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

v3.9.5.3

Browse Source
This commit is contained in:
Jay D Dee
2019-07-12 10:42:38 -04:00
parent 9abc19a30a
commit e625ed5420
31 changed files with 1269 additions and 1188 deletions
Download Patch File Download Diff File Expand all files Collapse all files

234
algo/blake/blake256-hash-4way.c
View File

@@ -412,34 +412,16 @@ do { \
V5 = H5; \
V6 = H6; \
V7 = H7; \
V8 = _mm_xor_si128( S0, _mm_set_epi32( CS0, CS0, CS0, CS0 ) ); \
V9 = _mm_xor_si128( S1, _mm_set_epi32( CS1, CS1, CS1, CS1 ) ); \
VA = _mm_xor_si128( S2, _mm_set_epi32( CS2, CS2, CS2, CS2 ) ); \
VB = _mm_xor_si128( S3, _mm_set_epi32( CS3, CS3, CS3, CS3 ) ); \
VC = _mm_xor_si128( _mm_set_epi32( T0, T0, T0, T0 ), \
_mm_set_epi32( CS4, CS4, CS4, CS4 ) ); \
VD = _mm_xor_si128( _mm_set_epi32( T0, T0, T0, T0 ), \
_mm_set_epi32( CS5, CS5, CS5, CS5 ) ); \
VE = _mm_xor_si128( _mm_set_epi32( T1, T1, T1, T1 ) \
, _mm_set_epi32( CS6, CS6, CS6, CS6 ) ); \
VF = _mm_xor_si128( _mm_set_epi32( T1, T1, T1, T1 ), \
_mm_set_epi32( CS7, CS7, CS7, CS7 ) ); \
M[0x0] = mm128_bswap_32( *(buf + 0) ); \
M[0x1] = mm128_bswap_32( *(buf + 1) ); \
M[0x2] = mm128_bswap_32( *(buf + 2) ); \
M[0x3] = mm128_bswap_32( *(buf + 3) ); \
M[0x4] = mm128_bswap_32( *(buf + 4) ); \
M[0x5] = mm128_bswap_32( *(buf + 5) ); \
M[0x6] = mm128_bswap_32( *(buf + 6) ); \
M[0x7] = mm128_bswap_32( *(buf + 7) ); \
M[0x8] = mm128_bswap_32( *(buf + 8) ); \
M[0x9] = mm128_bswap_32( *(buf + 9) ); \
M[0xA] = mm128_bswap_32( *(buf + 10) ); \
M[0xB] = mm128_bswap_32( *(buf + 11) ); \
M[0xC] = mm128_bswap_32( *(buf + 12) ); \
M[0xD] = mm128_bswap_32( *(buf + 13) ); \
M[0xE] = mm128_bswap_32( *(buf + 14) ); \
M[0xF] = mm128_bswap_32( *(buf + 15) ); \
V8 = _mm_xor_si128( S0, _mm_set1_epi32( CS0 ) ); \
V9 = _mm_xor_si128( S1, _mm_set1_epi32( CS1 ) ); \
VA = _mm_xor_si128( S2, _mm_set1_epi32( CS2 ) ); \
VB = _mm_xor_si128( S3, _mm_set1_epi32( CS3 ) ); \
VC = _mm_xor_si128( _mm_set1_epi32( T0 ), _mm_set1_epi32( CS4 ) ); \
VD = _mm_xor_si128( _mm_set1_epi32( T0 ), _mm_set1_epi32( CS5 ) ); \
VE = _mm_xor_si128( _mm_set1_epi32( T1 ), _mm_set1_epi32( CS6 ) ); \
VF = _mm_xor_si128( _mm_set1_epi32( T1 ), _mm_set1_epi32( CS7 ) ); \
mm128_block_bswap_32( M, buf ); \
mm128_block_bswap_32( M+8, buf+8 ); \
for (r = 0; r < rounds; r ++) \
ROUND_S_4WAY(r); \
H0 = _mm_xor_si128( _mm_xor_si128( \
@@ -464,6 +446,54 @@ do { \
// current impl
#if defined(__SSSE3__)
#define BLAKE256_4WAY_BLOCK_BSWAP32 do \
{ \
__m128i shuf_bswap32 = _mm_set_epi64x( 0x0c0d0e0f08090a0b, \
0x0405060700010203 ); \
M0 = _mm_shuffle_epi8( buf[ 0], shuf_bswap32 ); \
