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

v3.23.2

Browse Source
This commit is contained in:
Jay D Dee
2023-09-21 12:34:06 -04:00
parent d6b5750362
commit be88afc349
113 changed files with 3349 additions and 2920 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -41,8 +41,6 @@ extern "C"{
#endif
#include <stddef.h>
#include "algo/sha/sph_types.h"
#include "simd-utils.h"
#define SPH_SIZE_bmw256 256
@@ -57,7 +55,7 @@ typedef struct {
__m128i buf[64];
__m128i H[16];
size_t ptr;
sph_u32 bit_count; // assume bit_count fits in 32 bits
uint32_t bit_count; // assume bit_count fits in 32 bits
} bmw_4way_small_context;
typedef bmw_4way_small_context bmw256_4way_context;
@@ -144,7 +142,7 @@ typedef struct {
__m256i buf[16];
__m256i H[16];
size_t ptr;
sph_u64 bit_count;
uint64_t bit_count;
} bmw_4way_big_context __attribute__((aligned(128)));
typedef bmw_4way_big_context bmw512_4way_context;

4
algo/bmw/bmw256-hash-4way.c
View File

@@ -109,7 +109,7 @@ static const uint32_t IV256[] = {
_mm_sub_epi32( _mm_add_epi32( rol_off_32( M, j, 0 ), \
rol_off_32( M, j, 3 ) ), \
rol_off_32( M, j, 10 ) ), \
_mm_set1_epi32( ( (j)+16 ) * SPH_C32(0x05555555UL) ) ), \
_mm_set1_epi32( ( (j)+16 ) * 0x05555555UL ) ), \
H[ ( (j)+7 ) & 0xF ] )
@@ -485,7 +485,7 @@ bmw32_4way(bmw_4way_small_context *sc, const void *data, size_t len)
size_t ptr;
const int buf_size = 64; // bytes of one lane, compatible with len
sc->bit_count += (sph_u32)len << 3;
sc->bit_count += (uint32_t)len << 3;
buf = sc->buf;
ptr = sc->ptr;
h1 = sc->H;

24
algo/bmw/bmw512-hash-4way.c
View File

@@ -45,15 +45,15 @@ extern "C"{
#define LPAR (
static const sph_u64 IV512[] = {
SPH_C64(0x8081828384858687), SPH_C64(0x88898A8B8C8D8E8F),
SPH_C64(0x9091929394959697), SPH_C64(0x98999A9B9C9D9E9F),
SPH_C64(0xA0A1A2A3A4A5A6A7), SPH_C64(0xA8A9AAABACADAEAF),
SPH_C64(0xB0B1B2B3B4B5B6B7), SPH_C64(0xB8B9BABBBCBDBEBF),
SPH_C64(0xC0C1C2C3C4C5C6C7), SPH_C64(0xC8C9CACBCCCDCECF),
SPH_C64(0xD0D1D2D3D4D5D6D7), SPH_C64(0xD8D9DADBDCDDDEDF),
SPH_C64(0xE0E1E2E3E4E5E6E7), SPH_C64(0xE8E9EAEBECEDEEEF),
SPH_C64(0xF0F1F2F3F4F5F6F7), SPH_C64(0xF8F9FAFBFCFDFEFF)
static const uint64_t IV512[] = {
0x8081828384858687, 0x88898A8B8C8D8E8F,
0x9091929394959697, 0x98999A9B9C9D9E9F,
0xA0A1A2A3A4A5A6A7, 0xA8A9AAABACADAEAF,
0xB0B1B2B3B4B5B6B7, 0xB8B9BABBBCBDBEBF,
0xC0C1C2C3C4C5C6C7, 0xC8C9CACBCCCDCECF,
0xD0D1D2D3D4D5D6D7, 0xD8D9DADBDCDDDEDF,
0xE0E1E2E3E4E5E6E7, 0xE8E9EAEBECEDEEEF,
0xF0F1F2F3F4F5F6F7, 0xF8F9FAFBFCFDFEFF
};
#if defined(__SSE2__)
@@ -894,7 +894,7 @@ static const __m256i final_b[16] =
};
static void
bmw64_4way_init( bmw_4way_big_context *sc, const sph_u64 *iv )
bmw64_4way_init( bmw_4way_big_context *sc, const uint64_t *iv )
{
sc->H[ 0] = _mm256_set1_epi64x( 0x8081828384858687 );
sc->H[ 1] = _mm256_set1_epi64x( 0x88898A8B8C8D8E8F );
@@ -926,7 +926,7 @@ bmw64_4way( bmw_4way_big_context *sc, const void *data, size_t len )
size_t ptr;
const int buf_size = 128; // bytes of one lane, compatible with len
sc->bit_count += (sph_u64)len << 3;
sc->bit_count += (uint64_t)len << 3;
buf = sc->buf;
ptr = sc->ptr;
h1 = sc->H;
@@ -1377,7 +1377,7 @@ static const __m512i final_b8[16] =
void bmw512_8way_init( bmw512_8way_context *ctx )
//bmw64_4way_init( bmw_4way_big_context *sc, const sph_u64 *iv )
//bmw64_4way_init( bmw_4way_big_context *sc, const uint64_t *iv )
{
ctx->H[ 0] = _mm512_set1_epi64( 0x8081828384858687 );
ctx->H[ 1] = _mm512_set1_epi64( 0x88898A8B8C8D8E8F );

2
algo/bmw/sph_bmw.h
View File

@@ -41,7 +41,7 @@ extern "C"{
#endif
#include <stddef.h>
#include "algo/sha/sph_types.h"
#include "compat/sph_types.h"
/**
* Output size (in bits) for BMW-224.