coffee/fast_float
1
0
Fork 0
mirror of https://github.com/fastfloat/fast_float.git synced 2025-12-06 16:56:57 +08:00
Code Issues Packages Projects Releases Wiki Activity

Correcting the previous implementation's erorr and also improving loop_parse_if_eight_digits.

Browse Source
This commit is contained in:
RealTimeChris 2024-11-22 20:17:14 -05:00
parent 04034f79c6
commit dd484ae31e
1 changed files with 16 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
include/fast_float/ascii_number.h
View File

@ -118,11 +118,10 @@ uint64_t simd_read8_to_u64(UC const *) {
// credit @aqrit // credit @aqrit
fastfloat_really_inline FASTFLOAT_CONSTEXPR14 uint32_t fastfloat_really_inline FASTFLOAT_CONSTEXPR14 uint32_t
parse_eight_digits_unrolled(uint64_t val) { parse_eight_digits_unrolled(uint64_t val) {
const uint64_t mask = 0x000000FF000000FF; constexpr uint64_t mask = 0x000000FF000000FF;
const uint64_t mul1 = 0x000F424000000064; // 100 + (1000000ULL << 32) constexpr uint64_t mul1 = 0x000F424000000064; // 100 + (1000000ULL << 32)
const uint64_t mul2 = 0x0000271000000001; // 1 + (10000ULL << 32) constexpr uint64_t mul2 = 0x0000271000000001; // 1 + (10000ULL << 32)
val -= 0x3030303030303030; val = (val * 10) + (val >> 8); // val = (val * 2561) >> 8;
val = (val * 10) + (val >> 8); // val = (val * 2561) >> 8;
val = (((val & mask) * mul1) + (((val >> 16) & mask) * mul2)) >> 32; val = (((val & mask) * mul1) + (((val >> 16) & mask) * mul2)) >> 32;
return uint32_t(val); return uint32_t(val);
} }
@ -137,18 +136,13 @@ parse_eight_digits_unrolled(UC const *chars) noexcept {
return parse_eight_digits_unrolled(simd_read8_to_u64(chars)); return parse_eight_digits_unrolled(simd_read8_to_u64(chars));
} }
// credit @realtimechris // credit @realtimechris
fastfloat_really_inline constexpr bool fastfloat_really_inline constexpr bool
is_made_of_eight_digits_fast(uint64_t val) noexcept { is_made_of_eight_digits_fast(uint64_t val) noexcept {
constexpr uint64_t byte_mask = ~uint64_t(0) / 255ull; constexpr uint64_t byte_mask = ~uint64_t(0) / 255ull;
constexpr uint64_t msb_mask = byte_mask * 128ull; constexpr uint64_t msb_mask = byte_mask * 128ull;
constexpr uint64_t sub_mask = constexpr uint64_t threshold_byte_mask = byte_mask * (127ull - 10ull);
byte_mask * (127ull - 10ull) - 0x3030303030303030ull; return !((val + threshold_byte_mask | val) & msb_mask);
#if !defined(__clang__)
return !bool((val + sub_mask | val) & msb_mask);
#else
return ((val + sub_mask | val) & msb_mask) == 0;
#endif
} }
#ifdef FASTFLOAT_HAS_SIMD #ifdef FASTFLOAT_HAS_SIMD
@ -229,13 +223,15 @@ loop_parse_if_eight_digits(const UC *&p, const UC *const pend, uint64_t &i) {
fastfloat_really_inline FASTFLOAT_CONSTEXPR20 void fastfloat_really_inline FASTFLOAT_CONSTEXPR20 void
loop_parse_if_eight_digits(const char *&p, const char *const pend, loop_parse_if_eight_digits(const char *&p, const char *const pend,
uint64_t &i) { uint64_t &i) {
// optimizes better than parse_if_eight_digits_unrolled() for UC = char. if (pend - p >= 8) {
while ((std::distance(p, pend) >= 8) && uint64_t val{read8_to_u64(p) - 0x3030303030303030};
is_made_of_eight_digits_fast(read8_to_u64(p))) { if (is_made_of_eight_digits_fast(val)) {
i = i * 100000000 + do {
parse_eight_digits_unrolled(read8_to_u64( i = i * 100000000 + parse_eight_digits_unrolled(val);
p)); // in rare cases, this will overflow, but that's ok p += 8;
p += 8; val = read8_to_u64(p) - 0x3030303030303030;
} while ((pend - p) >= 8 && is_made_of_eight_digits_fast(val));
}
} }
} }