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

v3.23.1

Browse Source
This commit is contained in:
Jay D Dee
2023-09-13 11:48:52 -04:00
parent 4378d2f841
commit d6b5750362
28 changed files with 1626 additions and 1327 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
algo/sha/sha-hash-4way.h
View File

@@ -67,7 +67,7 @@ void sha256_4way_prehash_3rounds( __m128i *state_mid, __m128i *X,
void sha256_4way_final_rounds( __m128i *state_out, const __m128i *data,
const __m128i *state_in, const __m128i *state_mid, const __m128i *X );
int sha256_4way_transform_le_short( __m128i *state_out, const __m128i *data,
const __m128i *state_in );
const __m128i *state_in, const uint32_t *target );
#endif // SSE2
@@ -95,7 +95,7 @@ void sha256_8way_prehash_3rounds( __m256i *state_mid, __m256i *X,
void sha256_8way_final_rounds( __m256i *state_out, const __m256i *data,
const __m256i *state_in, const __m256i *state_mid, const __m256i *X );
int sha256_8way_transform_le_short( __m256i *state_out, const __m256i *data,
const __m256i *state_in );
const __m256i *state_in, const uint32_t *target );
#endif // AVX2
@@ -123,7 +123,7 @@ void sha256_16way_final_rounds( __m512i *state_out, const __m512i *data,
const __m512i *state_in, const __m512i *state_mid, const __m512i *X );
int sha256_16way_transform_le_short( __m512i *state_out, const __m512i *data,
const __m512i *state_in );
const __m512i *state_in, const uint32_t *target );
#endif // AVX512

35
algo/sha/sha2.c
View File

@@ -658,43 +658,14 @@ int scanhash_sha256d_pooler( struct work *work, uint32_t max_nonce,
return 0;
}
/*
int scanhash_SHA256d( struct work *work, const uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t _ALIGN(128) hash[8];
uint32_t _ALIGN(64) data[20];
uint32_t *pdata = work->data;
const uint32_t *ptarget = work->target;
uint32_t n = pdata[19] - 1;
const uint32_t first_nonce = pdata[19];
const uint32_t Htarg = ptarget[7];
int thr_id = mythr->id;
memcpy( data, pdata, 80 );
do {
data[19] = ++n;
sha256d( (unsigned char*)hash, (const unsigned char*)data, 80 );
if ( unlikely( swab32( hash[7] ) <= Htarg ) )
{
pdata[19] = n;
sha256d_80_swap(hash, pdata);
if ( fulltest( hash, ptarget ) && !opt_benchmark )
submit_solution( work, hash, mythr );
}
} while ( likely( n < max_nonce && !work_restart[thr_id].restart ) );
*hashes_done = n - first_nonce + 1;
pdata[19] = n;
return 0;
}
*/
bool register_sha256d_algo( algo_gate_t* gate )
{
gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT;
#if defined(SHA256D_16WAY)
gate->scanhash = (void*)&scanhash_sha256d_16way;
#elif defined(SHA256D_SHA)
gate->optimizations = SHA_OPT;
gate->scanhash = (void*)&scanhash_sha256d_sha;
//#elif defined(SHA256D_8WAY)
// gate->scanhash = (void*)&scanhash_sha256d_8way;
#else

1028
algo/sha/sha256-hash-4way.c
View File

File diff suppressed because it is too large Load Diff

59
algo/sha/sha256-hash.c
View File

@@ -50,65 +50,6 @@ void sha256_update( sha256_context *ctx, const void *data, size_t len )
memcpy( ctx->buf, src, len );
}
#if 0
void sha256_final( sha256_context *ctx, uint32_t *hash )
{
size_t r;
/* Figure out how many bytes we have buffered. */
r = ctx->count & 0x3f;
// r = ( ctx->count >> 3 ) & 0x3f;
//printf("final: count= %d, r= %d\n", ctx->count, r );
/* Pad to 56 mod 64, transforming if we finish a block en route. */
if ( r < 56 )
{
/* Pad to 56 mod 64. */
memcpy( &ctx->buf[r], SHA256_PAD, 56 - r );
}
else
{
/* Finish the current block and mix. */
memcpy( &ctx->buf[r], SHA256_PAD, 64 - r );
sha256_transform_be( ctx->state, (uint32_t*)ctx->buf, ctx->state );
// SHA256_Transform(ctx->state, ctx->buf, &tmp32[0], &tmp32[64]);
/* The start of the final block is all zeroes. */
memset( &ctx->buf[0], 0, 56 );
}
/* Add the terminating bit-count. */
ctx->buf[56] = bswap_64( ctx->count << 3 );
// ctx->buf[56] = bswap_64( ctx->count );
// be64enc( &ctx->buf[56], ctx->count );
/* Mix in the final block. */
sha256_transform_be( ctx->state, (uint32_t*)ctx->buf, ctx->state );
// SHA256_Transform(ctx->state, ctx->buf, &tmp32[0], &tmp32[64]);
for ( int i = 0; i < 8; i++ ) hash[i] = bswap_32( ctx->state[i] );
// for ( int i = 0; i < 8; i++ ) be32enc( hash + 4*i, ctx->state + i );
/*
// be32enc_vect(digest, ctx->state, 4);
// be32enc_vect(uint8_t * dst, const uint32_t * src, size_t len)
// Encode vector, two words at a time.
do {
be32enc(&dst[0], src[0]);
be32enc(&dst[4], src[1]);
src += 2;
dst += 8;
} while (--len);
*/
}
#endif
void sha256_final( sha256_context *ctx, void *hash )
{
int ptr = ctx->count & 0x3f;

