Skip to content

Commit

Permalink
v24.3
Browse files Browse the repository at this point in the history
  • Loading branch information
@JayDDee
JayDDee committed May 28, 2024
1 parent 042d13d commit c47c4a8
Show file tree
Hide file tree
Showing 36 changed files with 473 additions and 463 deletions.
7 changes: 7 additions & 0 deletions RELEASE_NOTES
View file
Original file line number Diff line number Diff line change
Expand Up @@ -75,6 +75,13 @@ If not what makes it happen or not happen?
Change Log
----------

v24.3

ARM: CPU feature detection and reporting is now working.
ARM: Verthash is now working.
ARM: Small speedup for yescrypt, yespower & argon2d.
Code cleanup.

v24.2

x86_64: Fixed blakes2s for AVX2 & AVX512, x25x for AVX512, broken in v3.23.4.
Expand Down
1 change: 0 additions & 1 deletion algo-gate-api.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -98,7 +98,6 @@ typedef uint32_t set_t;
#define AVX512_OPT 1 << 6 // Skylake-X, Zen4 (AVX512[F,VL,DQ,BW])
#define AES_OPT 1 << 7 // Intel Westmere, AArch64
#define VAES_OPT 1 << 8 // Icelake, Zen3
#define SHA_OPT 1 << 9 // Zen1, Icelake, AArch64
#define SHA256_OPT 1 << 9 // Zen1, Icelake, AArch64
#define SHA512_OPT 1 << 10 // Intel Arrow Lake, AArch64
#define NEON_OPT 1 << 11 // AArch64
Expand Down
24 changes: 12 additions & 12 deletions algo/blake/blake2b-hash.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -240,7 +240,7 @@ static const uint8_t sigma[12][16] =
v[b] = mm512_ror_64( _mm512_xor_si512( v[b], v[c] ), 63 ); \
}

static void blake2b_8way_compress( blake2b_8way_ctx *ctx, int last )
static void blake2b_8x64_compress( blake2b_8x64_ctx *ctx, int last )
{
__m512i v[16], m[16];

Expand Down Expand Up @@ -306,7 +306,7 @@ static void blake2b_8way_compress( blake2b_8way_ctx *ctx, int last )
ctx->h[7] = mm512_xor3( ctx->h[7], v[7], v[15] );
}

int blake2b_8way_init( blake2b_8way_ctx *ctx )
int blake2b_8x64_init( blake2b_8x64_ctx *ctx )
{
size_t i;

Expand All @@ -333,7 +333,7 @@ int blake2b_8way_init( blake2b_8way_ctx *ctx )
}


void blake2b_8way_update( blake2b_8way_ctx *ctx, const void *input,
void blake2b_8x64_update( blake2b_8x64_ctx *ctx, const void *input,
size_t inlen )
{
__m512i* in =(__m512i*)input;
Expand All @@ -348,15 +348,15 @@ void blake2b_8way_update( blake2b_8way_ctx *ctx, const void *input,
ctx->t[0] += ctx->c;
if ( ctx->t[0] < ctx->c )
ctx->t[1]++;
blake2b_8way_compress( ctx, 0 );
blake2b_8x64_compress( ctx, 0 );
ctx->c = 0;
}
ctx->b[ c++ ] = in[i];
ctx->c += 8;
}
}

void blake2b_8way_final( blake2b_8way_ctx *ctx, void *out )
void blake2b_8x64_final( blake2b_8x64_ctx *ctx, void *out )
{
size_t c;
c = ctx->c >> 3;
Expand All @@ -371,7 +371,7 @@ void blake2b_8way_final( blake2b_8way_ctx *ctx, void *out )
ctx->c += 8;
}

blake2b_8way_compress( ctx, 1 ); // final block flag = 1
blake2b_8x64_compress( ctx, 1 ); // final block flag = 1

casti_m512i( out, 0 ) = ctx->h[0];
casti_m512i( out, 1 ) = ctx->h[1];
Expand Down Expand Up @@ -407,7 +407,7 @@ static const uint64_t blake2b_iv[8] = {
};
*/

static void blake2b_4way_compress( blake2b_4way_ctx *ctx, int last )
static void blake2b_4x64_compress( blake2b_4x64_ctx *ctx, int last )
{
__m256i v[16], m[16];

Expand Down Expand Up @@ -473,7 +473,7 @@ static void blake2b_4way_compress( blake2b_4way_ctx *ctx, int last )
ctx->h[7] = _mm256_xor_si256( _mm256_xor_si256( ctx->h[7], v[7] ), v[15] );
}

int blake2b_4way_init( blake2b_4way_ctx *ctx )
int blake2b_4x64_init( blake2b_4x64_ctx *ctx )
{
size_t i;

