPKDsmPc: 620| 112| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 621| 112| std::pair ret = 622| 112| sse_convert_utf16_to_latin1(buf, len, latin1_output); 623| 112| if (ret.first == nullptr) { ------------------ | Branch (623:7): [True: 40, False: 72] ------------------ 624| 40| return 0; 625| 40| } 626| 72| size_t saved_bytes = ret.second - latin1_output; 627| | 628| 72| if (ret.first != buf + len) { ------------------ | Branch (628:7): [True: 54, False: 18] ------------------ 629| 54| const size_t scalar_saved_bytes = 630| 54| scalar::utf16_to_latin1::convert( 631| 54| ret.first, len - (ret.first - buf), ret.second); 632| 54| if (scalar_saved_bytes == 0) { ------------------ | Branch (632:9): [True: 28, False: 26] ------------------ 633| 28| return 0; 634| 28| } 635| 26| saved_bytes += scalar_saved_bytes; 636| 26| } 637| 44| return saved_bytes; 638| 72|} _ZNK7simdutf8westmere14implementation37convert_utf16le_to_latin1_with_errorsEPKDsmPc: 642| 226| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 643| 226| std::pair ret = 644| 226| sse_convert_utf16_to_latin1_with_errors( 645| 226| buf, len, latin1_output); 646| 226| if (ret.first.error) { ------------------ | Branch (646:7): [True: 98, False: 128] ------------------ 647| 98| return ret.first; 648| 98| } // Can return directly since scalar fallback already found correct 649| | // ret.first.count 650| 128| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (650:7): [True: 88, False: 40] ------------------ 651| 88| result scalar_res = 652| 88| scalar::utf16_to_latin1::convert_with_errors( 653| 88| buf + ret.first.count, len - ret.first.count, ret.second); 654| 88| if (scalar_res.error) { ------------------ | Branch (654:9): [True: 34, False: 54] ------------------ 655| 34| scalar_res.count += ret.first.count; 656| 34| return scalar_res; 657| 54| } else { 658| 54| ret.second += scalar_res.count; 659| 54| } 660| 88| } 661| 94| ret.first.count = 662| 94| ret.second - 663| 94| latin1_output; // Set count to the number of 8-bit code units written 664| 94| return ret.first; 665| 128|} _ZNK7simdutf8westmere14implementation37convert_utf16be_to_latin1_with_errorsEPKDsmPc: 669| 230| const char16_t *buf, size_t len, char *latin1_output) const noexcept { 670| 230| std::pair ret = 671| 230| sse_convert_utf16_to_latin1_with_errors(buf, len, 672| 230| latin1_output); 673| 230| if (ret.first.error) { ------------------ | Branch (673:7): [True: 116, False: 114] ------------------ 674| 116| return ret.first; 675| 116| } // Can return directly since scalar fallback already found correct 676| | // ret.first.count 677| 114| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (677:7): [True: 80, False: 34] ------------------ 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: 40, False: 40] ------------------ 682| 40| scalar_res.count += ret.first.count; 683| 40| return scalar_res; 684| 40| } else { 685| 40| ret.second += scalar_res.count; 686| 40| } 687| 80| } 688| 74| ret.first.count = 689| 74| ret.second - 690| 74| latin1_output; // Set count to the number of 8-bit code units written 691| 74| return ret.first; 692| 114|} _ZNK7simdutf8westmere14implementation23convert_utf16le_to_utf8EPKDsmPc: 709| 876| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 710| 876| std::pair ret = 711| 876| sse_convert_utf16_to_utf8(buf, len, utf8_output); 712| 876| if (ret.first == nullptr) { ------------------ | Branch (712:7): [True: 152, False: 724] ------------------ 713| 152| return 0; 714| 152| } 715| 724| size_t saved_bytes = ret.second - utf8_output; 716| 724| if (ret.first != buf + len) { ------------------ | Branch (716:7): [True: 722, False: 2] ------------------ 717| 722| const size_t scalar_saved_bytes = 718| 722| scalar::utf16_to_utf8::convert( 719| 722| ret.first, len - (ret.first - buf), ret.second); 720| 722| if (scalar_saved_bytes == 0) { ------------------ | Branch (720:9): [True: 90, False: 632] ------------------ 721| 90| return 0; 722| 90| } 723| 632| saved_bytes += scalar_saved_bytes; 724| 632| } 725| 634| return saved_bytes; 726| 724|} _ZNK7simdutf8westmere14implementation23convert_utf16be_to_utf8EPKDsmPc: 729| 772| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 730| 772| std::pair ret = 731| 772| sse_convert_utf16_to_utf8(buf, len, utf8_output); 732| 772| if (ret.first == nullptr) { ------------------ | Branch (732:7): [True: 74, False: 698] ------------------ 733| 74| return 0; 734| 74| } 735| 698| size_t saved_bytes = ret.second - utf8_output; 736| 698| if (ret.first != buf + len) { ------------------ | Branch (736:7): [True: 696, False: 2] ------------------ 737| 696| const size_t scalar_saved_bytes = 738| 696| scalar::utf16_to_utf8::convert( 739| 696| ret.first, len - (ret.first - buf), ret.second); 740| 696| if (scalar_saved_bytes == 0) { ------------------ | Branch (740:9): [True: 68, False: 628] ------------------ 741| 68| return 0; 742| 68| } 743| 628| saved_bytes += scalar_saved_bytes; 744| 628| } 745| 630| return saved_bytes; 746| 698|} _ZNK7simdutf8westmere14implementation35convert_utf16le_to_utf8_with_errorsEPKDsmPc: 770| 654| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 771| 654| return convert_utf16_to_utf8_with_details(buf, len, 772| 654| utf8_output); 773| 654|} _ZNK7simdutf8westmere14implementation35convert_utf16be_to_utf8_with_errorsEPKDsmPc: 776| 648| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 777| 648| return convert_utf16_to_utf8_with_details(buf, len, 778| 648| utf8_output); 779| 648|} _ZNK7simdutf8westmere14implementation29convert_valid_utf16le_to_utf8EPKDsmPc: 782| 384| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 783| 384| return convert_utf16le_to_utf8(buf, len, utf8_output); 784| 384|} _ZNK7simdutf8westmere14implementation29convert_valid_utf16be_to_utf8EPKDsmPc: 787| 414| const char16_t *buf, size_t len, char *utf8_output) const noexcept { 788| 414| return convert_utf16be_to_utf8(buf, len, utf8_output); 789| 414|} _ZNK7simdutf8westmere14implementation23convert_utf32_to_latin1EPKDimPc: 794| 206| const char32_t *buf, size_t len, char *latin1_output) const noexcept { 795| 206| std::pair ret = 796| 206| sse_convert_utf32_to_latin1(buf, len, latin1_output); 797| 206| if (ret.first == nullptr) { ------------------ | Branch (797:7): [True: 54, False: 152] ------------------ 798| 54| return 0; 799| 54| } 800| 152| size_t saved_bytes = ret.second - latin1_output; 801| | // if (ret.first != buf + len) { 802| 152| if (ret.first < buf + len) { ------------------ | Branch (802:7): [True: 132, False: 20] ------------------ 803| 132| const size_t scalar_saved_bytes = scalar::utf32_to_latin1::convert( 804| 132| ret.first, len - (ret.first - buf), ret.second); 805| 132| if (scalar_saved_bytes == 0) { ------------------ | Branch (805:9): [True: 102, False: 30] ------------------ 806| 102| return 0; 807| 102| } 808| 30| saved_bytes += scalar_saved_bytes; 809| 30| } 810| 50| return saved_bytes; 811| 152|} _ZNK7simdutf8westmere14implementation35convert_utf32_to_latin1_with_errorsEPKDimPc: 814| 454| 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| 454| std::pair ret = 818| 454| westmere::sse_convert_utf32_to_latin1_with_errors(buf, len, 819| 454| latin1_output); 820| 454| if (ret.first.count != len) { ------------------ | Branch (820:7): [True: 430, False: 24] ------------------ 821| 430| result scalar_res = scalar::utf32_to_latin1::convert_with_errors( 822| 430| buf + ret.first.count, len - ret.first.count, ret.second); 823| 430| if (scalar_res.error) { ------------------ | Branch (823:9): [True: 404, False: 26] ------------------ 824| 404| scalar_res.count += ret.first.count; 825| 404| return scalar_res; 826| 404| } else { 827| 26| ret.second += scalar_res.count; 828| 26| } 829| 430| } 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| 454|} _ZNK7simdutf8westmere14implementation21convert_utf32_to_utf8EPKDimPc: 845| 872| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 846| 872| std::pair ret = 847| 872| sse_convert_utf32_to_utf8(buf, len, utf8_output); 848| 872| if (ret.first == nullptr) { ------------------ | Branch (848:7): [True: 282, False: 590] ------------------ 849| 282| return 0; 850| 282| } 851| 590| size_t saved_bytes = ret.second - utf8_output; 852| 590| if (ret.first != buf + len) { ------------------ | Branch (852:7): [True: 588, False: 2] ------------------ 853| 588| const size_t scalar_saved_bytes = scalar::utf32_to_utf8::convert( 854| 588| ret.first, len - (ret.first - buf), ret.second); 855| 588| if (scalar_saved_bytes == 0) { ------------------ | Branch (855:9): [True: 296, False: 292] ------------------ 856| 296| return 0; 857| 296| } 858| 292| saved_bytes += scalar_saved_bytes; 859| 292| } 860| 294| return saved_bytes; 861| 590|} _ZNK7simdutf8westmere14implementation33convert_utf32_to_utf8_with_errorsEPKDimPc: 864| 924| 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| 924| std::pair ret = 868| 924| westmere::sse_convert_utf32_to_utf8_with_errors(buf, len, utf8_output); 869| 924| if (ret.first.count != len) { ------------------ | Branch (869:7): [True: 920, False: 4] ------------------ 870| 920| result scalar_res = scalar::utf32_to_utf8::convert_with_errors( 871| 920| buf + ret.first.count, len - ret.first.count, ret.second); 872| 920| if (scalar_res.error) { ------------------ | Branch (872:9): [True: 690, False: 230] ------------------ 873| 690| scalar_res.count += ret.first.count; 874| 690| return scalar_res; 875| 690| } else { 876| 230| ret.second += scalar_res.count; 877| 230| } 878| 920| } 879| 234| ret.first.count = 880| 234| ret.second - 881| 234| utf8_output; // Set count to the number of 8-bit code units written 882| 234| return ret.first; 883| 924|} _ZNK7simdutf8westmere14implementation24convert_utf16le_to_utf32EPKDsmPDi: 888| 428| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 889| 428| std::pair ret = 890| 428| sse_convert_utf16_to_utf32(buf, len, utf32_output); 891| 428| if (ret.first == nullptr) { ------------------ | Branch (891:7): [True: 136, False: 292] ------------------ 892| 136| return 0; 893| 136| } 894| 292| size_t saved_bytes = ret.second - utf32_output; 895| 292| if (ret.first != buf + len) { ------------------ | Branch (895:7): [True: 216, False: 76] ------------------ 896| 216| const size_t scalar_saved_bytes = 897| 216| scalar::utf16_to_utf32::convert( 898| 216| ret.first, len - (ret.first - buf), ret.second); 899| 216| if (scalar_saved_bytes == 0) { ------------------ | Branch (899:9): [True: 62, False: 154] ------------------ 900| 62| return 0; 901| 62| } 902| 154| saved_bytes += scalar_saved_bytes; 903| 154| } 904| 230| return saved_bytes; 905| 292|} _ZNK7simdutf8westmere14implementation24convert_utf16be_to_utf32EPKDsmPDi: 908| 448| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 909| 448| std::pair ret = 910| 448| sse_convert_utf16_to_utf32(buf, len, utf32_output); 911| 448| if (ret.first == nullptr) { ------------------ | Branch (911:7): [True: 108, False: 340] ------------------ 912| 108| return 0; 913| 108| } 914| 340| size_t saved_bytes = ret.second - utf32_output; 915| 340| if (ret.first != buf + len) { ------------------ | Branch (915:7): [True: 280, False: 60] ------------------ 916| 280| const size_t scalar_saved_bytes = 917| 280| scalar::utf16_to_utf32::convert( 918| 280| ret.first, len - (ret.first - buf), ret.second); 919| 280| if (scalar_saved_bytes == 0) { ------------------ | Branch (919:9): [True: 74, False: 206] ------------------ 920| 74| return 0; 921| 74| } 922| 206| saved_bytes += scalar_saved_bytes; 923| 206| } 924| 266| return saved_bytes; 925| 340|} _ZNK7simdutf8westmere14implementation36convert_utf16le_to_utf32_with_errorsEPKDsmPDi: 928| 402| 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| 402| std::pair ret = 932| 402| westmere::sse_convert_utf16_to_utf32_with_errors( 933| 402| buf, len, utf32_output); 934| 402| if (ret.first.error) { ------------------ | Branch (934:7): [True: 140, False: 262] ------------------ 935| 140| return ret.first; 936| 140| } // Can return directly since scalar fallback already found correct 937| | // ret.first.count 938| 262| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (938:7): [True: 216, False: 46] ------------------ 939| 216| result scalar_res = 940| 216| scalar::utf16_to_utf32::convert_with_errors( 941| 216| buf + ret.first.count, len - ret.first.count, ret.second); 942| 216| if (scalar_res.error) { ------------------ | Branch (942:9): [True: 62, False: 154] ------------------ 943| 62| scalar_res.count += ret.first.count; 944| 62| return scalar_res; 945| 154| } else { 946| 154| ret.second += scalar_res.count; 947| 154| } 948| 216| } 949| 200| ret.first.count = 950| 200| ret.second - 951| 200| utf32_output; // Set count to the number of 8-bit code units written 952| 200| return ret.first; 953| 262|} _ZNK7simdutf8westmere14implementation36convert_utf16be_to_utf32_with_errorsEPKDsmPDi: 956| 422| 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| 422| std::pair ret = 960| 422| westmere::sse_convert_utf16_to_utf32_with_errors( 961| 422| buf, len, utf32_output); 962| 422| if (ret.first.error) { ------------------ | Branch (962:7): [True: 166, False: 256] ------------------ 963| 166| return ret.first; 964| 166| } // Can return directly since scalar fallback already found correct 965| | // ret.first.count 966| 256| if (ret.first.count != len) { // All good so far, but not finished ------------------ | Branch (966:7): [True: 222, False: 34] ------------------ 967| 222| result scalar_res = 968| 222| scalar::utf16_to_utf32::convert_with_errors( 969| 222| buf + ret.first.count, len - ret.first.count, ret.second); 970| 222| if (scalar_res.error) { ------------------ | Branch (970:9): [True: 80, False: 142] ------------------ 971| 80| scalar_res.count += ret.first.count; 972| 80| return scalar_res; 973| 142| } else { 974| 142| ret.second += scalar_res.count; 975| 142| } 976| 222| } 977| 176| ret.first.count = 978| 176| ret.second - 979| 176| utf32_output; // Set count to the number of 8-bit code units written 980| 176| return ret.first; 981| 256|} _ZNK7simdutf8westmere14implementation27convert_valid_utf32_to_utf8EPKDimPc: 986| 188| const char32_t *buf, size_t len, char *utf8_output) const noexcept { 987| 188| return convert_utf32_to_utf8(buf, len, utf8_output); 988| 188|} _ZNK7simdutf8westmere14implementation24convert_utf32_to_utf16leEPKDimPDs: 993| 732| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 994| 732| std::pair ret = 995| 732| sse_convert_utf32_to_utf16(buf, len, utf16_output); 996| 732| if (ret.first == nullptr) { ------------------ | Branch (996:7): [True: 72, False: 660] ------------------ 997| 72| return 0; 998| 72| } 999| 660| size_t saved_bytes = ret.second - utf16_output; 1000| 660| if (ret.first != buf + len) { ------------------ | Branch (1000:7): [True: 656, False: 4] ------------------ 1001| 656| const size_t scalar_saved_bytes = 1002| 656| scalar::utf32_to_utf16::convert( 1003| 656| ret.first, len - (ret.first - buf), ret.second); 1004| 656| if (scalar_saved_bytes == 0) { ------------------ | Branch (1004:9): [True: 284, False: 372] ------------------ 1005| 284| return 0; 1006| 284| } 1007| 372| saved_bytes += scalar_saved_bytes; 1008| 372| } 1009| 376| return saved_bytes; 1010| 660|} _ZNK7simdutf8westmere14implementation24convert_utf32_to_utf16beEPKDimPDs: 1013| 540| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1014| 540| std::pair ret = 1015| 540| sse_convert_utf32_to_utf16(buf, len, utf16_output); 1016| 540| if (ret.first == nullptr) { ------------------ | Branch (1016:7): [True: 46, False: 494] ------------------ 1017| 46| return 0; 1018| 46| } 1019| 494| size_t saved_bytes = ret.second - utf16_output; 1020| 494| if (ret.first != buf + len) { ------------------ | Branch (1020:7): [True: 492, False: 2] ------------------ 1021| 492| const size_t scalar_saved_bytes = 1022| 492| scalar::utf32_to_utf16::convert( 1023| 492| ret.first, len - (ret.first - buf), ret.second); 1024| 492| if (scalar_saved_bytes == 0) { ------------------ | Branch (1024:9): [True: 246, False: 246] ------------------ 1025| 246| return 0; 1026| 246| } 1027| 246| saved_bytes += scalar_saved_bytes; 1028| 246| } 1029| 248| return saved_bytes; 1030| 494|} _ZNK7simdutf8westmere14implementation36convert_utf32_to_utf16le_with_errorsEPKDimPDs: 1033| 714| 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| 714| std::pair ret = 1037| 714| westmere::sse_convert_utf32_to_utf16_with_errors( 1038| 714| buf, len, utf16_output); 1039| 714| if (ret.first.count != len) { ------------------ | Branch (1039:7): [True: 688, False: 26] ------------------ 1040| 688| result scalar_res = 1041| 688| scalar::utf32_to_utf16::convert_with_errors( 1042| 688| buf + ret.first.count, len - ret.first.count, ret.second); 1043| 688| if (scalar_res.error) { ------------------ | Branch (1043:9): [True: 536, False: 152] ------------------ 1044| 536| scalar_res.count += ret.first.count; 1045| 536| return scalar_res; 1046| 536| } else { 1047| 152| ret.second += scalar_res.count; 1048| 152| } 1049| 688| } 1050| 178| ret.first.count = 1051| 178| ret.second - 1052| 178| utf16_output; // Set count to the number of 8-bit code units written 1053| 178| return ret.first; 1054| 714|} _ZNK7simdutf8westmere14implementation36convert_utf32_to_utf16be_with_errorsEPKDimPDs: 1057| 660| 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| 660| std::pair ret = 1061| 660| westmere::sse_convert_utf32_to_utf16_with_errors( 1062| 660| buf, len, utf16_output); 1063| 660| if (ret.first.count != len) { ------------------ | Branch (1063:7): [True: 638, False: 22] ------------------ 1064| 638| result scalar_res = 1065| 638| scalar::utf32_to_utf16::convert_with_errors( 1066| 638| buf + ret.first.count, len - ret.first.count, ret.second); 1067| 638| if (scalar_res.error) { ------------------ | Branch (1067:9): [True: 478, False: 160] ------------------ 1068| 478| scalar_res.count += ret.first.count; 1069| 478| return scalar_res; 1070| 478| } else { 1071| 160| ret.second += scalar_res.count; 1072| 160| } 1073| 638| } 1074| 182| ret.first.count = 1075| 182| ret.second - 1076| 182| utf16_output; // Set count to the number of 8-bit code units written 1077| 182| return ret.first; 1078| 660|} _ZNK7simdutf8westmere14implementation30convert_valid_utf32_to_utf16leEPKDimPDs: 1081| 252| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1082| 252| return convert_utf32_to_utf16le(buf, len, utf16_output); 1083| 252|} _ZNK7simdutf8westmere14implementation30convert_valid_utf32_to_utf16beEPKDimPDs: 1086| 114| const char32_t *buf, size_t len, char16_t *utf16_output) const noexcept { 1087| 114| return convert_utf32_to_utf16be(buf, len, utf16_output); 1088| 114|} _ZNK7simdutf8westmere14implementation30convert_valid_utf16le_to_utf32EPKDsmPDi: 1091| 120| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1092| 120| return convert_utf16le_to_utf32(buf, len, utf32_output); 1093| 120|} _ZNK7simdutf8westmere14implementation30convert_valid_utf16be_to_utf32EPKDsmPDi: 1096| 166| const char16_t *buf, size_t len, char32_t *utf32_output) const noexcept { 1097| 166| return convert_utf16be_to_utf32(buf, len, utf32_output); 1098| 166|} _ZNK7simdutf8westmere14implementation13count_utf16leEPKDsm: 1109| 1.38k| const char16_t *input, size_t length) const noexcept { 1110| 1.38k| return utf16::count_code_points(input, length); 1111| 1.38k|} _ZNK7simdutf8westmere14implementation13count_utf16beEPKDsm: 1114| 1.34k| const char16_t *input, size_t length) const noexcept { 1115| 1.34k| return utf16::count_code_points(input, length); 1116| 1.34k|} _ZNK7simdutf8westmere14implementation10count_utf8EPKcm: 1121| 4.08k|implementation::count_utf8(const char *input, size_t length) const noexcept { 1122| 4.08k| return utf8::count_code_points_bytemask(input, length); 1123| 4.08k|} _ZNK7simdutf8westmere14implementation23latin1_length_from_utf8EPKcm: 1128| 567| const char *buf, size_t len) const noexcept { 1129| 567| return count_utf8(buf, len); 1130| 567|} _ZNK7simdutf8westmere14implementation24utf8_length_from_utf16leEPKDsm: 1135| 778| const char16_t *input, size_t length) const noexcept { 1136| 778| return utf16::utf8_length_from_utf16_bytemask(input, 1137| 778| length); 1138| 778|} _ZNK7simdutf8westmere14implementation24utf8_length_from_utf16beEPKDsm: 1141| 727| const char16_t *input, size_t length) const noexcept { 1142| 727| return utf16::utf8_length_from_utf16_bytemask(input, length); 1143| 727|} _ZNK7simdutf8westmere14implementation23utf8_length_from_latin1EPKcm: 1148| 242| const char *input, size_t len) const noexcept { 1149| 242| const uint8_t *str = reinterpret_cast(input); 1150| 242| size_t answer = len / sizeof(__m128i) * sizeof(__m128i); 1151| 242| size_t i = 0; 1152| 242| if (answer >= 2048) { // long strings optimization ------------------ | Branch (1152:7): [True: 101, False: 