280
algo/sha/sha256d-4way.c
View File

@@ -3,10 +3,194 @@
#include <stdint.h>
#include <string.h>
#include <stdio.h>
#include "sha256-hash.h"
#include "sha-hash-4way.h"
static const uint32_t sha256_iv[8] __attribute__ ((aligned (32))) =
{
0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A,
0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19
};
#if defined(SHA256D_SHA)
int scanhash_sha256d_sha( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t block0[16] __attribute__ ((aligned (64)));
uint32_t block1[16] __attribute__ ((aligned (64)));
uint32_t hash0[8] __attribute__ ((aligned (32)));
uint32_t hash1[8] __attribute__ ((aligned (32)));
uint32_t mstate[8] __attribute__ ((aligned (32)));
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 2;
uint32_t n = first_nonce;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
const __m128i shuf_bswap32 =
_mm_set_epi64x( 0x0c0d0e0f08090a0bULL, 0x0405060700010203ULL );
// hash first 64 bytes of data
sha256_opt_transform_le( mstate, pdata, sha256_iv );
do
{
// 1. final 16 bytes of data, with padding
memcpy( block0, pdata + 16, 16 );
memcpy( block1, pdata + 16, 16 );
block0[ 3] = n;
block1[ 3] = n+1;
block0[ 4] = block1[ 4] = 0x80000000;
memset( block0 + 5, 0, 40 );
memset( block1 + 5, 0, 40 );
block0[15] = block1[15] = 80*8; // bit count
sha256_ni2way_transform_le( hash0, hash1, block0, block1,
mstate, mstate );
// 2. 32 byte hash from 1.
memcpy( block0, hash0, 32 );
memcpy( block1, hash1, 32 );
block0[ 8] = block1[ 8] = 0x80000000;
memset( block0 + 9, 0, 24 );
memset( block1 + 9, 0, 24 );
block0[15] = block1[15] = 32*8; // bit count
sha256_ni2way_transform_le( hash0, hash1, block0, block1,
sha256_iv, sha256_iv );
if ( unlikely( bswap_32( hash0[7] ) <= ptarget[7] ) )
{
casti_m128i( hash0, 0 ) =
_mm_shuffle_epi8( casti_m128i( hash0, 0 ), shuf_bswap32 );
casti_m128i( hash0, 1 ) =
_mm_shuffle_epi8( casti_m128i( hash0, 1 ), shuf_bswap32 );
if ( likely( valid_hash( hash0, ptarget ) && !bench ) )
{
pdata[19] = n;
submit_solution( work, hash0, mythr );
}
}
if ( unlikely( bswap_32( hash1[7] ) <= ptarget[7] ) )
{
casti_m128i( hash1, 0 ) =
_mm_shuffle_epi8( casti_m128i( hash1, 0 ), shuf_bswap32 );
casti_m128i( hash1, 1 ) =
_mm_shuffle_epi8( casti_m128i( hash1, 1 ), shuf_bswap32 );
if ( likely( valid_hash( hash1, ptarget ) && !bench ) )
{
pdata[19] = n+1;
submit_solution( work, hash1, mythr );
}
}
n += 2;
} while ( (n < last_nonce) && !work_restart[thr_id].restart );
pdata[19] = n;
*hashes_done = n - first_nonce;
return 0;
}
#endif
#if defined(SHA256D_16WAY)
int scanhash_sha256d_16way( struct work *work, const uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
__m512i hash32[8] __attribute__ ((aligned (128)));
__m512i block[16] __attribute__ ((aligned (64)));
__m512i buf[16] __attribute__ ((aligned (64)));
__m512i mstate1[8] __attribute__ ((aligned (64)));
__m512i mstate2[8] __attribute__ ((aligned (64)));
__m512i istate[8] __attribute__ ((aligned (64)));
__m512i mexp_pre[8] __attribute__ ((aligned (64)));
uint32_t phash[8] __attribute__ ((aligned (32)));
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t *hash32_d7 = (uint32_t*)&(hash32[7]);
const uint32_t targ32_d7 = ptarget[7];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 16;
const __m512i last_byte = _mm512_set1_epi32( 0x80000000 );
uint32_t n = first_nonce;
const int thr_id = mythr->id;
const __m512i sixteen = _mm512_set1_epi32( 16 );
const bool bench = opt_benchmark;
const __m256i bswap_shuf = mm256_bcast_m128( _mm_set_epi64x(
0x0c0d0e0f08090a0b, 0x0405060700010203 ) );
// prehash first block directly from pdata
sha256_transform_le( phash, pdata, sha256_iv );
// vectorize block 0 hash for second block
mstate1[0] = _mm512_set1_epi32( phash[0] );
mstate1[1] = _mm512_set1_epi32( phash[1] );
mstate1[2] = _mm512_set1_epi32( phash[2] );
mstate1[3] = _mm512_set1_epi32( phash[3] );
mstate1[4] = _mm512_set1_epi32( phash[4] );
mstate1[5] = _mm512_set1_epi32( phash[5] );
mstate1[6] = _mm512_set1_epi32( phash[6] );
mstate1[7] = _mm512_set1_epi32( phash[7] );
// second message block data, with nonce & padding
buf[0] = _mm512_set1_epi32( pdata[16] );
buf[1] = _mm512_set1_epi32( pdata[17] );
buf[2] = _mm512_set1_epi32( pdata[18] );
buf[3] = _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 );
buf[4] = last_byte;
memset_zero_512( buf+5, 10 );
buf[15] = _mm512_set1_epi32( 80*8 ); // bit count
// partially pre-expand & prehash second message block, avoiding the nonces
sha256_16way_prehash_3rounds( mstate2, mexp_pre, buf, mstate1 );
// vectorize IV for 2nd & 3rd sha256
istate[0] = _mm512_set1_epi32( sha256_iv[0] );
istate[1] = _mm512_set1_epi32( sha256_iv[1] );
istate[2] = _mm512_set1_epi32( sha256_iv[2] );
istate[3] = _mm512_set1_epi32( sha256_iv[3] );
istate[4] = _mm512_set1_epi32( sha256_iv[4] );
istate[5] = _mm512_set1_epi32( sha256_iv[5] );
istate[6] = _mm512_set1_epi32( sha256_iv[6] );
istate[7] = _mm512_set1_epi32( sha256_iv[7] );
// initialize padding for 2nd sha256
block[ 8] = last_byte;
memset_zero_512( block + 9, 6 );
block[15] = _mm512_set1_epi32( 32*8 ); // bit count
do
{
sha256_16way_final_rounds( block, buf, mstate1, mstate2, mexp_pre );
if ( sha256_16way_transform_le_short( hash32, block, istate, ptarget ) )
{
for ( int lane = 0; lane < 16; lane++ )