M1 = _mm_shuffle_epi8( buf[ 1], shuf_bswap32 ); \
M2 = _mm_shuffle_epi8( buf[ 2], shuf_bswap32 ); \
M3 = _mm_shuffle_epi8( buf[ 3], shuf_bswap32 ); \
M4 = _mm_shuffle_epi8( buf[ 4], shuf_bswap32 ); \
M5 = _mm_shuffle_epi8( buf[ 5], shuf_bswap32 ); \
M6 = _mm_shuffle_epi8( buf[ 6], shuf_bswap32 ); \
M7 = _mm_shuffle_epi8( buf[ 7], shuf_bswap32 ); \
M8 = _mm_shuffle_epi8( buf[ 8], shuf_bswap32 ); \
M9 = _mm_shuffle_epi8( buf[ 9], shuf_bswap32 ); \
MA = _mm_shuffle_epi8( buf[10], shuf_bswap32 ); \
MB = _mm_shuffle_epi8( buf[11], shuf_bswap32 ); \
MC = _mm_shuffle_epi8( buf[12], shuf_bswap32 ); \
MD = _mm_shuffle_epi8( buf[13], shuf_bswap32 ); \
ME = _mm_shuffle_epi8( buf[14], shuf_bswap32 ); \
MF = _mm_shuffle_epi8( buf[15], shuf_bswap32 ); \
} while(0)
#else // SSE2
#define BLAKE256_4WAY_BLOCK_BSWAP32 do \
{ \
M0 = mm128_bswap_32( buf[0] ); \
M1 = mm128_bswap_32( buf[1] ); \
M2 = mm128_bswap_32( buf[2] ); \
M3 = mm128_bswap_32( buf[3] ); \
M4 = mm128_bswap_32( buf[4] ); \
M5 = mm128_bswap_32( buf[5] ); \
M6 = mm128_bswap_32( buf[6] ); \
M7 = mm128_bswap_32( buf[7] ); \
M8 = mm128_bswap_32( buf[8] ); \
M9 = mm128_bswap_32( buf[9] ); \
MA = mm128_bswap_32( buf[10] ); \
MB = mm128_bswap_32( buf[11] ); \
MC = mm128_bswap_32( buf[12] ); \
MD = mm128_bswap_32( buf[13] ); \
ME = mm128_bswap_32( buf[14] ); \
MF = mm128_bswap_32( buf[15] ); \
} while(0)
#endif // SSSE3 else SSE2
#define COMPRESS32_4WAY( rounds ) \
do { \
__m128i M0, M1, M2, M3, M4, M5, M6, M7; \
@@ -486,22 +516,7 @@ do { \
VD = _mm_xor_si128( _mm_set1_epi32( T0 ), _mm_set1_epi32( CS5 ) ); \
VE = _mm_xor_si128( _mm_set1_epi32( T1 ), _mm_set1_epi32( CS6 ) ); \
VF = _mm_xor_si128( _mm_set1_epi32( T1 ), _mm_set1_epi32( CS7 ) ); \
M0 = mm128_bswap_32( buf[ 0] ); \
M1 = mm128_bswap_32( buf[ 1] ); \
M2 = mm128_bswap_32( buf[ 2] ); \
M3 = mm128_bswap_32( buf[ 3] ); \
M4 = mm128_bswap_32( buf[ 4] ); \
M5 = mm128_bswap_32( buf[ 5] ); \
M6 = mm128_bswap_32( buf[ 6] ); \
M7 = mm128_bswap_32( buf[ 7] ); \
M8 = mm128_bswap_32( buf[ 8] ); \
M9 = mm128_bswap_32( buf[ 9] ); \
MA = mm128_bswap_32( buf[10] ); \
MB = mm128_bswap_32( buf[11] ); \
MC = mm128_bswap_32( buf[12] ); \
MD = mm128_bswap_32( buf[13] ); \
ME = mm128_bswap_32( buf[14] ); \
MF = mm128_bswap_32( buf[15] ); \
BLAKE256_4WAY_BLOCK_BSWAP32; \
ROUND_S_4WAY(0); \
ROUND_S_4WAY(1); \
ROUND_S_4WAY(2); \
@@ -519,14 +534,14 @@ do { \
ROUND_S_4WAY(2); \
ROUND_S_4WAY(3); \
} \
H0 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( V8, V0 ), S0 ), H0 ); \
H1 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( V9, V1 ), S1 ), H1 ); \