141] ------------------ 1153| 101| __m128i two_64bits = _mm_setzero_si128(); 1154| 4.34k| while (i + sizeof(__m128i) <= len) { ------------------ | Branch (1154:12): [True: 4.24k, False: 101] ------------------ 1155| 4.24k| __m128i runner = _mm_setzero_si128(); 1156| 4.24k| size_t iterations = (len - i) / sizeof(__m128i); 1157| 4.24k| if (iterations > 255) { ------------------ | Branch (1157:11): [True: 4.14k, False: 101] ------------------ 1158| 4.14k| iterations = 255; 1159| 4.14k| } 1160| 4.24k| size_t max_i = i + iterations * sizeof(__m128i) - sizeof(__m128i); 1161| 266k| for (; i + 4 * sizeof(__m128i) <= max_i; i += 4 * sizeof(__m128i)) { ------------------ | Branch (1161:14): [True: 262k, False: 4.24k] ------------------ 1162| 262k| __m128i input1 = _mm_loadu_si128((const __m128i *)(str + i)); 1163| 262k| __m128i input2 = 1164| 262k| _mm_loadu_si128((const __m128i *)(str + i + sizeof(__m128i))); 1165| 262k| __m128i input3 = 1166| 262k| _mm_loadu_si128((const __m128i *)(str + i + 2 * sizeof(__m128i))); 1167| 262k| __m128i input4 = 1168| 262k| _mm_loadu_si128((const __m128i *)(str + i + 3 * sizeof(__m128i))); 1169| 262k| __m128i input12 = 1170| 262k| _mm_add_epi8(_mm_cmpgt_epi8(_mm_setzero_si128(), input1), 1171| 262k| _mm_cmpgt_epi8(_mm_setzero_si128(), input2)); 1172| 262k| __m128i input34 = 1173| 262k| _mm_add_epi8(_mm_cmpgt_epi8(_mm_setzero_si128(), input3), 1174| 262k| _mm_cmpgt_epi8(_mm_setzero_si128(), input4)); 1175| 262k| __m128i input1234 = _mm_add_epi8(input12, input34); 1176| 262k| runner = _mm_sub_epi8(runner, input1234); 1177| 262k| } 1178| 16.8k| for (; i <= max_i; i += sizeof(__m128i)) { ------------------ | Branch (1178:14): [True: 12.6k, False: 4.24k] ------------------ 1179| 12.6k| __m128i more_input = _mm_loadu_si128((const __m128i *)(str + i)); 1180| 12.6k| runner = _mm_sub_epi8(runner, 1181| 12.6k| _mm_cmpgt_epi8(_mm_setzero_si128(), more_input)); 1182| 12.6k| } 1183| 4.24k| two_64bits = 1184| 4.24k| _mm_add_epi64(two_64bits, _mm_sad_epu8(runner, _mm_setzero_si128())); 1185| 4.24k| } 1186| 101| answer += 1187| 101| _mm_extract_epi64(two_64bits, 0) + _mm_extract_epi64(two_64bits, 1); 1188| 141| } else if (answer > 0) { // short string optimization ------------------ | Branch (1188:14): [True: 93, False: 48] ------------------ 1189| 1.02k| for (; i + 2 * sizeof(__m128i) <= len; i += 2 * sizeof(__m128i)) { ------------------ | Branch (1189:12): [True: 932, False: 93] ------------------ 1190| 932| __m128i latin = _mm_loadu_si128((const __m128i *)(input + i)); 1191| 932| uint16_t non_ascii = (uint16_t)_mm_movemask_epi8(latin); 1192| 932| answer += count_ones(non_ascii); 1193| 932| latin = _mm_loadu_si128((const __m128i *)(input + i) + 1); 1194| 932| non_ascii = (uint16_t)_mm_movemask_epi8(latin); 1195| 932| answer += count_ones(non_ascii); 1196| 932| } 1197| 141| for (; i + sizeof(__m128i) <= len; i += sizeof(__m128i)) { ------------------ | Branch (1197:12): [True: 48, False: 93] ------------------ 1198| 48| __m128i latin = _mm_loadu_si128((const __m128i *)(input + i)); 1199| 48| uint16_t non_ascii = (uint16_t)_mm_movemask_epi8(latin); 1200| 48| answer += count_ones(non_ascii); 1201| 48| } 1202| 93| } 1203| 242| return answer + scalar::latin1::utf8_length_from_latin1( 1204| 242| reinterpret_cast(str + i), len - i); 1205| 242|} _ZNK7simdutf8westmere14implementation25utf32_length_from_utf16leEPKDsm: 1210| 420| const char16_t *input, size_t length) const noexcept { 1211| 420| return utf16::utf32_length_from_utf16(input, length); 1212| 420|} _ZNK7simdutf8westmere14implementation25utf32_length_from_utf16beEPKDsm: 1215| 448| const char16_t *input, size_t length) const noexcept { 1216| 448| return utf16::utf32_length_from_utf16(input, length); 1217| 448|} _ZNK7simdutf8westmere14implementation22utf16_length_from_utf8EPKcm: 1222| 1.92k| const char *input, size_t length) const noexcept { 1223| 1.92k| return utf8::utf16_length_from_utf8_bytemask(input, length); 1224| 1.92k|} _ZNK7simdutf8westmere14implementation22utf8_length_from_utf32EPKDim: 1263| 904| const char32_t *input, size_t length) const noexcept { 1264| 904| return utf32::utf8_length_from_utf32(input, length); 1265| 904|} _ZNK7simdutf8westmere14implementation23utf16_length_from_utf32EPKDim: 1270| 1.33k| const char32_t *input, size_t length) const noexcept { 1271| 1.33k| const __m128i v_00000000 = _mm_setzero_si128(); 1272| 1.33k| const __m128i v_ffff0000 = _mm_set1_epi32((uint32_t)0xffff0000); 1273| 1.33k| size_t pos = 0; 1274| 1.33k| size_t count = 0; 1275| 1.06M| for (; pos + 4 <= length; pos += 4) { ------------------ | Branch (1275:10): [True: 1.05M, False: 1.33k] ------------------ 1276| 1.05M| __m128i in = _mm_loadu_si128((__m128i *)(input + pos)); 1277| 1.05M| const __m128i surrogate_bytemask = 1278| 1.05M| _mm_cmpeq_epi32(_mm_and_si128(in, v_ffff0000), v_00000000); 1279| 1.05M| const uint16_t surrogate_bitmask = 1280| 1.05M| static_cast(_mm_movemask_epi8(surrogate_bytemask)); 1281| 1.05M| size_t surrogate_count = (16 - count_ones(surrogate_bitmask)) / 4; 1282| 1.05M| count += 4 + surrogate_count; 1283| 1.05M| } 1284| 1.33k| return count + 1285| 1.33k| scalar::utf32::utf16_length_from_utf32(input + pos, length - pos); 1286| 1.33k|} _ZNK7simdutf8westmere14implementation22utf32_length_from_utf8EPKcm: 1291| 1.03k| const char *input, size_t length) const noexcept { 1292| 1.03k| return utf8::count_code_points(input, length); 1293| 1.03k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_18is_asciiERKNS1_4simd8simd8x64IhEE: 13| 4.75M|simdutf_really_inline bool is_ascii(const simd8x64 &input) { 14| 4.75M| return input.reduce_or().is_ascii(); 15| 4.75M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_124must_be_2_3_continuationENS1_4simd5simd8IhEES4_: 22| 6.46M| const simd8 prev3) { 23| 6.46M| simd8 is_third_byte = 24| 6.46M| prev2.saturating_sub(0xe0u - 0x80); // Only 111_____ will be >= 0x80 25| 6.46M| simd8 is_fourth_byte = 26| 6.46M| prev3.saturating_sub(0xf0u - 0x80); // Only 1111____ will be >= 0x80 27| 6.46M| return simd8(is_third_byte | is_fourth_byte); 28| 6.46M|} _ZN7simdutf8westmere34convert_utf16_to_utf8_with_detailsILNS_10endiannessE0EEENS_11full_resultEPKDsmPc: 750| 654| const char16_t *buf, size_t len, char *utf8_output) { 751| 654| std::pair ret = 752| 654| westmere::sse_convert_utf16_to_utf8_with_errors(buf, len, 753| 654| utf8_output); 754| 654| if (ret.first.error) { ------------------ | Branch (754:7): [True: 122, False: 532] ------------------ 755| 122| return full_result(ret.first.error, ret.first.count, 756| 122| size_t(ret.second - utf8_output)); 757| 122| } 758| 532| if (ret.first.count != len) { ------------------ | Branch (758:7): [True: 528, False: 4] ------------------ 759| 528| full_result sres = 760| 528| scalar::utf16_to_utf8::convert_with_errors( 761| 528| buf + ret.first.count, len - ret.first.count, ret.second, 0); 762| 528| return full_result(sres.error, ret.first.count + sres.input_count, 763| 528| size_t(ret.second - utf8_output) + sres.output_count); 764| 528| } 765| 4| return full_result(error_code::SUCCESS, len, 766| 4| size_t(ret.second - utf8_output)); 767| 532|} _ZN7simdutf8westmere34convert_utf16_to_utf8_with_detailsILNS_10endiannessE1EEENS_11full_resultEPKDsmPc: 750| 648| const char16_t *buf, size_t len, char *utf8_output) { 751| 648| std::pair ret = 752| 648| westmere::sse_convert_utf16_to_utf8_with_errors(buf, len, 753| 648| utf8_output); 754| 648| if (ret.first.error) { ------------------ | Branch (754:7): [True: 154, False: 494] ------------------ 755| 154| return full_result(ret.first.error, ret.first.count, 756| 154| size_t(ret.second - utf8_output)); 757| 154| } 758| 494| if (ret.first.count != len) { ------------------ | Branch (758:7): [True: 490, False: 4] ------------------ 759| 490| full_result sres = 760| 490| scalar::utf16_to_utf8::convert_with_errors( 761| 490| buf + ret.first.count, len - ret.first.count, ret.second, 0); 762| 490| return full_result(sres.error, ret.first.count + sres.input_count, 763| 490| size_t(ret.second - utf8_output) + sres.output_count); 764| 490| } 765| 4| return full_result(error_code::SUCCESS, len, 766| 4| size_t(ret.second - utf8_output)); 767| 494|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_18internal8westmere26write_v_u16_11bits_to_utf8EDv2_xRPcS4_S4_: 57| 382k| const __m128i v_ff80) { 58| | // no bits set above 7th bit 59| 382k| const __m128i one_byte_bytemask = 60| 382k| _mm_cmpeq_epi16(_mm_and_si128(v_u16, v_ff80), v_0000); 61| 382k| const uint16_t one_byte_bitmask = 62| 382k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 63| | 64| 382k| write_v_u16_11bits_to_utf8(v_u16, utf8_output, one_byte_bytemask, 65| 382k| one_byte_bitmask); 66| 382k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_18internal8westmere26write_v_u16_11bits_to_utf8EDv2_xRPcS4_t: 10| 714k| const uint16_t one_byte_bitmask) { 11| | // 0b1100_0000_1000_0000 12| 714k| const __m128i v_c080 = _mm_set1_epi16((int16_t)0xc080); 13| | // 0b0001_1111_0000_0000 14| 714k| const __m128i v_1f00 = _mm_set1_epi16((int16_t)0x1f00); 15| | // 0b0000_0000_0011_1111 16| 714k| 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| 714k| const __m128i t0 = _mm_slli_epi16(v_u16, 2); 24| | // t1 = [000a|aaaa|0000|0000] 25| 714k| const __m128i t1 = _mm_and_si128(t0, v_1f00); 26| | // t2 = [0000|0000|00bb|bbbb] 27| 714k| const __m128i t2 = _mm_and_si128(v_u16, v_003f); 28| | // t3 = [000a|aaaa|00bb|bbbb] 29| 714k| const __m128i t3 = _mm_or_si128(t1, t2); 30| | // t4 = [110a|aaaa|10bb|bbbb] 31| 714k| const __m128i t4 = _mm_or_si128(t3, v_c080); 32| | 33| | // 2. merge ASCII and 2-byte codewords 34| 714k| 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| 714k| const uint16_t m0 = one_byte_bitmask & 0x5555; // m0 = 0h0g0f0e0d0c0b0a 40| 714k| const uint16_t m1 = static_cast(m0 >> 7); // m1 = 00000000h0g0f0e0 41| 714k| const uint8_t m2 = static_cast((m0 | m1) & 0xff); // m2 = hdgcfbea 42| | // 4. pack the bytes 43| 714k| const uint8_t *row = 44| 714k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[m2][0]; 45| 714k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 46| 714k| const __m128i utf8_packed = _mm_shuffle_epi8(utf8_unpacked, shuffle); 47| | 48| | // 5. store bytes 49| 714k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 50| | 51| | // 6. adjust pointers 52| 714k| utf8_output += row[0]; 53| 714k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_latin1_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 88| char16_t *utf16_output) { 5| 88| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| 175k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (6:22): [True: 175k, False: 88] ------------------ 7| | // Load 16 Latin1 characters into a 128-bit register 8| 175k| __m128i in = 9| 175k| _mm_loadu_si128(reinterpret_cast(&latin1_input[i])); 10| 175k| __m128i out1 = big_endian ? _mm_unpacklo_epi8(_mm_setzero_si128(), in) ------------------ | Branch (10:20): [Folded, False: 175k] ------------------ 11| 175k| : _mm_unpacklo_epi8(in, _mm_setzero_si128()); 12| 175k| __m128i out2 = big_endian ? _mm_unpackhi_epi8(_mm_setzero_si128(), in) ------------------ | Branch (12:20): [Folded, False: 175k] ------------------ 13| 175k| : _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| 175k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i]), out1); 17| 175k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i + 8]), out2); 18| 175k| } 19| | // return pointers pointing to where we left off 20| 88| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 21| 88|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_latin1_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKcPDsEES7_mS8_: 4| 110| char16_t *utf16_output) { 5| 110| size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 6| 180k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (6:22): [True: 180k, False: 110] ------------------ 7| | // Load 16 Latin1 characters into a 128-bit register 8| 180k| __m128i in = 9| 180k| _mm_loadu_si128(reinterpret_cast(&latin1_input[i])); 10| 180k| __m128i out1 = big_endian ? _mm_unpacklo_epi8(_mm_setzero_si128(), in) ------------------ | Branch (10:20): [True: 180k, Folded] ------------------ 11| 180k| : _mm_unpacklo_epi8(in, _mm_setzero_si128()); 12| 180k| __m128i out2 = big_endian ? _mm_unpackhi_epi8(_mm_setzero_si128(), in) ------------------ | Branch (12:20): [True: 180k, Folded] ------------------ 13| 180k| : _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| 180k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i]), out1); 17| 180k| _mm_storeu_si128(reinterpret_cast<__m128i *>(&utf16_output[i + 8]), out2); 18| 180k| } 19| | // return pointers pointing to where we left off 20| 110| return std::make_pair(latin1_input + rounded_len, utf16_output + rounded_len); 21| 110|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_latin1_to_utf32EPKcmPDi: 3| 100| char32_t *utf32_output) { 4| 100| const char *end = buf + len; 5| | 6| 168k| while (end - buf >= 16) { ------------------ | Branch (6:10): [True: 168k, False: 100] ------------------ 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| 100| return std::make_pair(buf, utf32_output); 31| 100|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_latin1_to_utf8EPKcmPc: 3| 484| const size_t latin_input_length, char *utf8_output) { 4| 484| const char *end = latin_input + latin_input_length; 5| | 6| 484| const __m128i v_0000 = _mm_setzero_si128(); 7| | // 0b1000_0000 8| 484| const __m128i v_80 = _mm_set1_epi8((uint8_t)0x80); 9| | // 0b1111_1111_1000_0000 10| 484| const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80); 11| | 12| 484| const __m128i latin_1_half_into_u16_byte_mask = 13| 484| _mm_setr_epi8(0, '\x80', 1, '\x80', 2, '\x80', 3, '\x80', 4, '\x80', 5, 14| 484| '\x80', 6, '\x80', 7, '\x80'); 15| | 16| 484| const __m128i latin_2_half_into_u16_byte_mask = 17| 484| _mm_setr_epi8(8, '\x80', 9, '\x80', 10, '\x80', 11, '\x80', 12, '\x80', 18| 484| 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| 2.12M| while (end - latin_input >= 16 + 8) { ------------------ | Branch (25:10): [True: 2.12M, False: 484] ------------------ 26| | // Load 16 Latin1 characters (16 bytes) into a 128-bit register 27| 2.12M| __m128i v_latin = _mm_loadu_si128((__m128i *)latin_input); 28| | 29| 2.12M| if (_mm_testz_si128(v_latin, v_80)) { // ASCII fast path!!!! ------------------ | Branch (29:9): [True: 1.93M, False: 191k] ------------------ 30| 1.93M| _mm_storeu_si128((__m128i *)utf8_output, v_latin); 31| 1.93M| latin_input += 16; 32| 1.93M| utf8_output += 16; 33| 1.93M| continue; 34| 1.93M| } 35| | 36| | // assuming a/b are bytes and A/B are uint16 of the same value 37| | // aaaa_aaaa_bbbb_bbbb -> AAAA_AAAA 38| 191k| __m128i v_u16_latin_1_half = 39| 191k| _mm_shuffle_epi8(v_latin, latin_1_half_into_u16_byte_mask); 40| | // aaaa_aaaa_bbbb_bbbb -> BBBB_BBBB 41| 191k| __m128i v_u16_latin_2_half = 42| 191k| _mm_shuffle_epi8(v_latin, latin_2_half_into_u16_byte_mask); 43| | 44| 191k| internal::westmere::write_v_u16_11bits_to_utf8(v_u16_latin_1_half, 45| 191k| utf8_output, v_0000, v_ff80); 46| 191k| internal::westmere::write_v_u16_11bits_to_utf8(v_u16_latin_2_half, 47| 191k| utf8_output, v_0000, v_ff80); 48| 191k| latin_input += 16; 49| 191k| } 50| | 51| 484| if (end - latin_input >= 16) { ------------------ | Branch (51:7): [True: 256, False: 228] ------------------ 52| | // Load 16 Latin1 characters (16 bytes) into a 128-bit register 53| 256| __m128i v_latin = _mm_loadu_si128((__m128i *)latin_input); 54| | 55| 256| if (_mm_testz_si128(v_latin, v_80)) { // ASCII fast path!!!! ------------------ | Branch (55:9): [True: 88, False: 168] ------------------ 56| 88| _mm_storeu_si128((__m128i *)utf8_output, v_latin); 57| 88| latin_input += 16; 58| 88| utf8_output += 16; 59| 168| } else { 60| | // assuming a/b are bytes and A/B are uint16 of the same value 61| | // aaaa_aaaa_bbbb_bbbb -> AAAA_AAAA 62| 168| __m128i v_u16_latin_1_half = 63| 168| _mm_shuffle_epi8(v_latin, latin_1_half_into_u16_byte_mask); 64| 168| internal::westmere::write_v_u16_11bits_to_utf8( 65| 168| v_u16_latin_1_half, utf8_output, v_0000, v_ff80); 66| 168| latin_input += 8; 67| 168| } 68| 256| } 69| | 70| 484| return std::make_pair(latin_input, utf8_output); 71| 484|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_utf16_to_latin1ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 150| char *latin1_output) { 5| 150| const char16_t *end = buf + len; 6| 33.2k| while (end - buf >= 8) { ------------------ | Branch (6:10): [True: 33.2k, False: 86] ------------------ 7| | // Load 8 UTF-16 characters into 128-bit SSE register 8| 33.2k| __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| 33.2k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 17| 33.2k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (17:9): [True: 33.1k, False: 64] ------------------ 18| | // Pack 16-bit characters into 8-bit and store in latin1_output 19| 33.1k| __m128i latin1_packed = _mm_packus_epi16(in, in); 20| 33.1k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 21| 33.1k| latin1_packed); 22| | // Adjust pointers for next iteration 23| 33.1k| buf += 8; 24| 33.1k| latin1_output += 8; 25| 33.1k| } else { 26| 64| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 27| 64| } 28| 33.2k| } // while 29| 86| return std::make_pair(buf, latin1_output); 30| 150|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_utf16_to_latin1ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 4| 112| char *latin1_output) { 5| 112| const char16_t *end = buf + len; 6| 2.04k| while (end - buf >= 8) { ------------------ | Branch (6:10): [True: 1.96k, False: 76] ------------------ 7| | // Load 8 UTF-16 characters into 128-bit SSE register 8| 1.96k| __m128i in = _mm_loadu_si128(reinterpret_cast(buf)); 9| | 10| 1.96k| if constexpr (!match_system(big_endian)) { 11| 1.96k| const __m128i swap = 12| 1.96k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 13| 1.96k| in = _mm_shuffle_epi8(in, swap); 14| 1.96k| } 15| | 16| 1.96k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 17| 1.96k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (17:9): [True: 1.93k, False: 36] ------------------ 18| | // Pack 16-bit characters into 8-bit and store in latin1_output 19| 1.93k| __m128i latin1_packed = _mm_packus_epi16(in, in); 20| 1.93k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 21| 1.93k| latin1_packed); 22| | // Adjust pointers for next iteration 23| 1.93k| buf += 8; 24| 1.93k| latin1_output += 8; 25| 1.93k| } else { 26| 36| return std::make_pair(nullptr, reinterpret_cast(latin1_output)); 27| 36| } 28| 1.96k| } // while 29| 76| return std::make_pair(buf, latin1_output); 30| 112|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_139sse_convert_utf16_to_latin1_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 35| 226| char *latin1_output) { 36| 226| const char16_t *start = buf; 37| 226| const char16_t *end = buf + len; 38| 786k| while (end - buf >= 8) { ------------------ | Branch (38:10): [True: 786k, False: 128] ------------------ 39| 786k| __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| 786k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 48| 786k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (48:9): [True: 786k, False: 98] ------------------ 49| 786k| __m128i latin1_packed = _mm_packus_epi16(in, in); 50| 786k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 51| 786k| latin1_packed); 52| 786k| buf += 8; 53| 786k| latin1_output += 8; 54| 786k| } else { 55| | // Fallback to scalar code for handling errors 56| 408| for (int k = 0; k < 8; k++) { ------------------ | Branch (56:23): [True: 408, False: 0] ------------------ 57| 408| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 58| 408| if (word <= 0xff) { ------------------ | Branch (58:13): [True: 310, False: 98] ------------------ 59| 310| *latin1_output++ = char(word); 60| 310| } else { 61| 98| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 62| 98| latin1_output); 63| 98| } 64| 408| } 65| 0| buf += 8; 66| 0| } 67| 786k| } // while 68| 128| return std::make_pair(result(error_code::SUCCESS, buf - start), 69| 128| latin1_output); 70| 226|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_139sse_convert_utf16_to_latin1_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 35| 230| char *latin1_output) { 36| 230| const char16_t *start = buf; 37| 230| const char16_t *end = buf + len; 38| 915k| while (end - buf >= 8) { ------------------ | Branch (38:10): [True: 914k, False: 114] ------------------ 39| 914k| __m128i in = _mm_loadu_si128(reinterpret_cast(buf)); 40| | 41| 914k| if constexpr (!match_system(big_endian)) { 42| 914k| const __m128i swap = 43| 914k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 44| 914k| in = _mm_shuffle_epi8(in, swap); 45| 914k| } 46| | 47| 914k| __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00); 48| 914k| if (_mm_testz_si128(in, high_byte_mask)) { ------------------ | Branch (48:9): [True: 914k, False: 116] ------------------ 49| 914k| __m128i latin1_packed = _mm_packus_epi16(in, in); 50| 914k| _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output), 51| 914k| latin1_packed); 52| 914k| buf += 8; 53| 914k| latin1_output += 8; 54| 914k| } else { 55| | // Fallback to scalar code for handling errors 56| 370| for (int k = 0; k < 8; k++) { ------------------ | Branch (56:23): [True: 370, False: 0] ------------------ 57| 370| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 58| 370| if (word <= 0xff) { ------------------ | Branch (58:13): [True: 254, False: 116] ------------------ 59| 254| *latin1_output++ = char(word); 60| 254| } else { 61| 116| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 62| 116| latin1_output); 63| 116| } 64| 370| } 65| 0| buf += 8; 66| 0| } 67| 914k| } // while 68| 114| return std::make_pair(result(error_code::SUCCESS, buf - start), 69| 114| latin1_output); 70| 230|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf16_to_utf32ILNS_10endiannessE0EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 428| char32_t *utf32_output) { 58| 428| const char16_t *end = buf + len; 59| | 60| 428| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 61| 428| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 62| | 63| 154k| while (end - buf >= 8) { ------------------ | Branch (63:10): [True: 154k, False: 298] ------------------ 64| 154k| __m128i in = _mm_loadu_si128((__m128i *)buf); 65| | 66| 154k| if (big_endian) { ------------------ | Branch (66:9): [Folded, False: 154k] ------------------ 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| 154k| const __m128i surrogates_bytemask = 76| 154k| _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| 154k| const uint16_t surrogates_bitmask = 81| 154k| 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| 154k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (84:9): [True: 149k, False: 4.64k] ------------------ 85| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 86| 149k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 87| 149k| _mm_cvtepu16_epi32(in)); 88| 149k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 89| 149k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 90| 149k| utf32_output += 8; 91| 149k| buf += 8; 92| | // surrogate pair(s) in a register 93| 149k| } else { 94| | // Let us do a scalar fallback. 