if ( bswap_32( hash32_d7[ lane ] ) <= targ32_d7 )
{
extr_lane_16x32( phash, hash32, lane, 256 );
casti_m256i( phash, 0 ) =
_mm256_shuffle_epi8( casti_m256i( phash, 0 ), bswap_shuf );
if ( likely( valid_hash( phash, ptarget ) && !bench ) )
{
pdata[19] = n + lane;
submit_solution( work, phash, mythr );
}
}
}
buf[3] = _mm512_add_epi32( buf[3], sixteen );
n += 16;
} while ( (n < last_nonce) && !work_restart[thr_id].restart );
pdata[19] = n;
*hashes_done = n - first_nonce;
return 0;
}
/*
int scanhash_sha256d_16way( struct work *work, const uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
@@ -67,20 +251,18 @@ int scanhash_sha256d_16way( struct work *work, const uint32_t max_nonce,
mexp_pre );
// 2. 32 byte hash from 1.
if ( sha256_16way_transform_le_short( hash32, block, initstate ) )
{
// byte swap final hash for testing
mm512_block_bswap_32( hash32, hash32 );
sha256_16way_transform_le( hash32, block, initstate );
// byte swap final hash for testing
mm512_block_bswap_32( hash32, hash32 );
for ( int lane = 0; lane < 16; lane++ )
if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
for ( int lane = 0; lane < 16; lane++ )
if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
{
extr_lane_16x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
{
extr_lane_16x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
{
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
}
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
}
}
*noncev = _mm512_add_epi32( *noncev, sixteen );
@@ -90,6 +272,7 @@ int scanhash_sha256d_16way( struct work *work, const uint32_t max_nonce,
*hashes_done = n - first_nonce;
return 0;
}
*/
#endif
@@ -104,7 +287,7 @@ int scanhash_sha256d_8way( struct work *work, const uint32_t max_nonce,
__m256i initstate[8] __attribute__ ((aligned (32)));
__m256i midstate1[8] __attribute__ ((aligned (32)));
__m256i midstate2[8] __attribute__ ((aligned (32)));
__m256i mexp_pre[16] __attribute__ ((aligned (32)));
__m256i mexp_pre[8] __attribute__ ((aligned (32)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *hash32_d7 = (uint32_t*)&( hash32[7] );
uint32_t *pdata = work->data;
@@ -154,21 +337,18 @@ int scanhash_sha256d_8way( struct work *work, const uint32_t max_nonce,
mexp_pre );
// 2. 32 byte hash from 1.
if ( unlikely(
sha256_8way_transform_le_short( hash32, block, initstate ) ) )
{
// byte swap final hash for testing
mm256_block_bswap_32( hash32, hash32 );
sha256_8way_transform_le( hash32, block, initstate );
// byte swap final hash for testing
mm256_block_bswap_32( hash32, hash32 );
for ( int lane = 0; lane < 8; lane++ )
if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
for ( int lane = 0; lane < 8; lane++ )
if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
{
extr_lane_8x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
{
extr_lane_8x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
{
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
}
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
}
}
*noncev = _mm256_add_epi32( *noncev, eight );
@@ -191,8 +371,6 @@ int scanhash_sha256d_4way( struct work *work, const uint32_t max_nonce,
__m128i hash32[8] __attribute__ ((aligned (32)));
__m128i initstate[8] __attribute__ ((aligned (32)));
__m128i midstate1[8] __attribute__ ((aligned (32)));
__m128i midstate2[8] __attribute__ ((aligned (32)));
__m128i mexp_pre[16] __attribute__ ((aligned (32)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *hash32_d7 = (uint32_t*)&( hash32[7] );
uint32_t *pdata = work->data;
@@ -232,31 +410,25 @@ int scanhash_sha256d_4way( struct work *work, const uint32_t max_nonce,
// hash first 64 bytes of data
sha256_4way_transform_le( midstate1, vdata, initstate );
// Do 3 rounds on the first 12 bytes of the next block
sha256_4way_prehash_3rounds( midstate2, mexp_pre, vdata + 16, midstate1 );
do
{
// 1. final 16 bytes of data, with padding
sha256_4way_final_rounds( block, vdata+16, midstate1, midstate2,
mexp_pre );
sha256_4way_transform_le( block, vdata+16, initstate );
// 2. 32 byte hash from 1.
if ( unlikely(
sha256_4way_transform_le_short( hash32, block, initstate ) ) )
{
// byte swap final hash for testing
mm128_block_bswap_32( hash32, hash32 );
sha256_4way_transform_le( hash32, block, initstate );
// byte swap final hash for testing
mm128_block_bswap_32( hash32, hash32 );
for ( int lane = 0; lane < 4; lane++ )
if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
for ( int lane = 0; lane < 4; lane++ )
if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
{
extr_lane_4x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
{
extr_lane_4x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
{
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
}
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
}
}
*noncev = _mm_add_epi32( *noncev, four );
@@ -268,21 +440,3 @@ int scanhash_sha256d_4way( struct work *work, const uint32_t max_nonce,
}
#endif
/*
bool register_sha256d_algo( algo_gate_t* gate )
{
gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT;
#if defined(SHA256D_16WAY)
gate->scanhash = (void*)&scanhash_sha256d_16way;
#elif defined(SHA256D_8WAY)
gate->scanhash = (void*)&scanhash_sha256d_8way;
#elif defined(SHA256D_4WAY)
gate->scanhash = (void*)&scanhash_sha256d_4way;
#endif
// gate->hash = (void*)&sha256d;
return true;
};
*/

