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

v3.9.11

Browse Source
This commit is contained in:
Jay D Dee
2019-11-26 09:22:03 -05:00
parent a52c5eccf7
commit 91ec6f1771
28 changed files with 7609 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

168
algo/blake/blake2s-hash-4way.c
View File

@@ -20,12 +20,13 @@
//#if defined(__SSE4_2__)
#if defined(__SSE2__)
/*
static const uint32_t blake2s_IV[8] =
{
0x6A09E667UL, 0xBB67AE85UL, 0x3C6EF372UL, 0xA54FF53AUL,
0x510E527FUL, 0x9B05688CUL, 0x1F83D9ABUL, 0x5BE0CD19UL
};
*/
static const uint8_t blake2s_sigma[10][16] =
{
@@ -41,6 +42,7 @@ static const uint8_t blake2s_sigma[10][16] =
{ 10, 2, 8, 4, 7, 6, 1, 5, 15, 11, 9, 14, 3, 12, 13 , 0 } ,
};
// define a constant for initial param.
int blake2s_4way_init( blake2s_4way_state *S, const uint8_t outlen )
@@ -88,41 +90,45 @@ int blake2s_4way_compress( blake2s_4way_state *S, const __m128i* block )
memcpy_128( m, block, 16 );
memcpy_128( v, S->h, 8 );
v[ 8] = _mm_set1_epi32( blake2s_IV[0] );
v[ 9] = _mm_set1_epi32( blake2s_IV[1] );
v[10] = _mm_set1_epi32( blake2s_IV[2] );
v[11] = _mm_set1_epi32( blake2s_IV[3] );
v[ 8] = m128_const1_64( 0x6A09E6676A09E667ULL );
v[ 9] = m128_const1_64( 0xBB67AE85BB67AE85ULL );
v[10] = m128_const1_64( 0x3C6EF3723C6EF372ULL );
v[11] = m128_const1_64( 0xA54FF53AA54FF53AULL );
v[12] = _mm_xor_si128( _mm_set1_epi32( S->t[0] ),
_mm_set1_epi32( blake2s_IV[4] ) );
m128_const1_64( 0x510E527F510E527FULL ) );
v[13] = _mm_xor_si128( _mm_set1_epi32( S->t[1] ),
_mm_set1_epi32( blake2s_IV[5] ) );
m128_const1_64( 0x9B05688C9B05688CULL ) );
v[14] = _mm_xor_si128( _mm_set1_epi32( S->f[0] ),
_mm_set1_epi32( blake2s_IV[6] ) );
m128_const1_64( 0x1F83D9AB1F83D9ABULL ) );
v[15] = _mm_xor_si128( _mm_set1_epi32( S->f[1] ),
_mm_set1_epi32( blake2s_IV[7] ) );
m128_const1_64( 0x5BE0CD195BE0CD19ULL ) );
#define G4W(r,i,a,b,c,d) \
#define G4W( sigma0, sigma1, a, b, c, d ) \
do { \
a = _mm_add_epi32( _mm_add_epi32( a, b ), m[ blake2s_sigma[r][2*i+0] ] ); \
uint8_t s0 = sigma0; \
uint8_t s1 = sigma1; \
a = _mm_add_epi32( _mm_add_epi32( a, b ), m[ s0 ] ); \
d = mm128_ror_32( _mm_xor_si128( d, a ), 16 ); \
c = _mm_add_epi32( c, d ); \
b = mm128_ror_32( _mm_xor_si128( b, c ), 12 ); \
a = _mm_add_epi32( _mm_add_epi32( a, b ), m[ blake2s_sigma[r][2*i+1] ] ); \
a = _mm_add_epi32( _mm_add_epi32( a, b ), m[ s1 ] ); \
d = mm128_ror_32( _mm_xor_si128( d, a ), 8 ); \
c = _mm_add_epi32( c, d ); \
b = mm128_ror_32( _mm_xor_si128( b, c ), 7 ); \
} while(0)
#define ROUND4W(r) \
do { \
G4W( r, 0, v[ 0], v[ 4], v[ 8], v[12] ); \
G4W( r, 1, v[ 1], v[ 5], v[ 9], v[13] ); \
G4W( r, 2, v[ 2], v[ 6], v[10], v[14] ); \
G4W( r, 3, v[ 3], v[ 7], v[11], v[15] ); \
G4W( r, 4, v[ 0], v[ 5], v[10], v[15] ); \
G4W( r, 5, v[ 1], v[ 6], v[11], v[12] ); \
G4W( r, 6, v[ 2], v[ 7], v[ 8], v[13] ); \
G4W( r, 7, v[ 3], v[ 4], v[ 9], v[14] ); \
uint8_t *sigma = (uint8_t*)&blake2s_sigma[r]; \
G4W( sigma[ 0], sigma[ 1], v[ 0], v[ 4], v[ 8], v[12] ); \
G4W( sigma[ 2], sigma[ 3], v[ 1], v[ 5], v[ 9], v[13] ); \
G4W( sigma[ 4], sigma[ 5], v[ 2], v[ 6], v[10], v[14] ); \
G4W( sigma[ 6], sigma[ 7], v[ 3], v[ 7], v[11], v[15] ); \
G4W( sigma[ 8], sigma[ 9], v[ 0], v[ 5], v[10], v[15] ); \
G4W( sigma[10], sigma[11], v[ 1], v[ 6], v[11], v[12] ); \
G4W( sigma[12], sigma[13], v[ 2], v[ 7], v[ 8], v[13] ); \
G4W( sigma[14], sigma[15], v[ 3], v[ 4], v[ 9], v[14] ); \
} while(0)
ROUND4W( 0 );
@@ -144,26 +150,47 @@ do { \
return 0;
}
// There is a problem that can't be resolved internally.