95| | // It may seem wasteful to use scalar code, but being efficient with SIMD 96| | // in the presence of surrogate pairs may require non-trivial tables. 97| 4.64k| size_t forward = 15; 98| 4.64k| size_t k = 0; 99| 4.64k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (99:11): [True: 92, False: 4.55k] ------------------ 100| 92| forward = size_t(end - buf - 1); 101| 92| } 102| 67.0k| for (; k < forward; k++) { ------------------ | Branch (102:14): [True: 62.5k, False: 4.51k] ------------------ 103| 62.5k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 104| 62.5k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (104:13): [True: 56.8k, False: 5.68k] ------------------ 105| 56.8k| *utf32_output++ = char32_t(word); 106| 56.8k| } else { 107| | // must be a surrogate pair 108| 5.68k| uint16_t diff = uint16_t(word - 0xD800); 109| 5.68k| uint16_t next_word = 110| 5.68k| scalar::utf16::swap_if_needed(buf[k + 1]); 111| 5.68k| k++; 112| 5.68k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 113| 5.68k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (113:15): [True: 130, False: 5.55k] ------------------ 114| 130| return std::make_pair(nullptr, utf32_output); 115| 130| } 116| 5.55k| uint32_t value = (diff << 10) + diff2 + 0x10000; 117| 5.55k| *utf32_output++ = char32_t(value); 118| 5.55k| } 119| 62.5k| } 120| 4.51k| buf += k; 121| 4.51k| } 122| 154k| } // while 123| 298| return std::make_pair(buf, utf32_output); 124| 428|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf16_to_utf32ILNS_10endiannessE1EEENSt3__14pairIPKDsPDiEES7_mS8_: 57| 448| char32_t *utf32_output) { 58| 448| const char16_t *end = buf + len; 59| | 60| 448| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 61| 448| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 62| | 63| 756k| while (end - buf >= 8) { ------------------ | Branch (63:10): [True: 756k, False: 346] ------------------ 64| 756k| __m128i in = _mm_loadu_si128((__m128i *)buf); 65| | 66| 756k| if (big_endian) { ------------------ | Branch (66:9): [True: 756k, Folded] ------------------ 67| 756k| const __m128i swap = 68| 756k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 69| 756k| in = _mm_shuffle_epi8(in, swap); 70| 756k| } 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| 756k| const __m128i surrogates_bytemask = 76| 756k| _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| 756k| const uint16_t surrogates_bitmask = 81| 756k| 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| 756k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (84:9): [True: 754k, False: 1.87k] ------------------ 85| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 86| 754k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 87| 754k| _mm_cvtepu16_epi32(in)); 88| 754k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 89| 754k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 90| 754k| utf32_output += 8; 91| 754k| buf += 8; 92| | // surrogate pair(s) in a register 93| 754k| } 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| 1.87k| size_t forward = 15; 98| 1.87k| size_t k = 0; 99| 1.87k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (99:11): [True: 96, False: 1.78k] ------------------ 100| 96| forward = size_t(end - buf - 1); 101| 96| } 102| 25.4k| for (; k < forward; k++) { ------------------ | Branch (102:14): [True: 23.6k, False: 1.77k] ------------------ 103| 23.6k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 104| 23.6k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (104:13): [True: 20.5k, False: 3.19k] ------------------ 105| 20.5k| *utf32_output++ = char32_t(word); 106| 20.5k| } else { 107| | // must be a surrogate pair 108| 3.19k| uint16_t diff = uint16_t(word - 0xD800); 109| 3.19k| uint16_t next_word = 110| 3.19k| scalar::utf16::swap_if_needed(buf[k + 1]); 111| 3.19k| k++; 112| 3.19k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 113| 3.19k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (113:15): [True: 102, False: 3.09k] ------------------ 114| 102| return std::make_pair(nullptr, utf32_output); 115| 102| } 116| 3.09k| uint32_t value = (diff << 10) + diff2 + 0x10000; 117| 3.09k| *utf32_output++ = char32_t(value); 118| 3.09k| } 119| 23.6k| } 120| 1.77k| buf += k; 121| 1.77k| } 122| 756k| } // while 123| 346| return std::make_pair(buf, utf32_output); 124| 448|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf16_to_utf32_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 136| 402| char32_t *utf32_output) { 137| 402| const char16_t *start = buf; 138| 402| const char16_t *end = buf + len; 139| | 140| 402| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 141| 402| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 142| | 143| 310k| while (end - buf >= 8) { ------------------ | Branch (143:10): [True: 309k, False: 262] ------------------ 144| 309k| __m128i in = _mm_loadu_si128((__m128i *)buf); 145| | 146| 309k| if (big_endian) { ------------------ | Branch (146:9): [Folded, False: 309k] ------------------ 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| 309k| const __m128i surrogates_bytemask = 156| 309k| _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| 309k| const uint16_t surrogates_bitmask = 161| 309k| 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| 309k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (164:9): [True: 303k, False: 6.16k] ------------------ 165| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 166| 303k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 167| 303k| _mm_cvtepu16_epi32(in)); 168| 303k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 169| 303k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 170| 303k| utf32_output += 8; 171| 303k| buf += 8; 172| | // surrogate pair(s) in a register 173| 303k| } 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| 6.16k| size_t forward = 15; 178| 6.16k| size_t k = 0; 179| 6.16k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (179:11): [True: 88, False: 6.07k] ------------------ 180| 88| forward = size_t(end - buf - 1); 181| 88| } 182| 86.8k| for (; k < forward; k++) { ------------------ | Branch (182:14): [True: 80.7k, False: 6.02k] ------------------ 183| 80.7k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 184| 80.7k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (184:13): [True: 70.0k, False: 10.7k] ------------------ 185| 70.0k| *utf32_output++ = char32_t(word); 186| 70.0k| } else { 187| | // must be a surrogate pair 188| 10.7k| uint16_t diff = uint16_t(word - 0xD800); 189| 10.7k| uint16_t next_word = 190| 10.7k| scalar::utf16::swap_if_needed(buf[k + 1]); 191| 10.7k| k++; 192| 10.7k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 193| 10.7k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (193:15): [True: 140, False: 10.5k] ------------------ 194| 140| return std::make_pair( 195| 140| result(error_code::SURROGATE, buf - start + k - 1), 196| 140| utf32_output); 197| 140| } 198| 10.5k| uint32_t value = (diff << 10) + diff2 + 0x10000; 199| 10.5k| *utf32_output++ = char32_t(value); 200| 10.5k| } 201| 80.7k| } 202| 6.02k| buf += k; 203| 6.02k| } 204| 309k| } // while 205| 262| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 206| 402|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf16_to_utf32_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDiEEPKDsmS7_: 136| 422| char32_t *utf32_output) { 137| 422| const char16_t *start = buf; 138| 422| const char16_t *end = buf + len; 139| | 140| 422| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 141| 422| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 142| | 143| 313k| while (end - buf >= 8) { ------------------ | Branch (143:10): [True: 313k, False: 256] ------------------ 144| 313k| __m128i in = _mm_loadu_si128((__m128i *)buf); 145| | 146| 313k| if (big_endian) { ------------------ | Branch (146:9): [True: 313k, Folded] ------------------ 147| 313k| const __m128i swap = 148| 313k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 149| 313k| in = _mm_shuffle_epi8(in, swap); 150| 313k| } 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| 313k| const __m128i surrogates_bytemask = 156| 313k| _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| 313k| const uint16_t surrogates_bitmask = 161| 313k| 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| 313k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (164:9): [True: 308k, False: 5.04k] ------------------ 165| | // case: no surrogate pair, extend 16-bit code units to 32-bit code units 166| 308k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 167| 308k| _mm_cvtepu16_epi32(in)); 168| 308k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 169| 308k| _mm_cvtepu16_epi32(_mm_srli_si128(in, 8))); 170| 308k| utf32_output += 8; 171| 308k| buf += 8; 172| | // surrogate pair(s) in a register 173| 308k| } else { 174| | // Let us do a scalar fallback. 175| | // It may seem wasteful to use scalar code, but being efficient with SIMD 176| | // in the presence of surrogate pairs may require non-trivial tables. 177| 5.04k| size_t forward = 15; 178| 5.04k| size_t k = 0; 179| 5.04k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (179:11): [True: 98, False: 4.94k] ------------------ 180| 98| forward = size_t(end - buf - 1); 181| 98| } 182| 68.4k| for (; k < forward; k++) { ------------------ | Branch (182:14): [True: 63.5k, False: 4.87k] ------------------ 183| 63.5k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 184| 63.5k| if ((word & 0xF800) != 0xD800) { ------------------ | Branch (184:13): [True: 52.5k, False: 11.0k] ------------------ 185| 52.5k| *utf32_output++ = char32_t(word); 186| 52.5k| } else { 187| | // must be a surrogate pair 188| 11.0k| uint16_t diff = uint16_t(word - 0xD800); 189| 11.0k| uint16_t next_word = 190| 11.0k| scalar::utf16::swap_if_needed(buf[k + 1]); 191| 11.0k| k++; 192| 11.0k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 193| 11.0k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (193:15): [True: 166, False: 10.9k] ------------------ 194| 166| return std::make_pair( 195| 166| result(error_code::SURROGATE, buf - start + k - 1), 196| 166| utf32_output); 197| 166| } 198| 10.9k| uint32_t value = (diff << 10) + diff2 + 0x10000; 199| 10.9k| *utf32_output++ = char32_t(value); 200| 10.9k| } 201| 63.5k| } 202| 4.87k| buf += k; 203| 4.87k| } 204| 313k| } // while 205| 256| return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output); 206| 422|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_125sse_convert_utf16_to_utf8ILNS_10endiannessE0EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 876|sse_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| | 58| 876| const char16_t *end = buf + len; 59| | 60| 876| const __m128i v_0000 = _mm_setzero_si128(); 61| 876| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 62| 876| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 63| 876| const size_t safety_margin = 64| 876| 12; // to avoid overruns, see issue 65| | // https://github.com/simdutf/simdutf/issues/92 66| | 67| 981k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (67:10): [True: 980k, False: 728] ------------------ 68| 980k| __m128i in = _mm_loadu_si128((__m128i *)buf); 69| 980k| if (big_endian) { ------------------ | Branch (69:9): [Folded, False: 980k] ------------------ 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| 980k| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 76| 980k| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 799k, False: 181k] ------------------ 77| 799k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 78| 799k| if (big_endian) { ------------------ | Branch (78:11): [Folded, False: 799k] ------------------ 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| 799k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (83:11): [True: 35.5k, False: 763k] ------------------ 84| | // 1. pack the bytes 85| | // obviously suboptimal. 86| 35.5k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 87| | // 2. store (16 bytes) 88| 35.5k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 89| | // 3. adjust pointers 90| 35.5k| buf += 8; 91| 35.5k| utf8_output += 8; 92| 35.5k| in = nextin; 93| 763k| } else { 94| | // 1. pack the bytes 95| | // obviously suboptimal. 96| 763k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 97| | // 2. store (16 bytes) 98| 763k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 99| | // 3. adjust pointers 100| 763k| buf += 16; 101| 763k| utf8_output += 16; 102| 763k| continue; // we are done for this round! 103| 763k| } 104| 799k| } 105| | 106| | // no bits set above 7th bit 107| 216k| const __m128i one_byte_bytemask = 108| 216k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 109| 216k| const uint16_t one_byte_bitmask = 110| 216k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 111| | 112| | // no bits set above 11th bit 113| 216k| const __m128i one_or_two_bytes_bytemask = 114| 216k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 115| 216k| const uint16_t one_or_two_bytes_bitmask = 116| 216k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 117| | 118| 216k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (118:9): [True: 84.6k, False: 131k] ------------------ 119| 84.6k| internal::westmere::write_v_u16_11bits_to_utf8( 120| 84.6k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 121| 84.6k| buf += 8; 122| 84.6k| continue; 123| 84.6k| } 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| 131k| const __m128i surrogates_bytemask = 129| 131k| _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| 131k| const uint16_t surrogates_bitmask = 134| 131k| 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| 131k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (137:9): [True: 125k, False: 6.76k] ------------------ 138| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 139| 125k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 140| 125k| 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| 125k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 170| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 171| 125k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 172| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 173| 125k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 169| 125k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 174| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 175| 125k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 169| 125k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 176| | 177| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 178| 125k| const __m128i s0 = _mm_srli_epi16(in, 4); 179| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 180| 125k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 169| 125k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 181| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 182| 125k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 169| 125k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 183| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 184| 125k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 169| 125k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 185| 125k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 186| 125k| simdutf_vec(0b0100000000000000)); ------------------ | | 169| 125k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 187| 125k| const __m128i s4 = _mm_xor_si128(s3, m0); 188| 125k|#undef simdutf_vec 189| | 190| | // 4. expand code units 16-bit => 32-bit 191| 125k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 192| 125k| 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| 125k| const uint16_t mask = 196| 125k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 197| 125k| if (mask == 0) { ------------------ | Branch (197:11): [True: 30.0k, False: 95.0k] ------------------ 198| | // We only have three-byte code units. Use fast path. 199| 30.0k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 200| 30.0k| 15, 13, -1, -1, -1, -1); 201| 30.0k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 202| 30.0k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 203| 30.0k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 204| 30.0k| utf8_output += 12; 205| 30.0k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 206| 30.0k| utf8_output += 12; 207| 30.0k| buf += 8; 208| 30.0k| continue; 209| 30.0k| } 210| 95.0k| const uint8_t mask0 = uint8_t(mask); 211| | 212| 95.0k| const uint8_t *row0 = 213| 95.0k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 214| 95.0k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 215| 95.0k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 216| | 217| 95.0k| const uint8_t mask1 = static_cast(mask >> 8); 218| | 219| 95.0k| const uint8_t *row1 = 220| 95.0k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 221| 95.0k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 222| 95.0k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 223| | 224| 95.0k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 225| 95.0k| utf8_output += row0[0]; 226| 95.0k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 227| 95.0k| utf8_output += row1[0]; 228| | 229| 95.0k| buf += 8; 230| | // surrogate pair(s) in a register 231| 95.0k| } 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| 6.76k| size_t forward = 15; 236| 6.76k| size_t k = 0; 237| 6.76k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (237:11): [True: 0, False: 6.76k] ------------------ 238| 0| forward = size_t(end - buf - 1); 239| 0| } 240| 98.6k| for (; k < forward; k++) { ------------------ | Branch (240:14): [True: 92.0k, False: 6.62k] ------------------ 241| 92.0k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 242| 92.0k| if ((word & 0xFF80) == 0) { ------------------ | Branch (242:13): [True: 12.7k, False: 79.3k] ------------------ 243| 12.7k| *utf8_output++ = char(word); 244| 79.3k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (244:20): [True: 4.96k, False: 74.3k] ------------------ 245| 4.96k| *utf8_output++ = char((word >> 6) | 0b11000000); 246| 4.96k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 247| 74.3k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (247:20): [True: 65.9k, False: 8.43k] ------------------ 248| 65.9k| *utf8_output++ = char((word >> 12) | 0b11100000); 249| 65.9k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 250| 65.9k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 251| 65.9k| } else { 252| | // must be a surrogate pair 253| 8.43k| uint16_t diff = uint16_t(word - 0xD800); 254| 8.43k| uint16_t next_word = 255| 8.43k| scalar::utf16::swap_if_needed(buf[k + 1]); 256| 8.43k| k++; 257| 8.43k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 258| 8.43k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (258:15): [True: 148, False: 8.28k] ------------------ 259| 148| return std::make_pair(nullptr, utf8_output); 260| 148| } 261| 8.28k| uint32_t value = (diff << 10) + diff2 + 0x10000; 262| 8.28k| *utf8_output++ = char((value >> 18) | 0b11110000); 263| 8.28k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 264| 8.28k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 265| 8.28k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 266| 8.28k| } 267| 92.0k| } 268| 6.62k| buf += k; 269| 6.62k| } 270| 131k| } // while 271| | 272| 728| return std::make_pair(buf, utf8_output); 273| 876|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_125sse_convert_utf16_to_utf8ILNS_10endiannessE1EEENSt3__14pairIPKDsPcEES7_mS8_: 56| 772|sse_convert_utf16_to_utf8(const char16_t *buf, size_t len, char *utf8_output) { 57| | 58| 772| const char16_t *end = buf + len; 59| | 60| 772| const __m128i v_0000 = _mm_setzero_si128(); 61| 772| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 62| 772| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 63| 772| const size_t safety_margin = 64| 772| 12; // to avoid overruns, see issue 65| | // https://github.com/simdutf/simdutf/issues/92 66| | 67| 1.34M| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (67:10): [True: 1.34M, False: 702] ------------------ 68| 1.34M| __m128i in = _mm_loadu_si128((__m128i *)buf); 69| 1.34M| if (big_endian) { ------------------ | Branch (69:9): [True: 1.34M, Folded] ------------------ 70| 1.34M| const __m128i swap = 71| 1.34M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 72| 1.34M| in = _mm_shuffle_epi8(in, swap); 73| 1.34M| } 74| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 75| 1.34M| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 76| 1.34M| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (76:9): [True: 837k, False: 506k] ------------------ 77| 837k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 78| 837k| if (big_endian) { ------------------ | Branch (78:11): [True: 837k, Folded] ------------------ 79| 837k| const __m128i swap = 80| 837k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 81| 837k| nextin = _mm_shuffle_epi8(nextin, swap); 82| 837k| } 83| 837k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (83:11): [True: 55.3k, False: 782k] ------------------ 84| | // 1. pack the bytes 85| | // obviously suboptimal. 