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

v3.7.3

Browse Source
This commit is contained in:
Jay D Dee
2017-11-20 21:19:15 -05:00
parent ab39e88318
commit 6d1361c87f
46 changed files with 6314 additions and 141 deletions
Download Patch File Download Diff File Expand all files Collapse all files

105
algo/keccak/keccak-4way.c Normal file
View File

@@ -0,0 +1,105 @@
#include "keccak-gate.h"
#include <stdlib.h>
#include <string.h>
#include <stdint.h>
#include "sph_keccak.h"
#include "keccak-hash-4way.h"
#ifdef __AVX2__
void keccakhash_4way(void *state, const void *input)
{
uint64_t hash0[8] __attribute__ ((aligned (64)));
uint64_t hash1[8] __attribute__ ((aligned (64)));
uint64_t hash2[8] __attribute__ ((aligned (64)));
uint64_t hash3[8] __attribute__ ((aligned (64)));
uint64_t vhash[8*4] __attribute__ ((aligned (64)));
keccak256_4way_context ctx;
keccak256_4way_init( &ctx );
keccak256_4way( &ctx, input, 80 );
keccak256_4way_close( &ctx, vhash );
m256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
memcpy( state, hash0, 32 );
memcpy( state+32, hash1, 32 );
memcpy( state+64, hash2, 32 );
memcpy( state+96, hash3, 32 );
}
int scanhash_keccak_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done)
{
uint32_t hash[4*8] __attribute__ ((aligned (64)));
uint32_t vdata[20*4] __attribute__ ((aligned (64)));
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t n = pdata[19];
const uint32_t first_nonce = pdata[19];
const uint32_t Htarg = ptarget[7];
uint32_t endiandata[20];
uint32_t *nonces = work->nonces;
bool *found = work->nfound;
int num_found;
uint32_t *noncep0 = vdata + 73; // 9*8 + 1
uint32_t *noncep1 = vdata + 75;
uint32_t *noncep2 = vdata + 77;
uint32_t *noncep3 = vdata + 79;
for ( int i=0; i < 19; i++ )
be32enc( &endiandata[i], pdata[i] );
uint64_t *edata = (uint64_t*)endiandata;
m256_interleave_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
do {
found[0] = found[1] = found[2] = found[3] = false;
num_found = 0;
be32enc( noncep0, n );
be32enc( noncep1, n+1 );
be32enc( noncep2, n+2 );
be32enc( noncep3, n+3 );
keccakhash_4way( hash, vdata );
if ( ( ( hash[7] & 0xFFFFFF00 ) == 0 )
&& fulltest( hash, ptarget) )
{
found[0] = true;
num_found++;
nonces[0] = n;
pdata[19] = n;
}
if ( ( ( (hash+8)[7] & 0xFFFFFF00 ) == 0 )
&& fulltest( hash+8, ptarget) )
{
found[1] = true;
num_found++;
nonces[1] = n+1;
}
if ( ( ( (hash+16) [7] & 0xFFFFFF00 ) == 0 )
&& fulltest( hash+16, ptarget) )
{
found[2] = true;
num_found++;
nonces[2] = n+2;
}
if ( ( ( (hash+24)[7] & 0xFFFFFF00 ) == 0 )
&& fulltest( hash+24, ptarget) )
{
found[3] = true;
num_found++;
nonces[3] = n+3;
}
n += 4;
} while ( (num_found == 0) && (n < max_nonce-4)
&& !work_restart[thr_id].restart);
*hashes_done = n - first_nonce + 1;
return num_found;
}
#endif