|
35 | 35 | #include "v8-fast-api-calls.h" |
36 | 36 | #include "v8.h" |
37 | 37 |
|
| 38 | +#include <stdint.h> |
38 | 39 | #include <climits> |
39 | 40 | #include <cstring> |
40 | 41 | #include "nbytes.h" |
@@ -741,13 +742,40 @@ uint32_t FastByteLengthUtf8(Local<Value> receiver, |
741 | 742 | if (source.length > 128) { |
742 | 743 | return simdutf::utf8_length_from_latin1(source.data, source.length); |
743 | 744 | } |
| 745 | + |
744 | 746 | uint32_t length = source.length; |
745 | | - uint32_t result = length; |
746 | | - const uint8_t* data = reinterpret_cast<const uint8_t*>(source.data); |
747 | | - for (uint32_t i = 0; i < length; ++i) { |
748 | | - result += (data[i] >> 7); |
| 747 | + const auto input = reinterpret_cast<const uint8_t*>(source.data); |
| 748 | + |
| 749 | + uint32_t answer = length; |
| 750 | + uint32_t i = 0; |
| 751 | + |
| 752 | + auto pop = [](uint64_t v) { |
| 753 | + return static_cast<size_t>(((v >> 7) & UINT64_C(0x0101010101010101)) * |
| 754 | + UINT64_C(0x0101010101010101) >> |
| 755 | + 56); |
| 756 | + }; |
| 757 | + |
| 758 | + for (; i + 32 <= length; i += 32) { |
| 759 | + uint64_t v; |
| 760 | + memcpy(&v, input + i, 8); |
| 761 | + answer += pop(v); |
| 762 | + memcpy(&v, input + i + 8, 8); |
| 763 | + answer += pop(v); |
| 764 | + memcpy(&v, input + i + 16, 8); |
| 765 | + answer += pop(v); |
| 766 | + memcpy(&v, input + i + 24, 8); |
| 767 | + answer += pop(v); |
| 768 | + } |
| 769 | + for (; i + 8 <= length; i += 8) { |
| 770 | + uint64_t v; |
| 771 | + memcpy(&v, input + i, 8); |
| 772 | + answer += pop(v); |
749 | 773 | } |
750 | | - return result; |
| 774 | + for (; i + 1 <= length; i += 1) { |
| 775 | + answer += input[i] >> 7; |
| 776 | + } |
| 777 | + |
| 778 | + return answer; |
751 | 779 | } |
752 | 780 |
|
753 | 781 | static v8::CFunction fast_byte_length_utf8( |
|