86| 55.3k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 87| | // 2. store (16 bytes) 88| 55.3k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 89| | // 3. adjust pointers 90| 55.3k| buf += 8; 91| 55.3k| utf8_output += 8; 92| 55.3k| in = nextin; 93| 782k| } else { 94| | // 1. pack the bytes 95| | // obviously suboptimal. 96| 782k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 97| | // 2. store (16 bytes) 98| 782k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 99| | // 3. adjust pointers 100| 782k| buf += 16; 101| 782k| utf8_output += 16; 102| 782k| continue; // we are done for this round! 103| 782k| } 104| 837k| } 105| | 106| | // no bits set above 7th bit 107| 561k| const __m128i one_byte_bytemask = 108| 561k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 109| 561k| const uint16_t one_byte_bitmask = 110| 561k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 111| | 112| | // no bits set above 11th bit 113| 561k| const __m128i one_or_two_bytes_bytemask = 114| 561k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 115| 561k| const uint16_t one_or_two_bytes_bitmask = 116| 561k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 117| | 118| 561k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (118:9): [True: 56.3k, False: 505k] ------------------ 119| 56.3k| internal::westmere::write_v_u16_11bits_to_utf8( 120| 56.3k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 121| 56.3k| buf += 8; 122| 56.3k| continue; 123| 56.3k| } 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| 505k| const __m128i surrogates_bytemask = 129| 505k| _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| 505k| const uint16_t surrogates_bitmask = 134| 505k| 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| 505k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (137:9): [True: 499k, False: 5.94k] ------------------ 138| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 139| 499k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 140| 499k| 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| 499k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 170| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 171| 499k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 172| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 173| 499k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 169| 499k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 174| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 175| 499k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 169| 499k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 176| | 177| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 178| 499k| const __m128i s0 = _mm_srli_epi16(in, 4); 179| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 180| 499k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 169| 499k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 181| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 182| 499k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 169| 499k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 183| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 184| 499k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 169| 499k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 185| 499k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 186| 499k| simdutf_vec(0b0100000000000000)); ------------------ | | 169| 499k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 187| 499k| const __m128i s4 = _mm_xor_si128(s3, m0); 188| 499k|#undef simdutf_vec 189| | 190| | // 4. expand code units 16-bit => 32-bit 191| 499k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 192| 499k| 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| 499k| const uint16_t mask = 196| 499k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 197| 499k| if (mask == 0) { ------------------ | Branch (197:11): [True: 276k, False: 223k] ------------------ 198| | // We only have three-byte code units. Use fast path. 199| 276k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 200| 276k| 15, 13, -1, -1, -1, -1); 201| 276k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 202| 276k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 203| 276k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 204| 276k| utf8_output += 12; 205| 276k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 206| 276k| utf8_output += 12; 207| 276k| buf += 8; 208| 276k| continue; 209| 276k| } 210| 223k| const uint8_t mask0 = uint8_t(mask); 211| | 212| 223k| const uint8_t *row0 = 213| 223k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 214| 223k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 215| 223k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 216| | 217| 223k| const uint8_t mask1 = static_cast(mask >> 8); 218| | 219| 223k| const uint8_t *row1 = 220| 223k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 221| 223k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 222| 223k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 223| | 224| 223k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 225| 223k| utf8_output += row0[0]; 226| 223k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 227| 223k| utf8_output += row1[0]; 228| | 229| 223k| buf += 8; 230| | // surrogate pair(s) in a register 231| 223k| } 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| 5.94k| size_t forward = 15; 236| 5.94k| size_t k = 0; 237| 5.94k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (237:11): [True: 0, False: 5.94k] ------------------ 238| 0| forward = size_t(end - buf - 1); 239| 0| } 240| 87.3k| for (; k < forward; k++) { ------------------ | Branch (240:14): [True: 81.4k, False: 5.87k] ------------------ 241| 81.4k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 242| 81.4k| if ((word & 0xFF80) == 0) { ------------------ | Branch (242:13): [True: 12.1k, False: 69.2k] ------------------ 243| 12.1k| *utf8_output++ = char(word); 244| 69.2k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (244:20): [True: 2.44k, False: 66.8k] ------------------ 245| 2.44k| *utf8_output++ = char((word >> 6) | 0b11000000); 246| 2.44k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 247| 66.8k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (247:20): [True: 59.6k, False: 7.21k] ------------------ 248| 59.6k| *utf8_output++ = char((word >> 12) | 0b11100000); 249| 59.6k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 250| 59.6k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 251| 59.6k| } else { 252| | // must be a surrogate pair 253| 7.21k| uint16_t diff = uint16_t(word - 0xD800); 254| 7.21k| uint16_t next_word = 255| 7.21k| scalar::utf16::swap_if_needed(buf[k + 1]); 256| 7.21k| k++; 257| 7.21k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 258| 7.21k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (258:15): [True: 70, False: 7.14k] ------------------ 259| 70| return std::make_pair(nullptr, utf8_output); 260| 70| } 261| 7.14k| uint32_t value = (diff << 10) + diff2 + 0x10000; 262| 7.14k| *utf8_output++ = char((value >> 18) | 0b11110000); 263| 7.14k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 264| 7.14k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 265| 7.14k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 266| 7.14k| } 267| 81.4k| } 268| 5.87k| buf += k; 269| 5.87k| } 270| 505k| } // while 271| | 272| 702| return std::make_pair(buf, utf8_output); 273| 772|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_137sse_convert_utf16_to_utf8_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 285| 654| char *utf8_output) { 286| 654| const char16_t *start = buf; 287| 654| const char16_t *end = buf + len; 288| | 289| 654| const __m128i v_0000 = _mm_setzero_si128(); 290| 654| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 291| 654| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 292| 654| const size_t safety_margin = 293| 654| 12; // to avoid overruns, see issue 294| | // https://github.com/simdutf/simdutf/issues/92 295| | 296| 917k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (296:10): [True: 916k, False: 532] ------------------ 297| 916k| __m128i in = _mm_loadu_si128((__m128i *)buf); 298| 916k| if (big_endian) { ------------------ | Branch (298:9): [Folded, False: 916k] ------------------ 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| 916k| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 305| 916k| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (305:9): [True: 372k, False: 543k] ------------------ 306| 372k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 307| 372k| if (big_endian) { ------------------ | Branch (307:11): [Folded, False: 372k] ------------------ 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| 372k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (312:11): [True: 14.6k, False: 358k] ------------------ 313| | // 1. pack the bytes 314| | // obviously suboptimal. 315| 14.6k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 316| | // 2. store (16 bytes) 317| 14.6k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 318| | // 3. adjust pointers 319| 14.6k| buf += 8; 320| 14.6k| utf8_output += 8; 321| 14.6k| in = nextin; 322| 358k| } else { 323| | // 1. pack the bytes 324| | // obviously suboptimal. 325| 358k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 326| | // 2. store (16 bytes) 327| 358k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 328| | // 3. adjust pointers 329| 358k| buf += 16; 330| 358k| utf8_output += 16; 331| 358k| continue; // we are done for this round! 332| 358k| } 333| 372k| } 334| | 335| | // no bits set above 7th bit 336| 558k| const __m128i one_byte_bytemask = 337| 558k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 338| 558k| const uint16_t one_byte_bitmask = 339| 558k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 340| | 341| | // no bits set above 11th bit 342| 558k| const __m128i one_or_two_bytes_bytemask = 343| 558k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 344| 558k| const uint16_t one_or_two_bytes_bitmask = 345| 558k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 346| | 347| 558k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (347:9): [True: 146k, False: 411k] ------------------ 348| 146k| internal::westmere::write_v_u16_11bits_to_utf8( 349| 146k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 350| 146k| buf += 8; 351| 146k| continue; 352| 146k| } 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| 411k| const __m128i surrogates_bytemask = 358| 411k| _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| 411k| const uint16_t surrogates_bitmask = 363| 411k| 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| 411k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (366:9): [True: 407k, False: 3.81k] ------------------ 367| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 368| 407k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 369| 407k| 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| 407k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 399| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 400| 407k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 401| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 402| 407k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 398| 407k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 403| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 404| 407k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 398| 407k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 405| | 406| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 407| 407k| const __m128i s0 = _mm_srli_epi16(in, 4); 408| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 409| 407k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 398| 407k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 410| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 411| 407k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 398| 407k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 412| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 413| 407k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 398| 407k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 414| 407k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 415| 407k| simdutf_vec(0b0100000000000000)); ------------------ | | 398| 407k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 416| 407k| const __m128i s4 = _mm_xor_si128(s3, m0); 417| 407k|#undef simdutf_vec 418| | 419| | // 4. expand code units 16-bit => 32-bit 420| 407k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 421| 407k| 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| 407k| const uint16_t mask = 425| 407k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 426| 407k| if (mask == 0) { ------------------ | Branch (426:11): [True: 316k, False: 91.0k] ------------------ 427| | // We only have three-byte code units. Use fast path. 428| 316k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 429| 316k| 15, 13, -1, -1, -1, -1); 430| 316k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 431| 316k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 432| 316k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 433| 316k| utf8_output += 12; 434| 316k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 435| 316k| utf8_output += 12; 436| 316k| buf += 8; 437| 316k| continue; 438| 316k| } 439| 91.0k| const uint8_t mask0 = uint8_t(mask); 440| | 441| 91.0k| const uint8_t *row0 = 442| 91.0k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 443| 91.0k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 444| 91.0k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 445| | 446| 91.0k| const uint8_t mask1 = static_cast(mask >> 8); 447| | 448| 91.0k| const uint8_t *row1 = 449| 91.0k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 450| 91.0k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 451| 91.0k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 452| | 453| 91.0k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 454| 91.0k| utf8_output += row0[0]; 455| 91.0k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 456| 91.0k| utf8_output += row1[0]; 457| | 458| 91.0k| buf += 8; 459| | // surrogate pair(s) in a register 460| 91.0k| } 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| 3.81k| size_t forward = 15; 465| 3.81k| size_t k = 0; 466| 3.81k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (466:11): [True: 0, False: 3.81k] ------------------ 467| 0| forward = size_t(end - buf - 1); 468| 0| } 469| 55.3k| for (; k < forward; k++) { ------------------ | Branch (469:14): [True: 51.6k, False: 3.69k] ------------------ 470| 51.6k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 471| 51.6k| if ((word & 0xFF80) == 0) { ------------------ | Branch (471:13): [True: 10.0k, False: 41.6k] ------------------ 472| 10.0k| *utf8_output++ = char(word); 473| 41.6k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (473:20): [True: 1.92k, False: 39.7k] ------------------ 474| 1.92k| *utf8_output++ = char((word >> 6) | 0b11000000); 475| 1.92k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 476| 39.7k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (476:20): [True: 35.2k, False: 4.47k] ------------------ 477| 35.2k| *utf8_output++ = char((word >> 12) | 0b11100000); 478| 35.2k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 479| 35.2k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 480| 35.2k| } else { 481| | // must be a surrogate pair 482| 4.47k| uint16_t diff = uint16_t(word - 0xD800); 483| 4.47k| uint16_t next_word = 484| 4.47k| scalar::utf16::swap_if_needed(buf[k + 1]); 485| 4.47k| k++; 486| 4.47k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 487| 4.47k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (487:15): [True: 122, False: 4.35k] ------------------ 488| 122| return std::make_pair( 489| 122| result(error_code::SURROGATE, buf - start + k - 1), 490| 122| utf8_output); 491| 122| } 492| 4.35k| uint32_t value = (diff << 10) + diff2 + 0x10000; 493| 4.35k| *utf8_output++ = char((value >> 18) | 0b11110000); 494| 4.35k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 495| 4.35k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 496| 4.35k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 497| 4.35k| } 498| 51.6k| } 499| 3.69k| buf += k; 500| 3.69k| } 501| 411k| } // while 502| | 503| 532| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 504| 654|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_137sse_convert_utf16_to_utf8_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPcEEPKDsmS7_: 285| 648| char *utf8_output) { 286| 648| const char16_t *start = buf; 287| 648| const char16_t *end = buf + len; 288| | 289| 648| const __m128i v_0000 = _mm_setzero_si128(); 290| 648| const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800); 291| 648| const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800); 292| 648| const size_t safety_margin = 293| 648| 12; // to avoid overruns, see issue 294| | // https://github.com/simdutf/simdutf/issues/92 295| | 296| 676k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (296:10): [True: 676k, False: 494] ------------------ 297| 676k| __m128i in = _mm_loadu_si128((__m128i *)buf); 298| 676k| if (big_endian) { ------------------ | Branch (298:9): [True: 676k, Folded] ------------------ 299| 676k| const __m128i swap = 300| 676k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 301| 676k| in = _mm_shuffle_epi8(in, swap); 302| 676k| } 303| | // a single 16-bit UTF-16 word can yield 1, 2 or 3 UTF-8 bytes 304| 676k| const __m128i v_ff80 = _mm_set1_epi16((int16_t)0xff80); 305| 676k| if (_mm_testz_si128(in, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (305:9): [True: 469k, False: 207k] ------------------ 306| 469k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 307| 469k| if (big_endian) { ------------------ | Branch (307:11): [True: 469k, Folded] ------------------ 308| 469k| const __m128i swap = 309| 469k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 310| 469k| nextin = _mm_shuffle_epi8(nextin, swap); 311| 469k| } 312| 469k| if (!