15
algo/sha/sha256d-4way.h
View File

@@ -6,6 +6,8 @@
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
#define SHA256D_16WAY 1
#elif defined(__SHA__)
#define SHA256D_SHA 1
#elif defined(__AVX2__)
#define SHA256D_8WAY 1
#else
@@ -32,15 +34,12 @@ int scanhash_sha256d_4way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr );
#endif
#if defined(SHA256D_SHA)
/*
#if defined(__SHA__)
int scanhash_sha256d( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr );
#endif
*/
int scanhash_sha256d_sha( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr );
#endif
#endif

299
algo/sha/sha256dt.c
View File

@@ -3,99 +3,201 @@
#include <stdint.h>
#include <string.h>
#include <stdio.h>
#include "sha256-hash.h"
#include "sha-hash-4way.h"
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
#define SHA256DT_16WAY 1
#elif defined(__SHA__)
#define SHA256DT_SHA 1
#elif defined(__AVX2__)
#define SHA256DT_8WAY 1
#else
#define SHA256DT_4WAY 1
#endif
static const uint32_t sha256dt_iv[8] __attribute__ ((aligned (32))) =
{
0xdfa9bf2c, 0xb72074d4, 0x6bb01122, 0xd338e869,
0xaa3ff126, 0x475bbf30, 0x8fd52e5b, 0x9f75c9ad
};
#if defined(SHA256DT_16WAY)
int scanhash_sha256dt_16way( struct work *work, const uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
__m512i vdata[32] __attribute__ ((aligned (128)));
__m512i hash32[8] __attribute__ ((aligned (128)));
__m512i block[16] __attribute__ ((aligned (64)));
__m512i hash32[8] __attribute__ ((aligned (64)));
__m512i initstate[8] __attribute__ ((aligned (64)));
__m512i midstate1[8] __attribute__ ((aligned (64)));
__m512i midstate2[8] __attribute__ ((aligned (64)));
__m512i mexp_pre[16] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (64)));
uint32_t *hash32_d7 = (uint32_t*)&( hash32[7] );
__m512i buf[16] __attribute__ ((aligned (64)));
__m512i mstate1[8] __attribute__ ((aligned (64)));
__m512i mstate2[8] __attribute__ ((aligned (64)));
__m512i istate[8] __attribute__ ((aligned (64)));
__m512i mexp_pre[8] __attribute__ ((aligned (64)));
uint32_t phash[8] __attribute__ ((aligned (32)));
uint32_t *pdata = work->data;
const uint32_t *ptarget = work->target;
const uint32_t targ32_d7 = ptarget[7];
// uint32_t *hash32_d7 = (uint32_t*)&(hash32[7]);
// const uint32_t targ32_d7 = ptarget[7];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 16;
uint32_t n = first_nonce;
__m512i *noncev = vdata + 19;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
const __m512i last_byte = _mm512_set1_epi32( 0x80000000 );
uint32_t n = first_nonce;
const int thr_id = mythr->id;
const __m512i sixteen = _mm512_set1_epi32( 16 );
const bool bench = opt_benchmark;
const __m256i bswap_shuf = mm256_bcast_m128( _mm_set_epi64x(
0x0c0d0e0f08090a0b, 0x0405060700010203 ) );
for ( int i = 0; i < 19; i++ )
vdata[i] = _mm512_set1_epi32( pdata[i] );
// prehash first block directly from pdata
sha256_transform_le( phash, pdata, sha256dt_iv );
*noncev = _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 );
// vectorize block 0 hash for second block
mstate1[0] = _mm512_set1_epi32( phash[0] );
mstate1[1] = _mm512_set1_epi32( phash[1] );
mstate1[2] = _mm512_set1_epi32( phash[2] );
mstate1[3] = _mm512_set1_epi32( phash[3] );
mstate1[4] = _mm512_set1_epi32( phash[4] );
mstate1[5] = _mm512_set1_epi32( phash[5] );
mstate1[6] = _mm512_set1_epi32( phash[6] );
mstate1[7] = _mm512_set1_epi32( phash[7] );
vdata[16+4] = last_byte;
memset_zero_512( vdata+16 + 5, 10 );
vdata[16+15] = _mm512_set1_epi32( 0x480 );
// second message block data, with nonce & padding
buf[0] = _mm512_set1_epi32( pdata[16] );
buf[1] = _mm512_set1_epi32( pdata[17] );
buf[2] = _mm512_set1_epi32( pdata[18] );
buf[3] = _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 );
buf[4] = last_byte;
memset_zero_512( buf+5, 10 );
buf[15] = _mm512_set1_epi32( 0x480 ); // sha256dt funky bit count
// partially pre-expand & prehash second message block, avoiding the nonces
sha256_16way_prehash_3rounds( mstate2, mexp_pre, buf, mstate1 );
// vectorize IV for 2nd sha256
istate[0] = _mm512_set1_epi32( sha256dt_iv[0] );
istate[1] = _mm512_set1_epi32( sha256dt_iv[1] );
istate[2] = _mm512_set1_epi32( sha256dt_iv[2] );
istate[3] = _mm512_set1_epi32( sha256dt_iv[3] );
istate[4] = _mm512_set1_epi32( sha256dt_iv[4] );
istate[5] = _mm512_set1_epi32( sha256dt_iv[5] );
istate[6] = _mm512_set1_epi32( sha256dt_iv[6] );
istate[7] = _mm512_set1_epi32( sha256dt_iv[7] );
// initialize padding for 2nd sha256
block[ 8] = last_byte;
memset_zero_512( block + 9, 6 );
block[15] = _mm512_set1_epi32( 0x300 );
initstate[0] = _mm512_set1_epi64( 0xdfa9bf2cdfa9bf2c );
initstate[1] = _mm512_set1_epi64( 0xb72074d4b72074d4 );
initstate[2] = _mm512_set1_epi64( 0x6bb011226bb01122 );
initstate[3] = _mm512_set1_epi64( 0xd338e869d338e869 );
initstate[4] = _mm512_set1_epi64( 0xaa3ff126aa3ff126 );
initstate[5] = _mm512_set1_epi64( 0x475bbf30475bbf30 );
initstate[6] = _mm512_set1_epi64( 0x8fd52e5b8fd52e5b );
initstate[7] = _mm512_set1_epi64( 0x9f75c9ad9f75c9ad );
sha256_16way_transform_le( midstate1, vdata, initstate );
// Do 3 rounds on the first 12 bytes of the next block
sha256_16way_prehash_3rounds( midstate2, mexp_pre, vdata+16, midstate1 );
memset_zero_512( block+9, 6 );
block[15] = _mm512_set1_epi32( 0x300 ); // bit count
do
{
sha256_16way_final_rounds( block, vdata+16, midstate1, midstate2,
mexp_pre );
sha256_16way_transform_le( hash32, block, initstate );
mm512_block_bswap_32( hash32, hash32 );
for ( int lane = 0; lane < 16; lane++ )
if ( hash32_d7[ lane ] <= targ32_d7 )
// finish second block with nonces
sha256_16way_final_rounds( block, buf, mstate1, mstate2, mexp_pre );
if ( unlikely( sha256_16way_transform_le_short(
hash32, block, istate, ptarget ) ) )
{
extr_lane_16x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
for ( int lane = 0; lane < 16; lane++ )
// if ( bswap_32( hash32_d7[ lane ] ) <= targ32_d7 )
{
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
extr_lane_16x32( phash, hash32, lane, 256 );
casti_m256i( phash, 0 ) =
_mm256_shuffle_epi8( casti_m256i( phash, 0 ), bswap_shuf );
if ( likely( valid_hash( phash, ptarget ) && !bench ) )
{
pdata[19] = n + lane;
submit_solution( work, phash, mythr );
}
}
}
*noncev = _mm512_add_epi32( *noncev, sixteen );
buf[3] = _mm512_add_epi32( buf[3], sixteen );
n += 16;
} while ( (n < last_nonce) && !work_restart[thr_id].restart );
pdata[19] = n;
*hashes_done = n - first_nonce;
return 0;
}
#elif defined(SHA256DT_SHA)
int scanhash_sha256dt_sha( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t block0[16] __attribute__ ((aligned (64)));
uint32_t block1[16] __attribute__ ((aligned (64)));
uint32_t hash0[8] __attribute__ ((aligned (32)));
uint32_t hash1[8] __attribute__ ((aligned (32)));
uint32_t mstate[8] __attribute__ ((aligned (32)));
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 2;
uint32_t n = first_nonce;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
const __m128i shuf_bswap32 =
_mm_set_epi64x( 0x0c0d0e0f08090a0bULL, 0x0405060700010203ULL );
#endif
// hash first 64 bytes of data
sha256_opt_transform_le( mstate, pdata, sha256dt_iv );
#if defined(SHA256DT_8WAY)
do
{
// 1. final 16 bytes of data, with padding