H2 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VA, V2 ), S2 ), H2 ); \
H3 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VB, V3 ), S3 ), H3 ); \
H4 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VC, V4 ), S0 ), H4 ); \
H5 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VD, V5 ), S1 ), H5 ); \
H6 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VE, V6 ), S2 ), H6 ); \
H7 = _mm_xor_si128( _mm_xor_si128( _mm_xor_si128( VF, V7 ), S3 ), H7 ); \
H0 = mm128_xor4( V8, V0, S0, H0 ); \
H1 = mm128_xor4( V9, V1, S1, H1 ); \
H2 = mm128_xor4( VA, V2, S2, H2 ); \
H3 = mm128_xor4( VB, V3, S3, H3 ); \
H4 = mm128_xor4( VC, V4, S0, H4 ); \
H5 = mm128_xor4( VD, V5, S1, H5 ); \
H6 = mm128_xor4( VE, V6, S2, H6 ); \
H7 = mm128_xor4( VF, V7, S3, H7 ); \
} while (0)
#endif
@@ -607,6 +622,7 @@ do { \
__m256i M8, M9, MA, MB, MC, MD, ME, MF; \
__m256i V0, V1, V2, V3, V4, V5, V6, V7; \
__m256i V8, V9, VA, VB, VC, VD, VE, VF; \
__m256i shuf_bswap32; \
V0 = H0; \
V1 = H1; \
V2 = H2; \
@@ -623,22 +639,24 @@ do { \
VD = _mm256_xor_si256( _mm256_set1_epi32( T0 ), _mm256_set1_epi32( CS5 ) ); \
VE = _mm256_xor_si256( _mm256_set1_epi32( T1 ), _mm256_set1_epi32( CS6 ) ); \
VF = _mm256_xor_si256( _mm256_set1_epi32( T1 ), _mm256_set1_epi32( CS7 ) ); \
M0 = mm256_bswap_32( * buf ); \
M1 = mm256_bswap_32( *(buf+1) ); \
M2 = mm256_bswap_32( *(buf+2) ); \
M3 = mm256_bswap_32( *(buf+3) ); \
M4 = mm256_bswap_32( *(buf+4) ); \
M5 = mm256_bswap_32( *(buf+5) ); \
M6 = mm256_bswap_32( *(buf+6) ); \
M7 = mm256_bswap_32( *(buf+7) ); \
M8 = mm256_bswap_32( *(buf+8) ); \
M9 = mm256_bswap_32( *(buf+9) ); \
MA = mm256_bswap_32( *(buf+10) ); \
MB = mm256_bswap_32( *(buf+11) ); \
MC = mm256_bswap_32( *(buf+12) ); \
MD = mm256_bswap_32( *(buf+13) ); \
ME = mm256_bswap_32( *(buf+14) ); \
MF = mm256_bswap_32( *(buf+15) ); \
shuf_bswap32 = _mm256_set_epi64x( 0x0c0d0e0f08090a0b, 0x0405060700010203, \
0x0c0d0e0f08090a0b, 0x0405060700010203 ); \
M0 = _mm256_shuffle_epi8( * buf , shuf_bswap32 ); \
M1 = _mm256_shuffle_epi8( *(buf+ 1), shuf_bswap32 ); \
M2 = _mm256_shuffle_epi8( *(buf+ 2), shuf_bswap32 ); \
M3 = _mm256_shuffle_epi8( *(buf+ 3), shuf_bswap32 ); \
M4 = _mm256_shuffle_epi8( *(buf+ 4), shuf_bswap32 ); \
M5 = _mm256_shuffle_epi8( *(buf+ 5), shuf_bswap32 ); \
M6 = _mm256_shuffle_epi8( *(buf+ 6), shuf_bswap32 ); \
M7 = _mm256_shuffle_epi8( *(buf+ 7), shuf_bswap32 ); \
M8 = _mm256_shuffle_epi8( *(buf+ 8), shuf_bswap32 ); \
M9 = _mm256_shuffle_epi8( *(buf+ 9), shuf_bswap32 ); \
MA = _mm256_shuffle_epi8( *(buf+10), shuf_bswap32 ); \
MB = _mm256_shuffle_epi8( *(buf+11), shuf_bswap32 ); \
