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

v3.21.3 Unreleased

Browse Source
This commit is contained in:
Jay D Dee
2023-03-13 03:20:13 -04:00
parent b339450898
commit c6bc9d67fb
49 changed files with 1126 additions and 1111 deletions
Download Patch File Download Diff File Expand all files Collapse all files

85
algo/skein/skein-4way.c
View File

@@ -7,16 +7,8 @@
#if defined (SKEIN_8WAY)
static skein512_8way_context skein512_8way_ctx
static __thread skein512_8way_context skein512_8way_ctx
__attribute__ ((aligned (64)));
static uint32_t skein_8way_vdata[20*8] __attribute__ ((aligned (64)));
int skein_8way_prehash( struct work *work )
{
mm512_bswap32_intrlv80_8x64( skein_8way_vdata, work->data );
skein512_8way_prehash64( &skein512_8way_ctx, skein_8way_vdata );
return 1;
}
void skeinhash_8way( void *state, const void *input )
{
@@ -37,27 +29,25 @@ void skeinhash_8way( void *state, const void *input )
int scanhash_skein_8way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t vdata[20*8] __attribute__ ((aligned (128)));
uint32_t hash[8*8] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (64)));
uint32_t *hash_d7 = &(hash[7*8]);
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t targ_d7 = ptarget[7];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 8;
uint32_t n = first_nonce;
__m512i *noncev = (__m512i*)vdata + 9;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
pthread_rwlock_rdlock( &g_work_lock );
memcpy( vdata, skein_8way_vdata, sizeof vdata );
pthread_rwlock_unlock( &g_work_lock );
uint32_t vdata[20*8] __attribute__ ((aligned (128)));
uint32_t hash[8*8] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (64)));
uint32_t *hash_d7 = &(hash[7*8]);
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t targ_d7 = ptarget[7];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 8;
uint32_t n = first_nonce;
__m512i *noncev = (__m512i*)vdata + 9;
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 );
skein512_8way_prehash64( &skein512_8way_ctx, vdata );
do
{
skeinhash_8way( hash, vdata );
@@ -84,16 +74,8 @@ int scanhash_skein_8way( struct work *work, uint32_t max_nonce,
#elif defined (SKEIN_4WAY)
static skein512_4way_context skein512_4way_ctx
static __thread skein512_4way_context skein512_4way_ctx
__attribute__ ((aligned (64)));
static uint32_t skein_4way_vdata[20*4] __attribute__ ((aligned (64)));
int skein_4way_prehash( struct work *work )
{
mm256_bswap32_intrlv80_4x64( skein_4way_vdata, work->data );
skein512_4way_prehash64( &skein512_4way_ctx, skein_4way_vdata );
return 1;
}
void skeinhash_4way( void *state, const void *input )
{
@@ -136,24 +118,23 @@ void skeinhash_4way( void *state, const void *input )
int scanhash_skein_4way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t vdata[20*4] __attribute__ ((aligned (64)));
uint32_t hash[8*4] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *hash_d7 = &(hash[7<<2]);
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t targ_d7 = ptarget[7];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 4;
uint32_t n = first_nonce;
__m256i *noncev = (__m256i*)vdata + 9;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
uint32_t vdata[20*4] __attribute__ ((aligned (64)));
uint32_t hash[8*4] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *hash_d7 = &(hash[7<<2]);
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t targ_d7 = ptarget[7];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 4;
uint32_t n = first_nonce;
__m256i *noncev = (__m256i*)vdata + 9;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
mm256_bswap32_intrlv80_4x64( vdata, pdata );
skein512_4way_prehash64( &skein512_4way_ctx, vdata );
pthread_rwlock_rdlock( &g_work_lock );
memcpy( vdata, skein_4way_vdata, sizeof vdata );
pthread_rwlock_unlock( &g_work_lock );
*noncev = mm256_intrlv_blend_32(
_mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ), *noncev );
do

8
algo/skein/skein-gate.c
View File

@@ -7,12 +7,10 @@ bool register_skein_algo( algo_gate_t* gate )
#if defined (SKEIN_8WAY)
gate->optimizations = AVX2_OPT | AVX512_OPT;
gate->scanhash = (void*)&scanhash_skein_8way;
gate->prehash = (void*)&skein_8way_prehash;
gate->hash = (void*)&skeinhash_8way;
#elif defined (SKEIN_4WAY)
gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
gate->scanhash = (void*)&scanhash_skein_4way;
gate->prehash = (void*)&skein_4way_prehash;
gate->hash = (void*)&skeinhash_4way;
#else
gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
@@ -27,12 +25,10 @@ bool register_skein2_algo( algo_gate_t* gate )
gate->optimizations = AVX2_OPT | AVX512_OPT;
#if defined (SKEIN_8WAY)
gate->scanhash = (void*)&scanhash_skein2_8way;
// gate->hash = (void*)&skein2hash_8way;
gate->prehash = (void*)&skein2_8way_prehash;
gate->hash = (void*)&skein2hash_8way;
#elif defined (SKEIN_4WAY)
gate->scanhash = (void*)&scanhash_skein2_4way;
// gate->hash = (void*)&skein2hash_4way;
gate->prehash = (void*)&skein2_4way_prehash;
gate->hash = (void*)&skein2hash_4way;
#else
gate->scanhash = (void*)&scanhash_skein2;
gate->hash = (void*)&skein2hash;