Expand All @@ -499,7 +499,7 @@ int blake2b_4way_init( blake2b_4way_ctx *ctx )
return 0;
}

void blake2b_4way_update( blake2b_4way_ctx *ctx, const void *input,
void blake2b_4x64_update( blake2b_4x64_ctx *ctx, const void *input,
size_t inlen )
{
__m256i* in =(__m256i*)input;
Expand All @@ -514,15 +514,15 @@ void blake2b_4way_update( blake2b_4way_ctx *ctx, const void *input,
ctx->t[0] += ctx->c;
if ( ctx->t[0] < ctx->c )
ctx->t[1]++;
blake2b_4way_compress( ctx, 0 );
blake2b_4x64_compress( ctx, 0 );
ctx->c = 0;
}
ctx->b[ c++ ] = in[i];
ctx->c += 8;
}
}

void blake2b_4way_final( blake2b_4way_ctx *ctx, void *out )
void blake2b_4x64_final( blake2b_4x64_ctx *ctx, void *out )
{
size_t c;
c = ctx->c >> 3;
Expand All @@ -537,7 +537,7 @@ void blake2b_4way_final( blake2b_4way_ctx *ctx, void *out )
ctx->c += 8;
}

blake2b_4way_compress( ctx, 1 ); // final block flag = 1
blake2b_4x64_compress( ctx, 1 ); // final block flag = 1

casti_m256i( out, 0 ) = ctx->h[0];
casti_m256i( out, 1 ) = ctx->h[1];
Expand Down
30 changes: 20 additions & 10 deletions algo/blake/blake2b-hash.h
View file
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
#pragma once
#ifndef __BLAKE2B_HASH_4WAY_H__
#define __BLAKE2B_HASH_4WAY_H__
#ifndef BLAKE2B_HASH_4WAY_H__
#define BLAKE2B_HASH_4WAY_H__

#include "simd-utils.h"
#include <stddef.h>
Expand All @@ -23,12 +23,17 @@ typedef struct ALIGN( 64 ) {
uint64_t t[2]; // total number of bytes
size_t c; // pointer for b[]
size_t outlen; // digest size
} blake2b_8way_ctx;
} blake2b_8x64_ctx;

int blake2b_8way_init( blake2b_8way_ctx *ctx );
void blake2b_8way_update( blake2b_8way_ctx *ctx, const void *input,
int blake2b_8x64_init( blake2b_8x64_ctx *ctx );
void blake2b_8x64_update( blake2b_8x64_ctx *ctx, const void *input,
size_t inlen );
void blake2b_8way_final( blake2b_8way_ctx *ctx, void *out );
void blake2b_8x64_final( blake2b_8x64_ctx *ctx, void *out );

#define blake2b_8way_ctx blake2b_8x64_ctx
#define blake2b_8way_init blake2b_8x64_init
#define blake2b_8way_update blake2b_8x64_update
#define blake2b_8way_final blake2b_8x64_final

#endif

Expand All @@ -41,12 +46,17 @@ typedef struct ALIGN( 64 ) {
uint64_t t[2]; // total number of bytes
size_t c; // pointer for b[]
size_t outlen; // digest size
} blake2b_4way_ctx;
} blake2b_4x64_ctx;

int blake2b_4way_init( blake2b_4way_ctx *ctx );
void blake2b_4way_update( blake2b_4way_ctx *ctx, const void *input,
int blake2b_4x64_init( blake2b_4x64_ctx *ctx );
void blake2b_4x64_update( blake2b_4x64_ctx *ctx, const void *input,
size_t inlen );
void blake2b_4way_final( blake2b_4way_ctx *ctx, void *out );
void blake2b_4x64_final( blake2b_4x64_ctx *ctx, void *out );

#define blake2b_4way_ctx blake2b_4x64_ctx
#define blake2b_4way_init blake2b_4x64_init
#define blake2b_4way_update blake2b_4x64_update
#define blake2b_4way_final blake2b_4x64_final

#endif

Expand Down
4 changes: 2 additions & 2 deletions algo/blake/blake2s-hash.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -11,8 +11,8 @@
* this software. If not, see <http://creativecommons.org/publicdomain/zero/1.0/>.
*/
//#pragma once
#ifndef __BLAKE2S_HASH_4WAY_H__
#define __BLAKE2S_HASH_4WAY_H__ 1
#ifndef BLAKE2S_HASH_4WAY_H__
#define BLAKE2S_HASH_4WAY_H__ 1

#if defined(__SSE2__) || defined(__ARM_NEON)

Expand Down
2 changes: 1 addition & 1 deletion algo/groestl/myrgr-gate.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@ bool register_myriad_algo( algo_gate_t* gate )
init_myrgr_ctx();
gate->scanhash = (void*)&scanhash_myriad;
gate->hash = (void*)&myriad_hash;
gate->optimizations = AES_OPT | SSE2_OPT | AVX2_OPT | SHA_OPT | VAES_OPT;
gate->optimizations = AES_OPT | SSE2_OPT | AVX2_OPT | SHA256_OPT | VAES_OPT;
#endif
return true;
};
Expand Down
4 changes: 0 additions & 4 deletions algo/lyra2/sponge.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -195,10 +195,6 @@ static const uint64_t blake2b_IV[8] =

