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

252
algo/argon2d/blake2/blamka-round-opt.h
View File

@@ -66,82 +66,60 @@ static BLAKE2_INLINE v128_t fBlaMka(v128_t x, v128_t y)
#if defined(__SSSE3__) || defined(__ARM_NEON)
#define DIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1) \
do { \
v128_t t0 = v128_alignr8(B1, B0, 8); \
v128_t t1 = v128_alignr8(B0, B1, 8); \
B0 = t0; \
B1 = t1; \
\
t0 = C0; \
C0 = C1; \
C1 = t0; \
\
t0 = v128_alignr8(D1, D0, 8); \
t1 = v128_alignr8(D0, D1, 8); \
D0 = t1; \
D1 = t0; \
} while ((void)0, 0)
#define DIAGONALIZE( A0, B0, C0, D0, A1, B1, C1, D1 ) \
{ \
v128_t t = v128_alignr8( B1, B0, 8 ); \
B1 = v128_alignr8( B0, B1, 8 ); \
B0 = t; \
t = v128_alignr8( D1, D0, 8 ); \
D0 = v128_alignr8( D0, D1, 8 ); \
D1 = t; \
}
#define UNDIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1) \
do { \
v128_t t0 = v128_alignr8(B0, B1, 8); \
v128_t t1 = v128_alignr8(B1, B0, 8); \
B0 = t0; \
B1 = t1; \
\
t0 = C0; \
C0 = C1; \
C1 = t0; \
\
t0 = v128_alignr8(D0, D1, 8); \
t1 = v128_alignr8(D1, D0, 8); \
D0 = t1; \
D1 = t0; \
} while ((void)0, 0)
#define UNDIAGONALIZE( A0, B0, C0, D0, A1, B1, C1, D1 ) \
{ \
v128_t t = v128_alignr8( B0, B1, 8 ); \
B1 = v128_alignr8( B1, B0, 8 ); \
B0 = t; \
t = v128_alignr8( D0, D1, 8 ); \
D0 = v128_alignr8( D1, D0, 8 ); \
D1 = t; \
}
#else /* SSE2 */
#define DIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1) \
do { \
v128_t t0 = D0; \
v128_t t1 = B0; \
D0 = C0; \
C0 = C1; \
C1 = D0; \
D0 = v128_unpackhi64(D1, v128_unpacklo64(t0, t0)); \
D1 = v128_unpackhi64(t0, v128_unpacklo64(D1, D1)); \
B0 = v128_unpackhi64(B0, v128_unpacklo64(B1, B1)); \
B1 = v128_unpackhi64(B1, v128_unpacklo64(t1, t1)); \
} while ((void)0, 0)
#define DIAGONALIZE( A0, B0, C0, D0, A1, B1, C1, D1 ) \
{ \
v128_t t = D0; \
D0 = v128_unpackhi64( D1, v128_unpacklo64( D0, D0 ) ); \
D1 = v128_unpackhi64( t, v128_unpacklo64( D1, D1 ) ); \
t = B0; \
B0 = v128_unpackhi64( B0, v128_unpacklo64( B1, B1 ) ); \
B1 = v128_unpackhi64( B1, v128_unpacklo64( t, t ) ); \
}
#define UNDIAGONALIZE( A0, B0, C0, D0, A1, B1, C1, D1 ) \
{ \
v128_t t = B0; \
B0 = v128_unpackhi64( B1, v128_unpacklo64( B0, B0 ) ); \
B1 = v128_unpackhi64( t, v128_unpacklo64( B1, B1 ) ); \
t = D0; \
D0 = v128_unpackhi64( D0, v128_unpacklo64( D1, D1 ) ); \
D1 = v128_unpackhi64( D1, v128_unpacklo64( t, t ) ); \
}
