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

v3.7.6

Browse Source
This commit is contained in:
Jay D Dee
2017-12-14 18:28:51 -05:00
parent af1c940919
commit 7a1389998b
31 changed files with 1285 additions and 377 deletions
Download Patch File Download Diff File Expand all files Collapse all files

97
algo/blake/decred-4way.c
View File

@@ -13,46 +13,35 @@ static __thread bool ctx_midstate_done = false;
void decred_hash_4way( void *state, const void *input )
{
uint32_t vhash[4*4] __attribute__ ((aligned (64)));
uint32_t hash0[4] __attribute__ ((aligned (32)));
uint32_t hash1[4] __attribute__ ((aligned (32)));
uint32_t hash2[4] __attribute__ ((aligned (32)));
uint32_t hash3[4] __attribute__ ((aligned (32)));
uint32_t vhash[8*4] __attribute__ ((aligned (64)));
uint32_t hash0[8] __attribute__ ((aligned (32)));
uint32_t hash1[8] __attribute__ ((aligned (32)));
uint32_t hash2[8] __attribute__ ((aligned (32)));
uint32_t hash3[8] __attribute__ ((aligned (32)));
blake256_4way_context ctx __attribute__ ((aligned (64)));
sph_blake256_context ctx2 __attribute__ ((aligned (64)));
uint32_t hash[16] __attribute__ ((aligned (64)));
uint32_t sin0[45], sin1[45], sin2[45], sin3[45];
mm_deinterleave_4x32x( sin0, sin1, sin2, sin3, input, 180*8 );
void *tail = input + DECRED_MIDSTATE_LEN;
void *tail = input + ( DECRED_MIDSTATE_LEN << 2 );
int tail_len = 180 - DECRED_MIDSTATE_LEN;
// #define MIDSTATE_LEN 128
/*
uint8_t *ending = (uint8_t*) input;
ending += MIDSTATE_LEN;
if ( !ctx_midstate_done )
{
blake256_4way_init( &blake_mid );
blake256_4way( &blake_mid, input, DECRED_MIDSTATE_LEN );
ctx_midstate_done = true;
}
memcpy( &ctx, &blake_mid, sizeof(blake_mid) );
blake256_4way( &ctx, tail, tail_len );
blake256_4way_close( &ctx, vhash );
*/
/*
sph_blake256_init( &ctx2 );
sph_blake256( &ctx2, sin0, 180 );
sph_blake256_close( &ctx2, hash );
*/
/*
blake256_4way_init( &ctx );
blake256_4way( &ctx, input, 180 );
blake256_4way_close( &ctx, vhash );
*/
mm_deinterleave_4x32( hash0, hash1, hash2, hash3, vhash, 256 );
/*
for ( int i = 0; i < 8; i++ )
@@ -66,22 +55,21 @@ printf("hash0: %08lx %08lx %08lx %08lx\n", *hash0, *(hash0+1),
printf("\n");
*/
// memcpy( state, hash0, 32 );
// memcpy( state+32, hash1, 32 );
// memcpy( state+64, hash1, 32 );
// memcpy( state+96, hash1, 32 );
memcpy( state, hash0, 32 );
memcpy( state+32, hash1, 32 );
memcpy( state+64, hash2, 32 );
memcpy( state+96, hash3, 32 );
memcpy( state, hash, 32 );
// memcpy( state, hash, 32 );
}
int scanhash_decred_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done)
{
uint32_t vdata[45*4] __attribute__ ((aligned (64)));
uint32_t hash[4*4] __attribute__ ((aligned (32)));
uint32_t _ALIGN(64) endiandata[48];
// uint32_t _ALIGN(64) hash32[8];
uint32_t vdata[48*4] __attribute__ ((aligned (64)));
uint32_t hash[8*4] __attribute__ ((aligned (32)));
uint32_t _ALIGN(64) edata[48];
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[DECRED_NONCE_INDEX];
@@ -91,28 +79,25 @@ int scanhash_decred_4way( int thr_id, struct work *work, uint32_t max_nonce,
bool *found = work->nfound;
int num_found = 0;
// #define DCR_NONCE_OFT32 35
ctx_midstate_done = false;
// memcpy(endiandata, pdata, 180);
memcpy( edata, pdata, 180 );
// use the old way until new way updated for size.
mm_interleave_4x32x( vdata, pdata, pdata, pdata, pdata, 180*8 );
mm_interleave_4x32( vdata, edata, edata, edata, edata, 180*8 );
blake256_4way_init( &blake_mid );
blake256_4way( &blake_mid, vdata, DECRED_MIDSTATE_LEN );
uint32_t *noncep = vdata + DECRED_NONCE_INDEX * 4;
do {
found[0] = found[1] = found[2] = found[3] = false;
* noncep = n;
*(noncep+2) = n+1;
*(noncep+4) = n+2;
*(noncep+6) = n+3;
*(noncep+1) = n+1;
*(noncep+2) = n+2;
*(noncep+3) = n+3;
decred_hash_4way( hash, vdata );
// endiandata[DCR_NONCE_OFT32] = n;
// decred_hash(hash32, endiandata);
if ( hash[7] <= HTarget && fulltest( hash, ptarget ) )
{
work_set_target_ratio( work, hash );
@@ -121,29 +106,47 @@ int scanhash_decred_4way( int thr_id, struct work *work, uint32_t max_nonce,
nonces[0] = n;
pdata[DECRED_NONCE_INDEX] = n;
}
/* if ( (hash+8)[7] <= HTarget && fulltest( hash+8, ptarget ) )
/*
if ( (hash+8)[7] <= HTarget && fulltest( hash+8, ptarget ) )
{
printf("found 1\n");
printf("vhash: %08lx %08lx %08lx %08lx\n", hash[8], hash[9], hash[10],hash[11] );
printf("vhash: %08lx %08lx %08lx %08lx\n", hash[12], hash[13], hash[14],hash[15] );
printf("shash: %08lx %08lx %08lx %08lx\n", shash[0], shash[1], shash[2],shash[3] );
printf("shash: %08lx %08lx %08lx %08lx\n\n", shash[4], shash[5], shash[6],shash[7] );
work_set_target_ratio( work, hash+8 );
found[1] = true;
num_found++;
nonces[1] = n;
nonces[1] = n+1;
}
*/
if ( (hash+16)[7] <= HTarget && fulltest( hash+16, ptarget ) )
{
work_set_target_ratio( work, hash+16 );
found[2] = true;
num_found++;
nonces[2] = n;
nonces[2] = n+2;
}
/*
if ( (hash+24)[7] <= HTarget && fulltest( hash+24, ptarget ) )
{
printf("found 3\n");
printf("vhash: %08lx %08lx %08lx %08lx\n", hash[0], hash[1], hash[2],hash[3] );
printf("vhash: %08lx %08lx %08lx %08lx\n", hash[4], hash[5], hash[6],hash[7] );
printf("shash: %08lx %08lx %08lx %08lx\n", shash[0], shash[1], shash[2],shash[3] );
printf("shash: %08lx %08lx %08lx %08lx\n\n", shash[4], shash[5], shash[6],shash[7] );
work_set_target_ratio( work, hash+24 );
found[3] = true;
num_found++;
nonces[3] = n;
nonces[3] = n+3;
}
*/
n += 4;
n += 2;
// n += 4;
} while ( (num_found == 0) && (n < max_nonce)
&& !work_restart[thr_id].restart );