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

18
algo/sha/sha256t-4way.c
View File

@@ -72,7 +72,7 @@ int scanhash_sha256t_11way( struct work *work, uint32_t max_nonce,
casti_m256i( dataz, 1 ) = mm256_bswap_32( casti_m256i( pdata, 1 ) );
casti_m128i( dataz, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
mm256_intrlv_8x32( datax, dataz, dataz, dataz, dataz,
intrlv_8x32( datax, dataz, dataz, dataz, dataz,
dataz, dataz, dataz, dataz, 640 );
mm64_interleave_2x32( datay, dataz, dataz, 640 );
@@ -99,7 +99,7 @@ int scanhash_sha256t_11way( struct work *work, uint32_t max_nonce,
for ( i = 0; i < 8; i++ ) if ( !( hash7[ i ] & mask ) )
{
// deinterleave hash for lane
mm256_extr_lane_8x32( lane_hash, hashx, i, 256 );
extr_lane_8x32( lane_hash, hashx, i, 256 );
if ( fulltest( lane_hash, ptarget ) )
{
pdata[19] = n + i;
@@ -163,7 +163,6 @@ int scanhash_sha256t_8way( struct work *work, uint32_t max_nonce,
{
uint32_t vdata[20*8] __attribute__ ((aligned (64)));
uint32_t hash[8*8] __attribute__ ((aligned (32)));
uint32_t edata[20] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *hash7 = &(hash[7<<3]);
uint32_t *pdata = work->data;
@@ -187,12 +186,9 @@ int scanhash_sha256t_8way( struct work *work, uint32_t max_nonce,
0xFFFF0000,
0 };
swab32_array( edata, pdata, 20 );
mm256_intrlv_8x32( vdata, edata, edata, edata, edata,
edata, edata, edata, edata, 640 );
// Need big endian data
// mm256_bswap_intrlv80_8x32( vdata, pdata );
mm256_bswap32_intrlv80_8x32( vdata, pdata );
sha256_8way_init( &sha256_ctx8 );
sha256_8way( &sha256_ctx8, vdata, 64 );
@@ -209,7 +205,7 @@ int scanhash_sha256t_8way( struct work *work, uint32_t max_nonce,
if ( !( hash7[ lane ] & mask ) )
{
// deinterleave hash for lane
mm256_extr_lane_8x32( lane_hash, hash, lane, 256 );
extr_lane_8x32( lane_hash, hash, lane, 256 );
if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
{
pdata[19] = n + lane;
@@ -253,7 +249,6 @@ int scanhash_sha256t_4way( struct work *work, uint32_t max_nonce,
{
uint32_t vdata[20*4] __attribute__ ((aligned (64)));
uint32_t hash[8*4] __attribute__ ((aligned (32)));
uint32_t edata[20] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (64)));
uint32_t *hash7 = &(hash[7<<2]);
uint32_t *pdata = work->data;
@@ -277,10 +272,7 @@ int scanhash_sha256t_4way( struct work *work, uint32_t max_nonce,
0xFFFF0000,
0 };
swab32_array( edata, pdata, 20 );
mm128_intrlv_4x32( vdata, edata, edata, edata, edata, 640 );
// mm128_bswap_intrlv80_4x32( vdata, pdata );
mm128_bswap32_intrlv80_4x32( vdata, pdata );
sha256_4way_init( &sha256_ctx4 );
sha256_4way( &sha256_ctx4, vdata, 64 );