#define UNDIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1) \
do { \
v128_t t0, t1; \
t0 = C0; \
C0 = C1; \
C1 = t0; \
t0 = B0; \
t1 = D0; \
B0 = v128_unpackhi64(B1, v128_unpacklo64(B0, B0)); \
B1 = v128_unpackhi64(t0, v128_unpacklo64(B1, B1)); \
D0 = v128_unpackhi64(D0, v128_unpacklo64(D1, D1)); \
D1 = v128_unpackhi64(D1, v128_unpacklo64(t1, t1)); \
} while ((void)0, 0)
#endif
#define BLAKE2_ROUND(A0, A1, B0, B1, C0, C1, D0, D1) \
do { \
G1(A0, B0, C0, D0, A1, B1, C1, D1); \
G2(A0, B0, C0, D0, A1, B1, C1, D1); \
\
DIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1); \
\
G1(A0, B0, C0, D0, A1, B1, C1, D1); \
G2(A0, B0, C0, D0, A1, B1, C1, D1); \
\
UNDIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1); \
} while ((void)0, 0)
#define BLAKE2_ROUND( A0, A1, B0, B1, C0, C1, D0, D1 ) \
{ \
G1( A0, B0, C0, D0, A1, B1, C1, D1 ); \
G2( A0, B0, C0, D0, A1, B1, C1, D1 ); \
DIAGONALIZE( A0, B0, C0, D0, A1, B1, C1, D1 ); \
G1( A0, B0, C1, D0, A1, B1, C0, D1 ); \
G2( A0, B0, C1, D0, A1, B1, C0, D1 ); \
UNDIAGONALIZE( A0, B0, C0, D0, A1, B1, C1, D1 ); \
}
#else /* __AVX2__ */
#include <immintrin.h>
@@ -211,7 +189,6 @@ static BLAKE2_INLINE v128_t fBlaMka(v128_t x, v128_t y)
B0 = _mm256_permute4x64_epi64(B0, _MM_SHUFFLE(0, 3, 2, 1)); \
C0 = _mm256_permute4x64_epi64(C0, _MM_SHUFFLE(1, 0, 3, 2)); \
D0 = _mm256_permute4x64_epi64(D0, _MM_SHUFFLE(2, 1, 0, 3)); \
\
B1 = _mm256_permute4x64_epi64(B1, _MM_SHUFFLE(0, 3, 2, 1)); \
C1 = _mm256_permute4x64_epi64(C1, _MM_SHUFFLE(1, 0, 3, 2)); \
D1 = _mm256_permute4x64_epi64(D1, _MM_SHUFFLE(2, 1, 0, 3)); \
@@ -219,17 +196,14 @@ static BLAKE2_INLINE v128_t fBlaMka(v128_t x, v128_t y)
#define DIAGONALIZE_2(A0, A1, B0, B1, C0, C1, D0, D1) \
do { \
__m256i tmp1 = _mm256_blend_epi32(B0, B1, 0xCC); \
__m256i tmp2 = _mm256_blend_epi32(B0, B1, 0x33); \
B1 = _mm256_permute4x64_epi64(tmp1, _MM_SHUFFLE(2,3,0,1)); \
tmp1 = C0; \
B0 = _mm256_permute4x64_epi64(tmp2, _MM_SHUFFLE(2,3,0,1)); \
C0 = C1; \
tmp2 = _mm256_blend_epi32(D0, D1, 0x33); \
C1 = tmp1; \
__m256i tmp1 = _mm256_blend_epi32(B0, B1, 0x33); \
__m256i tmp2 = _mm256_blend_epi32(B0, B1, 0xCC); \
B0 = _mm256_shuffle_epi32( tmp1, 0x4e ); \
B1 = _mm256_shuffle_epi32( tmp2, 0x4e ); \
tmp1 = _mm256_blend_epi32(D0, D1, 0xCC); \
D1 = _mm256_permute4x64_epi64(tmp2, _MM_SHUFFLE(2,3,0,1)); \
D0 = _mm256_permute4x64_epi64(tmp1, _MM_SHUFFLE(2,3,0,1)); \