_mm_testz_si128(nextin, v_ff80)) { ------------------ | Branch (312:11): [True: 33.2k, False: 435k] ------------------ 313| | // 1. pack the bytes 314| | // obviously suboptimal. 315| 33.2k| const __m128i utf8_packed = _mm_packus_epi16(in, in); 316| | // 2. store (16 bytes) 317| 33.2k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 318| | // 3. adjust pointers 319| 33.2k| buf += 8; 320| 33.2k| utf8_output += 8; 321| 33.2k| in = nextin; 322| 435k| } else { 323| | // 1. pack the bytes 324| | // obviously suboptimal. 325| 435k| const __m128i utf8_packed = _mm_packus_epi16(in, nextin); 326| | // 2. store (16 bytes) 327| 435k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 328| | // 3. adjust pointers 329| 435k| buf += 16; 330| 435k| utf8_output += 16; 331| 435k| continue; // we are done for this round! 332| 435k| } 333| 469k| } 334| | 335| | // no bits set above 7th bit 336| 240k| const __m128i one_byte_bytemask = 337| 240k| _mm_cmpeq_epi16(_mm_and_si128(in, v_ff80), v_0000); 338| 240k| const uint16_t one_byte_bitmask = 339| 240k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 340| | 341| | // no bits set above 11th bit 342| 240k| const __m128i one_or_two_bytes_bytemask = 343| 240k| _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_0000); 344| 240k| const uint16_t one_or_two_bytes_bitmask = 345| 240k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 346| | 347| 240k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (347:9): [True: 44.3k, False: 195k] ------------------ 348| 44.3k| internal::westmere::write_v_u16_11bits_to_utf8( 349| 44.3k| in, utf8_output, one_byte_bytemask, one_byte_bitmask); 350| 44.3k| buf += 8; 351| 44.3k| continue; 352| 44.3k| } 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| 195k| const __m128i surrogates_bytemask = 358| 195k| _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| 195k| const uint16_t surrogates_bitmask = 363| 195k| 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| 195k| if (surrogates_bitmask == 0x0000) { ------------------ | Branch (366:9): [True: 189k, False: 6.06k] ------------------ 367| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 368| 189k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 369| 189k| 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| 189k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 399| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 400| 189k| const __m128i t0 = _mm_shuffle_epi8(in, dup_even); 401| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 402| 189k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 398| 189k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 403| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 404| 189k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 398| 189k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 405| | 406| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 407| 189k| const __m128i s0 = _mm_srli_epi16(in, 4); 408| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 409| 189k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 398| 189k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 410| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 411| 189k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 398| 189k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 412| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 413| 189k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 398| 189k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 414| 189k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 415| 189k| simdutf_vec(0b0100000000000000)); ------------------ | | 398| 189k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 416| 189k| const __m128i s4 = _mm_xor_si128(s3, m0); 417| 189k|#undef simdutf_vec 418| | 419| | // 4. expand code units 16-bit => 32-bit 420| 189k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 421| 189k| 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| 189k| const uint16_t mask = 425| 189k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 426| 189k| if (mask == 0) { ------------------ | Branch (426:11): [True: 111k, False: 78.4k] ------------------ 427| | // We only have three-byte code units. Use fast path. 428| 111k| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 429| 111k| 15, 13, -1, -1, -1, -1); 430| 111k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 431| 111k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 432| 111k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 433| 111k| utf8_output += 12; 434| 111k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 435| 111k| utf8_output += 12; 436| 111k| buf += 8; 437| 111k| continue; 438| 111k| } 439| 78.4k| const uint8_t mask0 = uint8_t(mask); 440| | 441| 78.4k| const uint8_t *row0 = 442| 78.4k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 443| 78.4k| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 444| 78.4k| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 445| | 446| 78.4k| const uint8_t mask1 = static_cast(mask >> 8); 447| | 448| 78.4k| const uint8_t *row1 = 449| 78.4k| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 450| 78.4k| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 451| 78.4k| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 452| | 453| 78.4k| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 454| 78.4k| utf8_output += row0[0]; 455| 78.4k| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 456| 78.4k| utf8_output += row1[0]; 457| | 458| 78.4k| buf += 8; 459| | // surrogate pair(s) in a register 460| 78.4k| } else { 461| | // Let us do a scalar fallback. 462| | // It may seem wasteful to use scalar code, but being efficient with SIMD 463| | // in the presence of surrogate pairs may require non-trivial tables. 464| 6.06k| size_t forward = 15; 465| 6.06k| size_t k = 0; 466| 6.06k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (466:11): [True: 0, False: 6.06k] ------------------ 467| 0| forward = size_t(end - buf - 1); 468| 0| } 469| 88.4k| for (; k < forward; k++) { ------------------ | Branch (469:14): [True: 82.5k, False: 5.91k] ------------------ 470| 82.5k| uint16_t word = scalar::utf16::swap_if_needed(buf[k]); 471| 82.5k| if ((word & 0xFF80) == 0) { ------------------ | Branch (471:13): [True: 23.1k, False: 59.3k] ------------------ 472| 23.1k| *utf8_output++ = char(word); 473| 59.3k| } else if ((word & 0xF800) == 0) { ------------------ | Branch (473:20): [True: 3.85k, False: 55.5k] ------------------ 474| 3.85k| *utf8_output++ = char((word >> 6) | 0b11000000); 475| 3.85k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 476| 55.5k| } else if ((word & 0xF800) != 0xD800) { ------------------ | Branch (476:20): [True: 48.3k, False: 7.14k] ------------------ 477| 48.3k| *utf8_output++ = char((word >> 12) | 0b11100000); 478| 48.3k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 479| 48.3k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 480| 48.3k| } else { 481| | // must be a surrogate pair 482| 7.14k| uint16_t diff = uint16_t(word - 0xD800); 483| 7.14k| uint16_t next_word = 484| 7.14k| scalar::utf16::swap_if_needed(buf[k + 1]); 485| 7.14k| k++; 486| 7.14k| uint16_t diff2 = uint16_t(next_word - 0xDC00); 487| 7.14k| if ((diff | diff2) > 0x3FF) { ------------------ | Branch (487:15): [True: 154, False: 6.99k] ------------------ 488| 154| return std::make_pair( 489| 154| result(error_code::SURROGATE, buf - start + k - 1), 490| 154| utf8_output); 491| 154| } 492| 6.99k| uint32_t value = (diff << 10) + diff2 + 0x10000; 493| 6.99k| *utf8_output++ = char((value >> 18) | 0b11110000); 494| 6.99k| *utf8_output++ = char(((value >> 12) & 0b111111) | 0b10000000); 495| 6.99k| *utf8_output++ = char(((value >> 6) & 0b111111) | 0b10000000); 496| 6.99k| *utf8_output++ = char((value & 0b111111) | 0b10000000); 497| 6.99k| } 498| 82.5k| } 499| 5.91k| buf += k; 500| 5.91k| } 501| 195k| } // while 502| | 503| 494| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 504| 648|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_127sse_convert_utf32_to_latin1EPKDimPc: 3| 206| char *latin1_output) { 4| 206| const size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 5| | 6| 206| __m128i high_bytes_mask = _mm_set1_epi32(0xFFFFFF00); 7| 206| __m128i shufmask = 8| 206| _mm_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 12, 8, 4, 0); 9| | 10| 1.42k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (10:22): [True: 1.26k, False: 156] ------------------ 11| 1.26k| __m128i in1 = _mm_loadu_si128((__m128i *)buf); 12| 1.26k| __m128i in2 = _mm_loadu_si128((__m128i *)(buf + 4)); 13| 1.26k| __m128i in3 = _mm_loadu_si128((__m128i *)(buf + 8)); 14| 1.26k| __m128i in4 = _mm_loadu_si128((__m128i *)(buf + 12)); 15| | 16| 1.26k| __m128i check_combined = _mm_or_si128(in1, in2); 17| 1.26k| check_combined = _mm_or_si128(check_combined, in3); 18| 1.26k| check_combined = _mm_or_si128(check_combined, in4); 19| | 20| 1.26k| if (!_mm_testz_si128(check_combined, high_bytes_mask)) { ------------------ | Branch (20:9): [True: 50, False: 1.21k] ------------------ 21| 50| return std::make_pair(nullptr, latin1_output); 22| 50| } 23| 1.21k| __m128i pack1 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in1, shufmask), 24| 1.21k| _mm_shuffle_epi8(in2, shufmask)); 25| 1.21k| __m128i pack2 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in3, shufmask), 26| 1.21k| _mm_shuffle_epi8(in4, shufmask)); 27| 1.21k| __m128i pack = _mm_unpacklo_epi64(pack1, pack2); 28| 1.21k| _mm_storeu_si128((__m128i *)latin1_output, pack); 29| 1.21k| latin1_output += 16; 30| 1.21k| buf += 16; 31| 1.21k| } 32| | 33| 156| return std::make_pair(buf, latin1_output); 34| 206|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_139sse_convert_utf32_to_latin1_with_errorsEPKDimPc: 38| 454| char *latin1_output) { 39| 454| const char32_t *start = buf; 40| 454| const size_t rounded_len = len & ~0xF; // Round down to nearest multiple of 16 41| | 42| 454| __m128i high_bytes_mask = _mm_set1_epi32(0xFFFFFF00); 43| 454| __m128i shufmask = 44| 454| _mm_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 12, 8, 4, 0); 45| | 46| 26.8k| for (size_t i = 0; i < rounded_len; i += 16) { ------------------ | Branch (46:22): [True: 26.6k, False: 256] ------------------ 47| 26.6k| __m128i in1 = _mm_loadu_si128((__m128i *)buf); 48| 26.6k| __m128i in2 = _mm_loadu_si128((__m128i *)(buf + 4)); 49| 26.6k| __m128i in3 = _mm_loadu_si128((__m128i *)(buf + 8)); 50| 26.6k| __m128i in4 = _mm_loadu_si128((__m128i *)(buf + 12)); 51| | 52| 26.6k| __m128i check_combined = _mm_or_si128(in1, in2); 53| 26.6k| check_combined = _mm_or_si128(check_combined, in3); 54| 26.6k| check_combined = _mm_or_si128(check_combined, in4); 55| | 56| 26.6k| if (!_mm_testz_si128(check_combined, high_bytes_mask)) { ------------------ | Branch (56:9): [True: 198, False: 26.4k] ------------------ 57| | // Fallback to scalar code for handling errors 58| 820| for (int k = 0; k < 16; k++) { ------------------ | Branch (58:23): [True: 820, False: 0] ------------------ 59| 820| char32_t codepoint = buf[k]; 60| 820| if (codepoint <= 0xff) { ------------------ | Branch (60:13): [True: 622, False: 198] ------------------ 61| 622| *latin1_output++ = char(codepoint); 62| 622| } else { 63| 198| return std::make_pair(result(error_code::TOO_LARGE, buf - start + k), 64| 198| latin1_output); 65| 198| } 66| 820| } 67| 0| buf += 16; 68| 0| continue; 69| 198| } 70| 26.4k| __m128i pack1 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in1, shufmask), 71| 26.4k| _mm_shuffle_epi8(in2, shufmask)); 72| 26.4k| __m128i pack2 = _mm_unpacklo_epi32(_mm_shuffle_epi8(in3, shufmask), 73| 26.4k| _mm_shuffle_epi8(in4, shufmask)); 74| 26.4k| __m128i pack = _mm_unpacklo_epi64(pack1, pack2); 75| 26.4k| _mm_storeu_si128((__m128i *)latin1_output, pack); 76| 26.4k| latin1_output += 16; 77| 26.4k| buf += 16; 78| 26.4k| } 79| | 80| 256| return std::make_pair(result(error_code::SUCCESS, buf - start), 81| 256| latin1_output); 82| 454|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf32_to_utf16ILNS_10endiannessE0EEENSt3__14pairIPKDiPDsEES7_mS8_: 58| 732| char16_t *utf16_output) { 59| | 60| 732| const char32_t *end = buf + len; 61| | 62| 732| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 63| 732| __m128i forbidden_bytemask = _mm_setzero_si128(); 64| | 65| 50.1k| while (end - buf >= 16 + 8) { ------------------ | Branch (65:10): [True: 49.4k, False: 688] ------------------ 66| 49.4k| const __m128i *ptr = reinterpret_cast(buf); 67| 49.4k| const __m128i in0 = _mm_loadu_si128(ptr + 0); 68| 49.4k| const __m128i in1 = _mm_loadu_si128(ptr + 1); 69| 49.4k| const __m128i in2 = _mm_loadu_si128(ptr + 2); 70| 49.4k| const __m128i in3 = _mm_loadu_si128(ptr + 3); 71| | 72| 49.4k| const __m128i combined = 73| 49.4k| _mm_or_si128(_mm_or_si128(in2, in3), _mm_or_si128(in0, in1)); 74| 49.4k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 49.4k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 48.0k, False: 1.39k] | | ------------------ ------------------ 75| | // No bits set above 16th, directly pack UTF-32 to UTF-16 76| 48.0k| __m128i utf16_packed0 = _mm_packus_epi32(in0, in1); 77| 48.0k| __m128i utf16_packed1 = _mm_packus_epi32(in2, in3); 78| | 79| 48.0k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 80| 48.0k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 81| 48.0k| forbidden_bytemask = _mm_or_si128( 82| 48.0k| forbidden_bytemask, 83| 48.0k| _mm_or_si128( 84| 48.0k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed0, v_f800), v_d800), 85| 48.0k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed1, v_f800), v_d800))); 86| | 87| 48.0k| if (big_endian) { ------------------ | Branch (87:11): [Folded, False: 48.0k] ------------------ 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| 48.0k| _mm_storeu_si128((__m128i *)utf16_output + 0, utf16_packed0); 95| 48.0k| _mm_storeu_si128((__m128i *)utf16_output + 1, utf16_packed1); 96| 48.0k| utf16_output += 16; 97| 48.0k| buf += 16; 98| 48.0k| } else { 99| 1.39k| if (!validate_utf32(in0, in1) || !validate_utf32(in2, in3)) { ------------------ | Branch (99:11): [True: 20, False: 1.37k] | Branch (99:40): [True: 24, False: 1.35k] ------------------ 100| 44| return std::make_pair(nullptr, utf16_output); 101| 44| } 102| | 103| 1.35k| const auto ret0 = sse_expand_surrogate(in0); 104| 1.35k| _mm_storeu_si128((__m128i *)utf16_output, ret0.compressed); 105| 1.35k| utf16_output += ret0.u16count; 106| | 107| 1.35k| const auto ret1 = sse_expand_surrogate(in1); 108| 1.35k| _mm_storeu_si128((__m128i *)utf16_output, ret1.compressed); 109| 1.35k| utf16_output += ret1.u16count; 110| | 111| 1.35k| const auto ret2 = sse_expand_surrogate(in2); 112| 1.35k| _mm_storeu_si128((__m128i *)utf16_output, ret2.compressed); 113| 1.35k| utf16_output += ret2.u16count; 114| | 115| 1.35k| const auto ret3 = sse_expand_surrogate(in3); 116| 1.35k| _mm_storeu_si128((__m128i *)utf16_output, ret3.compressed); 117| 1.35k| utf16_output += ret3.u16count; 118| | 119| 1.35k| buf += 16; 120| 1.35k| } 121| 49.4k| } 122| | 123| | // check for invalid input 124| 688| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (124:7): [True: 22, False: 666] ------------------ 125| 22| return std::make_pair(nullptr, utf16_output); 126| 22| } 127| | 128| 666| return std::make_pair(buf, utf16_output); 129| 688|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_114validate_utf32EDv2_xS2_: 37| 3.87k|simdutf_really_inline bool validate_utf32(const __m128i a, const __m128i b) { 38| 3.87k| using vector_u32 = simd32; 39| | 40| 3.87k| const auto in0 = vector_u32(a); 41| 3.87k| const auto in1 = vector_u32(b); 42| | 43| 3.87k| const auto standardmax = vector_u32::splat(0x10ffff); 44| 3.87k| const auto offset = vector_u32::splat(0xffff2000); 45| 3.87k| const auto standardoffsetmax = vector_u32::splat(0xfffff7ff); 46| | 47| 3.87k| const auto too_large = max(in0, in1) > standardmax; 48| 3.87k| const auto surrogate0 = (in0 + offset) > standardoffsetmax; 49| 3.87k| const auto surrogate1 = (in1 + offset) > standardoffsetmax; 50| | 51| 3.87k| const auto combined = too_large | surrogate0 | surrogate1; 52| 3.87k| return !combined.any(); 53| 3.87k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_120sse_expand_surrogateILNS_10endiannessE0EEENS1_18expansion_result_tEDv2_x: 9| 5.41k|expansion_result_t sse_expand_surrogate(const __m128i x) { 10| 5.41k| using vector_u32 = simd32; 11| 5.41k| using vector_u8 = simd8; 12| | 13| 5.41k| const auto in = vector_u32(x); 14| | 15| 5.41k| const auto non_surrogate_mask = (in & uint32_t(0xffff0000)) == uint32_t(0); 16| 5.41k| const auto mask = (~non_surrogate_mask.to_4bit_bitmask()) & 0xf; 17| | 18| 5.41k| const auto t0 = in - uint32_t(0x00010000); 19| 5.41k| const auto hi = t0.shr<10>() & uint32_t(0x000003ff); 20| 5.41k| const auto lo = t0.shl<16>() & uint32_t(0x03ff0000); 21| 5.41k| const auto surrogates = (lo | hi) | uint32_t(0xdc00d800); 22| | 23| 5.41k| const auto merged = as_vector_u8(select(non_surrogate_mask, in, surrogates)); 24| | 25| 5.41k| const auto shuffle = vector_u8::load( 26| 5.41k| (byte_order == endianness::LITTLE) ------------------ | Branch (26:7): [True: 5.41k, Folded] ------------------ 27| 5.41k| ? tables::utf32_to_utf16::pack_utf32_to_utf16le[mask] 28| 5.41k| : tables::utf32_to_utf16::pack_utf32_to_utf16be[mask]); 29| | 30| 5.41k| const size_t u16count = (4 + count_ones(mask)); 31| 5.41k| const auto compressed = shuffle.lookup_16(merged); 32| | 33| 5.41k| return {u16count, compressed}; 34| 5.41k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_126sse_convert_utf32_to_utf16ILNS_10endiannessE1EEENSt3__14pairIPKDiPDsEES7_mS8_: 58| 540| char16_t *utf16_output) { 59| | 60| 540| const char32_t *end = buf + len; 61| | 62| 540| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 63| 540| __m128i forbidden_bytemask = _mm_setzero_si128(); 64| | 65| 2.99k| while (end - buf >= 16 + 8) { ------------------ | Branch (65:10): [True: 2.48k, False: 506] ------------------ 66| 2.48k| const __m128i *ptr = reinterpret_cast(buf); 67| 2.48k| const __m128i in0 = _mm_loadu_si128(ptr + 0); 68| 2.48k| const __m128i in1 = _mm_loadu_si128(ptr + 1); 69| 2.48k| const __m128i in2 = _mm_loadu_si128(ptr + 2); 70| 2.48k| const __m128i in3 = _mm_loadu_si128(ptr + 3); 71| | 72| 2.48k| const __m128i combined = 73| 2.48k| _mm_or_si128(_mm_or_si128(in2, in3), _mm_or_si128(in0, in1)); 74| 2.48k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 2.48k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 1.92k, False: 558] | | ------------------ ------------------ 75| | // No bits set above 16th, directly pack UTF-32 to UTF-16 76| 1.92k| __m128i utf16_packed0 = _mm_packus_epi32(in0, in1); 77| 1.92k| __m128i utf16_packed1 = _mm_packus_epi32(in2, in3); 78| | 79| 1.92k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 80| 1.92k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 81| 1.92k| forbidden_bytemask = _mm_or_si128( 82| 1.92k| forbidden_bytemask, 83| 1.92k| _mm_or_si128( 84| 1.92k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed0, v_f800), v_d800), 85| 1.92k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed1, v_f800), v_d800))); 86| | 87| 1.92k| if (big_endian) { ------------------ | Branch (87:11): [True: 1.92k, Folded] ------------------ 88| 1.92k| const __m128i swap = 89| 1.92k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 90| 1.92k| utf16_packed0 = _mm_shuffle_epi8(utf16_packed0, swap); 91| 1.92k| utf16_packed1 = _mm_shuffle_epi8(utf16_packed1, swap); 92| 1.92k| } 93| | 94| 1.92k| _mm_storeu_si128((__m128i *)utf16_output + 0, utf16_packed0); 95| 1.92k| _mm_storeu_si128((__m128i *)utf16_output + 1, utf16_packed1); 96| 1.92k| utf16_output += 16; 97| 1.92k| buf += 16; 98| 1.92k| } else { 99| 558| if (!validate_utf32(in0, in1) || !validate_utf32(in2, in3)) { ------------------ | Branch (99:11): [True: 20, False: 538] | Branch (99:40): [True: 14, False: 524] ------------------ 100| 34| return std::make_pair(nullptr, utf16_output); 101| 34| } 102| | 103| 524| const auto ret0 = sse_expand_surrogate(in0); 104| 524| _mm_storeu_si128((__m128i *)utf16_output, ret0.compressed); 105| 524| utf16_output += ret0.u16count; 106| | 107| 524| const auto ret1 = sse_expand_surrogate(in1); 108| 524| _mm_storeu_si128((__m128i *)utf16_output, ret1.compressed); 109| 524| utf16_output += ret1.u16count; 110| | 111| 524| const auto ret2 = sse_expand_surrogate(in2); 112| 524| _mm_storeu_si128((__m128i *)utf16_output, ret2.compressed); 113| 524| utf16_output += ret2.u16count; 114| | 115| 524| const auto ret3 = sse_expand_surrogate(in3); 116| 524| _mm_storeu_si128((__m128i *)utf16_output, ret3.compressed); 117| 524| utf16_output += ret3.u16count; 118| | 119| 524| buf += 16; 120| 524| } 121| 