// If the last block is a full 64 bytes it should not be compressed in
// update but left for final. However, when streaming, it isn't known
// which block is last. There may be a subsequent call to update to add
// more data.
//
// The reference code handled this by juggling 2 blocks at a time at
// a significant performance penalty.
//
// Instead a new function is introduced called full_blocks which combines
// update and final and is to be used in non-streaming mode where the data
// is a multiple of 64 bytes.
//
// Supported:
// 64 + 16 bytes (blake2s with midstate optimization)
// 80 bytes without midstate (blake2s without midstate optimization)
// Any multiple of 64 bytes in one shot (x25x)
//
// Unsupported:
// Stream of 64 byte blocks one at a time.
//
// use for part blocks or when streaming more data
int blake2s_4way_update( blake2s_4way_state *S, const void *in,
uint64_t inlen )
{
__m128i *input = (__m128i*)in;
__m128i *buf = (__m128i*)S->buf;
const int bsize = BLAKE2S_BLOCKBYTES;
__m128i *input = (__m128i*)in;
__m128i *buf = (__m128i*)S->buf;
while( inlen > 0 )
{
size_t left = S->buflen;
if( inlen >= bsize - left )
if( inlen >= BLAKE2S_BLOCKBYTES - left )
{
memcpy_128( buf + (left>>2), input, (bsize - left) >> 2 );
S->buflen += bsize - left;
memcpy_128( buf + (left>>2), input, (BLAKE2S_BLOCKBYTES - left) >> 2 );
S->buflen += BLAKE2S_BLOCKBYTES - left;
S->t[0] += BLAKE2S_BLOCKBYTES;
S->t[1] += ( S->t[0] < BLAKE2S_BLOCKBYTES );
blake2s_4way_compress( S, buf );
S->buflen = 0;
input += ( bsize >> 2 );
inlen -= bsize;
input += ( BLAKE2S_BLOCKBYTES >> 2 );
inlen -= BLAKE2S_BLOCKBYTES;
}
else
{
@@ -195,8 +222,45 @@ int blake2s_4way_final( blake2s_4way_state *S, void *out, uint8_t outlen )
return 0;
}
// Update and final when inlen is a multiple of 64 bytes
int blake2s_4way_full_blocks( blake2s_4way_state *S, void *out,
const void *input, uint64_t inlen )
{
__m128i *in = (__m128i*)input;
__m128i *buf = (__m128i*)S->buf;
while( inlen > BLAKE2S_BLOCKBYTES )
{
memcpy_128( buf, in, BLAKE2S_BLOCKBYTES >> 2 );
S->buflen = BLAKE2S_BLOCKBYTES;
inlen -= BLAKE2S_BLOCKBYTES;
S->t[0] += BLAKE2S_BLOCKBYTES;
S->t[1] += ( S->t[0] < BLAKE2S_BLOCKBYTES );
blake2s_4way_compress( S, buf );
S->buflen = 0;
in += ( BLAKE2S_BLOCKBYTES >> 2 );
}
// last block
memcpy_128( buf, in, BLAKE2S_BLOCKBYTES >> 2 );
S->buflen = BLAKE2S_BLOCKBYTES;
S->t[0] += S->buflen;
S->t[1] += ( S->t[0] < S->buflen );
if ( S->last_node ) S->f[1] = ~0U;
S->f[0] = ~0U;
blake2s_4way_compress( S, buf );
for ( int i = 0; i < 8; ++i )
casti_m128i( out, i ) = S->h[ i ];
return 0;
}
#if defined(__AVX2__)
// The commented code below is slower on Intel but faster on
// Zen1 AVX2. It's also faster than Zen1 AVX.
