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

v3.23.4

Browse Source
This commit is contained in:
Jay D Dee
2023-10-06 22:18:09 -04:00
parent bc5a5c6df8
commit 31c4dedf59
144 changed files with 5931 additions and 3746 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
algo/argon2/argon2a/ar2/opt.c
View File

@@ -17,6 +17,8 @@
#include <stdio.h>
#include <inttypes.h>
#if defined(__SSE2__)
#include <immintrin.h>
#include "argon2.h"
@@ -183,3 +185,5 @@ void ar2_fill_segment(const argon2_instance_t *instance,
free(pseudo_rands);
}
#endif

4
algo/argon2/argon2d/argon2d-gate.c
View File

@@ -114,7 +114,7 @@ int scanhash_argon2d_dyn( struct work *work, uint32_t max_nonce,
uint32_t nonce = first_nonce;
const bool bench = opt_benchmark;
mm128_bswap32_80( edata, pdata );
v128_bswap32_80( edata, pdata );
do
{
edata[19] = nonce;
@@ -160,7 +160,7 @@ int scanhash_argon2d4096( struct work *work, uint32_t max_nonce,
uint32_t parallelism = 1; // 1 thread, 2 lanes
const bool bench = opt_benchmark;
mm128_bswap32_80( edata, pdata );
v128_bswap32_80( edata, pdata );
do {
edata[19] = n;

26
algo/argon2/argon2d/argon2d/opt.c
View File

@@ -131,22 +131,22 @@ static void fill_block(__m256i *state, const block *ref_block,
#else // SSE2
static void fill_block(__m128i *state, const block *ref_block,
static void fill_block( v128_t *state, const block *ref_block,
block *next_block, int with_xor) {
__m128i block_XY[ARGON2_OWORDS_IN_BLOCK];
v128_t block_XY[ARGON2_OWORDS_IN_BLOCK];
unsigned int i;
if (with_xor) {
for (i = 0; i < ARGON2_OWORDS_IN_BLOCK; i++) {
state[i] = _mm_xor_si128(
state[i], _mm_load_si128((const __m128i *)ref_block->v + i));
block_XY[i] = _mm_xor_si128(
state[i], _mm_load_si128((const __m128i *)next_block->v + i));
state[i] = v128_xor(
state[i], v128_load((const v128_t *)ref_block->v + i));
block_XY[i] = v128_xor(
state[i], v128_load((const v128_t *)next_block->v + i));
}
} else {
for (i = 0; i < ARGON2_OWORDS_IN_BLOCK; i++) {
block_XY[i] = state[i] = _mm_xor_si128(
state[i], _mm_load_si128((const __m128i *)ref_block->v + i));
block_XY[i] = state[i] = v128_xor(
state[i], v128_load((const v128_t *)ref_block->v + i));
}
}
@@ -185,8 +185,8 @@ static void fill_block(__m128i *state, const block *ref_block,
state[39], state[47], state[55], state[63] );
for (i = 0; i < ARGON2_OWORDS_IN_BLOCK; i++) {
state[i] = _mm_xor_si128(state[i], block_XY[i]);
_mm_store_si128((__m128i *)next_block->v + i, state[i]);
state[i] = v128_xor(state[i], block_XY[i]);
v128_store((v128_t *)next_block->v + i, state[i]);
}
}
@@ -202,8 +202,8 @@ static void next_addresses(block *address_block, block *input_block) {
__m256i zero_block[ARGON2_HWORDS_IN_BLOCK];
__m256i zero2_block[ARGON2_HWORDS_IN_BLOCK];
#else
__m128i zero_block[ARGON2_OWORDS_IN_BLOCK];
__m128i zero2_block[ARGON2_OWORDS_IN_BLOCK];
v128_t zero_block[ARGON2_OWORDS_IN_BLOCK];
v128_t zero2_block[ARGON2_OWORDS_IN_BLOCK];
#endif
memset(zero_block, 0, sizeof(zero_block));
@@ -232,7 +232,7 @@ void fill_segment(const argon2_instance_t *instance,
#elif defined(__AVX2__)
__m256i state[ARGON2_HWORDS_IN_BLOCK];
#else
__m128i state[ARGON2_OWORDS_IN_BLOCK];
v128_t state[ARGON2_OWORDS_IN_BLOCK];
#endif
// int data_independent_addressing;