memcpy( block0, pdata + 16, 16 );
memcpy( block1, pdata + 16, 16 );
block0[ 3] = n;
block1[ 3] = n+1;
block0[ 4] = block1[ 4] = 0x80000000;
memset( block0 + 5, 0, 40 );
memset( block1 + 5, 0, 40 );
block0[15] = block1[15] = 0x480; // funky bit count
sha256_ni2way_transform_le( hash0, hash1, block0, block1,
mstate, mstate );
// 2. 32 byte hash from 1.
memcpy( block0, hash0, 32 );
memcpy( block1, hash1, 32 );
block0[ 8] = block1[ 8] = 0x80000000;
memset( block0 + 9, 0, 24 );
memset( block1 + 9, 0, 24 );
block0[15] = block1[15] = 0x300; // bit count
sha256_ni2way_transform_le( hash0, hash1, block0, block1,
sha256dt_iv, sha256dt_iv );
if ( unlikely( bswap_32( hash0[7] ) <= ptarget[7] ) )
{
casti_m128i( hash0, 0 ) =
_mm_shuffle_epi8( casti_m128i( hash0, 0 ), shuf_bswap32 );
casti_m128i( hash0, 1 ) =
_mm_shuffle_epi8( casti_m128i( hash0, 1 ), shuf_bswap32 );
if ( likely( valid_hash( hash0, ptarget ) && !bench ) )
{
pdata[19] = n;
submit_solution( work, hash0, mythr );
}
}
if ( unlikely( bswap_32( hash1[7] ) <= ptarget[7] ) )
{
casti_m128i( hash1, 0 ) =
_mm_shuffle_epi8( casti_m128i( hash1, 0 ), shuf_bswap32 );
casti_m128i( hash1, 1 ) =
_mm_shuffle_epi8( casti_m128i( hash1, 1 ), shuf_bswap32 );
if ( likely( valid_hash( hash1, ptarget ) && !bench ) )
{
pdata[19] = n+1;
submit_solution( work, hash1, mythr );
}
}
n += 2;
} while ( (n < last_nonce) && !work_restart[thr_id].restart );
pdata[19] = n;
*hashes_done = n - first_nonce;
return 0;
}
#elif defined(SHA256DT_8WAY)
int scanhash_sha256dt_8way( struct work *work, const uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
@@ -103,15 +205,13 @@ int scanhash_sha256dt_8way( struct work *work, const uint32_t max_nonce,
__m256i vdata[32] __attribute__ ((aligned (64)));
__m256i block[16] __attribute__ ((aligned (32)));
__m256i hash32[8] __attribute__ ((aligned (32)));
__m256i initstate[8] __attribute__ ((aligned (32)));
__m256i midstate1[8] __attribute__ ((aligned (32)));
__m256i midstate2[8] __attribute__ ((aligned (32)));
__m256i mexp_pre[16] __attribute__ ((aligned (32)));
__m256i istate[8] __attribute__ ((aligned (32)));
__m256i mstate1[8] __attribute__ ((aligned (32)));
__m256i mstate2[8] __attribute__ ((aligned (32)));
__m256i mexp_pre[8] __attribute__ ((aligned (32)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *hash32_d7 = (uint32_t*)&( hash32[7] );
uint32_t *pdata = work->data;
const uint32_t *ptarget = work->target;
const uint32_t targ32_d7 = ptarget[7];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 8;
uint32_t n = first_nonce;
@@ -120,6 +220,8 @@ int scanhash_sha256dt_8way( struct work *work, const uint32_t max_nonce,
const bool bench = opt_benchmark;
const __m256i last_byte = _mm256_set1_epi32( 0x80000000 );
const __m256i eight = _mm256_set1_epi32( 8 );
const __m256i bswap_shuf = mm256_bcast_m128( _mm_set_epi64x(
0x0c0d0e0f08090a0b, 0x0405060700010203 ) );
for ( int i = 0; i < 19; i++ )
vdata[i] = _mm256_set1_epi32( pdata[i] );
@@ -135,35 +237,38 @@ int scanhash_sha256dt_8way( struct work *work, const uint32_t max_nonce,
block[15] = _mm256_set1_epi32( 0x300 );
// initialize state
initstate[0] = _mm256_set1_epi64x( 0xdfa9bf2cdfa9bf2c );
initstate[1] = _mm256_set1_epi64x( 0xb72074d4b72074d4 );
initstate[2] = _mm256_set1_epi64x( 0x6bb011226bb01122 );
initstate[3] = _mm256_set1_epi64x( 0xd338e869d338e869 );
initstate[4] = _mm256_set1_epi64x( 0xaa3ff126aa3ff126 );
initstate[5] = _mm256_set1_epi64x( 0x475bbf30475bbf30 );
initstate[6] = _mm256_set1_epi64x( 0x8fd52e5b8fd52e5b );
initstate[7] = _mm256_set1_epi64x( 0x9f75c9ad9f75c9ad );
istate[0] = _mm256_set1_epi64x( 0xdfa9bf2cdfa9bf2c );
istate[1] = _mm256_set1_epi64x( 0xb72074d4b72074d4 );
istate[2] = _mm256_set1_epi64x( 0x6bb011226bb01122 );
istate[3] = _mm256_set1_epi64x( 0xd338e869d338e869 );
istate[4] = _mm256_set1_epi64x( 0xaa3ff126aa3ff126 );
istate[5] = _mm256_set1_epi64x( 0x475bbf30475bbf30 );
istate[6] = _mm256_set1_epi64x( 0x8fd52e5b8fd52e5b );
istate[7] = _mm256_set1_epi64x( 0x9f75c9ad9f75c9ad );
sha256_8way_transform_le( midstate1, vdata, initstate );
sha256_8way_transform_le( mstate1, vdata, istate );
// Do 3 rounds on the first 12 bytes of the next block
sha256_8way_prehash_3rounds( midstate2, mexp_pre, vdata + 16, midstate1 );
sha256_8way_prehash_3rounds( mstate2, mexp_pre, vdata + 16, mstate1 );
do
{
sha256_8way_final_rounds( block, vdata+16, midstate1, midstate2,
sha256_8way_final_rounds( block, vdata+16, mstate1, mstate2,
mexp_pre );
sha256_8way_transform_le( hash32, block, initstate );
mm256_block_bswap_32( hash32, hash32 );
for ( int lane = 0; lane < 8; lane++ )
if ( hash32_d7[ lane ] <= targ32_d7 )
if ( unlikely( sha256_8way_transform_le_short(
hash32, block, istate, ptarget ) ) )
{
extr_lane_8x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
for ( int lane = 0; lane < 8; lane++ )
{
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
extr_lane_8x32( lane_hash, hash32, lane, 256 );
casti_m256i( lane_hash, 0 ) =
_mm256_shuffle_epi8( casti_m256i( lane_hash, 0 ), bswap_shuf );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
{
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
}
}
}
*noncev = _mm256_add_epi32( *noncev, eight );
@@ -174,10 +279,7 @@ int scanhash_sha256dt_8way( struct work *work, const uint32_t max_nonce,
return 0;
}
#endif
#if defined(SHA256DT_4WAY)
#elif defined(SHA256DT_4WAY)
int scanhash_sha256dt_4way( struct work *work, const uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
@@ -230,21 +332,25 @@ int scanhash_sha256dt_4way( struct work *work, const uint32_t max_nonce,
do
{
sha256_4way_transform_le( block, vdata+16, midstate );
sha256_4way_transform_le( hash32, block, initstate );
mm128_block_bswap_32( hash32, hash32 );
sha256_4way_transform_le( hash32, block, initstate );
for ( int lane = 0; lane < 4; lane++ )
if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
{
extr_lane_4x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
// if ( sha256_4way_transform_le_short( hash32, block, initstate, ptarget ) )
// {
mm128_block_bswap_32( hash32, hash32 );
for ( int lane = 0; lane < 4; lane++ )
if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
{
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
extr_lane_4x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
{
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
}
}
}
*noncev = _mm_add_epi32( *noncev, four );
n += 4;
// }
*noncev = _mm_add_epi32( *noncev, four );
n += 4;
} while ( (n < last_nonce) && !work_restart[thr_id].restart );
pdata[19] = n;
*hashes_done = n - first_nonce;
@@ -257,11 +363,14 @@ bool register_sha256dt_algo( algo_gate_t* gate )
{
gate->optimizations = SSE2_OPT | AVX2_OPT | AVX512_OPT;
#if defined(SHA256DT_16WAY)
gate->scanhash = (void*)&scanhash_sha256dt_16way;
gate->scanhash = (void*)&scanhash_sha256dt_16way;
#elif defined(SHA256DT_SHA)
gate->optimizations = SHA_OPT;
gate->scanhash = (void*)&scanhash_sha256dt_sha;
#elif defined(SHA256DT_8WAY)
gate->scanhash = (void*)&scanhash_sha256dt_8way;
gate->scanhash = (void*)&scanhash_sha256dt_8way;
#else
gate->scanhash = (void*)&scanhash_sha256dt_4way;
gate->scanhash = (void*)&scanhash_sha256dt_4way;
#endif
return true;
}