// Ryzen gen2 is unknown at this time.
int blake2s_8way_compress( blake2s_8way_state *S, const __m256i *block )
{
__m256i m[16];
@@ -205,6 +269,23 @@ int blake2s_8way_compress( blake2s_8way_state *S, const __m256i *block )
memcpy_256( m, block, 16 );
memcpy_256( v, S->h, 8 );
v[ 8] = m256_const1_64( 0x6A09E6676A09E667ULL );
v[ 9] = m256_const1_64( 0xBB67AE85BB67AE85ULL );
v[10] = m256_const1_64( 0x3C6EF3723C6EF372ULL );
v[11] = m256_const1_64( 0xA54FF53AA54FF53AULL );
v[12] = _mm256_xor_si256( _mm256_set1_epi32( S->t[0] ),
m256_const1_64( 0x510E527F510E527FULL ) );
v[13] = _mm256_xor_si256( _mm256_set1_epi32( S->t[1] ),
m256_const1_64( 0x9B05688C9B05688CULL ) );
v[14] = _mm256_xor_si256( _mm256_set1_epi32( S->f[0] ),
m256_const1_64( 0x1F83D9AB1F83D9ABULL ) );
v[15] = _mm256_xor_si256( _mm256_set1_epi32( S->f[1] ),
m256_const1_64( 0x5BE0CD195BE0CD19ULL ) );
/*
v[ 8] = _mm256_set1_epi32( blake2s_IV[0] );
v[ 9] = _mm256_set1_epi32( blake2s_IV[1] );
v[10] = _mm256_set1_epi32( blake2s_IV[2] );
@@ -218,6 +299,7 @@ int blake2s_8way_compress( blake2s_8way_state *S, const __m256i *block )
v[15] = _mm256_xor_si256( _mm256_set1_epi32( S->f[1] ),
_mm256_set1_epi32( blake2s_IV[7] ) );
#define G8W(r,i,a,b,c,d) \
do { \
a = _mm256_add_epi32( _mm256_add_epi32( a, b ), \
@@ -231,7 +313,36 @@ do { \
c = _mm256_add_epi32( c, d ); \
b = mm256_ror_32( _mm256_xor_si256( b, c ), 7 ); \
} while(0)
*/
#define G8W( sigma0, sigma1, a, b, c, d) \
do { \
uint8_t s0 = sigma0; \
uint8_t s1 = sigma1; \
a = _mm256_add_epi32( _mm256_add_epi32( a, b ), m[ s0 ] ); \
d = mm256_ror_32( _mm256_xor_si256( d, a ), 16 ); \
c = _mm256_add_epi32( c, d ); \
b = mm256_ror_32( _mm256_xor_si256( b, c ), 12 ); \
a = _mm256_add_epi32( _mm256_add_epi32( a, b ), m[ s1 ] ); \
d = mm256_ror_32( _mm256_xor_si256( d, a ), 8 ); \
c = _mm256_add_epi32( c, d ); \
b = mm256_ror_32( _mm256_xor_si256( b, c ), 7 ); \
} while(0)
#define ROUND8W(r) \
do { \
uint8_t *sigma = (uint8_t*)&blake2s_sigma[r]; \
G8W( sigma[ 0], sigma[ 1], v[ 0], v[ 4], v[ 8], v[12] ); \
G8W( sigma[ 2], sigma[ 3], v[ 1], v[ 5], v[ 9], v[13] ); \
G8W( sigma[ 4], sigma[ 5], v[ 2], v[ 6], v[10], v[14] ); \
G8W( sigma[ 6], sigma[ 7], v[ 3], v[ 7], v[11], v[15] ); \
G8W( sigma[ 8], sigma[ 9], v[ 0], v[ 5], v[10], v[15] ); \
G8W( sigma[10], sigma[11], v[ 1], v[ 6], v[11], v[12] ); \
G8W( sigma[12], sigma[13], v[ 2], v[ 7], v[ 8], v[13] ); \
G8W( sigma[14], sigma[15], v[ 3], v[ 4], v[ 9], v[14] ); \
} while(0)
/*
#define ROUND8W(r) \
do { \
G8W( r, 0, v[ 0], v[ 4], v[ 8], v[12] ); \
@@ -243,6 +354,7 @@ do { \
G8W( r, 6, v[ 2], v[ 7], v[ 8], v[13] ); \
G8W( r, 7, v[ 3], v[ 4], v[ 9], v[14] ); \
} while(0)
*/
ROUND8W( 0 );
ROUND8W( 1 );