popov/cpuminer-opt-gpu
1
0
Fork 0
mirror of https://github.com/JayDDee/cpuminer-opt.git synced 2026-02-22 16:33:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

v25.7

Browse Source
This commit is contained in:
Jay D Dee
2025-11-15 10:44:32 -05:00
parent 12480a3ea5
commit 8f2f9ec3e9
26 changed files with 474 additions and 452 deletions
Download Patch File Download Diff File Expand all files Collapse all files

48
simd-utils/simd-neon.h
View File

@@ -126,7 +126,7 @@
#define v128_andnot( v1, v0 ) vbicq_u32( v0, v1 )
// ~( v1 ^ v0 ), same as (~v1) ^ v0
#define v128_xnor( v1, v0 ) v128_not( v128_xor( v1, v0 ) )
#define v128_nxor( v1, v0 ) v128_not( v128_xor( v1, v0 ) )
// ~v1 | v0, args reversed for consistency with x86_64
#define v128_ornot( v1, v0 ) vornq_u32( v0, v1 )
@@ -349,52 +349,6 @@ static inline void v128_memcpy( void *dst, const void *src, const int n )
vqtbl1q_u8( (uint8x16_t)(v), (uint8x16_t)(vmask) )
// Bit rotation
/*
#define v128_shuflr64_8( v ) v128_shuffle8( v, V128_SHUFLR64_8 )
#define v128_shufll64_8( v ) v128_shuffle8( v, V128_SHUFLL64_8 )
#define v128_shuflr64_16(v ) v128_shuffle8( v, V128_SHUFLR64_16 )
#define v128_shufll64_16(v ) v128_shuffle8( v, V128_SHUFLL64_16 )
#define v128_shuflr64_24(v ) v128_shuffle8( v, V128_SHUFLR64_24 )
#define v128_shufll64_24(v ) v128_shuffle8( v, V128_SHUFLL64_24 )
#define v128_shuflr32_8( v ) v128_shuffle8( v, V128_SHUFLR32_8 )
#define v128_shufll32_8( v ) v128_shuffle8( v, V128_SHUFLL32_8 )
#define v128_ror64( v, c ) \
( (c) == 8 ) ? v128_shuflr64_8( v ) \
: ( (c) == 16 ) ? v128_shuflr64_16( v ) \
: ( (c) == 24 ) ? v128_shuflr64_24( v ) \
: ( (c) == 32 ) ? (uint64x2_t)vrev64q_u32( ((uint32x4_t)(v)) ) \
: ( (c) == 40 ) ? v128_shufll64_24( v ) \
: ( (c) == 48 ) ? v128_shufll64_16( v ) \
: ( (c) == 56 ) ? v128_shufll64_8( v ) \
: vsriq_n_u64( vshlq_n_u64( ((uint64x2_t)(v)), 64-(c) ), \
((uint64x2_t)(v)), c )
#define v128_rol64( v, c ) \
( (c) == 8 ) ? v128_shufll64_8( v ) \
: ( (c) == 16 ) ? v128_shufll64_16( v ) \
: ( (c) == 24 ) ? v128_shufll64_24( v ) \
: ( (c) == 32 ) ? (uint64x2_t)vrev64q_u32( ((uint32x4_t)(v)) ) \
: ( (c) == 40 ) ? v128_shuflr64_24( v ) \
: ( (c) == 48 ) ? v128_shuflr64_16( v ) \
: ( (c) == 56 ) ? v128_shuflr64_8( v ) \
: vsliq_n_u64( vshrq_n_u64( ((uint64x2_t)(v)), 64-(c) ), \
((uint64x2_t)(v)), c )
#define v128_ror32( v, c ) \
( (c) == 8 ) ? v128_shuflr32_8( v ) \
: ( (c) == 16 ) ? (uint32x4_t)vrev32q_u16( ((uint16x8_t)(v)) ) \
: ( (c) == 24 ) ? v128_shufll32_8( v ) \
: vsriq_n_u32( vshlq_n_u32( ((uint32x4_t)(v)), 32-(c) ), \
((uint32x4_t)(v)), c )
#define v128_rol32( v, c ) \
( (c) == 8 ) ? v128_shufll32_8( v ) \
: ( (c) == 16 ) ? (uint32x4_t)vrev32q_u16( ((uint16x8_t)(v)) ) \
: ( (c) == 24 ) ? v128_shuflr32_8( v ) \
: vsliq_n_u32( vshrq_n_u32( ((uint32x4_t)(v)), 32-(c) ), \
((uint32x4_t)(v)), c )
*/
#define v128_ror64( v, c ) \
( (c) == 32 ) ? (uint64x2_t)vrev64q_u32( ((uint32x4_t)(v)) ) \