MC = _mm256_shuffle_epi8( *(buf+12), shuf_bswap32 ); \
MD = _mm256_shuffle_epi8( *(buf+13), shuf_bswap32 ); \
ME = _mm256_shuffle_epi8( *(buf+14), shuf_bswap32 ); \
MF = _mm256_shuffle_epi8( *(buf+15), shuf_bswap32 ); \
ROUND_S_8WAY(0); \
ROUND_S_8WAY(1); \
ROUND_S_8WAY(2); \
@@ -656,22 +674,14 @@ do { \
ROUND_S_8WAY(2); \
ROUND_S_8WAY(3); \
} \
H0 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( V8, V0 ), \
S0 ), H0 ); \
H1 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( V9, V1 ), \
S1 ), H1 ); \
H2 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VA, V2 ), \
S2 ), H2 ); \
H3 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VB, V3 ), \
S3 ), H3 ); \
H4 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VC, V4 ), \
S0 ), H4 ); \
H5 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VD, V5 ), \
S1 ), H5 ); \
H6 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VE, V6 ), \
S2 ), H6 ); \
H7 = _mm256_xor_si256( _mm256_xor_si256( _mm256_xor_si256( VF, V7 ), \
S3 ), H7 ); \
H0 = mm256_xor4( V8, V0, S0, H0 ); \
H1 = mm256_xor4( V9, V1, S1, H1 ); \
H2 = mm256_xor4( VA, V2, S2, H2 ); \
H3 = mm256_xor4( VB, V3, S3, H3 ); \
H4 = mm256_xor4( VC, V4, S0, H4 ); \
H5 = mm256_xor4( VD, V5, S1, H5 ); \
H6 = mm256_xor4( VE, V6, S2, H6 ); \
H7 = mm256_xor4( VF, V7, S3, H7 ); \
} while (0)
@@ -685,6 +695,7 @@ static void
blake32_4way_init( blake_4way_small_context *ctx, const uint32_t *iv,
const uint32_t *salt, int rounds )
{
__m128i zero = m128_zero;
casti_m128i( ctx->H, 0 ) = _mm_set1_epi32( iv[0] );
casti_m128i( ctx->H, 1 ) = _mm_set1_epi32( iv[1] );
casti_m128i( ctx->H, 2 ) = _mm_set1_epi32( iv[2] );
@@ -694,16 +705,10 @@ blake32_4way_init( blake_4way_small_context *ctx, const uint32_t *iv,
casti_m128i( ctx->H, 6 ) = _mm_set1_epi32( iv[6] );
casti_m128i( ctx->H, 7 ) = _mm_set1_epi32( iv[7] );
casti_m128i( ctx->S, 0 ) = m128_zero;
casti_m128i( ctx->S, 1 ) = m128_zero;
casti_m128i( ctx->S, 2 ) = m128_zero;
casti_m128i( ctx->S, 3 ) = m128_zero;
/*
sc->S[0] = _mm_set1_epi32( salt[0] );
sc->S[1] = _mm_set1_epi32( salt[1] );
sc->S[2] = _mm_set1_epi32( salt[2] );
sc->S[3] = _mm_set1_epi32( salt[3] );
*/
casti_m128i( ctx->S, 0 ) = zero;
casti_m128i( ctx->S, 1 ) = zero;
casti_m128i( ctx->S, 2 ) = zero;
casti_m128i( ctx->S, 3 ) = zero;
ctx->T0 = ctx->T1 = 0;
ctx->ptr = 0;
ctx->rounds = rounds;
@@ -796,14 +801,7 @@ blake32_4way_close( blake_4way_small_context *ctx, unsigned ub, unsigned n,
blake32_4way( ctx, buf, 64 );
}
casti_m128i( dst, 0 ) = mm128_bswap_32( casti_m128i( ctx->H, 0 ) );
casti_m128i( dst, 1 ) = mm128_bswap_32( casti_m128i( ctx->H, 1 ) );
