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

v3.9.5.4

Browse Source
This commit is contained in:
Jay D Dee
2019-07-15 17:00:26 -04:00
parent e625ed5420
commit e2d5762ef2
63 changed files with 1973 additions and 2980 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
algo/blake/blake2s-4way.c
View File

@@ -16,9 +16,9 @@ void blake2s_8way_hash( void *output, const void *input )
blake2s_8way_update( &ctx, input + (64<<3), 16 );
blake2s_8way_final( &ctx, vhash, BLAKE2S_OUTBYTES );
mm256_dintrlv_8x32( output, output+ 32, output+ 64, output+ 96,
output+128, output+160, output+192, output+224,
vhash, 256 );
dintrlv_8x32( output, output+ 32, output+ 64, output+ 96,
output+128, output+160, output+192, output+224,
vhash, 256 );
}
int scanhash_blake2s_8way( struct work *work, uint32_t max_nonce,
@@ -28,28 +28,19 @@ int scanhash_blake2s_8way( struct work *work, uint32_t max_nonce,
uint32_t hash[8*8] __attribute__ ((aligned (32)));
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t _ALIGN(64) edata[20];
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
__m256i *noncev = (__m256i*)vdata + 19; // aligned
uint32_t n = first_nonce;
uint32_t *noncep = vdata + 152; // 19*8
int thr_id = mythr->id; // thr_id arg is deprecated
swab32_array( edata, pdata, 20 );
mm256_intrlv_8x32( vdata, edata, edata, edata, edata,
edata, edata, edata, edata, 640 );
mm256_bswap32_intrlv80_8x32( vdata, pdata );
blake2s_8way_init( &blake2s_8w_ctx, BLAKE2S_OUTBYTES );
blake2s_8way_update( &blake2s_8w_ctx, vdata, 64 );
do {
be32enc( noncep, n );
be32enc( noncep +1, n+1 );
be32enc( noncep +2, n+2 );
be32enc( noncep +3, n+3 );
be32enc( noncep +4, n+4 );
be32enc( noncep +5, n+5 );
be32enc( noncep +6, n+6 );
be32enc( noncep +7, n+7 );
*noncev = mm256_bswap_32( _mm256_set_epi32( n+7, n+6, n+5, n+4,
n+3, n+2, n+1, n ) );
pdata[19] = n;
blake2s_8way_hash( hash, vdata );
@@ -94,23 +85,18 @@ int scanhash_blake2s_4way( struct work *work, uint32_t max_nonce,
uint32_t hash[8*4] __attribute__ ((aligned (32)));
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t _ALIGN(64) edata[20];
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
__m128i *noncev = (__m128i*)vdata + 19; // aligned
uint32_t n = first_nonce;
uint32_t *noncep = vdata + 76; // 19*4
int thr_id = mythr->id; // thr_id arg is deprecated
swab32_array( edata, pdata, 20 );
mm128_intrlv_4x32( vdata, edata, edata, edata, edata, 640 );
mm128_bswap32_intrlv80_4x32( vdata, pdata );
blake2s_4way_init( &blake2s_4w_ctx, BLAKE2S_OUTBYTES );
blake2s_4way_update( &blake2s_4w_ctx, vdata, 64 );
do {
be32enc( noncep, n );
be32enc( noncep +1, n+1 );
be32enc( noncep +2, n+2 );
be32enc( noncep +3, n+3 );
*noncev = mm128_bswap_32( _mm_set_epi32( n+3, n+2, n+1, n ) );
pdata[19] = n;
blake2s_4way_hash( hash, vdata );