2.48k| } 122| | 123| | // check for invalid input 124| 506| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (124:7): [True: 8, False: 498] ------------------ 125| 8| return std::make_pair(nullptr, utf16_output); 126| 8| } 127| | 128| 498| return std::make_pair(buf, utf16_output); 129| 506|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_120sse_expand_surrogateILNS_10endiannessE1EEENS1_18expansion_result_tEDv2_x: 9| 2.09k|expansion_result_t sse_expand_surrogate(const __m128i x) { 10| 2.09k| using vector_u32 = simd32; 11| 2.09k| using vector_u8 = simd8; 12| | 13| 2.09k| const auto in = vector_u32(x); 14| | 15| 2.09k| const auto non_surrogate_mask = (in & uint32_t(0xffff0000)) == uint32_t(0); 16| 2.09k| const auto mask = (~non_surrogate_mask.to_4bit_bitmask()) & 0xf; 17| | 18| 2.09k| const auto t0 = in - uint32_t(0x00010000); 19| 2.09k| const auto hi = t0.shr<10>() & uint32_t(0x000003ff); 20| 2.09k| const auto lo = t0.shl<16>() & uint32_t(0x03ff0000); 21| 2.09k| const auto surrogates = (lo | hi) | uint32_t(0xdc00d800); 22| | 23| 2.09k| const auto merged = as_vector_u8(select(non_surrogate_mask, in, surrogates)); 24| | 25| 2.09k| const auto shuffle = vector_u8::load( 26| 2.09k| (byte_order == endianness::LITTLE) ------------------ | Branch (26:7): [Folded, False: 2.09k] ------------------ 27| 2.09k| ? tables::utf32_to_utf16::pack_utf32_to_utf16le[mask] 28| 2.09k| : tables::utf32_to_utf16::pack_utf32_to_utf16be[mask]); 29| | 30| 2.09k| const size_t u16count = (4 + count_ones(mask)); 31| 2.09k| const auto compressed = shuffle.lookup_16(merged); 32| | 33| 2.09k| return {u16count, compressed}; 34| 2.09k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf32_to_utf16_with_errorsILNS_10endiannessE0EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 134| 714| char16_t *utf16_output) { 135| 714| const char32_t *start = buf; 136| 714| const char32_t *end = buf + len; 137| | 138| 714| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 139| | 140| 9.71k| while (end - buf >= 8) { ------------------ | Branch (140:10): [True: 9.39k, False: 326] ------------------ 141| 9.39k| const __m128i in = _mm_loadu_si128((__m128i *)buf); 142| 9.39k| const __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 143| | 144| 9.39k| const __m128i combined = _mm_or_si128(in, nextin); 145| 9.39k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 9.39k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 7.29k, False: 2.09k] | | ------------------ ------------------ 146| | // No bits set above 16th, directly pack UTF-32 to UTF-16 147| 7.29k| __m128i utf16_packed = _mm_packus_epi32(in, nextin); 148| | 149| 7.29k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 150| 7.29k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 151| 7.29k| const __m128i forbidden_bytemask = 152| 7.29k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed, v_f800), v_d800); 153| 7.29k| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (153:11): [True: 46, False: 7.25k] ------------------ 154| 46| return std::make_pair(result(error_code::SURROGATE, buf - start), 155| 46| utf16_output); 156| 46| } 157| | 158| 7.25k| if (big_endian) { ------------------ | Branch (158:11): [Folded, False: 7.25k] ------------------ 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| 7.25k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 165| 7.25k| utf16_output += 8; 166| 7.25k| buf += 8; 167| 7.25k| } else { 168| 2.09k| size_t forward = 7; 169| 2.09k| size_t k = 0; 170| 2.09k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (170:11): [True: 0, False: 2.09k] ------------------ 171| 0| forward = size_t(end - buf - 1); 172| 0| } 173| 15.0k| for (; k < forward; k++) { ------------------ | Branch (173:14): [True: 13.3k, False: 1.75k] ------------------ 174| 13.3k| uint32_t word = buf[k]; 175| 13.3k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (175:13): [True: 6.32k, False: 6.98k] ------------------ 176| | // will not generate a surrogate pair 177| 6.32k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (177:15): [True: 808, False: 5.51k] | Branch (177:33): [True: 34, False: 774] ------------------ 178| 34| return std::make_pair( 179| 34| result(error_code::SURROGATE, buf - start + k), utf16_output); 180| 34| } 181| 6.29k| *utf16_output++ = 182| 6.29k| big_endian ------------------ | Branch (182:15): [Folded, False: 6.29k] ------------------ 183| 6.29k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 184| 6.29k| : char16_t(word); 185| 6.98k| } else { 186| | // will generate a surrogate pair 187| 6.98k| if (word > 0x10FFFF) { ------------------ | Branch (187:15): [True: 308, False: 6.67k] ------------------ 188| 308| return std::make_pair( 189| 308| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 190| 308| } 191| 6.67k| word -= 0x10000; 192| 6.67k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 193| 6.67k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 194| 6.67k| if (big_endian) { ------------------ | Branch (194:15): [Folded, False: 6.67k] ------------------ 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| 6.67k| *utf16_output++ = char16_t(high_surrogate); 201| 6.67k| *utf16_output++ = char16_t(low_surrogate); 202| 6.67k| } 203| 13.3k| } 204| 1.75k| buf += k; 205| 1.75k| } 206| 9.39k| } 207| | 208| 326| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 209| 714|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_138sse_convert_utf32_to_utf16_with_errorsILNS_10endiannessE1EEENSt3__14pairINS_6resultEPDsEEPKDimS7_: 134| 660| char16_t *utf16_output) { 135| 660| const char32_t *start = buf; 136| 660| const char32_t *end = buf + len; 137| | 138| 660| const __m128i v_ffff0000 = _mm_set1_epi32((int32_t)0xffff0000); 139| | 140| 10.4k| while (end - buf >= 8) { ------------------ | Branch (140:10): [True: 10.1k, False: 310] ------------------ 141| 10.1k| const __m128i in = _mm_loadu_si128((__m128i *)buf); 142| 10.1k| const __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 143| | 144| 10.1k| const __m128i combined = _mm_or_si128(in, nextin); 145| 10.1k| if (simdutf_likely(_mm_testz_si128(combined, v_ffff0000))) { ------------------ | | 90| 10.1k| #define simdutf_likely(x) __builtin_expect(!!(x), 1) | | ------------------ | | | Branch (90:31): [True: 9.18k, False: 986] | | ------------------ ------------------ 146| | // No bits set above 16th, directly pack UTF-32 to UTF-16 147| 9.18k| __m128i utf16_packed = _mm_packus_epi32(in, nextin); 148| | 149| 9.18k| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 150| 9.18k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 151| 9.18k| const __m128i forbidden_bytemask = 152| 9.18k| _mm_cmpeq_epi16(_mm_and_si128(utf16_packed, v_f800), v_d800); 153| 9.18k| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (153:11): [True: 26, False: 9.15k] ------------------ 154| 26| return std::make_pair(result(error_code::SURROGATE, buf - start), 155| 26| utf16_output); 156| 26| } 157| | 158| 9.15k| if (big_endian) { ------------------ | Branch (158:11): [True: 9.15k, Folded] ------------------ 159| 9.15k| const __m128i swap = 160| 9.15k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 161| 9.15k| utf16_packed = _mm_shuffle_epi8(utf16_packed, swap); 162| 9.15k| } 163| | 164| 9.15k| _mm_storeu_si128((__m128i *)utf16_output, utf16_packed); 165| 9.15k| utf16_output += 8; 166| 9.15k| buf += 8; 167| 9.15k| } else { 168| 986| size_t forward = 7; 169| 986| size_t k = 0; 170| 986| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (170:11): [True: 0, False: 986] ------------------ 171| 0| forward = size_t(end - buf - 1); 172| 0| } 173| 6.29k| for (; k < forward; k++) { ------------------ | Branch (173:14): [True: 5.63k, False: 662] ------------------ 174| 5.63k| uint32_t word = buf[k]; 175| 5.63k| if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (175:13): [True: 3.07k, False: 2.56k] ------------------ 176| | // will not generate a surrogate pair 177| 3.07k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (177:15): [True: 916, False: 2.15k] | Branch (177:33): [True: 32, False: 884] ------------------ 178| 32| return std::make_pair( 179| 32| result(error_code::SURROGATE, buf - start + k), utf16_output); 180| 32| } 181| 3.04k| *utf16_output++ = 182| 3.04k| big_endian ------------------ | Branch (182:15): [True: 3.04k, Folded] ------------------ 183| 3.04k| ? char16_t((uint16_t(word) >> 8) | (uint16_t(word) << 8)) 184| 3.04k| : char16_t(word); 185| 3.04k| } else { 186| | // will generate a surrogate pair 187| 2.56k| if (word > 0x10FFFF) { ------------------ | Branch (187:15): [True: 292, False: 2.27k] ------------------ 188| 292| return std::make_pair( 189| 292| result(error_code::TOO_LARGE, buf - start + k), utf16_output); 190| 292| } 191| 2.27k| word -= 0x10000; 192| 2.27k| uint16_t high_surrogate = uint16_t(0xD800 + (word >> 10)); 193| 2.27k| uint16_t low_surrogate = uint16_t(0xDC00 + (word & 0x3FF)); 194| 2.27k| if (big_endian) { ------------------ | Branch (194:15): [True: 2.27k, Folded] ------------------ 195| 2.27k| high_surrogate = 196| 2.27k| uint16_t((high_surrogate >> 8) | (high_surrogate << 8)); 197| 2.27k| low_surrogate = 198| 2.27k| uint16_t((low_surrogate >> 8) | (low_surrogate << 8)); 199| 2.27k| } 200| 2.27k| *utf16_output++ = char16_t(high_surrogate); 201| 2.27k| *utf16_output++ = char16_t(low_surrogate); 202| 2.27k| } 203| 5.63k| } 204| 662| buf += k; 205| 662| } 206| 10.1k| } 207| | 208| 310| return std::make_pair(result(error_code::SUCCESS, buf - start), utf16_output); 209| 660|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_125sse_convert_utf32_to_utf8EPKDimPc: 2| 872|sse_convert_utf32_to_utf8(const char32_t *buf, size_t len, char *utf8_output) { 3| 872| const char32_t *end = buf + len; 4| | 5| 872| const __m128i v_0000 = _mm_setzero_si128(); //__m128 = 128 bits 6| 872| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); // 1111 1000 0000 7| | // 0000 8| 872| const __m128i v_c080 = _mm_set1_epi16((uint16_t)0xc080); // 1100 0000 1000 9| | // 0000 10| 872| const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80); // 1111 1111 1000 11| | // 0000 12| 872| const __m128i v_ffff0000 = _mm_set1_epi32( 13| 872| (uint32_t)0xffff0000); // 1111 1111 1111 1111 0000 0000 0000 0000 14| 872| const __m128i v_7fffffff = _mm_set1_epi32( 15| 872| (uint32_t)0x7fffffff); // 0111 1111 1111 1111 1111 1111 1111 1111 16| 872| __m128i running_max = _mm_setzero_si128(); 17| 872| __m128i forbidden_bytemask = _mm_setzero_si128(); 18| 872| const size_t safety_margin = 19| 872| 12; // to avoid overruns, see issue 20| | // https://github.com/simdutf/simdutf/issues/92 21| | 22| 20.4k| while (end - buf >= ------------------ | Branch (22:10): [True: 19.7k, False: 668] ------------------ 23| 20.4k| std::ptrdiff_t( 24| 20.4k| 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| 19.7k| __m128i in = _mm_loadu_si128((__m128i *)buf); 29| 19.7k| __m128i nextin = _mm_loadu_si128( 30| 19.7k| (__m128i *)buf + 1); // These two values can hold only 8 UTF32 chars 31| 19.7k| running_max = _mm_max_epu32( 32| 19.7k| _mm_max_epu32(in, running_max), // take element-wise max char32_t from 33| | // in and running_max vector 34| 19.7k| 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| 19.7k| __m128i in_16 = _mm_packus_epi32( 40| 19.7k| _mm_and_si128(in, v_7fffffff), 41| 19.7k| _mm_and_si128( 42| 19.7k| nextin, 43| 19.7k| 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| 19.7k| if (_mm_testz_si128(in_16, v_ff80)) { // if the first two blocks are ASCII ------------------ | Branch (60:9): [True: 15.2k, False: 4.56k] ------------------ 61| 15.2k| __m128i thirdin = _mm_loadu_si128((__m128i *)buf + 2); 62| 15.2k| __m128i fourthin = _mm_loadu_si128((__m128i *)buf + 3); 63| 15.2k| running_max = _mm_max_epu32( 64| 15.2k| _mm_max_epu32(thirdin, running_max), 65| 15.2k| fourthin); // take the running max of all 4 vectors thus far 66| 15.2k| __m128i nextin_16 = _mm_packus_epi32( 67| 15.2k| _mm_and_si128(thirdin, v_7fffffff), 68| 15.2k| _mm_and_si128(fourthin, 69| 15.2k| v_7fffffff)); // pack into 1 vector, now you have two 70| 15.2k| if (!_mm_testz_si128( ------------------ | Branch (70:11): [True: 774, False: 14.4k] ------------------ 71| 15.2k| nextin_16, 72| 15.2k| v_ff80)) { // checks if the second packed vector is ASCII, if not: 73| | // 1. pack the bytes 74| | // obviously suboptimal. 75| 774| const __m128i utf8_packed = _mm_packus_epi16( 76| 774| in_16, in_16); // creates two copy of in_16 in 1 vector 77| | // 2. store (16 bytes) 78| 774| _mm_storeu_si128((__m128i *)utf8_output, 79| 774| utf8_packed); // put them into the output 80| | // 3. adjust pointers 81| 774| buf += 8; // the char32_t buffer pointer goes up 8 char32_t chars* 32 82| | // bits = 256 bits 83| 774| utf8_output += 84| 774| 8; // same with output, e.g. lift the first two blocks alone. 85| | // Proceed with next input 86| 774| in_16 = nextin_16; 87| | // We need to update in and nextin because they are used later. 88| 774| in = thirdin; 89| 774| nextin = fourthin; 90| 14.4k| } else { 91| | // 1. pack the bytes 92| 14.4k| const __m128i utf8_packed = _mm_packus_epi16(in_16, nextin_16); 93| | // 2. store (16 bytes) 94| 14.4k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 95| | // 3. adjust pointers 96| 14.4k| buf += 16; 97| 14.4k| utf8_output += 16; 98| 14.4k| continue; // we are done for this round! 99| 14.4k| } 100| 15.2k| } 101| | 102| | // no bits set above 7th bit -- find out all the ASCII characters 103| 5.34k| const __m128i one_byte_bytemask = 104| 5.34k| _mm_cmpeq_epi16( // this takes four bytes at a time and compares: 105| 5.34k| _mm_and_si128(in_16, v_ff80), // the vector that get only the first 106| | // 9 bits of each 16-bit/2-byte units 107| 5.34k| v_0000 // 108| 5.34k| ); // they should be all zero if they are ASCII. E.g. ASCII in UTF32 is 109| | // of format 0000 0000 0000 0XXX XXXX 110| | // _mm_cmpeq_epi16 should now return a 1111 1111 1111 1111 for equals, and 111| | // 0000 0000 0000 0000 if not for each 16-bit/2-byte units 112| 5.34k| const uint16_t one_byte_bitmask = static_cast(_mm_movemask_epi8( 113| 5.34k| one_byte_bytemask)); // collect the MSB from previous vector and put 114| | // them into uint16_t mas 115| | 116| | // no bits set above 11th bit 117| 5.34k| const __m128i one_or_two_bytes_bytemask = 118| 5.34k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_0000); 119| 5.34k| const uint16_t one_or_two_bytes_bitmask = 120| 5.34k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 121| | 122| 5.34k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (122:9): [True: 2.68k, False: 2.65k] ------------------ 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| 2.68k| const __m128i v_1f00 = 129| 2.68k| _mm_set1_epi16((int16_t)0x1f00); // 0001 1111 0000 0000 130| 2.68k| const __m128i v_003f = 131| 2.68k| _mm_set1_epi16((int16_t)0x003f); // 0000 0000 0011 1111 132| | 133| | // t0 = [000a|aaaa|bbbb|bb00] 134| 2.68k| const __m128i t0 = _mm_slli_epi16(in_16, 2); // shift packed vector by two 135| | // t1 = [000a|aaaa|0000|0000] 136| 2.68k| const __m128i t1 = _mm_and_si128(t0, v_1f00); // potential first utf8 byte 137| | // t2 = [0000|0000|00bb|bbbb] 138| 2.68k| const __m128i t2 = 139| 2.68k| _mm_and_si128(in_16, v_003f); // potential second utf8 byte 140| | // t3 = [000a|aaaa|00bb|bbbb] 141| 2.68k| const __m128i t3 = 142| 2.68k| _mm_or_si128(t1, t2); // first and second potential utf8 byte together 143| | // t4 = [110a|aaaa|10bb|bbbb] 144| 2.68k| const __m128i t4 = _mm_or_si128( 145| 2.68k| t3, 146| 2.68k| v_c080); // t3 | 1100 0000 1000 0000 = full potential 2-byte utf8 unit 147| | 148| | // 2. merge ASCII and 2-byte codewords 149| 2.68k| const __m128i utf8_unpacked = 150| 2.68k| _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| 2.68k| const uint16_t m0 = one_byte_bitmask & 0x5555; // m0 = 0h0g0f0e0d0c0b0a 156| 2.68k| const uint16_t m1 = 157| 2.68k| static_cast(m0 >> 7); // m1 = 00000000h0g0f0e0 158| 2.68k| const uint8_t m2 = 159| 2.68k| static_cast((m0 | m1) & 0xff); // m2 = hdgcfbea 160| | // 4. pack the bytes 161| 2.68k| const uint8_t *row = 162| 2.68k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[m2][0]; 163| 2.68k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 164| 2.68k| const __m128i utf8_packed = _mm_shuffle_epi8(utf8_unpacked, shuffle); 165| | 166| | // 5. store bytes 167| 2.68k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 168| | 169| | // 6. adjust pointers 170| 2.68k| buf += 8; 171| 2.68k| utf8_output += row[0]; 172| 2.68k| continue; 173| 2.68k| } 174| | 175| | // Check for overflow in packing 176| | 177| 2.65k| const __m128i saturation_bytemask = _mm_cmpeq_epi32( 178| 2.65k| _mm_and_si128(_mm_or_si128(in, nextin), v_ffff0000), v_0000); 179| 2.65k| const uint32_t saturation_bitmask = 180| 2.65k| static_cast(_mm_movemask_epi8(saturation_bytemask)); 181| 2.65k| if (saturation_bitmask == 0xffff) { ------------------ | Branch (181:9): [True: 1.29k, False: 1.36k] ------------------ 182| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 183| 1.29k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 184| 1.29k| forbidden_bytemask = 185| 1.29k| _mm_or_si128(forbidden_bytemask, 186| 1.29k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_d800)); 187| | 188| 1.29k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 189| 1.29k| 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| 1.29k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 219| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 220| 1.29k| const __m128i t0 = _mm_shuffle_epi8(in_16, dup_even); 221| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 222| 1.29k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 218| 1.29k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 223| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 224| 1.29k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 218| 1.29k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 225| | 226| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 227| 1.29k| const __m128i s0 = _mm_srli_epi16(in_16, 4); 228| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 229| 1.29k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 218| 1.29k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 230| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 231| 1.29k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 218| 1.29k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 232| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 233| 1.29k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 218| 1.29k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 234| 1.29k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 235| 1.29k| simdutf_vec(0b0100000000000000)); ------------------ | | 218| 1.29k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 236| 1.29k| const __m128i s4 = _mm_xor_si128(s3, m0); 237| 1.29k|#undef simdutf_vec 238| | 239| | // 4. expand code units 16-bit => 32-bit 240| 1.29k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 241| 1.29k| 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| 1.29k| const uint16_t mask = 245| 1.29k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 246| 1.29k| if (mask == 0) { ------------------ | Branch (246:11): [True: 580, False: 712] ------------------ 247| | // We only have three-byte code units. Use fast path. 248| 580| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 249| 580| 15, 13, -1, -1, -1, -1); 250| 580| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 251| 580| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 252| 580| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 253| 580| utf8_output += 12; 254| 580| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 255| 580| utf8_output += 12; 256| 580| buf += 8; 257| 580| continue; 258| 580| } 259| 712| const uint8_t mask0 = uint8_t(mask); 260| | 261| 712| const uint8_t *row0 = 262| 712| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 263| 712| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 264| 712| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 265| | 266| 712| const uint8_t mask1 = static_cast(mask >> 8); 267| | 268| 712| const uint8_t *row1 = 269| 712| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 270| 712| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 271| 712| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 272| | 273| 712| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 274| 712| utf8_output += row0[0]; 275| 712| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 276| 712| utf8_output += row1[0]; 277| | 278| 712| buf += 8; 279| 1.36k| } 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| 1.36k| size_t forward = 15; 285| 1.36k| size_t k = 0; 286| 1.36k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (286:11): [True: 0, False: 1.36k] ------------------ 287| 0| forward = size_t(end - buf - 1); 288| 0| } 289| 19.4k| for (; k < forward; k++) { ------------------ | Branch (289:14): [True: 18.2k, False: 1.15k] ------------------ 290| 18.2k| uint32_t word = buf[k]; 291| 18.2k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (291:13): [True: 8.56k, False: 9.69k] ------------------ 292| 8.56k| *utf8_output++ = char(word); 293| 9.69k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (293:20): [True: 1.21k, False: 8.47k] ------------------ 294| 1.21k| *utf8_output++ = char((word >> 6) | 0b11000000); 295| 1.21k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 296| 8.47k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (296:20): [True: 2.61k, False: 5.85k] ------------------ 297| 2.61k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (297:15): [True: 966, False: 1.65k] | Branch (297:33): [True: 24, False: 942] ------------------ 298| 24| return std::make_pair(nullptr, utf8_output); 299| 24| } 300| 2.59k| *utf8_output++ = char((word >> 12) | 0b11100000); 301| 2.59k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 302| 2.59k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 303| 5.85k| } else { 304| 5.85k| if (word > 0x10FFFF) { ------------------ | Branch (304:15): [True: 180, False: 5.67k] ------------------ 305| 180| return std::make_pair(nullptr, utf8_output); 306| 180| } 307| 5.67k| *utf8_output++ = char((word >> 18) | 0b11110000); 308| 5.67k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 309| 5.67k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 310| 5.67k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 311| 5.67k| } 312| 18.2k| } 313| 1.15k| buf += k; 314| 1.15k| } 315| 2.65k| } // while 316| | 317| | // check for invalid input 318| 668| const __m128i v_10ffff = _mm_set1_epi32((uint32_t)0x10ffff); 319| 668| if (static_cast(_mm_movemask_epi8(_mm_cmpeq_epi32( ------------------ | Branch (319:7): [True: 28, False: 640] ------------------ 320| 668| _mm_max_epu32(running_max, v_10ffff), v_10ffff))) != 0xffff) { 321| 28| return std::make_pair(nullptr, utf8_output); 322| 28| } 323| | 324| 640| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (324:7): [True: 46, False: 594] ------------------ 325| 46| return std::make_pair(nullptr, utf8_output); 326| 46| } 327| | 328| 594| return std::make_pair(buf, utf8_output); 329| 640|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_137sse_convert_utf32_to_utf8_with_errorsEPKDimPc: 333| 924| char *utf8_output) { 334| 924| const char32_t *end = buf + len; 335| 924| const char32_t *start = buf; 336| | 337| 924| const __m128i v_0000 = _mm_setzero_si128(); 338| 924| const __m128i v_f800 = _mm_set1_epi16((uint16_t)0xf800); 339| 924| const __m128i v_c080 = _mm_set1_epi16((uint16_t)0xc080); 340| 924| const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80); 341| 924| const __m128i v_ffff0000 = _mm_set1_epi32((uint32_t)0xffff0000); 342| 924| const __m128i v_7fffffff = _mm_set1_epi32((uint32_t)0x7fffffff); 343| 924| const __m128i v_10ffff = _mm_set1_epi32((uint32_t)0x10ffff); 344| | 345| 924| const size_t safety_margin = 346| 924| 12; // to avoid overruns, see issue 347| | // https://github.com/simdutf/simdutf/issues/92 348| | 349| 59.6k| while (end - buf >= std::ptrdiff_t(16 + safety_margin)) { ------------------ | Branch (349:10): [True: 58.8k, False: 704] ------------------ 350| | // We load two 16 bytes registers for a total of 32 bytes or 8 characters. 351| 58.8k| __m128i in = _mm_loadu_si128((__m128i *)buf); 352| 58.8k| __m128i nextin = _mm_loadu_si128((__m128i *)buf + 1); 353| | // Check for too large input 354| 58.8k| __m128i max_input = _mm_max_epu32(_mm_max_epu32(in, nextin), v_10ffff); 355| 58.8k| if (static_cast(_mm_movemask_epi8( ------------------ | Branch (355:9): [True: 54, False: 58.8k] ------------------ 356| 58.8k| _mm_cmpeq_epi32(max_input, v_10ffff))) != 0xffff) { 357| 54| return std::make_pair(result(error_code::TOO_LARGE, buf - start), 358| 54| utf8_output); 359| 54| } 360| | 361| | // Pack 32-bit UTF-32 code units to 16-bit UTF-16 code units with unsigned 362| | // saturation 363| 58.8k| __m128i in_16 = _mm_packus_epi32(_mm_and_si128(in, v_7fffffff), 364| 58.8k| _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| 58.8k| if (_mm_testz_si128(in_16, v_ff80)) { // ASCII fast path!!!! ------------------ | Branch (369:9): [True: 51.5k, False: 7.27k] ------------------ 370| | // 1. pack the bytes 371| | // obviously suboptimal. 372| 51.5k| const __m128i utf8_packed = _mm_packus_epi16(in_16, in_16); 373| | // 2. store (16 bytes) 374| 51.5k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 375| | // 3. adjust pointers 376| 51.5k| buf += 8; 377| 51.5k| utf8_output += 8; 378| 51.5k| continue; 379| 51.5k| } 380| | 381| | // no bits set above 7th bit 382| 7.27k| const __m128i one_byte_bytemask = 383| 7.27k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_ff80), v_0000); 384| 7.27k| const uint16_t one_byte_bitmask = 385| 7.27k| static_cast(_mm_movemask_epi8(one_byte_bytemask)); 386| | 387| | // no bits set above 11th bit 388| 7.27k| const __m128i one_or_two_bytes_bytemask = 389| 7.27k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_0000); 390| 7.27k| const uint16_t one_or_two_bytes_bitmask = 391| 7.27k| static_cast(_mm_movemask_epi8(one_or_two_bytes_bytemask)); 392| | 393| 7.27k| if (one_or_two_bytes_bitmask == 0xffff) { ------------------ | Branch (393:9): [True: 4.82k, False: 2.45k] ------------------ 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| 4.82k| const __m128i v_1f00 = _mm_set1_epi16((int16_t)0x1f00); 400| 4.82k| const __m128i v_003f = _mm_set1_epi16((int16_t)0x003f); 401| | 402| | // t0 = [000a|aaaa|bbbb|bb00] 403| 4.82k| const __m128i t0 = _mm_slli_epi16(in_16, 2); 404| | // t1 = [000a|aaaa|0000|0000] 405| 4.82k| const __m128i t1 = _mm_and_si128(t0, v_1f00); 406| | // t2 = [0000|0000|00bb|bbbb] 407| 4.82k| const __m128i t2 = _mm_and_si128(in_16, v_003f); 408| | // t3 = [000a|aaaa|00bb|bbbb] 409| 4.82k| const __m128i t3 = _mm_or_si128(t1, t2); 410| | // t4 = [110a|aaaa|10bb|bbbb] 411| 4.82k| const __m128i t4 = _mm_or_si128(t3, v_c080); 412| | 413| | // 2. merge ASCII and 2-byte codewords 414| 4.82k| const __m128i utf8_unpacked = 415| 4.82k| _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| 4.82k| const uint16_t m0 = one_byte_bitmask & 0x5555; // m0 = 0h0g0f0e0d0c0b0a 421| 4.82k| const uint16_t m1 = 422| 4.82k| static_cast(m0 >> 7); // m1 = 00000000h0g0f0e0 423| 4.82k| const uint8_t m2 = 424| 4.82k| static_cast((m0 | m1) & 0xff); // m2 = hdgcfbea 425| | // 4. pack the bytes 426| 4.82k| const uint8_t *row = 427| 4.82k| &simdutf::tables::utf16_to_utf8::pack_1_2_utf8_bytes[m2][0]; 428| 4.82k| const __m128i shuffle = _mm_loadu_si128((__m128i *)(row + 1)); 429| 4.82k| const __m128i utf8_packed = _mm_shuffle_epi8(utf8_unpacked, shuffle); 430| | 431| | // 5. store bytes 432| 4.82k| _mm_storeu_si128((__m128i *)utf8_output, utf8_packed); 433| | 434| | // 6. adjust pointers 435| 4.82k| buf += 8; 436| 4.82k| utf8_output += row[0]; 437| 4.82k| continue; 438| 4.82k| } 439| | 440| | // Check for overflow in packing 441| 2.45k| const __m128i saturation_bytemask = _mm_cmpeq_epi32( 442| 2.45k| _mm_and_si128(_mm_or_si128(in, nextin), v_ffff0000), v_0000); 443| 2.45k| const uint32_t saturation_bitmask = 444| 2.45k| static_cast(_mm_movemask_epi8(saturation_bytemask)); 445| | 446| 2.45k| if (saturation_bitmask == 0xffff) { ------------------ | Branch (446:9): [True: 1.21k, False: 1.24k] ------------------ 447| | // case: code units from register produce either 1, 2 or 3 UTF-8 bytes 448| | 449| | // Check for illegal surrogate code units 450| 1.21k| const __m128i v_d800 = _mm_set1_epi16((uint16_t)0xd800); 451| 1.21k| const __m128i forbidden_bytemask = 452| 1.21k| _mm_cmpeq_epi16(_mm_and_si128(in_16, v_f800), v_d800); 453| 1.21k| if (static_cast(_mm_movemask_epi8(forbidden_bytemask)) != 0) { ------------------ | Branch (453:11): [True: 32, False: 1.18k] ------------------ 454| 32| return std::make_pair(result(error_code::SURROGATE, buf - start), 455| 32| utf8_output); 456| 32| } 457| | 458| 1.18k| const __m128i dup_even = _mm_setr_epi16(0x0000, 0x0202, 0x0404, 0x0606, 459| 1.18k| 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| 1.18k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) 489| | // [aaaa|bbbb|bbcc|cccc] => [bbcc|cccc|bbcc|cccc] 490| 1.18k| const __m128i t0 = _mm_shuffle_epi8(in_16, dup_even); 491| | // [bbcc|cccc|bbcc|cccc] => [00cc|cccc|0bcc|cccc] 492| 1.18k| const __m128i t1 = _mm_and_si128(t0, simdutf_vec(0b0011111101111111)); ------------------ | | 488| 1.18k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 493| | // [00cc|cccc|0bcc|cccc] => [10cc|cccc|0bcc|cccc] 494| 1.18k| const __m128i t2 = _mm_or_si128(t1, simdutf_vec(0b1000000000000000)); ------------------ | | 488| 1.18k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 495| | 496| | // [aaaa|bbbb|bbcc|cccc] => [0000|aaaa|bbbb|bbcc] 497| 1.18k| const __m128i s0 = _mm_srli_epi16(in_16, 4); 498| | // [0000|aaaa|bbbb|bbcc] => [0000|aaaa|bbbb|bb00] 499| 1.18k| const __m128i s1 = _mm_and_si128(s0, simdutf_vec(0b0000111111111100)); ------------------ | | 488| 1.18k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 500| | // [0000|aaaa|bbbb|bb00] => [00bb|bbbb|0000|aaaa] 501| 1.18k| const __m128i s2 = _mm_maddubs_epi16(s1, simdutf_vec(0x0140)); ------------------ | | 488| 1.18k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 502| | // [00bb|bbbb|0000|aaaa] => [11bb|bbbb|1110|aaaa] 503| 1.18k| const __m128i s3 = _mm_or_si128(s2, simdutf_vec(0b1100000011100000)); ------------------ | | 488| 1.18k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 504| 1.18k| const __m128i m0 = _mm_andnot_si128(one_or_two_bytes_bytemask, 505| 1.18k| simdutf_vec(0b0100000000000000)); ------------------ | | 488| 1.18k|#define simdutf_vec(x) _mm_set1_epi16(static_cast(x)) ------------------ 506| 1.18k| const __m128i s4 = _mm_xor_si128(s3, m0); 507| 1.18k|#undef simdutf_vec 508| | 509| | // 4. expand code units 16-bit => 32-bit 510| 1.18k| const __m128i out0 = _mm_unpacklo_epi16(t2, s4); 511| 1.18k| 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| 1.18k| const uint16_t mask = 515| 1.18k| (one_byte_bitmask & 0x5555) | (one_or_two_bytes_bitmask & 0xaaaa); 516| 1.18k| if (mask == 0) { ------------------ | Branch (516:11): [True: 466, False: 716] ------------------ 517| | // We only have three-byte code units. Use fast path. 518| 466| const __m128i shuffle = _mm_setr_epi8(2, 3, 1, 6, 7, 5, 10, 11, 9, 14, 519| 466| 15, 13, -1, -1, -1, -1); 520| 466| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle); 521| 466| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle); 522| 466| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 523| 466| utf8_output += 12; 524| 466| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 525| 466| utf8_output += 12; 526| 466| buf += 8; 527| 466| continue; 528| 466| } 529| 716| const uint8_t mask0 = uint8_t(mask); 530| | 531| 716| const uint8_t *row0 = 532| 716| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask0][0]; 533| 716| const __m128i shuffle0 = _mm_loadu_si128((__m128i *)(row0 + 1)); 534| 716| const __m128i utf8_0 = _mm_shuffle_epi8(out0, shuffle0); 535| | 536| 716| const uint8_t mask1 = static_cast(mask >> 8); 537| | 538| 716| const uint8_t *row1 = 539| 716| &simdutf::tables::utf16_to_utf8::pack_1_2_3_utf8_bytes[mask1][0]; 540| 716| const __m128i shuffle1 = _mm_loadu_si128((__m128i *)(row1 + 1)); 541| 716| const __m128i utf8_1 = _mm_shuffle_epi8(out1, shuffle1); 542| | 543| 716| _mm_storeu_si128((__m128i *)utf8_output, utf8_0); 544| 716| utf8_output += row0[0]; 545| 716| _mm_storeu_si128((__m128i *)utf8_output, utf8_1); 546| 716| utf8_output += row1[0]; 547| | 548| 716| buf += 8; 549| 1.24k| } 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| 1.24k| size_t forward = 15; 555| 1.24k| size_t k = 0; 556| 1.24k| if (size_t(end - buf) < forward + 1) { ------------------ | Branch (556:11): [True: 0, False: 1.24k] ------------------ 557| 0| forward = size_t(end - buf - 1); 558| 0| } 559| 19.1k| for (; k < forward; k++) { ------------------ | Branch (559:14): [True: 18.0k, False: 1.11k] ------------------ 560| 18.0k| uint32_t word = buf[k]; 561| 18.0k| if ((word & 0xFFFFFF80) == 0) { ------------------ | Branch (561:13): [True: 5.23k, False: 12.8k] ------------------ 562| 5.23k| *utf8_output++ = char(word); 563| 12.8k| } else if ((word & 0xFFFFF800) == 0) { ------------------ | Branch (563:20): [True: 1.50k, False: 11.3k] ------------------ 564| 1.50k| *utf8_output++ = char((word >> 6) | 0b11000000); 565| 1.50k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 566| 11.3k| } else if ((word & 0xFFFF0000) == 0) { ------------------ | Branch (566:20): [True: 2.48k, False: 8.86k] ------------------ 567| 2.48k| if (word >= 0xD800 && word <= 0xDFFF) { ------------------ | Branch (567:15): [True: 1.18k, False: 1.30k] | Branch (567:33): [True: 28, False: 1.15k] ------------------ 568| 28| return std::make_pair( 569| 28| result(error_code::SURROGATE, buf - start + k), utf8_output); 570| 28| } 571| 2.45k| *utf8_output++ = char((word >> 12) | 0b11100000); 572| 2.45k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 573| 2.45k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 574| 8.86k| } else { 575| 8.86k| if (word > 0x10FFFF) { ------------------ | Branch (575:15): [True: 106, False: 8.76k] ------------------ 576| 106| return std::make_pair( 577| 106| result(error_code::TOO_LARGE, buf - start + k), utf8_output); 578| 106| } 579| 8.76k| *utf8_output++ = char((word >> 18) | 0b11110000); 580| 8.76k| *utf8_output++ = char(((word >> 12) & 0b111111) | 0b10000000); 581| 8.76k| *utf8_output++ = char(((word >> 6) & 0b111111) | 0b10000000); 582| 8.76k| *utf8_output++ = char((word & 0b111111) | 0b10000000); 583| 8.76k| } 584| 18.0k| } 585| 1.11k| buf += k; 586| 1.11k| } 587| 2.45k| } // while 588| 704| return std::make_pair(result(error_code::SUCCESS, buf - start), utf8_output); 589| 924|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_129convert_masked_utf8_to_latin1EPKcmRPc: 9| 736k| 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| 736k| const __m128i in = _mm_loadu_si128((__m128i *)input); 21| 736k| const uint16_t input_utf8_end_of_code_point_mask = 22| 736k| utf8_end_of_code_point_mask & 23| 736k| 0xfff; // we are only processing 12 bytes in case it is not all ASCII 24| 736k| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (24:7): [True: 4.70k, False: 731k] ------------------ 25| | // We process the data in chunks of 12 bytes. 26| 4.70k| _mm_storeu_si128(reinterpret_cast<__m128i *>(latin1_output), in); 27| 4.70k| latin1_output += 12; // We wrote 12 characters. 28| 4.70k| return 12; // We consumed 12 bytes. 29| 4.70k| } 30| | /// We do not have a fast path available, so we fallback. 31| 731k| const uint8_t idx = 32| 731k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 33| 731k| const uint8_t consumed = 34| 731k| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 35| | // this indicates an invalid input: 36| 731k| if (idx >= 64) { ------------------ | Branch (36:7): [True: 24.9k, False: 706k] ------------------ 37| 24.9k| return consumed; 38| 24.9k| } 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| 706k| const __m128i sh = 46| 706k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 47| 706k| const __m128i perm = _mm_shuffle_epi8(in, sh); 48| 706k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 49| 706k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 50| 706k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 51| 706k| 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| 706k| _mm_storel_epi64((__m128i *)latin1_output, latin1_packed); 56| 706k| latin1_output += 6; // We wrote 6 bytes. 57| 706k| return consumed; 58| 731k|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE0EEEmPKcmRPDs: 10| 2.97M| 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| 2.97M| const __m128i swap = 23| 2.97M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 2.97M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 2.97M| const uint16_t input_utf8_end_of_code_point_mask = 26| 2.97M| utf8_end_of_code_point_mask & 0xfff; 27| 2.97M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 10.4k, False: 2.96M] ------------------ 28| | // We process the data in chunks of 12 bytes. 29| | // Note: using 16 bytes is unsafe, see issue_ossfuzz_71218 30| 10.4k| __m128i ascii_first = _mm_cvtepu8_epi16(in); 31| 10.4k| __m128i ascii_second = _mm_cvtepu8_epi16(_mm_srli_si128(in, 8)); 32| 10.4k| if (big_endian) { ------------------ | Branch (32:9): [Folded, False: 10.4k] ------------------ 33| 0| ascii_first = _mm_shuffle_epi8(ascii_first, swap); 34| 0| ascii_second = _mm_shuffle_epi8(ascii_second, swap); 35| 0| } 36| 10.4k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf16_output), ascii_first); 37| 10.4k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf16_output + 8), 38| 10.4k| ascii_second); 39| 10.4k| utf16_output += 12; // We wrote 12 16-bit characters. 40| 10.4k| return 12; // We consumed 12 bytes. 41| 10.4k| } 42| 2.96M| if (((utf8_end_of_code_point_mask & 0xFFFF) == 0xaaaa)) { ------------------ | Branch (42:7): [True: 11.5k, False: 2.95M] ------------------ 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| 11.5k| const __m128i sh = 47| 11.5k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 48| 11.5k| const __m128i perm = _mm_shuffle_epi8(in, sh); 49| 11.5k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 50| 11.5k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 51| 11.5k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 52| 11.5k| if (big_endian) ------------------ | Branch (52:9): [Folded, False: 11.5k] ------------------ 53| 0| composed = _mm_shuffle_epi8(composed, swap); 54| 11.5k| _mm_storeu_si128((__m128i *)utf16_output, composed); 55| 11.5k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 56| 11.5k| return 16; 57| 11.5k| } 58| 2.95M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (58:7): [True: 4.17k, False: 2.94M] ------------------ 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| 4.17k| const __m128i sh = 63| 4.17k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 64| 4.17k| const __m128i perm = _mm_shuffle_epi8(in, sh); 65| 4.17k| const __m128i ascii = 66| 4.17k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 67| 4.17k| const __m128i middlebyte = 68| 4.17k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 69| 4.17k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 70| 4.17k| const __m128i highbyte = 71| 4.17k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 72| 4.17k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 73| 4.17k| const __m128i composed = 74| 4.17k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 75| 4.17k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 76| 4.17k| if (big_endian) ------------------ | Branch (76:9): [Folded, False: 4.17k] ------------------ 77| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 78| 4.17k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 79| 4.17k| utf16_output += 4; 80| 4.17k| return 12; 81| 4.17k| } 82| | /// We do not have a fast path available, so we fallback. 83| | 84| 2.94M| const uint8_t idx = 85| 2.94M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 86| 2.94M| const uint8_t consumed = 87| 2.94M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 88| 2.94M| if (idx < 64) { ------------------ | Branch (88:7): [True: 2.50M, False: 441k] ------------------ 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| 2.50M| const __m128i sh = 96| 2.50M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 97| 2.50M| const __m128i perm = _mm_shuffle_epi8(in, sh); 98| 2.50M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 99| 2.50M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 100| 2.50M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 101| 2.50M| if (big_endian) ------------------ | Branch (101:9): [Folded, False: 2.50M] ------------------ 102| 0| composed = _mm_shuffle_epi8(composed, swap); 103| 2.50M| _mm_storeu_si128((__m128i *)utf16_output, composed); 104| 2.50M| utf16_output += 6; // We wrote 12 bytes, 6 code points. 