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

v3.10.7

Browse Source
This commit is contained in:
Jay D Dee
2019-12-28 15:00:29 -05:00
parent 241bc26767
commit b47cfaa720
75 changed files with 1884 additions and 773 deletions
Download Patch File Download Diff File Expand all files Collapse all files

54
algo/ripemd/lbry-4way.c
View File

@@ -7,7 +7,7 @@
#include "ripemd-hash-4way.h"
#define LBRY_INPUT_SIZE 112
#define LBRY_MIDSTATE 64
#define LBRY_MIDSTATE 96
#define LBRY_TAIL (LBRY_INPUT_SIZE) - (LBRY_MIDSTATE)
#if defined(LBRY_16WAY)
@@ -35,9 +35,9 @@ void lbry_16way_hash( void* output, const void* input )
uint32_t _ALIGN(64) h13[32];
uint32_t _ALIGN(64) h14[32];
uint32_t _ALIGN(64) h15[32];
sha256_16way_context ctx_sha256 __attribute__ ((aligned (64)));
sha256_16way_context ctx_sha256 __attribute__ ((aligned (64)));
sha512_8way_context ctx_sha512;
ripemd160_16way_context ctx_ripemd;
ripemd160_16way_context ctx_ripemd;
memcpy( &ctx_sha256, &sha256_16w_mid, sizeof(ctx_sha256) );
sha256_16way_update( &ctx_sha256, input + (LBRY_MIDSTATE<<4), LBRY_TAIL );
@@ -62,7 +62,7 @@ void lbry_16way_hash( void* output, const void* input )
sha512_8way_close( &ctx_sha512, vhashB );
// back to 8-way 32 bit
dintrlv_8x64( h0, h1, h2, h3,h4, h5, h6, h7, vhashA, 512 );
dintrlv_8x64( h0, h1, h2, h3, h4, h5, h6, h7, vhashA, 512 );
dintrlv_8x64( h8, h9, h10, h11, h12, h13, h14, h15, vhashB, 512 );
intrlv_16x32( vhashA, h0, h1, h2, h3, h4, h5, h6, h7,
h8, h9, h10, h11, h12, h13, h14, h15, 512 );
@@ -90,14 +90,15 @@ int scanhash_lbry_16way( struct work *work, uint32_t max_nonce,
{
uint32_t hash[8*16] __attribute__ ((aligned (128)));
uint32_t vdata[32*16] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t lane_hash[8] __attribute__ ((aligned (64)));
uint32_t edata[32] __attribute__ ((aligned (64)));
uint32_t *hash7 = &(hash[7<<4]);
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t n = pdata[27];
const uint32_t first_nonce = pdata[27];
const uint32_t last_nonce = max_nonce - 16;
const uint32_t Htarg = ptarget[7];
uint32_t edata[32] __attribute__ ((aligned (64)));
__m512i *noncev = (__m512i*)vdata + 27; // aligned
int thr_id = mythr->id; // thr_id arg is deprecated
@@ -114,14 +115,13 @@ int scanhash_lbry_16way( struct work *work, uint32_t max_nonce,
edata, edata, edata, edata, edata, edata, edata, edata, edata, 1024 );
sha256_16way_init( &sha256_16w_mid );
sha256_16way( &sha256_16w_mid, vdata, LBRY_MIDSTATE );
sha256_16way_update( &sha256_16w_mid, vdata, LBRY_MIDSTATE );
do
{
*noncev = mm512_bswap_32( _mm512_set_epi32( n+15, n+14, n+13, n+12,
n+11, n+10, n+ 9, n+ 8,
n+ 7, n+ 6, n+ 5, n+ 4,
n+ 3, n+ 2, n+ 1, n ) );
*noncev = mm512_bswap_32( _mm512_set_epi32(
n+15, n+14, n+13, n+12, n+11, n+10, n+ 9, n+ 8,
n+ 7, n+ 6, n+ 5, n+ 4, n+ 3, n+ 2, n+ 1, n ) );
lbry_16way_hash( hash, vdata );
for ( int i = 0; i < 16; i++ )
@@ -129,27 +129,25 @@ int scanhash_lbry_16way( struct work *work, uint32_t max_nonce,
{
// deinterleave hash for lane
extr_lane_16x32( lane_hash, hash, i, 256 );
if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
if ( likely( fulltest( lane_hash, ptarget ) && !opt_benchmark ) )
{
pdata[27] = n + i;
submit_lane_solution( work, lane_hash, mythr, i );
}
}
n += 16;
} while ( (n < max_nonce-16) && !work_restart[thr_id].restart );