tmp2 = _mm256_blend_epi32(D0, D1, 0x33); \
D0 = _mm256_shuffle_epi32( tmp1, 0x4e ); \
D1 = _mm256_shuffle_epi32( tmp2, 0x4e ); \
} while(0);
#define UNDIAGONALIZE_1(A0, B0, C0, D0, A1, B1, C1, D1) \
@@ -237,7 +211,6 @@ static BLAKE2_INLINE v128_t fBlaMka(v128_t x, v128_t y)
B0 = _mm256_permute4x64_epi64(B0, _MM_SHUFFLE(2, 1, 0, 3)); \
C0 = _mm256_permute4x64_epi64(C0, _MM_SHUFFLE(1, 0, 3, 2)); \
D0 = _mm256_permute4x64_epi64(D0, _MM_SHUFFLE(0, 3, 2, 1)); \
\
B1 = _mm256_permute4x64_epi64(B1, _MM_SHUFFLE(2, 1, 0, 3)); \
C1 = _mm256_permute4x64_epi64(C1, _MM_SHUFFLE(1, 0, 3, 2)); \
D1 = _mm256_permute4x64_epi64(D1, _MM_SHUFFLE(0, 3, 2, 1)); \
@@ -247,27 +220,21 @@ static BLAKE2_INLINE v128_t fBlaMka(v128_t x, v128_t y)
do { \
__m256i tmp1 = _mm256_blend_epi32(B0, B1, 0xCC); \
__m256i tmp2 = _mm256_blend_epi32(B0, B1, 0x33); \
B0 = _mm256_permute4x64_epi64(tmp1, _MM_SHUFFLE(2,3,0,1)); \
tmp1 = C0; \
B1 = _mm256_permute4x64_epi64(tmp2, _MM_SHUFFLE(2,3,0,1)); \
C0 = C1; \
B0 = _mm256_shuffle_epi32( tmp1, 0x4e ); \
B1 = _mm256_shuffle_epi32( tmp2, 0x4e ); \
tmp2 = _mm256_blend_epi32(D0, D1, 0xCC); \
C1 = tmp1; \
tmp1 = _mm256_blend_epi32(D0, D1, 0x33); \
D1 = _mm256_permute4x64_epi64(tmp2, _MM_SHUFFLE(2,3,0,1)); \
D0 = _mm256_permute4x64_epi64(tmp1, _MM_SHUFFLE(2,3,0,1)); \
D1 = _mm256_shuffle_epi32( tmp2, 0x4e ); \
D0 = _mm256_shuffle_epi32( tmp1, 0x4e ); \
} while((void)0, 0);
#define BLAKE2_ROUND_1(A0, A1, B0, B1, C0, C1, D0, D1) \
do{ \
G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
\
DIAGONALIZE_1(A0, B0, C0, D0, A1, B1, C1, D1) \
\
G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
\
UNDIAGONALIZE_1(A0, B0, C0, D0, A1, B1, C1, D1) \
} while((void)0, 0);
@@ -275,12 +242,9 @@ static BLAKE2_INLINE v128_t fBlaMka(v128_t x, v128_t y)
do{ \
G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
\
DIAGONALIZE_2(A0, A1, B0, B1, C0, C1, D0, D1) \
\
G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
\
G1_AVX2(A0, A1, B0, B1, C1, C0, D0, D1) \
G2_AVX2(A0, A1, B0, B1, C1, C0, D0, D1) \
UNDIAGONALIZE_2(A0, A1, B0, B1, C0, C1, D0, D1) \
} while((void)0, 0);
@@ -290,12 +254,73 @@ static BLAKE2_INLINE v128_t fBlaMka(v128_t x, v128_t y)
#include <immintrin.h>
/*
static inline __m512i muladd(__m512i x, __m512i y)
{
__m512i z = _mm512_mul_epu32(x, y);
return _mm512_add_epi64(_mm512_add_epi64(x, y), _mm512_add_epi64(z, z));
