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

v3.10.1

Browse Source
This commit is contained in:
Jay D Dee
2019-12-05 19:09:23 -05:00
parent 40039386a0
commit 73430b13b1
52 changed files with 4515 additions and 874 deletions
Download Patch File Download Diff File Expand all files Collapse all files

178
algo/cubehash/cube-hash-2way.c
View File

@@ -26,6 +26,180 @@ static const uint64_t IV512[] =
0xA5A70E75D65C8A2B, 0xBC796576B1C62456, 0xE7989AF11921C8F7, 0xD43E3B447795D246
};
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
static void transform_4way( cube_4way_context *sp )
{
int r;
const int rounds = sp->rounds;
__m512i x0, x1, x2, x3, x4, x5, x6, x7, y0, y1;
x0 = _mm512_load_si512( (__m512i*)sp->h );
x1 = _mm512_load_si512( (__m512i*)sp->h + 1 );
x2 = _mm512_load_si512( (__m512i*)sp->h + 2 );
x3 = _mm512_load_si512( (__m512i*)sp->h + 3 );
x4 = _mm512_load_si512( (__m512i*)sp->h + 4 );
x5 = _mm512_load_si512( (__m512i*)sp->h + 5 );
x6 = _mm512_load_si512( (__m512i*)sp->h + 6 );
x7 = _mm512_load_si512( (__m512i*)sp->h + 7 );
for ( r = 0; r < rounds; ++r )
{
x4 = _mm512_add_epi32( x0, x4 );
x5 = _mm512_add_epi32( x1, x5 );
x6 = _mm512_add_epi32( x2, x6 );
x7 = _mm512_add_epi32( x3, x7 );
y0 = x0;
y1 = x1;
x0 = mm512_rol_32( x2, 7 );
x1 = mm512_rol_32( x3, 7 );
x2 = mm512_rol_32( y0, 7 );
x3 = mm512_rol_32( y1, 7 );
x0 = _mm512_xor_si512( x0, x4 );
x1 = _mm512_xor_si512( x1, x5 );
x2 = _mm512_xor_si512( x2, x6 );
x3 = _mm512_xor_si512( x3, x7 );
x4 = mm512_swap64_128( x4 );
x5 = mm512_swap64_128( x5 );
x6 = mm512_swap64_128( x6 );
x7 = mm512_swap64_128( x7 );
x4 = _mm512_add_epi32( x0, x4 );
x5 = _mm512_add_epi32( x1, x5 );
x6 = _mm512_add_epi32( x2, x6 );
x7 = _mm512_add_epi32( x3, x7 );
y0 = x0;
y1 = x2;
x0 = mm512_rol_32( x1, 11 );
x1 = mm512_rol_32( y0, 11 );
x2 = mm512_rol_32( x3, 11 );
x3 = mm512_rol_32( y1, 11 );
x0 = _mm512_xor_si512( x0, x4 );
x1 = _mm512_xor_si512( x1, x5 );
x2 = _mm512_xor_si512( x2, x6 );
x3 = _mm512_xor_si512( x3, x7 );
x4 = mm512_swap32_64( x4 );
x5 = mm512_swap32_64( x5 );
x6 = mm512_swap32_64( x6 );
x7 = mm512_swap32_64( x7 );
}
_mm512_store_si512( (__m512i*)sp->h, x0 );
_mm512_store_si512( (__m512i*)sp->h + 1, x1 );
_mm512_store_si512( (__m512i*)sp->h + 2, x2 );
_mm512_store_si512( (__m512i*)sp->h + 3, x3 );
_mm512_store_si512( (__m512i*)sp->h + 4, x4 );
_mm512_store_si512( (__m512i*)sp->h + 5, x5 );
_mm512_store_si512( (__m512i*)sp->h + 6, x6 );
_mm512_store_si512( (__m512i*)sp->h + 7, x7 );
}
int cube_4way_init( cube_4way_context *sp, int hashbitlen, int rounds,
int blockbytes )
{
__m512i *h = (__m512i*)sp->h;
__m128i *iv = (__m128i*)( hashbitlen == 512 ? (__m128i*)IV512
: (__m128i*)IV256 );
sp->hashlen = hashbitlen/128;
sp->blocksize = blockbytes/16;
sp->rounds = rounds;
sp->pos = 0;
