coffee/fast_float
1
0
Fork 0
mirror of https://github.com/fastfloat/fast_float.git synced 2026-02-07 10:19:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: readjust full_multiplication() and leading_zeroes() on windows

Browse Source
This commit is contained in:
Joao Paulo Magalhaes 2020-11-20 00:16:10 +00:00
parent 5ce64de524
commit f7b13da349
1 changed files with 16 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
include/fast_float/float_common.h
View File

@ -96,16 +96,17 @@ fastfloat_really_inline int leading_zeroes(uint64_t input_num) {
unsigned long leading_zero = 0; unsigned long leading_zero = 0;
// Search the mask data from most significant bit (MSB) // Search the mask data from most significant bit (MSB)
// to least significant bit (LSB) for a set bit (1). // to least significant bit (LSB) for a set bit (1).
return (int)(_BitScanReverse64(&leading_zero, input_num) - leading_zero); _BitScanReverse64(&leading_zero, input_num);
return (int)(63 - leading_zero);
#else #else
int n = 0; int last_bit = 0;
if(input_num & uint64_t(0xffffffff00000000)) input_num >>= 32, n |= 32; if(input_num & uint64_t(0xffffffff00000000)) input_num >>= 32, last_bit |= 32;
if(input_num & uint64_t( 0xffff0000)) input_num >>= 16, n |= 16; if(input_num & uint64_t( 0xffff0000)) input_num >>= 16, last_bit |= 16;
if(input_num & uint64_t( 0xff00)) input_num >>= 8, n |= 8; if(input_num & uint64_t( 0xff00)) input_num >>= 8, last_bit |= 8;
if(input_num & uint64_t( 0xf0)) input_num >>= 4, n |= 4; if(input_num & uint64_t( 0xf0)) input_num >>= 4, last_bit |= 4;
if(input_num & uint64_t( 0xc)) input_num >>= 2, n |= 2; if(input_num & uint64_t( 0xc)) input_num >>= 2, last_bit |= 2;
if(input_num & uint64_t( 0x2)) input_num >>= 1, n |= 1; if(input_num & uint64_t( 0x2)) input_num >>= 1, last_bit |= 1;
return 63 - n; return 63 - last_bit;
#endif #endif
#else #else
return __builtin_clzll(input_num); return __builtin_clzll(input_num);
@ -124,15 +125,18 @@ fastfloat_really_inline int leading_zeroes(uint64_t input_num) {
#endif #endif
#ifdef FASTFLOAT_32BIT #if ((defined(_WIN32) || defined(_WIN64)) && !defined(__clang__))
#if (defined(_WIN32) && !defined(__clang__))
#include <intrin.h> #include <intrin.h>
#endif #endif
#ifdef FASTFLOAT_32BIT
#if !defined(_WIN32)
// slow emulation routine for 32-bit // slow emulation routine for 32-bit
fastfloat_really_inline uint64_t __emulu(uint32_t x, uint32_t y) { fastfloat_really_inline uint64_t __emulu(uint32_t x, uint32_t y) {
return x * (uint64_t)y; return x * (uint64_t)y;
} }
#endif
// slow emulation routine for 32-bit // slow emulation routine for 32-bit
fastfloat_really_inline uint64_t _umul128(uint64_t ab, uint64_t cd, fastfloat_really_inline uint64_t _umul128(uint64_t ab, uint64_t cd,
@ -157,7 +161,7 @@ fastfloat_really_inline value128 full_multiplication(uint64_t a,
// ARM64 has native support for 64-bit multiplications, no need to emulate // ARM64 has native support for 64-bit multiplications, no need to emulate
answer.high = __umulh(a, b); answer.high = __umulh(a, b);
answer.low = a * b; answer.low = a * b;
#elif defined(FASTFLOAT_32BIT) #elif defined(FASTFLOAT_32BIT) || (defined(_WIN64))
answer.low = _umul128(a, b, &answer.high); // _umul128 not available on ARM64 answer.low = _umul128(a, b, &answer.high); // _umul128 not available on ARM64
#elif defined(FASTFLOAT_64BIT) #elif defined(FASTFLOAT_64BIT)
__uint128_t r = ((__uint128_t)a) * b; __uint128_t r = ((__uint128_t)a) * b;