#endif // AVX2 else SSE2

static inline uint64_t rotr64( const uint64_t w, const unsigned c ){
return ( w >> c ) | ( w << ( 64 - c ) );
}

#define G( r, i, a, b, c, d ) \
{ \
a = a + b; \
Expand Down
2 changes: 1 addition & 1 deletion algo/m7m/m7m.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -306,7 +306,7 @@ bool register_m7m_algo( algo_gate_t *gate )
applog( LOG_ERR, "M7M algo is not supported on MacOS");
return false;
#else
gate->optimizations = SHA_OPT;
gate->optimizations = SHA256_OPT;
init_m7m_ctx();
gate->scanhash = (void*)&scanhash_m7m_hash;
gate->build_stratum_request = (void*)&std_be_build_stratum_request;
Expand Down
32 changes: 16 additions & 16 deletions algo/ripemd/lbry-4way.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -104,14 +104,14 @@ int scanhash_lbry_16way( struct work *work, uint32_t max_nonce,
int thr_id = mythr->id; // thr_id arg is deprecated

// we need bigendian data...
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_m128i( edata, 5 ) = mm128_bswap_32( casti_m128i( pdata, 5 ) );
casti_m128i( edata, 6 ) = mm128_bswap_32( casti_m128i( pdata, 6 ) );
casti_m128i( edata, 7 ) = mm128_bswap_32( casti_m128i( pdata, 7 ) );
casti_m128i( edata, 0 ) = v128_bswap32( casti_m128i( pdata, 0 ) );
casti_m128i( edata, 1 ) = v128_bswap32( casti_m128i( pdata, 1 ) );
casti_m128i( edata, 2 ) = v128_bswap32( casti_m128i( pdata, 2 ) );
casti_m128i( edata, 3 ) = v128_bswap32( casti_m128i( pdata, 3 ) );
casti_m128i( edata, 4 ) = v128_bswap32( casti_m128i( pdata, 4 ) );
casti_m128i( edata, 5 ) = v128_bswap32( casti_m128i( pdata, 5 ) );
casti_m128i( edata, 6 ) = v128_bswap32( casti_m128i( pdata, 6 ) );
casti_m128i( edata, 7 ) = v128_bswap32( casti_m128i( pdata, 7 ) );
intrlv_16x32( vdata, edata, edata, edata, edata, edata, edata, edata,
edata, edata, edata, edata, edata, edata, edata, edata, edata, 1024 );

Expand Down Expand Up @@ -224,14 +224,14 @@ int scanhash_lbry_8way( struct work *work, uint32_t max_nonce,
int thr_id = mythr->id; // thr_id arg is deprecated

// we need bigendian data...
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_m128i( edata, 5 ) = mm128_bswap_32( casti_m128i( pdata, 5 ) );
casti_m128i( edata, 6 ) = mm128_bswap_32( casti_m128i( pdata, 6 ) );
casti_m128i( edata, 7 ) = mm128_bswap_32( casti_m128i( pdata, 7 ) );
casti_m128i( edata, 0 ) = v128_bswap32( casti_m128i( pdata, 0 ) );
casti_m128i( edata, 1 ) = v128_bswap32( casti_m128i( pdata, 1 ) );
casti_m128i( edata, 2 ) = v128_bswap32( casti_m128i( pdata, 2 ) );
casti_m128i( edata, 3 ) = v128_bswap32( casti_m128i( pdata, 3 ) );
casti_m128i( edata, 4 ) = v128_bswap32( casti_m128i( pdata, 4 ) );
casti_m128i( edata, 5 ) = v128_bswap32( casti_m128i( pdata, 5 ) );
casti_m128i( edata, 6 ) = v128_bswap32( casti_m128i( pdata, 6 ) );
casti_m128i( edata, 7 ) = v128_bswap32( casti_m128i( pdata, 7 ) );
intrlv_8x32( vdata, edata, edata, edata, edata,
edata, edata, edata, edata, 1024 );

Expand Down
3 changes: 1 addition & 2 deletions algo/ripemd/lbry-gate.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -51,7 +51,6 @@ int lbry_get_work_data_size() { return LBRY_WORK_DATA_SIZE; }

bool register_lbry_algo( algo_gate_t* gate )
{
// gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
#if defined (LBRY_16WAY)
gate->scanhash = (void*)&scanhash_lbry_16way;
gate->hash = (void*)&lbry_16way_hash;
Expand All @@ -67,7 +66,7 @@ bool register_lbry_algo( algo_gate_t* gate )
#else
gate->scanhash = (void*)&scanhash_lbry;
gate->hash = (void*)&lbry_hash;
gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
gate->optimizations = AVX2_OPT | AVX512_OPT | SHA256_OPT;
#endif
gate->build_stratum_request = (void*)&lbry_le_build_stratum_request;
gate->build_extraheader = (void*)&lbry_build_extraheader;
Expand Down
Loading

0 comments on commit c47c4a8

Please sign in to comment.