}
*/
#define G1( A0, B0, C0, D0, A1, B1, C1, D1 ) \
{ \
__m512i z0, z1; \
z0 = _mm512_mul_epu32( A0, B0 ); \
z1 = _mm512_mul_epu32( A1, B1 ); \
A0 = _mm512_add_epi64( A0, B0 ); \
A1 = _mm512_add_epi64( A1, B1 ); \
z0 = _mm512_add_epi64( z0, z0 ); \
z1 = _mm512_add_epi64( z1, z1 ); \
A0 = _mm512_add_epi64( A0, z0 ); \
A1 = _mm512_add_epi64( A1, z1 ); \
D0 = _mm512_xor_si512(D0, A0); \
D1 = _mm512_xor_si512(D1, A1); \
D0 = _mm512_ror_epi64(D0, 32); \
D1 = _mm512_ror_epi64(D1, 32); \
z0 = _mm512_mul_epu32( C0, D0 ); \
z1 = _mm512_mul_epu32( C1, D1 ); \
C0 = _mm512_add_epi64( C0, D0 ); \
C1 = _mm512_add_epi64( C1, D1 ); \
z0 = _mm512_add_epi64( z0, z0 ); \
z1 = _mm512_add_epi64( z1, z1 ); \
C0 = _mm512_add_epi64( C0, z0 ); \
C1 = _mm512_add_epi64( C1, z1 ); \
B0 = _mm512_xor_si512(B0, C0); \
B1 = _mm512_xor_si512(B1, C1); \
B0 = _mm512_ror_epi64(B0, 24); \
B1 = _mm512_ror_epi64(B1, 24); \
}
#define G2( A0, B0, C0, D0, A1, B1, C1, D1 ) \
{ \
__m512i z0, z1; \
z0 = _mm512_mul_epu32( A0, B0 ); \
z1 = _mm512_mul_epu32( A1, B1 ); \
A0 = _mm512_add_epi64( A0, B0 ); \
A1 = _mm512_add_epi64( A1, B1 ); \
z0 = _mm512_add_epi64( z0, z0 ); \
z1 = _mm512_add_epi64( z1, z1 ); \
A0 = _mm512_add_epi64( A0, z0 ); \
A1 = _mm512_add_epi64( A1, z1 ); \
D0 = _mm512_xor_si512(D0, A0); \
D1 = _mm512_xor_si512(D1, A1); \
D0 = _mm512_ror_epi64(D0, 16); \
D1 = _mm512_ror_epi64(D1, 16); \
z0 = _mm512_mul_epu32( C0, D0 ); \
z1 = _mm512_mul_epu32( C1, D1 ); \
C0 = _mm512_add_epi64( C0, D0 ); \
C1 = _mm512_add_epi64( C1, D1 ); \
z0 = _mm512_add_epi64( z0, z0 ); \
z1 = _mm512_add_epi64( z1, z1 ); \
C0 = _mm512_add_epi64( C0, z0 ); \
C1 = _mm512_add_epi64( C1, z1 ); \
B0 = _mm512_xor_si512(B0, C0); \
B1 = _mm512_xor_si512(B1, C1); \
B0 = _mm512_ror_epi64(B0, 63); \
B1 = _mm512_ror_epi64(B1, 63); \
}
/*
#define G1(A0, B0, C0, D0, A1, B1, C1, D1) \
do { \
A0 = muladd(A0, B0); \
@@ -316,7 +341,8 @@ static inline __m512i muladd(__m512i x, __m512i y)
B0 = _mm512_ror_epi64(B0, 24); \
B1 = _mm512_ror_epi64(B1, 24); \
} while ((void)0, 0)
*/
/*
#define G2(A0, B0, C0, D0, A1, B1, C1, D1) \
do { \
A0 = muladd(A0, B0); \
@@ -337,15 +363,14 @@ static inline __m512i muladd(__m512i x, __m512i y)
B0 = _mm512_ror_epi64(B0, 63); \
B1 = _mm512_ror_epi64(B1, 63); \
} while ((void)0, 0)
*/
#define DIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1) \
do { \
