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

v3.12.8

Browse Source
This commit is contained in:
Jay D Dee
2020-04-09 12:56:18 -04:00
parent fb9163185a
commit e96a6bd699
73 changed files with 462 additions and 625 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
simd-utils/simd-128.h
View File

@@ -273,6 +273,20 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, const int n )
#define mm128_ror_1x32( v ) _mm_shuffle_epi32( v, 0x39 )
#define mm128_rol_1x32( v ) _mm_shuffle_epi32( v, 0x93 )
//#define mm128_swap_64( v ) _mm_alignr_epi8( v, v, 8 )
//#define mm128_ror_1x32( v ) _mm_alignr_epi8( v, v, 4 )
//#define mm128_rol_1x32( v ) _mm_alignr_epi8( v, v, 12 )
#define mm128_ror_1x16( v ) _mm_alignr_epi8( v, v, 2 )
#define mm128_rol_1x16( v ) _mm_alignr_epi8( v, v, 14 )
#define mm128_ror_1x8( v ) _mm_alignr_epi8( v, v, 1 )
#define mm128_rol_1x8( v ) _mm_alignr_epi8( v, v, 15 )
#define mm128_ror_x8( v, c ) _mm_alignr_epi8( v, c )
#define mm128_rol_x8( v, c ) _mm_alignr_epi8( v, 16-(c) )
/*
// Rotate 16 byte (128 bit) vector by c bytes.
// Less efficient using shift but more versatile. Use only for odd number
// byte rotations. Use shuffle above whenever possible.
@@ -312,6 +326,8 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, const int n )
_mm_or_si128( _mm_slli_si128( v, 1 ), _mm_srli_si128( v, 15 ) )
#endif // SSE3 else SSE2
*/
// Invert vector: {3,2,1,0} -> {0,1,2,3}
#define mm128_invert_32( v ) _mm_shuffle_epi32( v, 0x1b )
@@ -331,7 +347,7 @@ static inline void memcpy_128( __m128i *dst, const __m128i *src, const int n )
//
// Rotate elements within lanes.
#define mm128_swap_64_32( v ) _mm_shuffle_epi32( v, 0xb1 )
#define mm128_swap64_32( v ) _mm_shuffle_epi32( v, 0xb1 )
#define mm128_rol64_8( v, c ) \
_mm_or_si128( _mm_slli_epi64( v, ( ( (c)<<3 ) ), \

9
simd-utils/simd-256.h
View File

@@ -442,16 +442,19 @@ static inline void memcpy_256( __m256i *dst, const __m256i *src, const int n )
#define mm256_ror128_32( v ) _mm256_shuffle_epi32( v, 0x39 )
#define mm256_rol128_1x32( v ) _mm256_shuffle_epi32( v, 0x93 )
#define mm256_rol128_32( v ) _mm256_shuffle_epi32( v, 0x93 )
// Rotave each 128 bit lane by c elements.
#define mm256_ror128_x8( v, c ) _mm256_alignr_epi8( v, v, c )
/*
// Rotate each 128 bit lane by c elements.
#define mm256_ror128_8( v, c ) \
_mm256_or_si256( _mm256_bsrli_epi128( v, c ), \
_mm256_bslli_epi128( v, 16-(c) ) )
#define mm256_rol128_8( v, c ) \
_mm256_or_si256( _mm256_bslli_epi128( v, c ), \
_mm256_bsrli_epi128( v, 16-(c) ) )
*/
// Rotate elements in each 64 bit lane

4
simd-utils/simd-512.h
View File

@@ -511,7 +511,9 @@ static inline void memcpy_512( __m512i *dst, const __m512i *src, const int n )
#define mm512_ror128_32( v ) _mm512_shuffle_epi32( v, 0x39 )
#define mm512_rol128_32( v ) _mm512_shuffle_epi32( v, 0x93 )
#define mm512_ror128_x8( v, c ) _mm512_alignr_epi8( v, v, c )
/*
// Rotate 128 bit lanes by c bytes, faster than building that monstrous
// constant above.
#define mm512_ror128_8( v, c ) \
@@ -520,7 +522,7 @@ static inline void memcpy_512( __m512i *dst, const __m512i *src, const int n )
#define mm512_rol128_8( v, c ) \
_mm512_or_si512( _mm512_bslli_epi128( v, c ), \
_mm512_bsrli_epi128( v, 16-(c) ) )
*/
//
// Rotate elements within 64 bit lanes.