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

v3.16.2

Browse Source
This commit is contained in:
Jay D Dee
2021-04-08 18:09:31 -04:00
parent 902ec046dd
commit f3333b0070
17 changed files with 826 additions and 336 deletions
Download Patch File Download Diff File Expand all files Collapse all files

83
algo/verthash/verthash-gate.c
View File

@@ -1,6 +1,7 @@
#include "algo-gate-api.h"
#include "algo/sha/sph_sha2.h"
#include "Verthash.h"
#include "tiny_sha3/sha3-4way.h"
static verthash_info_t verthashInfo;
@@ -12,6 +13,82 @@ static const uint8_t verthashDatFileHash_bytes[32] =
0x29, 0xec, 0xf8, 0x8f, 0x8a, 0xd4, 0x76, 0x39,
0xb6, 0xed, 0xed, 0xaf, 0xd7, 0x21, 0xaa, 0x48 };
#if defined(__AVX2__)
static __thread sha3_4way_ctx_t sha3_mid_ctxA;
static __thread sha3_4way_ctx_t sha3_mid_ctxB;
#else
static __thread sha3_ctx_t sha3_mid_ctx[8];
#endif
void verthash_sha3_512_prehash_72( const void *input )
{
#if defined(__AVX2__)
__m256i vin[10];
mm256_intrlv80_4x64( vin, input );
sha3_4way_init( &sha3_mid_ctxA, 64 );
sha3_4way_init( &sha3_mid_ctxB, 64 );
vin[0] = _mm256_add_epi8( vin[0], _mm256_set_epi64x( 4,3,2,1 ) );
sha3_4way_update( &sha3_mid_ctxA, vin, 72 );
vin[0] = _mm256_add_epi8( vin[0], _mm256_set1_epi64x( 4 ) );
sha3_4way_update( &sha3_mid_ctxB, vin, 72 );
#else
char in[80] __attribute__ ((aligned (64)));
memcpy( in, input, 80 );
for ( int i = 0; i < 8; i++ )
{
in[0] += 1;
sha3_init( &sha3_mid_ctx[i], 64 );
sha3_update( &sha3_mid_ctx[i], in, 72 );
}
#endif
}
void verthash_sha3_512_final_8( void *hash, const uint64_t nonce )
{
#if defined(__AVX2__)
__m256i vhashA[ 10 ] __attribute__ ((aligned (64)));
__m256i vhashB[ 10 ] __attribute__ ((aligned (64)));
sha3_4way_ctx_t ctx;
__m256i vnonce = _mm256_set1_epi64x( nonce );
memcpy( &ctx, &sha3_mid_ctxA, sizeof ctx );
sha3_4way_update( &ctx, &vnonce, 8 );
sha3_4way_final( vhashA, &ctx );
memcpy( &ctx, &sha3_mid_ctxB, sizeof ctx );
sha3_4way_update( &ctx, &vnonce, 8 );
sha3_4way_final( vhashB, &ctx );
dintrlv_4x64( hash, hash+64, hash+128, hash+192, vhashA, 512 );
dintrlv_4x64( hash+256, hash+320, hash+384, hash+448, vhashB, 512 );
#else
for ( int i = 0; i < 8; i++ )
{
sha3_ctx_t ctx;
memcpy( &ctx, &sha3_mid_ctx[i], sizeof ctx );
sha3_update( &ctx, &nonce, 8 );
sha3_final( hash + i*64, &ctx );
}
#endif
}
int scanhash_verthash( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
@@ -26,6 +103,8 @@ int scanhash_verthash( struct work *work, uint32_t max_nonce,
const bool bench = opt_benchmark;
mm128_bswap32_80( edata, pdata );
verthash_sha3_512_prehash_72( edata );
do
{
edata[19] = n;
@@ -51,15 +130,14 @@ bool register_verthash_algo( algo_gate_t* gate )
opt_target_factor = 256.0;
gate->scanhash = (void*)&scanhash_verthash;
gate->optimizations = AVX2_OPT;
// verthash data file
char *verthash_data_file = opt_data_file ? opt_data_file
: default_verthash_data_file;
int vhLoadResult = verthash_info_init( &verthashInfo, verthash_data_file );
if (vhLoadResult == 0) // No Error
{
// and verify data file(if it was enabled)
if ( opt_verify )
{
uint8_t vhDataFileHash[32] = { 0 };
@@ -78,7 +156,6 @@ bool register_verthash_algo( algo_gate_t* gate )
}
}
else
{
// Handle Verthash error codes
if ( vhLoadResult == 1 )