B0 = _mm512_permutex_epi64(B0, _MM_SHUFFLE(0, 3, 2, 1)); \
B1 = _mm512_permutex_epi64(B1, _MM_SHUFFLE(0, 3, 2, 1)); \
\
C0 = _mm512_permutex_epi64(C0, _MM_SHUFFLE(1, 0, 3, 2)); \
C1 = _mm512_permutex_epi64(C1, _MM_SHUFFLE(1, 0, 3, 2)); \
\
D0 = _mm512_permutex_epi64(D0, _MM_SHUFFLE(2, 1, 0, 3)); \
D1 = _mm512_permutex_epi64(D1, _MM_SHUFFLE(2, 1, 0, 3)); \
} while ((void)0, 0)
@@ -354,10 +379,8 @@ static inline __m512i muladd(__m512i x, __m512i y)
do { \
B0 = _mm512_permutex_epi64(B0, _MM_SHUFFLE(2, 1, 0, 3)); \
B1 = _mm512_permutex_epi64(B1, _MM_SHUFFLE(2, 1, 0, 3)); \
\
C0 = _mm512_permutex_epi64(C0, _MM_SHUFFLE(1, 0, 3, 2)); \
C1 = _mm512_permutex_epi64(C1, _MM_SHUFFLE(1, 0, 3, 2)); \
\
D0 = _mm512_permutex_epi64(D0, _MM_SHUFFLE(0, 3, 2, 1)); \
D1 = _mm512_permutex_epi64(D1, _MM_SHUFFLE(0, 3, 2, 1)); \
} while ((void)0, 0)
@@ -366,15 +389,17 @@ static inline __m512i muladd(__m512i x, __m512i y)
do { \
G1(A0, B0, C0, D0, A1, B1, C1, D1); \
G2(A0, B0, C0, D0, A1, B1, C1, D1); \
\
DIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1); \
\
G1(A0, B0, C0, D0, A1, B1, C1, D1); \
G2(A0, B0, C0, D0, A1, B1, C1, D1); \
\
UNDIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1); \
} while ((void)0, 0)
static const __m512i swap_q0 = { 0,1, 8,9, 2,3, 10,11 };
static const __m512i swap_q1 = { 4,5, 12,13, 6,7, 14,15 };
static const __m512i uswap_q0 = { 0,1, 4,5, 8,9, 12,13 };
static const __m512i uswap_q1 = { 2,3, 6,7, 10,11, 14,15 };
#define SWAP_HALVES(A0, A1) \
do { \
__m512i t; \
@@ -383,19 +408,36 @@ static inline __m512i muladd(__m512i x, __m512i y)
A0 = t; \
} while((void)0, 0)
#define SWAP_QUARTERS(A0, A1) \
{ \
__m512i t = _mm512_permutex2var_epi64( A0, swap_q0, A1 ); \
A1 = _mm512_permutex2var_epi64( A0, swap_q1, A1 ); \
A0 = t; \
}
#define UNSWAP_QUARTERS(A0, A1) \
{ \
__m512i t = _mm512_permutex2var_epi64( A0, uswap_q0, A1 ); \
A1 = _mm512_permutex2var_epi64( A0, uswap_q1, A1 ); \
A0 = t; \
}
/*
#define SWAP_QUARTERS(A0, A1) \
do { \
SWAP_HALVES(A0, A1); \
A0 = _mm512_shuffle_i64x2( A0, A0, 0xd8 ); \
A1 = _mm512_shuffle_i64x2( A1, A1, 0xd8 ); \
} while((void)0, 0)
*/
/*
#define UNSWAP_QUARTERS(A0, A1) \
do { \
A0 = _mm512_shuffle_i64x2( A0, A0, 0xd8 ); \
A1 = _mm512_shuffle_i64x2( A1, A1, 0xd8 ); \
SWAP_HALVES(A0, A1); \
} while((void)0, 0)
*/
#define BLAKE2_ROUND_1(A0, C0, B0, D0, A1, C1, B1, D1) \
do { \