casti_m128i( dst, 2 ) = mm128_bswap_32( casti_m128i( ctx->H, 2 ) );
casti_m128i( dst, 3 ) = mm128_bswap_32( casti_m128i( ctx->H, 3 ) );
casti_m128i( dst, 4 ) = mm128_bswap_32( casti_m128i( ctx->H, 4 ) );
casti_m128i( dst, 5 ) = mm128_bswap_32( casti_m128i( ctx->H, 5 ) );
casti_m128i( dst, 6 ) = mm128_bswap_32( casti_m128i( ctx->H, 6 ) );
casti_m128i( dst, 7 ) = mm128_bswap_32( casti_m128i( ctx->H, 7 ) );
mm128_block_bswap_32( (__m128i*)dst, (__m128i*)ctx->H );
}
#if defined (__AVX2__)
@@ -816,11 +814,21 @@ static void
blake32_8way_init( blake_8way_small_context *sc, const sph_u32 *iv,
const sph_u32 *salt, int rounds )
{
int i;
for ( i = 0; i < 8; i++ )
sc->H[i] = _mm256_set1_epi32( iv[i] );
for ( i = 0; i < 4; i++ )
sc->S[i] = _mm256_set1_epi32( salt[i] );
__m256i zero = m256_zero;
casti_m256i( sc->H, 0 ) = _mm256_set1_epi32( iv[0] );
casti_m256i( sc->H, 1 ) = _mm256_set1_epi32( iv[1] );
casti_m256i( sc->H, 2 ) = _mm256_set1_epi32( iv[2] );
casti_m256i( sc->H, 3 ) = _mm256_set1_epi32( iv[3] );
casti_m256i( sc->H, 4 ) = _mm256_set1_epi32( iv[4] );
casti_m256i( sc->H, 5 ) = _mm256_set1_epi32( iv[5] );
casti_m256i( sc->H, 6 ) = _mm256_set1_epi32( iv[6] );
casti_m256i( sc->H, 7 ) = _mm256_set1_epi32( iv[7] );
casti_m256i( sc->S, 0 ) = zero;
casti_m256i( sc->S, 1 ) = zero;
casti_m256i( sc->S, 2 ) = zero;
casti_m256i( sc->S, 3 ) = zero;
sc->T0 = sc->T1 = 0;
sc->ptr = 0;
sc->rounds = rounds;
@@ -872,14 +880,10 @@ static void
blake32_8way_close( blake_8way_small_context *sc, unsigned ub, unsigned n,
void *dst, size_t out_size_w32 )
{
// union {
__m256i buf[16];
// sph_u32 dummy;
// } u;
size_t ptr, k;
__m256i buf[16];
size_t ptr;
unsigned bit_len;
sph_u32 th, tl;
__m256i *out;
ptr = sc->ptr;
bit_len = ((unsigned)ptr << 3);
@@ -923,9 +927,7 @@ blake32_8way_close( blake_8way_small_context *sc, unsigned ub, unsigned n,
*(buf+(60>>2)) = mm256_bswap_32( _mm256_set1_epi32( tl ) );
blake32_8way( sc, buf, 64 );
}
out = (__m256i*)dst;
for ( k = 0; k < out_size_w32; k++ )
out[k] = mm256_bswap_32( sc->H[k] );
mm256_block_bswap_32( (__m256i*)dst, (__m256i*)sc->H );
}
#endif

204
algo/blake/blake512-hash-4way.c
View File

@@ -412,18 +412,18 @@ static const sph_u64 CB[16] = {
V5 = H5; \
V6 = H6; \
V7 = H7; \
V8 = _mm256_xor_si256( S0, _mm256_set_epi64x( CB0, CB0, CB0, CB0 ) ); \
V9 = _mm256_xor_si256( S1, _mm256_set_epi64x( CB1, CB1, CB1, CB1 ) ); \
VA = _mm256_xor_si256( S2, _mm256_set_epi64x( CB2, CB2, CB2, CB2 ) ); \
VB = _mm256_xor_si256( S3, _mm256_set_epi64x( CB3, CB3, CB3, CB3 ) ); \
VC = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
_mm256_set_epi64x( CB4, CB4, CB4, CB4 ) ); \
