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

v23.10

Browse Source
This commit is contained in:
Jay D Dee
2023-11-15 11:05:41 -05:00
parent 0a78013cbe
commit f3fde95f27
32 changed files with 1074 additions and 1027 deletions
Download Patch File Download Diff File Expand all files Collapse all files

65
algo/skein/skein-4way.c
View File

@@ -159,4 +159,69 @@ int scanhash_skein_4way( struct work *work, uint32_t max_nonce,
return 0;
}
#elif defined(SKEIN_2WAY)
static __thread skein512_2x64_context skein512_2x64_ctx
__attribute__ ((aligned (64)));
void skeinhash_2x64( void *state, const void *input )
{
uint64_t vhash64[8*2] __attribute__ ((aligned (32)));
uint32_t hash0[16] __attribute__ ((aligned (32)));
uint32_t hash1[16] __attribute__ ((aligned (32)));
skein512_2x64_context ctx_skein;
memcpy( &ctx_skein, &skein512_2x64_ctx, sizeof( ctx_skein ) );
skein512_2x64_final16( &ctx_skein, vhash64, input + (64*2) );
dintrlv_2x64( hash0, hash1, vhash64, 512 );
sha256_full( hash0, hash0, 64 );
sha256_full( hash1, hash1, 64 );
intrlv_2x32( state, hash0, hash1, 256 );
}
int scanhash_skein_2x64( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t vdata[20*2] __attribute__ ((aligned (32)));
uint32_t hash[8*2] __attribute__ ((aligned (32)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *hash_d7 = &(hash[7<<1]);
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 - 2;
uint32_t n = first_nonce;
v128u32_t *noncev = (v128u32_t*)vdata + 9;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
v128_bswap32_intrlv80_2x64( vdata, pdata );
skein512_2x64_prehash64( &skein512_2x64_ctx, vdata );
*noncev = v128_intrlv_blend_32( v128_set32( n+1, 0, n, 0 ), *noncev );
do
{
skeinhash_2x64( hash, vdata );
for ( int lane = 0; lane < 2; lane++ )
if ( unlikely( ( hash_d7[ lane ] <= targ_d7 ) && !bench ) )
{
extr_lane_2x32( lane_hash, hash, lane, 256 );
if ( valid_hash( lane_hash, ptarget ) )
{
pdata[19] = bswap_32( n + lane );
submit_solution( work, lane_hash, mythr );
}
}
*noncev = v128_add32( *noncev, v128_64( 0x0000000200000000 ) );
n += 2;
} while ( likely( (n < last_nonce) && !work_restart[thr_id].restart ) );
pdata[19] = n;
*hashes_done = n - first_nonce;
return 0;
}
#endif

25
algo/skein/skein-gate.c
View File

@@ -3,16 +3,20 @@
bool register_skein_algo( algo_gate_t* gate )
{
#if defined (SKEIN_8WAY)
gate->optimizations = AVX2_OPT | AVX512_OPT;
#if defined(SKEIN_8WAY)
gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT;
gate->scanhash = (void*)&scanhash_skein_8way;
gate->hash = (void*)&skeinhash_8way;
#elif defined (SKEIN_4WAY)
gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
#elif defined(SKEIN_4WAY)
gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT | SHA_OPT | NEON_OPT;
gate->scanhash = (void*)&scanhash_skein_4way;
gate->hash = (void*)&skeinhash_4way;
#elif defined(SKEIN_2WAY)
gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT | SHA_OPT | NEON_OPT;
gate->scanhash = (void*)&scanhash_skein_2x64;
gate->hash = (void*)&skeinhash_2x64;
#else
gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT | SHA_OPT | NEON_OPT;
gate->scanhash = (void*)&scanhash_skein;
gate->hash = (void*)&skeinhash;
#endif
@@ -21,16 +25,15 @@ bool register_skein_algo( algo_gate_t* gate )
bool register_skein2_algo( algo_gate_t* gate )
{
gate->optimizations = AVX2_OPT | AVX512_OPT;
#if defined (SKEIN_8WAY)
gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT | NEON_OPT;
#if defined(SKEIN_8WAY)
gate->scanhash = (void*)&scanhash_skein2_8way;
gate->hash = (void*)&skein2hash_8way;
#elif defined (SKEIN_4WAY)
#elif defined(SKEIN_4WAY)
gate->scanhash = (void*)&scanhash_skein2_4way;
gate->hash = (void*)&skein2hash_4way;
#elif defined(SKEIN_2WAY)
gate->scanhash = (void*)&scanhash_skein2_2x64;
#else
gate->scanhash = (void*)&scanhash_skein2;
gate->hash = (void*)&skein2hash;
#endif
return true;
};

12
algo/skein/skein-gate.h
View File

@@ -7,6 +7,8 @@
#define SKEIN_8WAY 1
#elif defined(__AVX2__)
#define SKEIN_4WAY 1
#elif defined(__SSE2__) || defined(__ARM_NEON)
#define SKEIN_2WAY 1
#endif
#if defined(SKEIN_8WAY)
@@ -29,6 +31,16 @@ 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 );
#elif defined(SKEIN_2WAY)
void skeinhash_2x64( void *output, const void *input );
int scanhash_skein_2x64( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr );
void skein2hash_2x64( void *output, const void *input );
int scanhash_skein2_2x64( struct work *work, uint32_t max_nonce,
uint64_t* hashes_done, struct thr_info *mythr );
#else
void skeinhash( void *output, const void *input );