322
algo/sha/sha256t-4way.c
View File

@@ -3,6 +3,7 @@
#include <stdint.h>
#include <string.h>
#include <stdio.h>
#include "sha256-hash.h"
#include "sha-hash-4way.h"
#if defined(SHA256T_16WAY)
@@ -10,83 +11,96 @@
int scanhash_sha256t_16way( struct work *work, const uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
__m512i vdata[32] __attribute__ ((aligned (128)));
__m512i hash32[8] __attribute__ ((aligned (128)));
__m512i block[16] __attribute__ ((aligned (64)));
__m512i hash32[8] __attribute__ ((aligned (64)));
__m512i initstate[8] __attribute__ ((aligned (64)));
__m512i midstate1[8] __attribute__ ((aligned (64)));
__m512i midstate2[8] __attribute__ ((aligned (64)));
__m512i mexp_pre[16] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (64)));
uint32_t *hash32_d7 = (uint32_t*)&( hash32[7] );
__m512i buf[16] __attribute__ ((aligned (64)));
__m512i mstate1[8] __attribute__ ((aligned (64)));
__m512i mstate2[8] __attribute__ ((aligned (64)));
__m512i istate[8] __attribute__ ((aligned (64)));
__m512i mexp_pre[8] __attribute__ ((aligned (64)));
uint32_t phash[8] __attribute__ ((aligned (32)));
static const uint32_t IV[8] __attribute__ ((aligned (32))) =
{
0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A,
0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19
};
uint32_t *pdata = work->data;
const uint32_t *ptarget = work->target;
uint32_t *ptarget = work->target;
uint32_t *hash32_d7 = (uint32_t*)&(hash32[7]);
const uint32_t targ32_d7 = ptarget[7];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 16;
uint32_t n = first_nonce;
__m512i *noncev = vdata + 19;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
const __m512i last_byte = _mm512_set1_epi32( 0x80000000 );
uint32_t n = first_nonce;
const int thr_id = mythr->id;
const __m512i sixteen = _mm512_set1_epi32( 16 );
const bool bench = opt_benchmark;
const __m256i bswap_shuf = mm256_bcast_m128( _mm_set_epi64x(
0x0c0d0e0f08090a0b, 0x0405060700010203 ) );
for ( int i = 0; i < 19; i++ )
vdata[i] = _mm512_set1_epi32( pdata[i] );
// prehash first block directly from pdata
sha256_transform_le( phash, pdata, IV );
*noncev = _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 );
// vectorize block 0 hash for second block
mstate1[0] = _mm512_set1_epi32( phash[0] );
mstate1[1] = _mm512_set1_epi32( phash[1] );
mstate1[2] = _mm512_set1_epi32( phash[2] );
mstate1[3] = _mm512_set1_epi32( phash[3] );
mstate1[4] = _mm512_set1_epi32( phash[4] );
mstate1[5] = _mm512_set1_epi32( phash[5] );
mstate1[6] = _mm512_set1_epi32( phash[6] );
mstate1[7] = _mm512_set1_epi32( phash[7] );
vdata[16+4] = last_byte;
memset_zero_512( vdata+16 + 5, 10 );
vdata[16+15] = _mm512_set1_epi32( 80*8 ); // bit count
// second message block data, with nonce & padding
buf[0] = _mm512_set1_epi32( pdata[16] );
buf[1] = _mm512_set1_epi32( pdata[17] );
buf[2] = _mm512_set1_epi32( pdata[18] );
buf[3] = _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 );
buf[4] = last_byte;
memset_zero_512( buf+5, 10 );
buf[15] = _mm512_set1_epi32( 80*8 ); // bit count
// partially pre-expand & prehash second message block, avoiding the nonces
sha256_16way_prehash_3rounds( mstate2, mexp_pre, buf, mstate1 );
// vectorize IV for 2nd & 3rd sha256
istate[0] = _mm512_set1_epi32( IV[0] );
istate[1] = _mm512_set1_epi32( IV[1] );
istate[2] = _mm512_set1_epi32( IV[2] );
istate[3] = _mm512_set1_epi32( IV[3] );
istate[4] = _mm512_set1_epi32( IV[4] );
istate[5] = _mm512_set1_epi32( IV[5] );
istate[6] = _mm512_set1_epi32( IV[6] );
istate[7] = _mm512_set1_epi32( IV[7] );
// initialize padding for 2nd & 3rd sha256
block[ 8] = last_byte;
memset_zero_512( block + 9, 6 );
block[15] = _mm512_set1_epi32( 32*8 ); // bit count
initstate[0] = _mm512_set1_epi64( 0x6A09E6676A09E667 );
initstate[1] = _mm512_set1_epi64( 0xBB67AE85BB67AE85 );
initstate[2] = _mm512_set1_epi64( 0x3C6EF3723C6EF372 );
initstate[3] = _mm512_set1_epi64( 0xA54FF53AA54FF53A );
initstate[4] = _mm512_set1_epi64( 0x510E527F510E527F );
initstate[5] = _mm512_set1_epi64( 0x9B05688C9B05688C );
initstate[6] = _mm512_set1_epi64( 0x1F83D9AB1F83D9AB );
initstate[7] = _mm512_set1_epi64( 0x5BE0CD195BE0CD19 );
sha256_16way_transform_le( midstate1, vdata, initstate );
// Do 3 rounds on the first 12 bytes of the next block
sha256_16way_prehash_3rounds( midstate2, mexp_pre, vdata+16, midstate1 );
do
{
// 1. final 16 bytes of data, pre-padded
sha256_16way_final_rounds( block, vdata+16, midstate1, midstate2,
mexp_pre );
sha256_16way_final_rounds( block, buf, mstate1, mstate2, mexp_pre );
// 2. 32 byte hash from 1.
sha256_16way_transform_le( block, block, initstate );
sha256_16way_transform_le( block, block, istate );
// 3. 32 byte hash from 2.
if ( unlikely(
sha256_16way_transform_le_short( hash32, block, initstate ) ) )
if ( sha256_16way_transform_le_short( hash32, block, istate, ptarget ) )
{
// byte swap final hash for testing
mm512_block_bswap_32( hash32, hash32 );
for ( int lane = 0; lane < 16; lane++ )
if ( hash32_d7[ lane ] <= targ32_d7 )
if ( bswap_32( hash32_d7[ lane ] ) <= targ32_d7 )
{
extr_lane_16x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
extr_lane_16x32( phash, hash32, lane, 256 );
casti_m256i( phash, 0 ) =
_mm256_shuffle_epi8( casti_m256i( phash, 0 ), bswap_shuf );
if ( likely( valid_hash( phash, ptarget ) && !bench ) )
{
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
submit_solution( work, phash, mythr );
}
}
}
*noncev = _mm512_add_epi32( *noncev, sixteen );