VD = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
_mm256_set_epi64x( CB5, CB5, CB5, CB5 ) ); \
VE = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
_mm256_set_epi64x( CB6, CB6, CB6, CB6 ) ); \
VF = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
_mm256_set_epi64x( CB7, CB7, CB7, CB7 ) ); \
V8 = _mm256_xor_si256( S0, _mm256_set_epi64x( CB0, CB0, CB0, CB0 ) ); \
V9 = _mm256_xor_si256( S1, _mm256_set_epi64x( CB1, CB1, CB1, CB1 ) ); \
VA = _mm256_xor_si256( S2, _mm256_set_epi64x( CB2, CB2, CB2, CB2 ) ); \
VB = _mm256_xor_si256( S3, _mm256_set_epi64x( CB3, CB3, CB3, CB3 ) ); \
VC = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
_mm256_set_epi64x( CB4, CB4, CB4, CB4 ) ); \
VD = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
_mm256_set_epi64x( CB5, CB5, CB5, CB5 ) ); \
VE = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
_mm256_set_epi64x( CB6, CB6, CB6, CB6 ) ); \
VF = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
_mm256_set_epi64x( CB7, CB7, CB7, CB7 ) ); \
M[0x0] = mm256_bswap_64( *(buf+0) ); \
M[0x1] = mm256_bswap_64( *(buf+1) ); \
M[0x2] = mm256_bswap_64( *(buf+2) ); \
@@ -464,80 +464,76 @@ static const sph_u64 CB[16] = {
//current impl
#define COMPRESS64_4WAY do { \
__m256i M0, M1, M2, M3, M4, M5, M6, M7; \
__m256i M8, M9, MA, MB, MC, MD, ME, MF; \
__m256i V0, V1, V2, V3, V4, V5, V6, V7; \
__m256i V8, V9, VA, VB, VC, VD, VE, VF; \
V0 = H0; \
V1 = H1; \
V2 = H2; \
V3 = H3; \
V4 = H4; \
V5 = H5; \
V6 = H6; \
V7 = H7; \
V8 = _mm256_xor_si256( S0, _mm256_set_epi64x( CB0, CB0, CB0, CB0 ) ); \
V9 = _mm256_xor_si256( S1, _mm256_set_epi64x( CB1, CB1, CB1, CB1 ) ); \
VA = _mm256_xor_si256( S2, _mm256_set_epi64x( CB2, CB2, CB2, CB2 ) ); \
VB = _mm256_xor_si256( S3, _mm256_set_epi64x( CB3, CB3, CB3, CB3 ) ); \
VC = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
_mm256_set_epi64x( CB4, CB4, CB4, CB4 ) ); \
VD = _mm256_xor_si256( _mm256_set_epi64x( T0, T0, T0, T0 ), \
_mm256_set_epi64x( CB5, CB5, CB5, CB5 ) ); \
VE = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
_mm256_set_epi64x( CB6, CB6, CB6, CB6 ) ); \
VF = _mm256_xor_si256( _mm256_set_epi64x( T1, T1, T1, T1 ), \
_mm256_set_epi64x( CB7, CB7, CB7, CB7 ) ); \
M0 = mm256_bswap_64( *(buf + 0) ); \
M1 = mm256_bswap_64( *(buf + 1) ); \
M2 = mm256_bswap_64( *(buf + 2) ); \
M3 = mm256_bswap_64( *(buf + 3) ); \
M4 = mm256_bswap_64( *(buf + 4) ); \
M5 = mm256_bswap_64( *(buf + 5) ); \
M6 = mm256_bswap_64( *(buf + 6) ); \
M7 = mm256_bswap_64( *(buf + 7) ); \
M8 = mm256_bswap_64( *(buf + 8) ); \
M9 = mm256_bswap_64( *(buf + 9) ); \
MA = mm256_bswap_64( *(buf + 10) ); \