4
algo/skein/skein-gate.h
View File

@@ -14,24 +14,20 @@
void skeinhash_8way( void *output, const void *input );
int scanhash_skein_8way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr );
int skein_8way_prehash( struct work * );
void skein2hash_8way( void *output, const void *input );
int scanhash_skein2_8way( struct work *work, uint32_t max_nonce,
uint64_t* hashes_done, struct thr_info *mythr );
int skein2_8way_prehash( struct work * );
#elif defined(SKEIN_4WAY)
void skeinhash_4way( void *output, const void *input );
int scanhash_skein_4way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr );
int skein_4way_prehash( struct work * );
void skein2hash_4way( void *output, const void *input );
int scanhash_skein2_4way( struct work *work, uint32_t max_nonce,
uint64_t* hashes_done, struct thr_info *mythr );
int skein2_4way_prehash( struct work * );
#else

13
algo/skein/skein.c
View File

@@ -31,19 +31,18 @@ int scanhash_skein( struct work *work, uint32_t max_nonce,
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
int thr_id = mythr->id;
int thr_id = mythr->id; // thr_id arg is deprecated
swab32_array( endiandata, pdata, 20 );
do {
be32enc(&endiandata[19], n);
skeinhash(hash64, endiandata);
if (hash64[7] <= Htarg )
if ( fulltest(hash64, ptarget) && !opt_benchmark )
{
pdata[19] = n;
submit_solution( work, hash64, mythr );
}
if (hash64[7] < Htarg && fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
pdata[19] = n;
return true;
}
n++;
} while (n < max_nonce && !work_restart[thr_id].restart);

42
algo/skein/skein2-4way.c
View File

@@ -5,17 +5,9 @@
#if defined(SKEIN_8WAY)
static skein512_8way_context skein512_8way_ctx __attribute__ ((aligned (64)));
static uint32_t skein2_8way_vdata[20*8] __attribute__ ((aligned (64)));
static __thread skein512_8way_context skein512_8way_ctx
__attribute__ ((aligned (64)));
int skein2_8way_prehash( struct work *work )
{
mm512_bswap32_intrlv80_8x64( skein2_8way_vdata, work->data );
skein512_8way_prehash64( &skein512_8way_ctx, skein2_8way_vdata );
return 1;
}
/* not used
void skein2hash_8way( void *output, const void *input )
{
uint64_t hash[16*8] __attribute__ ((aligned (128)));
@@ -25,7 +17,6 @@ void skein2hash_8way( void *output, const void *input )
skein512_8way_final16( &ctx, hash, input + (64*8) );
skein512_8way_full( &ctx, output, hash, 64 );
}
*/
int scanhash_skein2_8way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
@@ -45,14 +36,11 @@ int scanhash_skein2_8way( struct work *work, uint32_t max_nonce,
const bool bench = opt_benchmark;
skein512_8way_context ctx;
pthread_rwlock_rdlock( &g_work_lock );
memcpy( vdata, skein2_8way_vdata, sizeof vdata );
memcpy( &ctx, &skein512_8way_ctx, sizeof ctx );
pthread_rwlock_unlock( &g_work_lock );
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 );
skein512_8way_prehash64( &ctx, vdata );
do
{
skein512_8way_final16( &ctx, hash, vdata + (16*8) );
@@ -79,18 +67,10 @@ int scanhash_skein2_8way( struct work *work, uint32_t max_nonce,
}
#elif defined(SKEIN_4WAY)
static skein512_4way_context skein512_4way_ctx __attribute__ ((aligned (64)));
static uint32_t skein2_4way_vdata[20*4] __attribute__ ((aligned (64)));
int skein2_4way_prehash( struct work *work )
{
mm256_bswap32_intrlv80_4x64( skein2_4way_vdata, work->data );
skein512_4way_prehash64( &skein512_4way_ctx, skein2_4way_vdata );
return 1;
}
/* not used
static __thread skein512_4way_context skein512_4way_ctx
__attribute__ ((aligned (64)));
void skein2hash_4way( void *output, const void *input )
{
skein512_4way_context ctx;
@@ -100,7 +80,6 @@ void skein2hash_4way( void *output, const void *input )
skein512_4way_final16( &ctx, hash, input + (64*4) );
skein512_4way_full( &ctx, output, hash, 64 );
}
*/
int scanhash_skein2_4way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
@@ -120,11 +99,8 @@ int scanhash_skein2_4way( struct work *work, uint32_t max_nonce,
const bool bench = opt_benchmark;
skein512_4way_context ctx;
pthread_rwlock_rdlock( &g_work_lock );
memcpy( vdata, skein2_4way_vdata, sizeof vdata );
memcpy( &ctx, &skein512_4way_ctx, sizeof ctx );
pthread_rwlock_unlock( &g_work_lock );
mm256_bswap32_intrlv80_4x64( vdata, pdata );
skein512_4way_prehash64( &ctx, vdata );
*noncev = mm256_intrlv_blend_32(
_mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ), *noncev );
do

20
algo/skein/skein2.c
View File

@@ -34,31 +34,31 @@ void skein2hash(void *output, const void *input)
sph_skein512_close(&ctx_skein, hash);
memcpy(output, hash, 32);
}
int scanhash_skein2( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t hash64[8] __attribute__ ((aligned (64)));
uint32_t endiandata[20] __attribute__ ((aligned (64)));
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
int thr_id = mythr->id;
int thr_id = mythr->id; // thr_id arg is deprecated
swab32_array( endiandata, pdata, 20 );
swab32_array( endiandata, pdata, 20 );
do {
be32enc(&endiandata[19], n);
skein2hash(hash64, endiandata);
if (hash64[7] <= Htarg )
if ( fulltest(hash64, ptarget) && !opt_benchmark )
{
pdata[19] = n;
submit_solution( work, hash64, mythr );
}
if (hash64[7] < Htarg && fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
pdata[19] = n;
return true;
}
n++;
} while (n < max_nonce && !work_restart[thr_id].restart);