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

v23.7

Browse Source
This commit is contained in:
Jay D Dee
2023-11-07 04:59:44 -05:00
parent 46dca7a493
commit e043698442
33 changed files with 3880 additions and 4763 deletions
Download Patch File Download Diff File Expand all files Collapse all files

76
algo/sha/sha512256d-4way.c
View File

@@ -5,9 +5,11 @@
#include <stdint.h>
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
#define SHA512256D_8WAY 1
#define SHA512256D_8WAY 1
#elif defined(__AVX2__)
#define SHA512256D_4WAY 1
#define SHA512256D_4WAY 1
#elif defined(__SSE2__) || defined(__ARM_NEON)
#define SHA512256D_2WAY 1
#endif
#if defined(SHA512256D_8WAY)
@@ -145,6 +147,74 @@ int scanhash_sha512256d_4way( struct work *work, uint32_t max_nonce,
return 0;
}
#elif defined(SHA512256D_2WAY)
static void sha512256d_2x64_init( sha512_2x64_context *ctx )
{
ctx->count = 0;
ctx->initialized = true;
ctx->val[0] = v128_64( 0x22312194FC2BF72C );
ctx->val[1] = v128_64( 0x9F555FA3C84C64C2 );
ctx->val[2] = v128_64( 0x2393B86B6F53B151 );
ctx->val[3] = v128_64( 0x963877195940EABD );
ctx->val[4] = v128_64( 0x96283EE2A88EFFE3 );
ctx->val[5] = v128_64( 0xBE5E1E2553863992 );
ctx->val[6] = v128_64( 0x2B0199FC2C85B8AA );
ctx->val[7] = v128_64( 0x0EB72DDC81C52CA2 );
}
int scanhash_sha512256d_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)));
sha512_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 );
*noncev = v128_add32( v128_set32( 1, 0, 0, 0 ), *noncev );
// *noncev = v128_intrlv_blend_32( v128_set32( n+1, 0, n, 0 ), *noncev );
do
{
sha512256d_2x64_init( &ctx );
sha512_2x64_update( &ctx, vdata, 80 );
sha512_2x64_close( &ctx, hash );
sha512256d_2x64_init( &ctx );
sha512_2x64_update( &ctx, hash, 32 );
sha512_2x64_close( &ctx, hash );
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;
}
#else
#include "sph_sha2.h"
@@ -214,6 +284,8 @@ bool register_sha512256d_algo( algo_gate_t* gate )
gate->scanhash = (void*)&scanhash_sha512256d_8way;
#elif defined(SHA512256D_4WAY)
gate->scanhash = (void*)&scanhash_sha512256d_4way;
#elif defined(SHA512256D_2WAY)
gate->scanhash = (void*)&scanhash_sha512256d_2x64;
#else
gate->scanhash = (void*)&scanhash_sha512256d;
#endif