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

v3.23.4

Browse Source
This commit is contained in:
Jay D Dee
2023-10-06 22:18:09 -04:00
parent bc5a5c6df8
commit 31c4dedf59
144 changed files with 5931 additions and 3746 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
algo/lyra2/lyra2-gate.c
View File

@@ -1,5 +1,5 @@
#include "lyra2-gate.h"
#include <mm_malloc.h>
// huge pages
//

2
algo/lyra2/lyra2h-4way.c
View File

@@ -63,7 +63,7 @@ int scanhash_lyra2h_4way( struct work *work, uint32_t max_nonce,
if ( opt_benchmark )
ptarget[7] = 0x0000ff;
mm128_bswap32_intrlv80_4x32( vdata, pdata );
v128_bswap32_intrlv80_4x32( vdata, pdata );
lyra2h_4way_midstate( vdata );
do {

7
algo/lyra2/lyra2rev2-4way.c
View File

@@ -353,9 +353,6 @@ int scanhash_lyra2rev2_8way( struct work *work, const uint32_t max_nonce,
return 0;
}
#endif
/*
#elif defined (LYRA2REV2_4WAY)
typedef struct {
@@ -452,7 +449,7 @@ int scanhash_lyra2rev2_4way( struct work *work, uint32_t max_nonce,
if ( opt_benchmark )
( (uint32_t*)ptarget )[7] = 0x0000ff;
mm128_bswap32_intrlv80_4x32( vdata, pdata );
v128_bswap32_intrlv80_4x32( vdata, pdata );
blake256_4way_init( &l2v2_4way_ctx.blake );
blake256_4way_update( &l2v2_4way_ctx.blake, vdata, 64 );
@@ -480,4 +477,4 @@ int scanhash_lyra2rev2_4way( struct work *work, uint32_t max_nonce,
}
#endif
*/

2
algo/lyra2/lyra2rev3-4way.c
View File

@@ -371,7 +371,7 @@ int scanhash_lyra2rev3_4way( struct work *work, const uint32_t max_nonce,
if ( opt_benchmark )
( (uint32_t*)ptarget )[7] = 0x0000ff;
mm128_bswap32_intrlv80_4x32( vdata, pdata );
v128_bswap32_intrlv80_4x32( vdata, pdata );
*noncev = _mm_set_epi32( n+3, n+2, n+1, n );
blake256_4way_init( &l2v3_4way_ctx.blake );

10
algo/lyra2/lyra2rev3.c
View File

@@ -75,11 +75,11 @@ int scanhash_lyra2rev3( struct work *work,
((uint32_t*)ptarget)[7] = 0x0000ff;
// need big endian data
casti_m128i( endiandata, 0 ) = mm128_bswap_32( casti_m128i( pdata, 0 ) );
casti_m128i( endiandata, 1 ) = mm128_bswap_32( casti_m128i( pdata, 1 ) );
casti_m128i( endiandata, 2 ) = mm128_bswap_32( casti_m128i( pdata, 2 ) );
casti_m128i( endiandata, 3 ) = mm128_bswap_32( casti_m128i( pdata, 3 ) );
casti_m128i( endiandata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
casti_v128( endiandata, 0 ) = v128_bswap32( casti_v128( pdata, 0 ) );
casti_v128( endiandata, 1 ) = v128_bswap32( casti_v128( pdata, 1 ) );
casti_v128( endiandata, 2 ) = v128_bswap32( casti_v128( pdata, 2 ) );
casti_v128( endiandata, 3 ) = v128_bswap32( casti_v128( pdata, 3 ) );
casti_v128( endiandata, 4 ) = v128_bswap32( casti_v128( pdata, 4 ) );
l2v3_blake256_midstate( endiandata );
do
{

2
algo/lyra2/lyra2z-4way.c
View File

@@ -312,7 +312,7 @@ int scanhash_lyra2z_4way( struct work *work, uint32_t max_nonce,
if ( bench ) ptarget[7] = 0x0000ff;
mm128_bswap32_intrlv80_4x32( vdata, pdata );
v128_bswap32_intrlv80_4x32( vdata, pdata );
*noncev = _mm_set_epi32( n+3, n+2, n+1, n );
lyra2z_4way_midstate( vdata );

1
algo/lyra2/lyra2z.c
View File

@@ -53,7 +53,6 @@ int scanhash_lyra2z( struct work *work, uint32_t max_nonce,
uint32_t _ALIGN(64) endiandata[20];
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t nonce = first_nonce;
int thr_id = mythr->id;

11
algo/lyra2/lyra2z330.c
View File

@@ -2,6 +2,7 @@
#include "algo-gate-api.h"
#include "lyra2.h"
#include "simd-utils.h"
#include <mm_malloc.h>
static __thread uint64_t* lyra2z330_wholeMatrix;
@@ -29,11 +30,11 @@ int scanhash_lyra2z330( struct work *work, uint32_t max_nonce,
if (opt_benchmark)
ptarget[7] = 0x0000ff;
casti_m128i( edata, 0 ) = mm128_bswap_32( casti_m128i( pdata, 0 ) );
casti_m128i( edata, 1 ) = mm128_bswap_32( casti_m128i( pdata, 1 ) );
casti_m128i( edata, 2 ) = mm128_bswap_32( casti_m128i( pdata, 2 ) );
casti_m128i( edata, 3 ) = mm128_bswap_32( casti_m128i( pdata, 3 ) );
casti_m128i( edata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
casti_v128( edata, 0 ) = v128_bswap32( casti_v128( pdata, 0 ) );
casti_v128( edata, 1 ) = v128_bswap32( casti_v128( pdata, 1 ) );
casti_v128( edata, 2 ) = v128_bswap32( casti_v128( pdata, 2 ) );
casti_v128( edata, 3 ) = v128_bswap32( casti_v128( pdata, 3 ) );
casti_v128( edata, 4 ) = v128_bswap32( casti_v128( pdata, 4 ) );
do
{

2
algo/lyra2/sponge-2way.c
View File

@@ -23,9 +23,9 @@
#include <string.h>
#include <stdio.h>
#include <time.h>
#include <immintrin.h>
#include "sponge.h"
#include "lyra2.h"
#include "simd-utils.h"
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)

2
algo/lyra2/sponge.c
View File

@@ -22,7 +22,7 @@
#include <string.h>
#include <stdio.h>
#include <time.h>
#include <immintrin.h>
#include "simd-utils.h"
#include "sponge.h"
#include "lyra2.h"

4
algo/lyra2/sponge.h
View File

@@ -195,7 +195,7 @@ static const uint64_t blake2b_IV[8] =
#endif // AVX2 else SSE2
/*
// Scalar, not used.
static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
@@ -223,7 +223,7 @@ static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
G(r,5,v[ 1],v[ 6],v[11],v[12]); \
G(r,6,v[ 2],v[ 7],v[ 8],v[13]); \
G(r,7,v[ 3],v[ 4],v[ 9],v[14]);
*/
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)