buf[3] = _mm512_add_epi32( buf[3], sixteen );
n += 16;
} while ( (n < last_nonce) && !work_restart[thr_id].restart );
pdata[19] = n;
@@ -94,26 +108,23 @@ int scanhash_sha256t_16way( struct work *work, const uint32_t max_nonce,
return 0;
}
#endif
#if defined(SHA256T_8WAY)
int scanhash_sha256t_8way( struct work *work, const uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
__m256i vdata[32] __attribute__ ((aligned (64)));
__m256i block[16] __attribute__ ((aligned (32)));
__m256i hash32[8] __attribute__ ((aligned (32)));
__m256i initstate[8] __attribute__ ((aligned (32)));
__m256i midstate1[8] __attribute__ ((aligned (32)));
__m256i midstate2[8] __attribute__ ((aligned (32)));
__m256i mexp_pre[16] __attribute__ ((aligned (32)));
__m256i istate[8] __attribute__ ((aligned (32)));
__m256i mstate1[8] __attribute__ ((aligned (32)));
__m256i mstate2[8] __attribute__ ((aligned (32)));
__m256i mexp_pre[8] __attribute__ ((aligned (32)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *hash32_d7 = (uint32_t*)&( hash32[7] );
uint32_t *pdata = work->data;
const uint32_t *ptarget = work->target;
const uint32_t targ32_d7 = ptarget[7];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 8;
uint32_t n = first_nonce;
@@ -122,6 +133,8 @@ int scanhash_sha256t_8way( struct work *work, const uint32_t max_nonce,
const bool bench = opt_benchmark;
const __m256i last_byte = _mm256_set1_epi32( 0x80000000 );
const __m256i eight = _mm256_set1_epi32( 8 );
const __m256i bswap_shuf = mm256_bcast_m128( _mm_set_epi64x(
0x0c0d0e0f08090a0b, 0x0405060700010203 ) );
for ( int i = 0; i < 19; i++ )
vdata[i] = _mm256_set1_epi32( pdata[i] );
@@ -135,42 +148,40 @@ int scanhash_sha256t_8way( struct work *work, const uint32_t max_nonce,
block[ 8] = last_byte;
memset_zero_256( block + 9, 6 );
block[15] = _mm256_set1_epi32( 32*8 ); // bit count
// initialize state
initstate[0] = _mm256_set1_epi64x( 0x6A09E6676A09E667 );
initstate[1] = _mm256_set1_epi64x( 0xBB67AE85BB67AE85 );
initstate[2] = _mm256_set1_epi64x( 0x3C6EF3723C6EF372 );
initstate[3] = _mm256_set1_epi64x( 0xA54FF53AA54FF53A );
initstate[4] = _mm256_set1_epi64x( 0x510E527F510E527F );
initstate[5] = _mm256_set1_epi64x( 0x9B05688C9B05688C );
initstate[6] = _mm256_set1_epi64x( 0x1F83D9AB1F83D9AB );
initstate[7] = _mm256_set1_epi64x( 0x5BE0CD195BE0CD19 );
sha256_8way_transform_le( midstate1, vdata, initstate );
// initialize state
istate[0] = _mm256_set1_epi64x( 0x6A09E6676A09E667 );
istate[1] = _mm256_set1_epi64x( 0xBB67AE85BB67AE85 );
istate[2] = _mm256_set1_epi64x( 0x3C6EF3723C6EF372 );
istate[3] = _mm256_set1_epi64x( 0xA54FF53AA54FF53A );
istate[4] = _mm256_set1_epi64x( 0x510E527F510E527F );
istate[5] = _mm256_set1_epi64x( 0x9B05688C9B05688C );
istate[6] = _mm256_set1_epi64x( 0x1F83D9AB1F83D9AB );
istate[7] = _mm256_set1_epi64x( 0x5BE0CD195BE0CD19 );
sha256_8way_transform_le( mstate1, vdata, istate );
// Do 3 rounds on the first 12 bytes of the next block
sha256_8way_prehash_3rounds( midstate2, mexp_pre, vdata + 16, midstate1 );
sha256_8way_prehash_3rounds( mstate2, mexp_pre, vdata + 16, mstate1 );
do
{
// 1. final 16 bytes of data, with padding
sha256_8way_final_rounds( block, vdata+16, midstate1, midstate2,
sha256_8way_final_rounds( block, vdata+16, mstate1, mstate2,
mexp_pre );
// 2. 32 byte hash from 1.
sha256_8way_transform_le( block, block, initstate );
sha256_8way_transform_le( block, block, istate );
// 3. 32 byte hash from 2.
if ( unlikely(
sha256_8way_transform_le_short( hash32, block, initstate ) ) )
if ( unlikely( sha256_8way_transform_le_short(
hash32, block, istate, ptarget ) ) )
{
// byte swap final hash for testing
mm256_block_bswap_32( hash32, hash32 );
for ( int lane = 0; lane < 8; lane++ )
if ( hash32_d7[ lane ] <= targ32_d7 )
{
extr_lane_8x32( lane_hash, hash32, lane, 256 );
casti_m256i( lane_hash, 0 ) =
_mm256_shuffle_epi8( casti_m256i( lane_hash, 0 ), bswap_shuf );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
{
pdata[19] = n + lane;
@@ -188,109 +199,18 @@ int scanhash_sha256t_8way( struct work *work, const uint32_t max_nonce,
#endif
#if defined(SHA256T_4WAY)
// Optimizations are slower with AVX/SSE2
// https://github.com/JayDDee/cpuminer-opt/issues/344
/*
int scanhash_sha256t_4way( struct work *work, const uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
__m128i vdata[32] __attribute__ ((aligned (64)));
__m128i block[16] __attribute__ ((aligned (32)));
__m128i hash32[8] __attribute__ ((aligned (32)));
__m128i initstate[8] __attribute__ ((aligned (32)));
__m128i midstate1[8] __attribute__ ((aligned (32)));
__m128i midstate2[8] __attribute__ ((aligned (32)));
__m128i mexp_pre[16] __attribute__ ((aligned (32)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *hash32_d7 = (uint32_t*)&( hash32[7] );
uint32_t *pdata = work->data;
const uint32_t *ptarget = work->target;
const uint32_t targ32_d7 = ptarget[7];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 4;
uint32_t n = first_nonce;
__m128i *noncev = vdata + 19;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
const __m128i last_byte = _mm_set1_epi32( 0x80000000 );
const __m128i four = _mm_set1_epi32( 4 );
for ( int i = 0; i < 19; i++ )
vdata[i] = _mm_set1_epi32( pdata[i] );
*noncev = _mm_set_epi32( n+ 3, n+ 2, n+1, n );
vdata[16+4] = last_byte;
memset_zero_128( vdata+16 + 5, 10 );
vdata[16+15] = _mm_set1_epi32( 80*8 ); // bit count
block[ 8] = last_byte;
memset_zero_128( block + 9, 6 );
block[15] = _mm_set1_epi32( 32*8 ); // bit count
