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

v3.23.0

Browse Source
This commit is contained in:
Jay D Dee
2023-08-30 20:15:48 -04:00
parent 57a6b7b58b
commit 4378d2f841
72 changed files with 10184 additions and 2182 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
algo/simd/simd-hash-2way.c
View File

@@ -212,14 +212,24 @@ do { \
// targetted
#define shufxor2w(x,s) _mm256_shuffle_epi32( x, XCAT( SHUFXOR_, s ))
#if defined(__AVX512VL__)
//TODO Enable for AVX10_256
#define REDUCE(x) \
_mm256_sub_epi16( _mm256_and_si256( x, m256_const1_64( \
_mm256_sub_epi16( _mm256_maskz_mov_epi8( 0x55555555, x ), \
_mm256_srai_epi16( x, 8 ) )
#else
#define REDUCE(x) \
_mm256_sub_epi16( _mm256_and_si256( x, _mm256_set1_epi64x( \
0x00ff00ff00ff00ff ) ), _mm256_srai_epi16( x, 8 ) )
#endif
#define EXTRA_REDUCE_S(x)\
_mm256_sub_epi16( x, _mm256_and_si256( \
m256_const1_64( 0x0101010101010101 ), \
_mm256_cmpgt_epi16( x, m256_const1_64( 0x0080008000800080 ) ) ) )
_mm256_set1_epi64x( 0x0101010101010101 ), \
_mm256_cmpgt_epi16( x, _mm256_set1_epi64x( 0x0080008000800080 ) ) ) )
#define REDUCE_FULL_S( x ) EXTRA_REDUCE_S( REDUCE (x ) )
@@ -387,17 +397,11 @@ static const m512_v16 FFT256_Twiddle4w[] =
_mm512_sub_epi16( _mm512_maskz_mov_epi8( 0x5555555555555555, x ), \
_mm512_srai_epi16( x, 8 ) )
/*
#define REDUCE4w(x) \
_mm512_sub_epi16( _mm512_and_si512( x, m512_const1_64( \
0x00ff00ff00ff00ff ) ), _mm512_srai_epi16( x, 8 ) )
*/
#define EXTRA_REDUCE_S4w(x) \
_mm512_sub_epi16( x, _mm512_and_si512( \
m512_const1_64( 0x0101010101010101 ), \
_mm512_set1_epi64( 0x0101010101010101 ), \
_mm512_movm_epi16( _mm512_cmpgt_epi16_mask( \
x, m512_const1_64( 0x0080008000800080 ) ) ) ) )
x, _mm512_set1_epi64( 0x0080008000800080 ) ) ) ) )
// generic, except it calls targetted macros
#define REDUCE_FULL_S4w( x ) EXTRA_REDUCE_S4w( REDUCE4w (x ) )