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

v3.7.9

Browse Source
This commit is contained in:
Jay D Dee
2018-01-08 22:04:43 -05:00
parent 2d2e54f001
commit bee78eac76
58 changed files with 2817 additions and 499 deletions
Download Patch File Download Diff File Expand all files Collapse all files

51
algo/lyra2/lyra2z-4way.c
View File

@@ -4,13 +4,10 @@
#include <memory.h>
#include <mm_malloc.h>
//#include "algo-gate-api.h"
#include "lyra2.h"
#include "algo/blake/sph_blake.h"
#include "algo/blake/blake-hash-4way.h"
//#include "avxdefs.h"
// same size, only difference is the name, lyra2 is done serially
__thread uint64_t* lyra2z_4way_matrix;
bool lyra2z_4way_thread_init()
@@ -26,12 +23,8 @@ void lyra2z_4way_midstate( const void* input )
blake256_4way( &l2z_4way_blake_mid, input, 64 );
}
// block 2050 new algo, blake plus new lyra parms. new input
// is power of 2 so normal lyra can be used
//void zcoin_hash(void *state, const void *input, uint32_t height)
void lyra2z_4way_hash( void *state, const void *input )
{
// uint32_t _ALIGN(64) hash[16];
uint32_t hash0[8] __attribute__ ((aligned (64)));
uint32_t hash1[8] __attribute__ ((aligned (64)));
uint32_t hash2[8] __attribute__ ((aligned (64)));
@@ -39,27 +32,21 @@ void lyra2z_4way_hash( void *state, const void *input )
uint32_t vhash[8*4] __attribute__ ((aligned (64)));
blake256_4way_context ctx_blake __attribute__ ((aligned (64)));
// memcpy( &ctx_blake, &l2z_4way_blake_mid, sizeof l2z_4way_blake_mid );
// blake256_4way( &ctx_blake, input + (64*4), 16 );
// blake256_4way_close( &ctx_blake, vhash );
blake256_4way_init( &ctx_blake );
blake256_4way( &ctx_blake, input, 80 );
memcpy( &ctx_blake, &l2z_4way_blake_mid, sizeof l2z_4way_blake_mid );
blake256_4way( &ctx_blake, input + (64*4), 16 );
blake256_4way_close( &ctx_blake, vhash );
mm_deinterleave_4x32( hash0, hash1, hash2, hash3, vhash, 256 );
LYRA2Z( lyra2z_4way_matrix, hash0, 32, hash0, 32, hash0, 32, 8, 8, 8 );
// LYRA2Z( lyra2z_4way_matrix, hash1, 32, hash1, 32, hash1, 32, 8, 8, 8 );
LYRA2Z( lyra2z_4way_matrix, hash1, 32, hash1, 32, hash1, 32, 8, 8, 8 );
LYRA2Z( lyra2z_4way_matrix, hash2, 32, hash2, 32, hash2, 32, 8, 8, 8 );
// LYRA2Z( lyra2z_4way_matrix, hash3, 32, hash3, 32, hash3, 32, 8, 8, 8 );
LYRA2Z( lyra2z_4way_matrix, hash3, 32, hash3, 32, hash3, 32, 8, 8, 8 );
memcpy( state, hash0, 32 );
memcpy( state+32, hash1, 32 );
memcpy( state+64, hash2, 32 );
memcpy( state+96, hash3, 32 );
// memcpy(state, hash, 32);
}
int scanhash_lyra2z_4way( int thr_id, struct work *work, uint32_t max_nonce,
@@ -67,7 +54,6 @@ int scanhash_lyra2z_4way( int thr_id, struct work *work, uint32_t max_nonce,
{
uint32_t hash[8*4] __attribute__ ((aligned (64)));
uint32_t vdata[20*4] __attribute__ ((aligned (64)));
// uint32_t _ALIGN(64) hash[8];
uint32_t _ALIGN(64) edata[20];
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
@@ -90,7 +76,7 @@ int scanhash_lyra2z_4way( int thr_id, struct work *work, uint32_t max_nonce,
mm_interleave_4x32( vdata, edata, edata, edata, edata, 640 );
// lyra2z_4way_midstate( vdata );
lyra2z_4way_midstate( vdata );
do {
found[0] = found[1] = found[2] = found[3] = false;
@@ -104,42 +90,33 @@ int scanhash_lyra2z_4way( int thr_id, struct work *work, uint32_t max_nonce,
if ( hash[7] <= Htarg && fulltest( hash, ptarget ) )
{
printf("found 0\n");
found[0] = true;
num_found++;
nonces[0] = pdata[19] = n;
work_set_target_ratio( work, hash );
}
/*
if ( (hash+8)[7] <= Htarg && fulltest( hash+8, ptarget ) )
{
printf("found 1\n");
found[1] = true;
num_found++;
nonces[1] = n+1;
work_set_target_ratio( work, hash+8 );
}
*/
if ( (hash+16)[7] <= Htarg && fulltest( hash+16, ptarget ) )
{
printf("found 2\n");
found[2] = true;
num_found++;
nonces[2] = n+2;
work_set_target_ratio( work, hash+16 );
}
/*
if ( (hash+24)[7] <= Htarg && fulltest( hash+24, ptarget ) )
{
printf("found 3\n");
found[3] = true;
num_found++;
nonces[3] = n+3;
work_set_target_ratio( work, hash+24 );
}
n += 4;
*/
n += 2;
} while ( (num_found == 0) && (n < max_nonce-4)
&& !work_restart[thr_id].restart);
@@ -149,21 +126,3 @@ printf("found 3\n");
#endif
/*
if (hash[7] <= Htarg && fulltest(hash, ptarget)) {
work_set_target_ratio(work, hash);
pdata[19] = nonce;
*hashes_done = pdata[19] - first_nonce;
return 1;
}
nonce++;
} while (nonce < max_nonce && !work_restart[thr_id].restart);
pdata[19] = nonce;
*hashes_done = pdata[19] - first_nonce + 1;
return 0;
}
*/

5
algo/lyra2/lyra2z-gate.c
View File

@@ -9,18 +9,15 @@ void lyra2z_set_target( struct work* work, double job_diff )
bool register_lyra2z_algo( algo_gate_t* gate )
{
#ifdef LYRA2Z_4WAY
four_way_not_tested();
gate->optimizations = AVX_OPT | AVX2_OPT | FOUR_WAY_OPT;
gate->miner_thread_init = (void*)&lyra2z_4way_thread_init;
gate->scanhash = (void*)&scanhash_lyra2z_4way;
gate->hash = (void*)&lyra2z_4way_hash;
#else
gate->optimizations = AVX_OPT | AVX2_OPT;
gate->miner_thread_init = (void*)&lyra2z_thread_init;
gate->scanhash = (void*)&scanhash_lyra2z;
gate->hash = (void*)&lyra2z_hash;
#endif
gate->optimizations = AVX_OPT | AVX2_OPT | FOUR_WAY_OPT;
gate->get_max64 = (void*)&get_max64_0xffffLL;
gate->set_target = (void*)&lyra2z_set_target;
return true;