MB = mm256_bswap_64( *(buf + 11) ); \
MC = mm256_bswap_64( *(buf + 12) ); \
MD = mm256_bswap_64( *(buf + 13) ); \
ME = mm256_bswap_64( *(buf + 14) ); \
MF = mm256_bswap_64( *(buf + 15) ); \
ROUND_B_4WAY(0); \
ROUND_B_4WAY(1); \
ROUND_B_4WAY(2); \
ROUND_B_4WAY(3); \
ROUND_B_4WAY(4); \
ROUND_B_4WAY(5); \
ROUND_B_4WAY(6); \
ROUND_B_4WAY(7); \
ROUND_B_4WAY(8); \
ROUND_B_4WAY(9); \
ROUND_B_4WAY(0); \
ROUND_B_4WAY(1); \
ROUND_B_4WAY(2); \
ROUND_B_4WAY(3); \
ROUND_B_4WAY(4); \
ROUND_B_4WAY(5); \
H0 = _mm256_xor_si256( _mm256_xor_si256( \
_mm256_xor_si256( S0, V0 ), V8 ), H0 ); \
H1 = _mm256_xor_si256( _mm256_xor_si256( \
_mm256_xor_si256( S1, V1 ), V9 ), H1 ); \
H2 = _mm256_xor_si256( _mm256_xor_si256( \
_mm256_xor_si256( S2, V2 ), VA ), H2 ); \
H3 = _mm256_xor_si256( _mm256_xor_si256( \
_mm256_xor_si256( S3, V3 ), VB ), H3 ); \
H4 = _mm256_xor_si256( _mm256_xor_si256( \
_mm256_xor_si256( S0, V4 ), VC ), H4 ); \
H5 = _mm256_xor_si256( _mm256_xor_si256( \
_mm256_xor_si256( S1, V5 ), VD ), H5 ); \
H6 = _mm256_xor_si256( _mm256_xor_si256( \
_mm256_xor_si256( S2, V6 ), VE ), H6 ); \
H7 = _mm256_xor_si256( _mm256_xor_si256( \
_mm256_xor_si256( S3, V7 ), VF ), H7 ); \
} while (0)
#define COMPRESS64_4WAY do \
{ \
__m256i M0, M1, M2, M3, M4, M5, M6, M7; \
__m256i M8, M9, MA, MB, MC, MD, ME, MF; \
__m256i V0, V1, V2, V3, V4, V5, V6, V7; \
__m256i V8, V9, VA, VB, VC, VD, VE, VF; \
__m256i shuf_bswap64; \
V0 = H0; \
V1 = H1; \
V2 = H2; \
V3 = H3; \
V4 = H4; \
V5 = H5; \
V6 = H6; \
V7 = H7; \
V8 = _mm256_xor_si256( S0, _mm256_set1_epi64x( CB0 ) ); \
V9 = _mm256_xor_si256( S1, _mm256_set1_epi64x( CB1 ) ); \
VA = _mm256_xor_si256( S2, _mm256_set1_epi64x( CB2 ) ); \
VB = _mm256_xor_si256( S3, _mm256_set1_epi64x( CB3 ) ); \
VC = _mm256_xor_si256( _mm256_set1_epi64x( T0 ), \
_mm256_set1_epi64x( CB4 ) ); \
VD = _mm256_xor_si256( _mm256_set1_epi64x( T0 ), \
_mm256_set1_epi64x( CB5 ) ); \
VE = _mm256_xor_si256( _mm256_set1_epi64x( T1 ), \
_mm256_set1_epi64x( CB6 ) ); \
VF = _mm256_xor_si256( _mm256_set1_epi64x( T1 ), \
_mm256_set1_epi64x( CB7 ) ); \
shuf_bswap64 = _mm256_set_epi64x( 0x08090a0b0c0d0e0f, 0x0001020304050607, \
0x08090a0b0c0d0e0f, 0x0001020304050607 ); \
M0 = _mm256_shuffle_epi8( *(buf+ 0), shuf_bswap64 ); \
M1 = _mm256_shuffle_epi8( *(buf+ 1), shuf_bswap64 ); \
M2 = _mm256_shuffle_epi8( *(buf+ 2), shuf_bswap64 ); \
M3 = _mm256_shuffle_epi8( *(buf+ 3), shuf_bswap64 ); \
M4 = _mm256_shuffle_epi8( *(buf+ 4), shuf_bswap64 ); \
M5 = _mm256_shuffle_epi8( *(buf+ 5), shuf_bswap64 ); \
M6 = _mm256_shuffle_epi8( *(buf+ 6), shuf_bswap64 ); \
M7 = _mm256_shuffle_epi8( *(buf+ 7), shuf_bswap64 ); \
