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

v3.9.5.1

Browse Source
This commit is contained in:
Jay D Dee
2019-07-02 15:10:38 -04:00
parent 0d48d573ce
commit 0d769ee0fe
53 changed files with 1755 additions and 1170 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
algo/skein/skein2-4way.c
View File

@@ -23,29 +23,41 @@ int scanhash_skein2_4way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t hash[8*4] __attribute__ ((aligned (64)));
uint32_t *hash7 = &(hash[25]);
uint32_t edata[20] __attribute__ ((aligned (64)));
uint32_t vdata[20*4] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (64)));
uint32_t *hash7 = &(hash[25]);
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
__m256i *noncev = (__m256i*)vdata + 9; // aligned
// __m256i *noncev = (__m256i*)vdata + 9; // aligned
int thr_id = mythr->id; // thr_id arg is deprecated
uint32_t *noncep = vdata + 73; // 9*8 + 1
mm256_bswap_intrlv80_4x64( vdata, pdata );
swab32_array( edata, pdata, 20 );
mm256_intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
// mm256_bswap_intrlv80_4x64( vdata, pdata );
do
{
*noncev = mm256_intrlv_blend_32( mm256_bswap_32(
_mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );
be32enc( noncep, n );
be32enc( noncep+2, n+1 );
be32enc( noncep+4, n+2 );
be32enc( noncep+6, n+3 );
// *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
// _mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );
skein2hash_4way( hash, vdata );
for ( int lane = 0; lane < 4; lane++ )
if ( hash7[ lane<<1 ] <= Htarg )
{
uint32_t lane_hash[8];
mm256_extract_lane_4x64( lane_hash, hash, lane, 256 );
mm256_extr_lane_4x64( lane_hash, hash, lane, 256 );
if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
{
pdata[19] = n + lane;