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

v3.11.7

Browse Source
This commit is contained in:
Jay D Dee
2020-01-26 04:33:39 -05:00
parent 103e6ad36c
commit 88f81fda0b
43 changed files with 861 additions and 1183 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
algo/keccak/keccak-4way.c
View File

@@ -28,26 +28,28 @@ int scanhash_keccak_8way( struct work *work, uint32_t max_nonce,
const uint32_t first_nonce = pdata[19];
__m512i *noncev = (__m512i*)vdata + 9; // aligned
const uint32_t Htarg = ptarget[7];
int thr_id = mythr->id;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
mm512_bswap32_intrlv80_8x64( vdata, pdata );
*noncev = mm512_intrlv_blend_32(
_mm512_set_epi32( n+7, 0, n+6, 0, n+5, 0, n+4, 0,
n+3, 0, n+2, 0, n+1, 0, n , 0 ), *noncev );
do {
*noncev = mm512_intrlv_blend_32( mm512_bswap_32(
_mm512_set_epi32( n+7, 0, n+6, 0, n+5, 0, n+4, 0,
n+3, 0, n+2, 0, n+1, 0, n , 0 ) ), *noncev );
keccakhash_8way( hash, vdata );
for ( int lane = 0; lane < 8; lane++ )
if ( hash7[ lane<<1 ] <= Htarg )
if unlikely( hash7[ lane<<1 ] <= Htarg && !bench )
{
extr_lane_8x64( lane_hash, hash, lane, 256 );
if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
if ( valid_hash( lane_hash, ptarget ) )
{
pdata[19] = n + lane;
pdata[19] = bswap_32( n + lane );
submit_lane_solution( work, lane_hash, mythr, lane );
}
}
*noncev = _mm512_add_epi32( *noncev,
m512_const1_64( 0x0000000800000000 ) );
n += 8;
} while ( (n < max_nonce-8) && !work_restart[thr_id].restart);
@@ -79,27 +81,28 @@ int scanhash_keccak_4way( struct work *work, uint32_t max_nonce,
const uint32_t first_nonce = pdata[19];
__m256i *noncev = (__m256i*)vdata + 9; // aligned
const uint32_t Htarg = ptarget[7];
int thr_id = mythr->id;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
mm256_bswap32_intrlv80_4x64( vdata, pdata );
*noncev = mm256_intrlv_blend_32(
_mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ), *noncev );
do {
*noncev = mm256_intrlv_blend_32( mm256_bswap_32(
_mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );
keccakhash_4way( hash, vdata );
for ( int lane = 0; lane < 4; lane++ )
if ( hash7[ lane<<1 ] <= Htarg )
if unlikely( hash7[ lane<<1 ] <= Htarg && !bench )
{
extr_lane_4x64( lane_hash, hash, lane, 256 );
if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
if ( valid_hash( lane_hash, ptarget ))
{
pdata[19] = n + lane;
pdata[19] = bswap_32( n + lane );
submit_lane_solution( work, lane_hash, mythr, lane );
}
}
*noncev = _mm256_add_epi32( *noncev,
m256_const1_64( 0x0000000400000000 ) );
n += 4;
} while ( (n < max_nonce-4) && !work_restart[thr_id].restart);
*hashes_done = n - first_nonce + 1;