101
algo/argon2/argon2d/blake2/blamka-round-opt.h
View File

@@ -19,16 +19,6 @@
#define BLAKE_ROUND_MKA_OPT_H
#include "blake2-impl.h"
#include <emmintrin.h>
#if defined(__SSSE3__)
#include <tmmintrin.h> /* for _mm_shuffle_epi8 and _mm_alignr_epi8 */
#endif
#if defined(__XOP__) && (defined(__GNUC__) || defined(__clang__))
#include <x86intrin.h>
#endif
#include "simd-utils.h"
#if !defined(__AVX512F__)
@@ -39,7 +29,7 @@
(_mm_setr_epi8(2, 3, 4, 5, 6, 7, 0, 1, 10, 11, 12, 13, 14, 15, 8, 9))
#define r24 \
(_mm_setr_epi8(3, 4, 5, 6, 7, 0, 1, 2, 11, 12, 13, 14, 15, 8, 9, 10))
#define _mm_roti_epi64(x, c) \
#define v128_ror64(x, c) \
(-(c) == 32) \
? _mm_shuffle_epi32((x), _MM_SHUFFLE(2, 3, 0, 1)) \
: (-(c) == 24) \
@@ -47,20 +37,20 @@
: (-(c) == 16) \
? _mm_shuffle_epi8((x), r16) \
: (-(c) == 63) \
? _mm_xor_si128(_mm_srli_epi64((x), -(c)), \
_mm_add_epi64((x), (x))) \
: _mm_xor_si128(_mm_srli_epi64((x), -(c)), \
_mm_slli_epi64((x), 64 - (-(c))))
? v128_xor(v128_sr64((x), -(c)), \
v128_add64((x), (x))) \
: v128_xor(v128_sr64((x), -(c)), \
v128_sl64((x), 64 - (-(c))))
#else /* defined(__SSE2__) */
#define _mm_roti_epi64(r, c) \
_mm_xor_si128(_mm_srli_epi64((r), -(c)), _mm_slli_epi64((r), 64 - (-(c))))
#define v128_ror64(r, c) \
v128_xor(v128_sr64((r), -(c)), v128_sl64((r), 64 - (-(c))))
#endif
#else
#endif
static BLAKE2_INLINE __m128i fBlaMka(__m128i x, __m128i y) {
const __m128i z = _mm_mul_epu32(x, y);
return _mm_add_epi64(_mm_add_epi64(x, y), _mm_add_epi64(z, z));
static BLAKE2_INLINE v128_t fBlaMka(v128_t x, v128_t y) {
const v128_t z = v128_mul32(x, y);
return v128_add64(v128_add64(x, y), v128_add64(z, z));
}
#define G1(A0, B0, C0, D0, A1, B1, C1, D1) \
@@ -68,20 +58,20 @@ static BLAKE2_INLINE __m128i fBlaMka(__m128i x, __m128i y) {
A0 = fBlaMka(A0, B0); \
A1 = fBlaMka(A1, B1); \
\
D0 = _mm_xor_si128(D0, A0); \
D1 = _mm_xor_si128(D1, A1); \
D0 = v128_xor(D0, A0); \
D1 = v128_xor(D1, A1); \
\
D0 = _mm_roti_epi64(D0, -32); \
D1 = _mm_roti_epi64(D1, -32); \
D0 = v128_ror64(D0, -32); \
D1 = v128_ror64(D1, -32); \
\
C0 = fBlaMka(C0, D0); \
C1 = fBlaMka(C1, D1); \
\
B0 = _mm_xor_si128(B0, C0); \
B1 = _mm_xor_si128(B1, C1); \
B0 = v128_xor(B0, C0); \
B1 = v128_xor(B1, C1); \
\
B0 = _mm_roti_epi64(B0, -24); \
B1 = _mm_roti_epi64(B1, -24); \
B0 = v128_ror64(B0, -24); \
B1 = v128_ror64(B1, -24); \
} while ((void)0, 0)
#define G2(A0, B0, C0, D0, A1, B1, C1, D1) \
@@ -89,27 +79,27 @@ static BLAKE2_INLINE __m128i fBlaMka(__m128i x, __m128i y) {
A0 = fBlaMka(A0, B0); \
A1 = fBlaMka(A1, B1); \
\
D0 = _mm_xor_si128(D0, A0); \
D1 = _mm_xor_si128(D1, A1); \
D0 = v128_xor(D0, A0); \
D1 = v128_xor(D1, A1); \