105| 2.50M| } else if (idx < 145) { ------------------ | Branch (105:14): [True: 388k, False: 53.1k] ------------------ 106| | // FOUR (4) input code-code units 107| 388k| const __m128i sh = 108| 388k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 109| 388k| const __m128i perm = _mm_shuffle_epi8(in, sh); 110| 388k| const __m128i ascii = 111| 388k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 112| 388k| const __m128i middlebyte = 113| 388k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 114| 388k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 115| 388k| const __m128i highbyte = 116| 388k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 117| 388k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 118| 388k| const __m128i composed = 119| 388k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 120| 388k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 121| 388k| if (big_endian) ------------------ | Branch (121:9): [Folded, False: 388k] ------------------ 122| 0| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 123| 388k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 124| 388k| utf16_output += 4; 125| 388k| } else if (idx < 209) { ------------------ | Branch (125:14): [True: 52.0k, False: 1.13k] ------------------ 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| 52.0k| const __m128i sh = 137| 52.0k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 138| 52.0k| const __m128i perm = _mm_shuffle_epi8(in, sh); 139| 52.0k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 140| 52.0k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 141| 52.0k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 142| 52.0k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 143| | // correct for spurious high bit 144| 52.0k| const __m128i correct = 145| 52.0k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 146| 52.0k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 147| 52.0k| 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| 52.0k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 151| 52.0k| 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| 52.0k| const __m128i composed = 157| 52.0k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 158| 52.0k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 159| 52.0k| const __m128i composedminus = 160| 52.0k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 161| 52.0k| const __m128i lowtenbits = 162| 52.0k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 163| | // Notice the 0x3ff mask: 164| 52.0k| const __m128i hightenbits = 165| 52.0k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 166| 52.0k| const __m128i lowtenbitsadd = 167| 52.0k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 168| 52.0k| const __m128i hightenbitsadd = 169| 52.0k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 170| 52.0k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 171| 52.0k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 172| 52.0k| uint32_t basic_buffer[4]; 173| 52.0k| uint32_t basic_buffer_swap[4]; 174| 52.0k| if (big_endian) { ------------------ | Branch (174:9): [Folded, False: 52.0k] ------------------ 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| 52.0k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 180| 52.0k| uint32_t surrogate_buffer[4]; 181| 52.0k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 182| 208k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (182:24): [True: 156k, False: 52.0k] ------------------ 183| 156k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (183:11): [True: 43.7k, False: 112k] ------------------ 184| 43.7k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 185| 43.7k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 186| 43.7k| utf16_output += 2; 187| 112k| } else { 188| 112k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (188:27): [Folded, False: 112k] ------------------ 189| 112k| : uint16_t(basic_buffer[i]); 190| 112k| utf16_output++; 191| 112k| } 192| 156k| } 193| 52.0k| } else { 194| | // here we know that there is an error but we do not handle errors 195| 1.13k| } 196| 2.94M| return consumed; 197| 2.95M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_128convert_masked_utf8_to_utf16ILNS_10endiannessE1EEEmPKcmRPDs: 10| 2.57M| 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| 2.57M| const __m128i swap = 23| 2.57M| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 24| 2.57M| const __m128i in = _mm_loadu_si128((__m128i *)input); 25| 2.57M| const uint16_t input_utf8_end_of_code_point_mask = 26| 2.57M| utf8_end_of_code_point_mask & 0xfff; 27| 2.57M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (27:7): [True: 7.29k, False: 2.56M] ------------------ 28| | // We process the data in chunks of 12 bytes. 29| | // Note: using 16 bytes is unsafe, see issue_ossfuzz_71218 30| 7.29k| __m128i ascii_first = _mm_cvtepu8_epi16(in); 31| 7.29k| __m128i ascii_second = _mm_cvtepu8_epi16(_mm_srli_si128(in, 8)); 32| 7.29k| if (big_endian) { ------------------ | Branch (32:9): [True: 7.29k, Folded] ------------------ 33| 7.29k| ascii_first = _mm_shuffle_epi8(ascii_first, swap); 34| 7.29k| ascii_second = _mm_shuffle_epi8(ascii_second, swap); 35| 7.29k| } 36| 7.29k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf16_output), ascii_first); 37| 7.29k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf16_output + 8), 38| 7.29k| ascii_second); 39| 7.29k| utf16_output += 12; // We wrote 12 16-bit characters. 40| 7.29k| return 12; // We consumed 12 bytes. 41| 7.29k| } 42| 2.56M| if (((utf8_end_of_code_point_mask & 0xFFFF) == 0xaaaa)) { ------------------ | Branch (42:7): [True: 10.5k, False: 2.55M] ------------------ 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| 10.5k| const __m128i sh = 47| 10.5k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 48| 10.5k| const __m128i perm = _mm_shuffle_epi8(in, sh); 49| 10.5k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 50| 10.5k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 51| 10.5k| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 52| 10.5k| if (big_endian) ------------------ | Branch (52:9): [True: 10.5k, Folded] ------------------ 53| 10.5k| composed = _mm_shuffle_epi8(composed, swap); 54| 10.5k| _mm_storeu_si128((__m128i *)utf16_output, composed); 55| 10.5k| utf16_output += 8; // We wrote 16 bytes, 8 code points. 56| 10.5k| return 16; 57| 10.5k| } 58| 2.55M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (58:7): [True: 3.38k, False: 2.55M] ------------------ 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| 3.38k| const __m128i sh = 63| 3.38k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 64| 3.38k| const __m128i perm = _mm_shuffle_epi8(in, sh); 65| 3.38k| const __m128i ascii = 66| 3.38k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 67| 3.38k| const __m128i middlebyte = 68| 3.38k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 69| 3.38k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 70| 3.38k| const __m128i highbyte = 71| 3.38k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 72| 3.38k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 73| 3.38k| const __m128i composed = 74| 3.38k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 75| 3.38k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 76| 3.38k| if (big_endian) ------------------ | Branch (76:9): [True: 3.38k, Folded] ------------------ 77| 3.38k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 78| 3.38k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 79| 3.38k| utf16_output += 4; 80| 3.38k| return 12; 81| 3.38k| } 82| | /// We do not have a fast path available, so we fallback. 83| | 84| 2.55M| const uint8_t idx = 85| 2.55M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 86| 2.55M| const uint8_t consumed = 87| 2.55M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 88| 2.55M| if (idx < 64) { ------------------ | Branch (88:7): [True: 2.32M, False: 227k] ------------------ 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| 2.32M| const __m128i sh = 96| 2.32M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 97| 2.32M| const __m128i perm = _mm_shuffle_epi8(in, sh); 98| 2.32M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 99| 2.32M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 100| 2.32M| __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 101| 2.32M| if (big_endian) ------------------ | Branch (101:9): [True: 2.32M, Folded] ------------------ 102| 2.32M| composed = _mm_shuffle_epi8(composed, swap); 103| 2.32M| _mm_storeu_si128((__m128i *)utf16_output, composed); 104| 2.32M| utf16_output += 6; // We wrote 12 bytes, 6 code points. 105| 2.32M| } else if (idx < 145) { ------------------ | Branch (105:14): [True: 197k, False: 30.0k] ------------------ 106| | // FOUR (4) input code-code units 107| 197k| const __m128i sh = 108| 197k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 109| 197k| const __m128i perm = _mm_shuffle_epi8(in, sh); 110| 197k| const __m128i ascii = 111| 197k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 112| 197k| const __m128i middlebyte = 113| 197k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 114| 197k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 115| 197k| const __m128i highbyte = 116| 197k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 117| 197k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 118| 197k| const __m128i composed = 119| 197k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 120| 197k| __m128i composed_repacked = _mm_packus_epi32(composed, composed); 121| 197k| if (big_endian) ------------------ | Branch (121:9): [True: 197k, Folded] ------------------ 122| 197k| composed_repacked = _mm_shuffle_epi8(composed_repacked, swap); 123| 197k| _mm_storeu_si128((__m128i *)utf16_output, composed_repacked); 124| 197k| utf16_output += 4; 125| 197k| } else if (idx < 209) { ------------------ | Branch (125:14): [True: 29.0k, False: 978] ------------------ 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| 29.0k| const __m128i sh = 137| 29.0k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 138| 29.0k| const __m128i perm = _mm_shuffle_epi8(in, sh); 139| 29.0k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 140| 29.0k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 141| 29.0k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 142| 29.0k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 143| | // correct for spurious high bit 144| 29.0k| const __m128i correct = 145| 29.0k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 146| 29.0k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 147| 29.0k| 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| 29.0k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0xff000000)); 151| 29.0k| 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| 29.0k| const __m128i composed = 157| 29.0k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 158| 29.0k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 159| 29.0k| const __m128i composedminus = 160| 29.0k| _mm_sub_epi32(composed, _mm_set1_epi32(0x10000)); 161| 29.0k| const __m128i lowtenbits = 162| 29.0k| _mm_and_si128(composedminus, _mm_set1_epi32(0x3ff)); 163| | // Notice the 0x3ff mask: 164| 29.0k| const __m128i hightenbits = 165| 29.0k| _mm_and_si128(_mm_srli_epi32(composedminus, 10), _mm_set1_epi32(0x3ff)); 166| 29.0k| const __m128i lowtenbitsadd = 167| 29.0k| _mm_add_epi32(lowtenbits, _mm_set1_epi32(0xDC00)); 168| 29.0k| const __m128i hightenbitsadd = 169| 29.0k| _mm_add_epi32(hightenbits, _mm_set1_epi32(0xD800)); 170| 29.0k| const __m128i lowtenbitsaddshifted = _mm_slli_epi32(lowtenbitsadd, 16); 171| 29.0k| __m128i surrogates = _mm_or_si128(hightenbitsadd, lowtenbitsaddshifted); 172| 29.0k| uint32_t basic_buffer[4]; 173| 29.0k| uint32_t basic_buffer_swap[4]; 174| 29.0k| if (big_endian) { ------------------ | Branch (174:9): [True: 29.0k, Folded] ------------------ 175| 29.0k| _mm_storeu_si128((__m128i *)basic_buffer_swap, 176| 29.0k| _mm_shuffle_epi8(composed, swap)); 177| 29.0k| surrogates = _mm_shuffle_epi8(surrogates, swap); 178| 29.0k| } 179| 29.0k| _mm_storeu_si128((__m128i *)basic_buffer, composed); 180| 29.0k| uint32_t surrogate_buffer[4]; 181| 29.0k| _mm_storeu_si128((__m128i *)surrogate_buffer, surrogates); 182| 116k| for (size_t i = 0; i < 3; i++) { ------------------ | Branch (182:24): [True: 87.1k, False: 29.0k] ------------------ 183| 87.1k| if (basic_buffer[i] > 0x3c00000) { ------------------ | Branch (183:11): [True: 24.2k, False: 62.9k] ------------------ 184| 24.2k| utf16_output[0] = uint16_t(surrogate_buffer[i] & 0xffff); 185| 24.2k| utf16_output[1] = uint16_t(surrogate_buffer[i] >> 16); 186| 24.2k| utf16_output += 2; 187| 62.9k| } else { 188| 62.9k| utf16_output[0] = big_endian ? uint16_t(basic_buffer_swap[i]) ------------------ | Branch (188:27): [True: 62.9k, Folded] ------------------ 189| 62.9k| : uint16_t(basic_buffer[i]); 190| 62.9k| utf16_output++; 191| 62.9k| } 192| 87.1k| } 193| 29.0k| } else { 194| | // here we know that there is an error but we do not handle errors 195| 978| } 196| 2.55M| return consumed; 197| 2.55M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_128convert_masked_utf8_to_utf32EPKcmRPDi: 9| 3.32M| char32_t *&utf32_output) { 10| | // we use an approach where we try to process up to 12 input bytes. 11| | // Why 12 input bytes and not 16? Because we are concerned with the size of 12| | // the lookup tables. Also 12 is nicely divisible by two and three. 13| | // 14| | // 15| | // Optimization note: our main path below is load-latency dependent. Thus it 16| | // is maybe beneficial to have fast paths that depend on branch prediction but 17| | // have less latency. This results in more instructions but, potentially, also 18| | // higher speeds. 19| | // 20| | // We first try a few fast paths. 21| 3.32M| const __m128i in = _mm_loadu_si128((__m128i *)input); 22| 3.32M| const uint16_t input_utf8_end_of_code_point_mask = 23| 3.32M| utf8_end_of_code_point_mask & 0xfff; 24| 3.32M| if (utf8_end_of_code_point_mask == 0xfff) { ------------------ | Branch (24:7): [True: 9.53k, False: 3.31M] ------------------ 25| | // We process the data in chunks of 12 bytes. 26| 9.53k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 27| 9.53k| _mm_cvtepu8_epi32(in)); 28| 9.53k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 29| 9.53k| _mm_cvtepu8_epi32(_mm_srli_si128(in, 4))); 30| 9.53k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 8), 31| 9.53k| _mm_cvtepu8_epi32(_mm_srli_si128(in, 8))); 32| 9.53k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 12), 33| 9.53k| _mm_cvtepu8_epi32(_mm_srli_si128(in, 12))); 34| 9.53k| utf32_output += 12; // We wrote 12 32-bit characters. 35| 9.53k| return 12; // We consumed 12 bytes. 36| 9.53k| } 37| 3.31M| if (((utf8_end_of_code_point_mask & 0xffff) == 0xaaaa)) { ------------------ | Branch (37:7): [True: 24.4k, False: 3.29M] ------------------ 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| 24.4k| const __m128i sh = 42| 24.4k| _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14); 43| 24.4k| const __m128i perm = _mm_shuffle_epi8(in, sh); 44| 24.4k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 45| 24.4k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 46| 24.4k| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 47| 24.4k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 48| 24.4k| _mm_cvtepu16_epi32(composed)); 49| 24.4k| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 50| 24.4k| _mm_cvtepu16_epi32(_mm_srli_si128(composed, 8))); 51| 24.4k| utf32_output += 8; // We wrote 32 bytes, 8 code points. 52| 24.4k| return 16; 53| 24.4k| } 54| 3.29M| if (input_utf8_end_of_code_point_mask == 0x924) { ------------------ | Branch (54:7): [True: 3.32k, False: 3.28M] ------------------ 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| 3.32k| const __m128i sh = 59| 3.32k| _mm_setr_epi8(2, 1, 0, -1, 5, 4, 3, -1, 8, 7, 6, -1, 11, 10, 9, -1); 60| 3.32k| const __m128i perm = _mm_shuffle_epi8(in, sh); 61| 3.32k| const __m128i ascii = 62| 3.32k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 63| 3.32k| const __m128i middlebyte = 64| 3.32k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 65| 3.32k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 66| 3.32k| const __m128i highbyte = 67| 3.32k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 68| 3.32k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 69| 3.32k| const __m128i composed = 70| 3.32k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 71| 3.32k| _mm_storeu_si128((__m128i *)utf32_output, composed); 72| 3.32k| utf32_output += 4; 73| 3.32k| return 12; 74| 3.32k| } 75| | /// We do not have a fast path available, so we fallback. 76| | 77| 3.28M| const uint8_t idx = 78| 3.28M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][0]; 79| 3.28M| const uint8_t consumed = 80| 3.28M| tables::utf8_to_utf16::utf8bigindex[input_utf8_end_of_code_point_mask][1]; 81| 3.28M| if (idx < 64) { ------------------ | Branch (81:7): [True: 2.99M, False: 294k] ------------------ 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| 2.99M| const __m128i sh = 89| 2.99M| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 90| 2.99M| const __m128i perm = _mm_shuffle_epi8(in, sh); 91| 2.99M| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi16(0x7f)); 92| 2.99M| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi16(0x1f00)); 93| 2.99M| const __m128i composed = _mm_or_si128(ascii, _mm_srli_epi16(highbyte, 2)); 94| 2.99M| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output), 95| 2.99M| _mm_cvtepu16_epi32(composed)); 96| 2.99M| _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4), 97| 2.99M| _mm_cvtepu16_epi32(_mm_srli_si128(composed, 8))); 98| 2.99M| utf32_output += 6; // We wrote 12 bytes, 6 code points. 99| 2.99M| } else if (idx < 145) { ------------------ | Branch (99:14): [True: 227k, False: 67.0k] ------------------ 100| | // FOUR (4) input code-code units 101| 227k| const __m128i sh = 102| 227k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 103| 227k| const __m128i perm = _mm_shuffle_epi8(in, sh); 104| 227k| const __m128i ascii = 105| 227k| _mm_and_si128(perm, _mm_set1_epi32(0x7f)); // 7 or 6 bits 106| 227k| const __m128i middlebyte = 107| 227k| _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); // 5 or 6 bits 108| 227k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 109| 227k| const __m128i highbyte = 110| 227k| _mm_and_si128(perm, _mm_set1_epi32(0x0f0000)); // 4 bits 111| 227k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 4); 112| 227k| const __m128i composed = 113| 227k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), highbyte_shifted); 114| 227k| _mm_storeu_si128((__m128i *)utf32_output, composed); 115| 227k| utf32_output += 4; 116| 227k| } else if (idx < 209) { ------------------ | Branch (116:14): [True: 65.7k, False: 1.32k] ------------------ 117| | // TWO (2) input code-code units 118| 65.7k| const __m128i sh = 119| 65.7k| _mm_loadu_si128((const __m128i *)tables::utf8_to_utf16::shufutf8[idx]); 120| 65.7k| const __m128i perm = _mm_shuffle_epi8(in, sh); 121| 65.7k| const __m128i ascii = _mm_and_si128(perm, _mm_set1_epi32(0x7f)); 122| 65.7k| const __m128i middlebyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f00)); 123| 65.7k| const __m128i middlebyte_shifted = _mm_srli_epi32(middlebyte, 2); 124| 65.7k| __m128i middlehighbyte = _mm_and_si128(perm, _mm_set1_epi32(0x3f0000)); 125| | // correct for spurious high bit 126| 65.7k| const __m128i correct = 127| 65.7k| _mm_srli_epi32(_mm_and_si128(perm, _mm_set1_epi32(0x400000)), 1); 128| 65.7k| middlehighbyte = _mm_xor_si128(correct, middlehighbyte); 129| 65.7k| const __m128i middlehighbyte_shifted = _mm_srli_epi32(middlehighbyte, 4); 130| 65.7k| const __m128i highbyte = _mm_and_si128(perm, _mm_set1_epi32(0x07000000)); 131| 65.7k| const __m128i highbyte_shifted = _mm_srli_epi32(highbyte, 6); 132| 65.7k| const __m128i composed = 133| 65.7k| _mm_or_si128(_mm_or_si128(ascii, middlebyte_shifted), 134| 65.7k| _mm_or_si128(highbyte_shifted, middlehighbyte_shifted)); 135| 65.7k| _mm_storeu_si128((__m128i *)utf32_output, composed); 136| 65.7k| utf32_output += 3; 137| 65.7k| } else { 138| | // here we know that there is an error but we do not handle errors 139| 1.32k| } 140| 3.28M| return consumed; 141| 3.29M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_123utf16_gather_high_bytesILNS_10endiannessE0EEENS1_4simd5simd8IhEENS4_6simd16ItEES8_: 3| 2.33M| const simd16 in1) { 4| 2.33M| if (big_endian) { ------------------ | Branch (4:7): [Folded, False: 2.33M] ------------------ 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| 2.33M| } else { 12| 2.33M| const auto t0 = in0.shr<8>(); 13| 2.33M| const auto t1 = in1.shr<8>(); 14| | 15| 2.33M| return simd16::pack(t0, t1); 16| 2.33M| } 17| 2.33M|} simdutf.cpp:_ZN7simdutf8westmere12_GLOBAL__N_123utf16_gather_high_bytesILNS_10endiannessE1EEENS1_4simd5simd8IhEENS4_6simd16ItEES8_: 3| 2.77M| const simd16 in1) { 4| 2.77M| if (big_endian) { ------------------ | Branch (4:7): [True: 2.77M, Folded] ------------------ 5| | // we want lower bytes 6| 2.77M| const auto mask = simd16(0x00ff); 7| 2.77M| const auto t0 = in0 & mask; 8| 2.77M| const auto t1 = in1 & mask; 9| | 10| 2.77M| return simd16::pack(t0, t1); 11| 2.77M| } 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| 2.77M|}