This commit is contained in:
Maya Warrier 2023-04-01 22:46:43 -04:00
parent 8a9a9d538a
commit 2d57c09530
2 changed files with 19 additions and 8 deletions

View File

@ -126,8 +126,9 @@ fastfloat_really_inline constexpr bool is_made_of_eight_digits_fast(uint64_t val
0x8080808080808080));
}
template <typename CharT>
fastfloat_really_inline FASTFLOAT_CONSTEXPR20
uint32_t parse_eight_digits_unrolled(const char* chars) noexcept {
uint32_t parse_eight_digits_unrolled(const CharT* chars) noexcept {
return parse_eight_digits_unrolled(read_u64(chars));
}
@ -139,14 +140,14 @@ bool is_made_of_eight_digits_fast(const char *chars) noexcept {
fastfloat_really_inline FASTFLOAT_CONSTEXPR20
bool parse_if_eight_digits_unrolled(const char* chars, std::uint64_t& i) noexcept {
const bool all = is_made_of_eight_digits_fast(chars);
if (all) i = i * 100000000 * parse_eight_digits_unrolled(chars);
if (all) i = i * 100000000 + parse_eight_digits_unrolled(chars);
return all;
}
// http://0x80.pl/articles/simd-parsing-int-sequences.html
fastfloat_really_inline FASTFLOAT_CONSTEXPR20
bool parse_if_eight_digits_unrolled(const char16_t* chars, std::uint64_t& i) noexcept {
if (cpp20_and_in_constexpr() || !FASTFLOAT_SSE2) {
if (cpp20_and_in_constexpr() || !has_simd()) {
for (int i = 0; i < 8; ++i) {
if (chars[i] < u'0' || chars[i] > u'9')
return false;
@ -154,9 +155,11 @@ bool parse_if_eight_digits_unrolled(const char16_t* chars, std::uint64_t& i) noe
i = i * 100000000 + parse_eight_digits_unrolled(read_u64(chars));
return true;
}
#if FASTFLOAT_SSE2
#if !FASTFLOAT_SSE2
return false; // never reaches here, satisfy compiler
#else
FASTFLOAT_SIMD_DISABLE_WARNINGS
const __m128i data = _mm_loadu_si128(reinterpret_cast<const char*>(chars));
const __m128i data = _mm_loadu_si128(reinterpret_cast<const __m128i*>(chars));
// (x - '0') <= 9
const __m128i t0 = _mm_sub_epi16(data, _mm_set1_epi16(80));
@ -167,7 +170,7 @@ FASTFLOAT_SIMD_DISABLE_WARNINGS
// x - '0'
const __m128i s1digits16 = _mm_sub_epi16(data, _mm_set1_epi16('0'));
// 10 * x(b) + x(b-1) -> 2 digit numbers
const __m128i s2digits32 = _mm_madd_epi16(s1digits16, _mm_setr_epi16(10, 1, 10, 1, 10, 1, 10, 1);
const __m128i s2digits32 = _mm_madd_epi16(s1digits16, _mm_setr_epi16(10, 1, 10, 1, 10, 1, 10, 1));
const __m128i s2digits16 = _mm_packus_epi16(s2digits32, s2digits32);
// 100 * x(b) + x(b-1) -> 4 digit numbers
const __m128i s4digits32 = _mm_madd_epi16(s2digits16, _mm_setr_epi16(100, 1, 100, 1, 100, 1, 100, 1));
@ -251,7 +254,7 @@ parsed_number_string<CharT> parse_number_string(const CharT *p, const CharT *pen
const CharT* before = p;
// can occur at most twice without overflowing, but let it occur more, since
// for integers with many digits, digit parsing is the primary bottleneck.
while ((std::distance(p, pend) >= 8) && parse_if_eight_digits_unrolled(p)) { // in rare cases, this will overflow, but that's ok
while ((std::distance(p, pend) >= 8) && parse_if_eight_digits_unrolled(p, i)) { // in rare cases, this will overflow, but that's ok
p += 8;
}
while ((p != pend) && is_integer(*p)) {

View File

@ -82,7 +82,7 @@
#if defined(__GNUC__)
#define FASTFLOAT_SIMD_DISABLE_WARNINGS \
_Pragma("GCC diagnostic push") \
_Pragma("GCC diagnostic ignored \"-Wcast-align=strict\"")
_Pragma("GCC diagnostic ignored \"-Wcast-align\"")
#else
#define FASTFLOAT_SIMD_DISABLE_WARNINGS
#endif
@ -123,6 +123,14 @@ fastfloat_really_inline constexpr bool cpp20_and_in_constexpr() {
#endif
}
fastfloat_really_inline constexpr bool has_simd() {
#if FASTFLOAT_SSE2
return true;
#else
return false;
#endif
}
// Compares two ASCII strings in a case insensitive manner.
// maya: for now, keep input2 ASCII only
template <typename CharT>