\
D0 = _mm_roti_epi64(D0, -16); \
D1 = _mm_roti_epi64(D1, -16); \
D0 = v128_ror64(D0, -16); \
D1 = v128_ror64(D1, -16); \
\
C0 = fBlaMka(C0, D0); \
C1 = fBlaMka(C1, D1); \
\
B0 = _mm_xor_si128(B0, C0); \
B1 = _mm_xor_si128(B1, C1); \
B0 = v128_xor(B0, C0); \
B1 = v128_xor(B1, C1); \
\
B0 = _mm_roti_epi64(B0, -63); \
B1 = _mm_roti_epi64(B1, -63); \
B0 = v128_ror64(B0, -63); \
B1 = v128_ror64(B1, -63); \
} while ((void)0, 0)
#if defined(__SSSE3__)
#define DIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1) \
do { \
__m128i t0 = _mm_alignr_epi8(B1, B0, 8); \
__m128i t1 = _mm_alignr_epi8(B0, B1, 8); \
v128_t t0 = v128_alignr8(B1, B0, 8); \
v128_t t1 = v128_alignr8(B0, B1, 8); \
B0 = t0; \
B1 = t1; \
\
@@ -117,16 +107,16 @@ static BLAKE2_INLINE __m128i fBlaMka(__m128i x, __m128i y) {
C0 = C1; \
C1 = t0; \
\
t0 = _mm_alignr_epi8(D1, D0, 8); \
t1 = _mm_alignr_epi8(D0, D1, 8); \
t0 = v128_alignr8(D1, D0, 8); \
t1 = v128_alignr8(D0, D1, 8); \
D0 = t1; \
D1 = t0; \
} while ((void)0, 0)
#define UNDIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1) \
do { \
__m128i t0 = _mm_alignr_epi8(B0, B1, 8); \
__m128i t1 = _mm_alignr_epi8(B1, B0, 8); \
v128_t t0 = v128_alignr8(B0, B1, 8); \
v128_t t1 = v128_alignr8(B1, B0, 8); \
B0 = t0; \
B1 = t1; \
\
@@ -134,37 +124,37 @@ static BLAKE2_INLINE __m128i fBlaMka(__m128i x, __m128i y) {
C0 = C1; \
C1 = t0; \
\
t0 = _mm_alignr_epi8(D0, D1, 8); \
t1 = _mm_alignr_epi8(D1, D0, 8); \
t0 = v128_alignr8(D0, D1, 8); \
t1 = v128_alignr8(D1, D0, 8); \
D0 = t1; \
D1 = t0; \
} while ((void)0, 0)
#else /* SSE2 */
#define DIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1) \
do { \
__m128i t0 = D0; \
__m128i t1 = B0; \
v128_t t0 = D0; \
v128_t t1 = B0; \
D0 = C0; \
C0 = C1; \
C1 = D0; \
D0 = _mm_unpackhi_epi64(D1, _mm_unpacklo_epi64(t0, t0)); \
D1 = _mm_unpackhi_epi64(t0, _mm_unpacklo_epi64(D1, D1)); \
B0 = _mm_unpackhi_epi64(B0, _mm_unpacklo_epi64(B1, B1)); \
B1 = _mm_unpackhi_epi64(B1, _mm_unpacklo_epi64(t1, t1)); \
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 UNDIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1) \
do { \
__m128i t0, t1; \
v128_t t0, t1; \
t0 = C0; \
C0 = C1; \
C1 = t0; \
t0 = B0; \
t1 = D0; \
B0 = _mm_unpackhi_epi64(B1, _mm_unpacklo_epi64(B0, B0)); \
B1 = _mm_unpackhi_epi64(t0, _mm_unpacklo_epi64(B1, B1)); \
D0 = _mm_unpackhi_epi64(D0, _mm_unpacklo_epi64(D1, D1)); \
D1 = _mm_unpackhi_epi64(D1, _mm_unpacklo_epi64(t1, t1)); \
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
@@ -462,4 +452,5 @@ static inline __m512i muladd(__m512i x, __m512i y)
} while ((void)0, 0)
#endif /* __AVX512F__ */
#endif /* BLAKE_ROUND_MKA_OPT_H */