coffee/fast_float
1
0
Fork 0
mirror of https://github.com/fastfloat/fast_float.git synced 2025-12-06 08:46:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

* fix for parse_eight_digits_unrolled

Browse Source
This commit is contained in:
IRainman 2025-10-22 02:28:31 +03:00
parent 1dfd486df4
commit 57f63fc327
1 changed files with 4 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
include/fast_float/ascii_number.h
View File

@ -10,9 +10,7 @@
#include "float_common.h"
#if defined(FASTFLOAT_SSE42)
#include <nmmintrin.h> // SSE4.2 intrinsics
#elif defined(FASTFLOAT_SSE2)
#if defined(FASTFLOAT_SSE2)
#include <emmintrin.h>
#elif defined(FASTFLOAT_NEON)
#include <arm_neon.h>
@ -124,7 +122,7 @@ uint64_t simd_read8_to_u64(UC const *) {
}
// credit @aqrit
fastfloat_really_inline FASTFLOAT_CONSTEXPR14 uint32_t
fastfloat_really_inline FASTFLOAT_CONSTEXPR14 uint64_t
parse_eight_digits_unrolled(uint64_t val) noexcept {
constexpr uint64_t mask = 0x000000FF000000FF;
constexpr uint64_t mul1 = 0x000F424000000064; // 100 + (1000000ULL << 32)
@ -132,12 +130,12 @@ parse_eight_digits_unrolled(uint64_t val) noexcept {
val -= 0x3030303030303030;
val = (val * 10) + (val >> 8); // val = (val * 2561) >> 8;
val = (((val & mask) * mul1) + (((val >> 16) & mask) * mul2)) >> 32;
return uint32_t(val);
return val;
}
// Call this if chars are definitely 8 digits.
template <typename UC>
fastfloat_really_inline FASTFLOAT_CONSTEXPR20 uint32_t
fastfloat_really_inline FASTFLOAT_CONSTEXPR20 uint64_t
parse_eight_digits_unrolled(UC const *chars) noexcept {
if (cpp20_and_in_constexpr() || !has_simd_opt<UC>()) {
return parse_eight_digits_unrolled(read8_to_u64(chars)); // truncation okay
@ -169,16 +167,6 @@ simd_parse_if_eight_digits_unrolled(char16_t const *chars,
__m128i const data =
_mm_loadu_si128(reinterpret_cast<__m128i const *>(chars));
#ifdef FASTFLOAT_SSE42
// --- Digit range check using SSE4.2 comparisons ---
// Validate: '0' (0x0030) ≤ x ≤ '9' (0x0039)
__m128i const below0 = _mm_cmplt_epi16(data, _mm_set1_epi16(u'0')); // x < '0'
__m128i const above9 = _mm_cmpgt_epi16(data, _mm_set1_epi16(u'9')); // x > '9'
__m128i const invalid = _mm_or_si128(below0, above9);
// Check if any invalid byte exists
if (_mm_testz_si128(invalid, invalid)) { // SSE4.1/4.2: zero flag test
#else
// Branchless "are all digits?" trick from Lemire:
// (x - '0') <= 9 <=> (x + 32720) <= 32729
// encoded as signed comparison: (x + 32720) > -32759 ? not digit : digit
@ -188,7 +176,6 @@ simd_parse_if_eight_digits_unrolled(char16_t const *chars,
// If mask == 0 → all digits valid.
if (_mm_movemask_epi8(mask) == 0) {
#endif
i = i * 100000000 + parse_eight_digits_unrolled(simd_read8_to_u64(data));
return true;
}