36
algo/skein/skein-hash-4way.c
View File

@@ -675,11 +675,13 @@ void skein512_8way_full( skein512_8way_context *sc, void *out, const void *data,
// Close
unsigned et;
memset_zero_512( buf + (ptr>>3), (buf_size - ptr) >> 3 );
et = 352 + ((bcount == 0) << 7);
UBI_BIG_8WAY( et, ptr );
if ( ptr )
{
unsigned et;
memset_zero_512( buf + (ptr>>3), (buf_size - ptr) >> 3 );
et = 352 + ((bcount == 0) << 7);
UBI_BIG_8WAY( et, ptr );
}
memset_zero_512( buf, buf_size >> 3 );
bcount = 0;
@@ -970,11 +972,13 @@ skein512_4way_full( skein512_4way_context *sc, void *out, const void *data,
// Close
unsigned et;
memset_zero_256( buf + (ptr>>3), (buf_size - ptr) >> 3 );
et = 352 + ((bcount == 0) << 7);
UBI_BIG_4WAY( et, ptr );
if ( ptr )
{
unsigned et;
memset_zero_256( buf + (ptr>>3), (buf_size - ptr) >> 3 );
et = 352 + ((bcount == 0) << 7);
UBI_BIG_4WAY( et, ptr );
}
memset_zero_256( buf, buf_size >> 3 );
bcount = 0;
@@ -1364,11 +1368,13 @@ skein512_2x64_full( skein512_2x64_context *sc, void *out, const void *data,
// Close
unsigned et;
v128_memset_zero( buf + (ptr>>3), (buf_size - ptr) >> 3 );
et = 352 + ((bcount == 0) << 7);
UBI_BIG_2WAY( et, ptr );
if ( ptr )
{
unsigned et;
v128_memset_zero( buf + (ptr>>3), (buf_size - ptr) >> 3 );
et = 352 + ((bcount == 0) << 7);
UBI_BIG_2WAY( et, ptr );
}
v128_memset_zero( buf, buf_size >> 3 );
bcount = 0;

75
algo/skein/skein2-4way.c
View File

@@ -5,19 +5,6 @@
#if defined(SKEIN_8WAY)
static __thread skein512_8way_context skein512_8way_ctx
__attribute__ ((aligned (64)));
void skein2hash_8way( void *output, const void *input )
{
uint64_t hash[16*8] __attribute__ ((aligned (128)));
skein512_8way_context ctx;
memcpy( &ctx, &skein512_8way_ctx, sizeof( ctx ) );
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 )
{
@@ -68,19 +55,6 @@ int scanhash_skein2_8way( struct work *work, uint32_t max_nonce,
#elif defined(SKEIN_4WAY)
static __thread skein512_4way_context skein512_4way_ctx
__attribute__ ((aligned (64)));
void skein2hash_4way( void *output, const void *input )
{
skein512_4way_context ctx;
memcpy( &ctx, &skein512_4way_ctx, sizeof( ctx ) );
uint64_t hash[16*4] __attribute__ ((aligned (64)));
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 )
{
@@ -128,4 +102,53 @@ int scanhash_skein2_4way( struct work *work, uint32_t max_nonce,
return 0;
}
#elif defined(SKEIN_2WAY)
int scanhash_skein2_2x64( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint64_t hash[8*2] __attribute__ ((aligned (64)));
uint32_t vdata[20*2] __attribute__ ((aligned (64)));
skein512_2x64_context ctx;
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint64_t *hash_q3 = &(hash[3*2]);
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint64_t targ_q3 = ((uint64_t*)ptarget)[3];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 4;
uint32_t n = first_nonce;
v128u64_t *noncev = (v128u64_t*)vdata + 9;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
const v128u64_t two = v128_64( 0x0000000200000000 );
v128_bswap32_intrlv80_2x64( vdata, pdata );
skein512_2x64_prehash64( &ctx, vdata );
*noncev = v128_intrlv_blend_32( v128_set32( n+1, 0, n, 0 ), *noncev );
do
{
skein512_2x64_final16( &ctx, hash, vdata + (16*2) );
skein512_2x64_full( &ctx, hash, hash, 64 );
for ( int lane = 0; lane < 2; lane++ )
if ( hash_q3[ lane ] <= targ_q3 )
{
extr_lane_2x64( lane_hash, hash, lane, 256 );
if ( valid_hash( lane_hash, ptarget ) && !bench )
{
pdata[19] = bswap_32( n + lane );
submit_solution( work, lane_hash, mythr );
}
}
*noncev = v128_add32( *noncev, two );
n += 2;
} while ( (n < last_nonce) && !work_restart[thr_id].restart );
pdata[19] = n;
*hashes_done = n - first_nonce;
return 0;
}
#endif