// initialize state
initstate[0] = _mm_set1_epi64x( 0x6A09E6676A09E667 );
initstate[1] = _mm_set1_epi64x( 0xBB67AE85BB67AE85 );
initstate[2] = _mm_set1_epi64x( 0x3C6EF3723C6EF372 );
initstate[3] = _mm_set1_epi64x( 0xA54FF53AA54FF53A );
initstate[4] = _mm_set1_epi64x( 0x510E527F510E527F );
initstate[5] = _mm_set1_epi64x( 0x9B05688C9B05688C );
initstate[6] = _mm_set1_epi64x( 0x1F83D9AB1F83D9AB );
initstate[7] = _mm_set1_epi64x( 0x5BE0CD195BE0CD19 );
// hash first 64 bytes of data
sha256_4way_transform_le( midstate1, vdata, initstate );
// Do 3 rounds on the first 12 bytes of the next block
sha256_4way_prehash_3rounds( midstate2, mexp_pre, vdata + 16, midstate1 );
do
{
// 1. final 16 bytes of data, with padding
sha256_4way_final_rounds( block, vdata+16, midstate1, midstate2,
mexp_pre );
// 2. 32 byte hash from 1.
sha256_4way_transform_le( block, block, initstate );
// 3. 32 byte hash from 2.
if ( unlikely(
sha256_4way_transform_le_short( hash32, block, initstate ) ) )
{
// byte swap final hash for testing
mm128_block_bswap_32( hash32, hash32 );
for ( int lane = 0; lane < 4; lane++ )
if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
{
extr_lane_4x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
{
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
}
}
}
*noncev = _mm_add_epi32( *noncev, four );
n += 4;
} while ( (n < last_nonce) && !work_restart[thr_id].restart );
pdata[19] = n;
*hashes_done = n - first_nonce;
return 0;
}
*/
int scanhash_sha256t_4way( struct work *work, const uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
__m128i vdata[32] __attribute__ ((aligned (64)));
__m128i block[16] __attribute__ ((aligned (32)));
__m128i hash32[8] __attribute__ ((aligned (32)));
__m128i initstate[8] __attribute__ ((aligned (32)));
__m128i midstate[8] __attribute__ ((aligned (32)));
__m128i istate[8] __attribute__ ((aligned (32)));
__m128i mstate[8] __attribute__ ((aligned (32)));
// __m128i mstate2[8] __attribute__ ((aligned (32)));
// __m128i mexp_pre[8] __attribute__ ((aligned (32)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *hash32_d7 = (uint32_t*)&( hash32[7] );
uint32_t *pdata = work->data;
@@ -319,35 +239,44 @@ int scanhash_sha256t_4way( struct work *work, const uint32_t max_nonce,
block[15] = _mm_set1_epi32( 32*8 ); // bit count
// initialize state
initstate[0] = _mm_set1_epi64x( 0x6A09E6676A09E667 );
initstate[1] = _mm_set1_epi64x( 0xBB67AE85BB67AE85 );
initstate[2] = _mm_set1_epi64x( 0x3C6EF3723C6EF372 );
initstate[3] = _mm_set1_epi64x( 0xA54FF53AA54FF53A );
initstate[4] = _mm_set1_epi64x( 0x510E527F510E527F );
initstate[5] = _mm_set1_epi64x( 0x9B05688C9B05688C );
initstate[6] = _mm_set1_epi64x( 0x1F83D9AB1F83D9AB );
initstate[7] = _mm_set1_epi64x( 0x5BE0CD195BE0CD19 );
istate[0] = _mm_set1_epi64x( 0x6A09E6676A09E667 );
istate[1] = _mm_set1_epi64x( 0xBB67AE85BB67AE85 );
istate[2] = _mm_set1_epi64x( 0x3C6EF3723C6EF372 );
istate[3] = _mm_set1_epi64x( 0xA54FF53AA54FF53A );
istate[4] = _mm_set1_epi64x( 0x510E527F510E527F );
istate[5] = _mm_set1_epi64x( 0x9B05688C9B05688C );
istate[6] = _mm_set1_epi64x( 0x1F83D9AB1F83D9AB );
istate[7] = _mm_set1_epi64x( 0x5BE0CD195BE0CD19 );
// hash first 64 bytes of data
sha256_4way_transform_le( midstate, vdata, initstate );
sha256_4way_transform_le( mstate, vdata, istate );
// sha256_4way_prehash_3rounds( mstate2, mexp_pre, vdata + 16, mstate1 );
do
{
sha256_4way_transform_le( block, vdata+16, midstate );
sha256_4way_transform_le( block, block, initstate );
sha256_4way_transform_le( hash32, block, initstate );
mm128_block_bswap_32( hash32, hash32 );
// sha256_4way_final_rounds( block, vdata+16, mstate1, mstate2,
// mexp_pre );
sha256_4way_transform_le( block, vdata+16, mstate );
sha256_4way_transform_le( block, block, istate );
sha256_4way_transform_le( hash32, block, istate );
for ( int lane = 0; lane < 4; lane++ )
if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
{
extr_lane_4x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
// if ( unlikely( sha256_4way_transform_le_short(
// hash32, block, initstate, ptarget ) ))
// {
mm128_block_bswap_32( hash32, hash32 );
for ( int lane = 0; lane < 4; lane++ )
if ( unlikely( hash32_d7[ lane ] <= targ32_d7 ) )
{
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
extr_lane_4x32( lane_hash, hash32, lane, 256 );
if ( likely( valid_hash( lane_hash, ptarget ) && !bench ) )
{
pdata[19] = n + lane;
submit_solution( work, lane_hash, mythr );
}
}
}
// }
*noncev = _mm_add_epi32( *noncev, four );
n += 4;
} while ( (n < last_nonce) && !work_restart[thr_id].restart );
@@ -356,6 +285,5 @@ int scanhash_sha256t_4way( struct work *work, const uint32_t max_nonce,
return 0;
}
#endif

2
algo/sha/sha256t.c
View File

@@ -23,7 +23,7 @@ int scanhash_sha256t( struct work *work, uint32_t max_nonce,
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 1;
const uint32_t last_nonce = max_nonce - 2;
uint32_t n = first_nonce;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;

4
algo/sha/sph_sha2.c
View File

@@ -39,9 +39,9 @@
#define SPH_SMALL_FOOTPRINT_SHA2 1
#endif
#define CH(X, Y, Z) ((((Y) ^ (Z)) & (X)) ^ (Z))
#define CH(X, Y, Z) ( ( ( (Y) ^ (Z) ) & (X)) ^ (Z) )
//#define MAJ(X, Y, Z) (((Y) & (Z)) | (((Y) | (Z)) & (X)))
#define MAJ( X, Y, Z ) ( Y ^ ( ( X_xor_Y = X ^ Y ) & ( Y_xor_Z ) ) )
#define MAJ( X, Y, Z ) ( (Y) ^ ( ( (X_xor_Y) = (X) ^ (Y) ) & (Y_xor_Z) ) )
#define ROTR SPH_ROTR32
#define BSG2_0(x) (ROTR(x, 2) ^ ROTR(x, 13) ^ ROTR(x, 22))