mirrors/xmrig
1
0
Fork 1
mirror of https://github.com/xmrig/xmrig.git synced 2025-12-11 17:32:47 -05:00
Code Issues Projects Releases Wiki Activity

Optimize cn-heavy in GCC builds

Browse Source 
+0.7% in GCC builds, but GCC is still slower than MSVC on cn-heavy.
This commit is contained in:
SChernykh
2021-03-28 16:12:09 +02:00
parent c5c958743e
commit dc70893e6b
1 changed files with 11 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
src/crypto/cn/CryptoNight_x86.h
View File

@@ -740,7 +740,17 @@ inline void cryptonight_single_hash(const uint8_t *__restrict__ input, size_t si
if (props.isHeavy()) { if (props.isHeavy()) {
int64_t n = ((int64_t*)&l0[interleaved_index<interleave>(idx0 & MASK)])[0]; int64_t n = ((int64_t*)&l0[interleaved_index<interleave>(idx0 & MASK)])[0];
int64_t d = ((int32_t*)&l0[interleaved_index<interleave>(idx0 & MASK)])[2]; int64_t d = ((int32_t*)&l0[interleaved_index<interleave>(idx0 & MASK)])[2];
int64_t q = n / (d | 0x5);
int64_t d5;
# if defined _MSC_VER
d5 = d | 5;
# else
// Workaround for stupid GCC which converts to 32 bit before doing "| 5" and then converts back to 64 bit
asm("mov %1, %0\n\tor $5, %0" : "=r"(d5) : "r"(d));
# endif
int64_t q = n / d5;
((int64_t*)&l0[interleaved_index<interleave>(idx0 & MASK)])[0] = n ^ q; ((int64_t*)&l0[interleaved_index<interleave>(idx0 & MASK)])[0] = n ^ q;