M8 = _mm256_shuffle_epi8( *(buf+ 8), shuf_bswap64 ); \
M9 = _mm256_shuffle_epi8( *(buf+ 9), shuf_bswap64 ); \
MA = _mm256_shuffle_epi8( *(buf+10), shuf_bswap64 ); \
MB = _mm256_shuffle_epi8( *(buf+11), shuf_bswap64 ); \
MC = _mm256_shuffle_epi8( *(buf+12), shuf_bswap64 ); \
MD = _mm256_shuffle_epi8( *(buf+13), shuf_bswap64 ); \
ME = _mm256_shuffle_epi8( *(buf+14), shuf_bswap64 ); \
MF = _mm256_shuffle_epi8( *(buf+15), shuf_bswap64 ); \
ROUND_B_4WAY(0); \
ROUND_B_4WAY(1); \
ROUND_B_4WAY(2); \
ROUND_B_4WAY(3); \
ROUND_B_4WAY(4); \
ROUND_B_4WAY(5); \
ROUND_B_4WAY(6); \
ROUND_B_4WAY(7); \
ROUND_B_4WAY(8); \
ROUND_B_4WAY(9); \
ROUND_B_4WAY(0); \
ROUND_B_4WAY(1); \
ROUND_B_4WAY(2); \
ROUND_B_4WAY(3); \
ROUND_B_4WAY(4); \
ROUND_B_4WAY(5); \
H0 = mm256_xor4( V8, V0, S0, H0 ); \
H1 = mm256_xor4( V9, V1, S1, H1 ); \
H2 = mm256_xor4( VA, V2, S2, H2 ); \
H3 = mm256_xor4( VB, V3, S3, H3 ); \
H4 = mm256_xor4( VC, V4, S0, H4 ); \
H5 = mm256_xor4( VD, V5, S1, H5 ); \
H6 = mm256_xor4( VE, V6, S2, H6 ); \
H7 = mm256_xor4( VF, V7, S3, H7 ); \
} while (0)
#endif
@@ -547,13 +543,23 @@ static void
blake64_4way_init( blake_4way_big_context *sc, const sph_u64 *iv,
const sph_u64 *salt )
{
int i;
for ( i = 0; i < 8; i++ )
sc->H[i] = _mm256_set1_epi64x( iv[i] );
for ( i = 0; i < 4; i++ )
sc->S[i] = _mm256_set1_epi64x( salt[i] );
sc->T0 = sc->T1 = 0;
sc->ptr = 0;
__m256i zero = m256_zero;
casti_m256i( sc->H, 0 ) = _mm256_set1_epi64x( iv[0] );
casti_m256i( sc->H, 1 ) = _mm256_set1_epi64x( iv[1] );
casti_m256i( sc->H, 2 ) = _mm256_set1_epi64x( iv[2] );
casti_m256i( sc->H, 3 ) = _mm256_set1_epi64x( iv[3] );
casti_m256i( sc->H, 4 ) = _mm256_set1_epi64x( iv[4] );
casti_m256i( sc->H, 5 ) = _mm256_set1_epi64x( iv[5] );
casti_m256i( sc->H, 6 ) = _mm256_set1_epi64x( iv[6] );
casti_m256i( sc->H, 7 ) = _mm256_set1_epi64x( iv[7] );
casti_m256i( sc->S, 0 ) = zero;
casti_m256i( sc->S, 1 ) = zero;
casti_m256i( sc->S, 2 ) = zero;
casti_m256i( sc->S, 3 ) = zero;
sc->T0 = sc->T1 = 0;
sc->ptr = 0;
}
static void
@@ -604,15 +610,11 @@ static void
blake64_4way_close( blake_4way_big_context *sc,
unsigned ub, unsigned n, void *dst, size_t out_size_w64)
{
// union {
__m256i buf[16];
// sph_u64 dummy;
// } u;
size_t ptr, k;
__m256i buf[16];
size_t ptr;
unsigned bit_len;
uint64_t z, zz;
sph_u64 th, tl;
__m256i *out;
ptr = sc->ptr;
bit_len = ((unsigned)ptr << 3);
@@ -665,9 +667,7 @@ blake64_4way_close( blake_4way_big_context *sc,
blake64_4way( sc, buf, 128 );
}
out = (__m256i*)dst;
for ( k = 0; k < out_size_w64; k++ )
out[k] = mm256_bswap_64( sc->H[k] );
mm256_block_bswap_64( (__m256i*)dst, sc->H );
}
void