h[ 0] = m512_const1_128( iv[0] );
h[ 1] = m512_const1_128( iv[1] );
h[ 2] = m512_const1_128( iv[2] );
h[ 3] = m512_const1_128( iv[3] );
h[ 4] = m512_const1_128( iv[4] );
h[ 5] = m512_const1_128( iv[5] );
h[ 6] = m512_const1_128( iv[6] );
h[ 7] = m512_const1_128( iv[7] );
h[ 0] = m512_const1_128( iv[0] );
h[ 1] = m512_const1_128( iv[1] );
h[ 2] = m512_const1_128( iv[2] );
h[ 3] = m512_const1_128( iv[3] );
h[ 4] = m512_const1_128( iv[4] );
h[ 5] = m512_const1_128( iv[5] );
h[ 6] = m512_const1_128( iv[6] );
h[ 7] = m512_const1_128( iv[7] );
return 0;
}
int cube_4way_update( cube_4way_context *sp, const void *data, size_t size )
{
const int len = size >> 4;
const __m512i *in = (__m512i*)data;
int i;
for ( i = 0; i < len; i++ )
{
sp->h[ sp->pos ] = _mm512_xor_si512( sp->h[ sp->pos ], in[i] );
sp->pos++;
if ( sp->pos == sp->blocksize )
{
transform_4way( sp );
sp->pos = 0;
}
}
return 0;
}
int cube_4way_close( cube_4way_context *sp, void *output )
{
__m512i *hash = (__m512i*)output;
int i;
// pos is zero for 64 byte data, 1 for 80 byte data.
sp->h[ sp->pos ] = _mm512_xor_si512( sp->h[ sp->pos ],
m512_const2_64( 0, 0x0000000000000080 ) );
transform_4way( sp );
sp->h[7] = _mm512_xor_si512( sp->h[7],
m512_const2_64( 0x0000000100000000, 0 ) );
for ( i = 0; i < 10; ++i )
transform_4way( sp );
memcpy( hash, sp->h, sp->hashlen<<6 );
return 0;
}
int cube_4way_update_close( cube_4way_context *sp, void *output,
const void *data, size_t size )
{
const int len = size >> 4;
const __m512i *in = (__m512i*)data;
__m512i *hash = (__m512i*)output;
int i;
for ( i = 0; i < len; i++ )
{
sp->h[ sp->pos ] = _mm512_xor_si512( sp->h[ sp->pos ], in[i] );
sp->pos++;
if ( sp->pos == sp->blocksize )
{
transform_4way( sp );
sp->pos = 0;
}
}
// pos is zero for 64 byte data, 1 for 80 byte data.
sp->h[ sp->pos ] = _mm512_xor_si512( sp->h[ sp->pos ],
m512_const2_64( 0, 0x0000000000000080 ) );
transform_4way( sp );
sp->h[7] = _mm512_xor_si512( sp->h[7],
m512_const2_64( 0x0000000100000000, 0 ) );
for ( i = 0; i < 10; ++i )
transform_4way( sp );
memcpy( hash, sp->h, sp->hashlen<<6);
return 0;
}
#endif // AVX512
static void transform_2way( cube_2way_context *sp )
{
@@ -91,7 +265,6 @@ static void transform_2way( cube_2way_context *sp )
_mm256_store_si256( (__m256i*)sp->h + 5, x5 );
_mm256_store_si256( (__m256i*)sp->h + 6, x6 );
_mm256_store_si256( (__m256i*)sp->h + 7, x7 );
}
int cube_2way_init( cube_2way_context *sp, int hashbitlen, int rounds,
@@ -132,9 +305,6 @@ int cube_2way_update( cube_2way_context *sp, const void *data, size_t size )
const __m256i *in = (__m256i*)data;
int i;
// It is assumed data is aligned to 256 bits and is a multiple of 128 bits.
// Current usage sata is either 64 or 80 bytes.
for ( i = 0; i < len; i++ )
{
sp->h[ sp->pos ] = _mm256_xor_si256( sp->h[ sp->pos ], in[i] );