*hashes_done = n - first_nonce + 1;
} while ( likely( (n < last_nonce) && !work_restart[thr_id].restart ) );
*hashes_done = n - first_nonce;
return 0;
}
#elif defined(LBRY_8WAY)
static __thread sha256_8way_context sha256_8w_mid;
void lbry_8way_hash( void* output, const void* input )
{
uint32_t _ALIGN(64) vhashA[16<<3];
uint32_t _ALIGN(128) vhashA[16<<3];
uint32_t _ALIGN(64) vhashB[16<<3];
uint32_t _ALIGN(64) vhashC[16<<3];
uint32_t _ALIGN(32) h0[32];
@@ -165,11 +163,11 @@ void lbry_8way_hash( void* output, const void* input )
ripemd160_8way_context ctx_ripemd;
memcpy( &ctx_sha256, &sha256_8w_mid, sizeof(ctx_sha256) );
sha256_8way( &ctx_sha256, input + (LBRY_MIDSTATE<<3), LBRY_TAIL );
sha256_8way_update( &ctx_sha256, input + (LBRY_MIDSTATE<<3), LBRY_TAIL );
sha256_8way_close( &ctx_sha256, vhashA );
sha256_8way_init( &ctx_sha256 );
sha256_8way( &ctx_sha256, vhashA, 32 );
sha256_8way_update( &ctx_sha256, vhashA, 32 );
sha256_8way_close( &ctx_sha256, vhashA );
// reinterleave to do sha512 4-way 64 bit twice.
@@ -178,11 +176,11 @@ void lbry_8way_hash( void* output, const void* input )
intrlv_4x64( vhashB, h4, h5, h6, h7, 256 );
sha512_4way_init( &ctx_sha512 );
sha512_4way( &ctx_sha512, vhashA, 32 );
sha512_4way_update( &ctx_sha512, vhashA, 32 );
sha512_4way_close( &ctx_sha512, vhashA );
sha512_4way_init( &ctx_sha512 );
sha512_4way( &ctx_sha512, vhashB, 32 );
sha512_4way_update( &ctx_sha512, vhashB, 32 );
sha512_4way_close( &ctx_sha512, vhashB );
// back to 8-way 32 bit
@@ -191,20 +189,20 @@ void lbry_8way_hash( void* output, const void* input )
intrlv_8x32( vhashA, h0, h1, h2, h3, h4, h5, h6, h7, 512 );
ripemd160_8way_init( &ctx_ripemd );
ripemd160_8way( &ctx_ripemd, vhashA, 32 );
ripemd160_8way_update( &ctx_ripemd, vhashA, 32 );
ripemd160_8way_close( &ctx_ripemd, vhashB );
ripemd160_8way_init( &ctx_ripemd );
ripemd160_8way( &ctx_ripemd, vhashA+(8<<3), 32 );
ripemd160_8way_update( &ctx_ripemd, vhashA+(8<<3), 32 );
ripemd160_8way_close( &ctx_ripemd, vhashC );
sha256_8way_init( &ctx_sha256 );
sha256_8way( &ctx_sha256, vhashB, 20 );
sha256_8way( &ctx_sha256, vhashC, 20 );
sha256_8way_update( &ctx_sha256, vhashB, 20 );
sha256_8way_update( &ctx_sha256, vhashC, 20 );
sha256_8way_close( &ctx_sha256, vhashA );
sha256_8way_init( &ctx_sha256 );
sha256_8way( &ctx_sha256, vhashA, 32 );
sha256_8way_update( &ctx_sha256, vhashA, 32 );
sha256_8way_close( &ctx_sha256, output );
}
@@ -214,13 +212,13 @@ int scanhash_lbry_8way( struct work *work, uint32_t max_nonce,
uint32_t hash[8*8] __attribute__ ((aligned (64)));
uint32_t vdata[32*8] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t edata[32] __attribute__ ((aligned (64)));
uint32_t *hash7 = &(hash[7<<3]);
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t n = pdata[27];
const uint32_t first_nonce = pdata[27];
const uint32_t Htarg = ptarget[7];
uint32_t edata[32] __attribute__ ((aligned (64)));
__m256i *noncev = (__m256i*)vdata + 27; // aligned
int thr_id = mythr->id; // thr_id arg is deprecated
@@ -237,7 +235,7 @@ int scanhash_lbry_8way( struct work *work, uint32_t max_nonce,
edata, edata, edata, edata, 1024 );
sha256_8way_init( &sha256_8w_mid );
sha256_8way( &sha256_8w_mid, vdata, LBRY_MIDSTATE );
sha256_8way_update( &sha256_8w_mid, vdata, LBRY_MIDSTATE );
do
{

