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

v23.5

Browse Source
This commit is contained in:
Jay D Dee
2023-10-25 20:36:20 -04:00
parent 31c4dedf59
commit 160608cce5
180 changed files with 10318 additions and 13097 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
algo/sha/sha256t-4way.c
View File

@@ -35,7 +35,7 @@ int scanhash_sha256t_16way( struct work *work, const uint32_t max_nonce,
const int thr_id = mythr->id;
const __m512i sixteen = v512_32( 16 );
const bool bench = opt_benchmark;
const __m256i bswap_shuf = mm256_bcast_m128( _mm_set_epi64x(
const __m256i bswap_shuf = mm256_bcast_m128( v128_set64(
0x0c0d0e0f08090a0b, 0x0405060700010203 ) );
// prehash first block directly from pdata
@@ -208,8 +208,7 @@ int scanhash_sha256t_neon_sha2( struct work *work, uint32_t max_nonce,
uint32_t block2b[16] __attribute__ ((aligned (64)));
uint32_t hasha[8] __attribute__ ((aligned (32)));
uint32_t hashb[8] __attribute__ ((aligned (32)));
uint32_t mstatea[8] __attribute__ ((aligned (32)));
uint32_t sstate[8] __attribute__ ((aligned (32)));
uint32_t mstate[8] __attribute__ ((aligned (32)));
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[19];
@@ -219,7 +218,7 @@ int scanhash_sha256t_neon_sha2( struct work *work, uint32_t max_nonce,
const bool bench = opt_benchmark;
// hash first 64 byte block of data
sha256_transform_le( mstatea, pdata, sha256_iv );
sha256_transform_le( mstate, pdata, sha256_iv );
// fill & pad second bock without nonce
memcpy( block1a, pdata + 16, 12 );
@@ -242,11 +241,11 @@ int scanhash_sha256t_neon_sha2( struct work *work, uint32_t max_nonce,
// Insert nonce for second block
block1a[3] = n;
block1b[3] = n+1;
sha256_neon2x_transform_le( block2a, block2b, block1a, block1b,
mstatea, mstatea );
sha256_neon2x_transform_le( block2a, block2b, block2a, block2b,
sha256_neon_x2sha_transform_le( block2a, block2b, block1a, block1b,
mstate, mstate );
sha256_neon_x2sha_transform_le( block2a, block2b, block2a, block2b,
sha256_iv, sha256_iv );
sha256_neon2x_transform_le( hasha, hashb, block2a, block2b,
sha256_neon_x2sha_transform_le( hasha, hashb, block2a, block2b,
sha256_iv, sha256_iv );
if ( unlikely( bswap_32( hasha[7] ) <= ptarget[7] ) )
@@ -397,7 +396,7 @@ int scanhash_sha256t_4way( struct work *work, const uint32_t max_nonce,
for ( int i = 0; i < 19; i++ )
vdata[i] = v128_32( pdata[i] );
*noncev = v128_set_32( n+ 3, n+ 2, n+1, n );
*noncev = v128_set32( n+ 3, n+ 2, n+1, n );
vdata[16+4] = last_byte;
v128_memset_zero( vdata+16 + 5, 10 );