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

v3.11.5

Browse Source
This commit is contained in:
Jay D Dee
2020-01-18 15:14:27 -05:00
parent 70089d1224
commit 1a7a573675
105 changed files with 3579 additions and 2833 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
simd-utils/intrlv.h
View File

@@ -2857,6 +2857,8 @@ static inline void rintrlv_2x256_8x64( void *dst, const void *src0,
//#define mm256_intrlv_blend_64( hi, lo ) _mm256_blend_epi32( hi, lo, 0x33 )
#define mm256_intrlv_blend_32( hi, lo ) _mm256_blend_epi32( hi, lo, 0x55 )
// change to _mm256_blend_epi32
//
// Select lanes of 32 byte hash from 2 sources according to control mask.
// macro due to 256 bit value arg.
#define mm256_blend_hash_4x64( dst, a, b, mask ) \

6
simd-utils/simd-512.h
View File

@@ -129,8 +129,8 @@ static inline __m512i m512_const_64( const uint64_t i7, const uint64_t i6,
}
// Equivalent of set1, broadcast 64 bit constant to all 64 bit elements.
#define m512_const1_256( i ) _mm512_broadcast_i64x4( i )
#define m512_const1_128( i ) _mm512_broadcast_i64x2( i )
#define m512_const1_256( v ) _mm512_broadcast_i64x4( v )
#define m512_const1_128( v ) _mm512_broadcast_i64x2( v )
#define m512_const1_64( i ) _mm512_broadcastq_epi64( mm128_mov64_128( i ) )
#define m512_const1_32( i ) _mm512_broadcastd_epi32( mm128_mov32_128( i ) )
#define m512_const1_16( i ) _mm512_broadcastw_epi16( mm128_mov32_128( i ) )
@@ -547,8 +547,6 @@ static inline void memcpy_512( __m512i *dst, const __m512i *src, const int n )
//
// Rotate elements from 2 512 bit vectors in place, source arguments
// are overwritten.
// These can all be done with 2 permutex2var instructions but they are
// slower than either xor or alignr and require AVX512VBMI.
#define mm512_swap1024_512(v1, v2) \
v1 = _mm512_xor_si512(v1, v2); \