2
algo/ripemd/lbry-gate.c
View File

@@ -98,7 +98,7 @@ int lbry_get_work_data_size() { return LBRY_WORK_DATA_SIZE; }
bool register_lbry_algo( algo_gate_t* gate )
{
gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
// gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
#if defined (LBRY_16WAY)
gate->scanhash = (void*)&scanhash_lbry_16way;
gate->hash = (void*)&lbry_16way_hash;

13
algo/ripemd/lbry-gate.h
View File

@@ -5,11 +5,10 @@
#include <stdint.h>
// 16 way needs sha256 16 way
//#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
// #define LBRY_16WAY
#if defined(__AVX2__)
#define LBRY_8WAY
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
#define LBRY_16WAY 1
#elif defined(__AVX2__)
#define LBRY_8WAY 1
#endif
/*
#if !defined(__SHA__)
@@ -37,13 +36,13 @@ int scanhash_lbry_16way( struct work *work, uint32_t max_nonce,
void lbry_8way_hash( void *state, const void *input );
int scanhash_lbry_8way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr );
/*
#elif defined(LBRY_4WAY)
void lbry_4way_hash( void *state, const void *input );
int scanhash_lbry_4way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done );
*/
#else
void lbry_hash( void *state, const void *input );

8
algo/ripemd/ripemd-hash-4way.c
View File

@@ -259,7 +259,8 @@ void ripemd160_4way_init( ripemd160_4way_context *sc )
sc->count_high = sc->count_low = 0;
}
void ripemd160_4way( ripemd160_4way_context *sc, const void *data, size_t len )
void ripemd160_4way_update( ripemd160_4way_context *sc, const void *data,
size_t len )
{
__m128i *vdata = (__m128i*)data;
size_t ptr;
@@ -559,7 +560,8 @@ void ripemd160_8way_init( ripemd160_8way_context *sc )
sc->count_high = sc->count_low = 0;
}
void ripemd160_8way( ripemd160_8way_context *sc, const void *data, size_t len )
void ripemd160_8way_update( ripemd160_8way_context *sc, const void *data,
size_t len )
{
__m256i *vdata = (__m256i*)data;
size_t ptr;
@@ -859,7 +861,7 @@ void ripemd160_16way_init( ripemd160_16way_context *sc )
sc->count_high = sc->count_low = 0;
}
void ripemd160_16way( ripemd160_16way_context *sc, const void *data,
void ripemd160_16way_update( ripemd160_16way_context *sc, const void *data,
size_t len )
{
__m512i *vdata = (__m512i*)data;

10
algo/ripemd/ripemd-hash-4way.h
View File

@@ -16,7 +16,8 @@ typedef struct
} __attribute__ ((aligned (64))) ripemd160_4way_context;
void ripemd160_4way_init( ripemd160_4way_context *sc );
void ripemd160_4way( ripemd160_4way_context *sc, const void *data, size_t len );
void ripemd160_4way_update( ripemd160_4way_context *sc, const void *data,
size_t len );
void ripemd160_4way_close( ripemd160_4way_context *sc, void *dst );
#if defined (__AVX2__)
@@ -26,10 +27,11 @@ typedef struct
__m256i buf[64>>2];
__m256i val[5];
uint32_t count_high, count_low;
} __attribute__ ((aligned (64))) ripemd160_8way_context;
} __attribute__ ((aligned (128))) ripemd160_8way_context;
void ripemd160_8way_init( ripemd160_8way_context *sc );
void ripemd160_8way( ripemd160_8way_context *sc, const void *data, size_t len );
void ripemd160_8way_update( ripemd160_8way_context *sc, const void *data,
size_t len );
void ripemd160_8way_close( ripemd160_8way_context *sc, void *dst );
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
@@ -42,7 +44,7 @@ typedef struct
} __attribute__ ((aligned (128))) ripemd160_16way_context;
void ripemd160_16way_init( ripemd160_16way_context *sc );
void ripemd160_16way( ripemd160_16way_context *sc, const void *data,
void ripemd160_16way_update( ripemd160_16way_context *sc, const void *data,
size_t len );
void ripemd160_16way_close( ripemd160_16way_context *sc, void *dst );