popov/cpuminer-opt-gpu
1
0
Fork 0
mirror of https://github.com/JayDDee/cpuminer-opt.git synced 2025-09-17 23:44:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

v3.19.2

Browse Source
This commit is contained in:
Jay D Dee
2021-12-30 16:28:24 -05:00
parent 7d2ef7973d
commit 0e3945ddb5
8 changed files with 155 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
algo/hamsi/hamsi-hash-4way.c
View File

@@ -550,16 +550,38 @@ static const sph_u32 T512[64][16] = {
// Hamsi 8 way AVX512
// Tested on i9-9940x movepi64_mask is slow, cmple_epi64_mask with zero
// produces the same result but is faster.
#define INPUT_BIG8 \
do { \
__m512i db = *buf; \
const uint64_t *tp = (uint64_t*)&T512[0][0]; \
__m512i db = _mm512_ror_epi64( *buf, 1 ); \
const uint64_t *tp = (const uint64_t*)T512; \
m0 = m1 = m2 = m3 = m4 = m5 = m6 = m7 = m512_zero; \
for ( int u = 0; u < 64; u++ ) \
{ \
__m512i dm = _mm512_and_si512( db, m512_one_64 ) ; \
dm = mm512_negate_32( _mm512_or_si512( dm, \
_mm512_slli_epi64( dm, 32 ) ) ); \
__mmask8 dm = _mm512_cmplt_epi64_mask( db, m512_zero ); \
m0 = _mm512_mask_xor_epi64( m0, dm, m0, m512_const1_64( tp[0] ) ); \
m1 = _mm512_mask_xor_epi64( m1, dm, m1, m512_const1_64( tp[1] ) ); \
m2 = _mm512_mask_xor_epi64( m2, dm, m2, m512_const1_64( tp[2] ) ); \
m3 = _mm512_mask_xor_epi64( m3, dm, m3, m512_const1_64( tp[3] ) ); \
m4 = _mm512_mask_xor_epi64( m4, dm, m4, m512_const1_64( tp[4] ) ); \
m5 = _mm512_mask_xor_epi64( m5, dm, m5, m512_const1_64( tp[5] ) ); \
m6 = _mm512_mask_xor_epi64( m6, dm, m6, m512_const1_64( tp[6] ) ); \
m7 = _mm512_mask_xor_epi64( m7, dm, m7, m512_const1_64( tp[7] ) ); \
db = _mm512_ror_epi64( db, 1 ); \
tp += 8; \
} \
} while (0)
/*
#define INPUT_BIG8 \
do { \
__m512i db = *buf; \
const uint64_t *tp = (const uint64_t*)T512; \
m0 = m1 = m2 = m3 = m4 = m5 = m6 = m7 = m512_zero; \
for ( int u = 0; u < 64; u++ ) \
{ \
__m512i dm = mm512_negate_64( _mm512_and_si512( db, m512_one_64 ) ); \
m0 = mm512_xorand( m0, dm, m512_const1_64( tp[0] ) ); \
m1 = mm512_xorand( m1, dm, m512_const1_64( tp[1] ) ); \
m2 = mm512_xorand( m2, dm, m512_const1_64( tp[2] ) ); \
@@ -572,6 +594,7 @@ do { \
db = _mm512_srli_epi64( db, 1 ); \
} \
} while (0)
*/
#define SBOX8( a, b, c, d ) \
do { \
@@ -888,13 +911,11 @@ void hamsi512_8way_close( hamsi_8way_big_context *sc, void *dst )
#define INPUT_BIG \
do { \
__m256i db = *buf; \
const uint64_t *tp = (uint64_t*)&T512[0][0]; \
const uint64_t *tp = (const uint64_t*)T512; \
m0 = m1 = m2 = m3 = m4 = m5 = m6 = m7 = m256_zero; \
for ( int u = 0; u < 64; u++ ) \
{ \
__m256i dm = _mm256_and_si256( db, m256_one_64 ) ; \
dm = mm256_negate_32( _mm256_or_si256( dm, \
_mm256_slli_epi64( dm, 32 ) ) ); \
__m256i dm = mm256_negate_64( _mm256_and_si256( db, m256_one_64 ) ); \
m0 = _mm256_xor_si256( m0, _mm256_and_si256( dm, \
m256_const1_64( tp[0] ) ) ); \
m1 = _mm256_xor_si256( m1, _mm256_and_si256( dm, \