Skip to content

Commit

Permalink
v3.7.10
Browse files Browse the repository at this point in the history
  • Loading branch information
@JayDDee
JayDDee committed Jan 16, 2018
1 parent bee78ea commit a90d75b
Show file tree
Hide file tree
Showing 77 changed files with 3,429 additions and 1,235 deletions.
19 changes: 17 additions & 2 deletions Makefile.am
View file
Original file line number Diff line number Diff line change
Expand Up @@ -46,8 +46,10 @@ cpuminer_SOURCES = \
algo/blake/sph_blake2b.c \
algo/blake/blake2b.c \
algo/blake/blake2s.c \
algo/blake/blakecoin-gate.c \
algo/blake/mod_blakecoin.c \
algo/blake/blakecoin.c \
algo/blake/blakecoin-4way.c \
algo/blake/decred-gate.c \
algo/blake/decred.c \
algo/blake/decred-4way.c \
Expand Down Expand Up @@ -99,21 +101,27 @@ cpuminer_SOURCES = \
algo/luffa/sse2/luffa_for_sse2.c \
algo/lyra2/lyra2.c \
algo/lyra2/sponge.c \
algo/lyra2/lyra2rev2-gate.c \
algo/lyra2/lyra2rev2.c \
algo/lyra2/lyra2rev2-4way.c \
algo/lyra2/lyra2re.c \
algo/lyra2/lyra2z-gate.c \
algo/lyra2/lyra2z.c \
algo/lyra2/lyra2z-4way.c \
algo/lyra2/lyra2z330.c \
algo/lyra2/lyra2h-gate.c \
algo/lyra2/lyra2h.c \
algo/lyra2/lyra2h-4way.c \
algo/m7m.c \
algo/neoscrypt/neoscrypt.c \
algo/nist5/nist5-gate.c \
algo/nist5/nist5-4way.c \
algo/nist5/nist5.c \
algo/nist5/zr5.c \
algo/pluck.c \
algo/quark/quark-gate.c \
algo/quark/quark.c \
algo/quark/quark-4way.c \
algo/qubit/qubit.c \
algo/qubit/deep.c \
algo/ripemd/sph_ripemd.c \
Expand All @@ -140,9 +148,8 @@ cpuminer_SOURCES = \
algo/skein/skein2-4way.c \
algo/skein/skein2-gate.c \
algo/sm3/sm3.c \
algo/sm3/sm3-hash-4way.c \
algo/tiger/sph_tiger.c \
algo/timetravel.c \
algo/timetravel10.c \
algo/whirlpool/sph_whirlpool.c \
algo/whirlpool/whirlpool-hash-4way.c \
algo/whirlpool/whirlpool-gate.c \
Expand All @@ -161,8 +168,16 @@ cpuminer_SOURCES = \
algo/x11/tribus-gate.c \
algo/x11/tribus.c \
algo/x11/tribus-4way.c \
algo/x11/timetravel-gate.c \
algo/x11/timetravel.c \
algo/x11/timetravel-4way.c \
algo/x11/timetravel10-gate.c \
algo/x11/timetravel10.c \
algo/x11/timetravel10-4way.c \
algo/x11/fresh.c \
algo/x11/x11evo.c \
algo/x11/x11evo-4way.c \
algo/x11/x11evo-gate.c \
algo/x13/x13-gate.c \
algo/x13/x13.c \
algo/x13/x13-4way.c \
Expand Down
7 changes: 7 additions & 0 deletions RELEASE_NOTES
View file
Original file line number Diff line number Diff line change
Expand Up @@ -165,6 +165,13 @@ Support for even older x86_64 without AES_NI or SSE2 is not availble.
Change Log
----------

v3.7.10

4way optimizations for lyra2rev2, lyra2h, quark, timetravel8, timetravel10
x11evo, blakecoin.
Faster x13sm3 (hsr).
Added share difficulty to accepted message.

v3.7.9

Partial 4way optimizations for veltor, skunk, polytimos, lyra2z.
Expand Down
91 changes: 38 additions & 53 deletions algo/blake/blake-4way.c
View file
Original file line number Diff line number Diff line change
@@ -1,31 +1,22 @@
#include "blake-gate.h"
#include "sph_blake.h"

#if defined (__AVX__)

#include "blake-hash-4way.h"
#include <string.h>
#include <stdint.h>
#include <memory.h>

#if defined (BLAKE_4WAY)
blake256r14_4way_context blake_ctx;

void blakehash_4way(void *state, const void *input)
{
uint32_t vhash[4*4] __attribute__ ((aligned (64)));
uint32_t hash0[4] __attribute__ ((aligned (32)));
uint32_t hash1[4] __attribute__ ((aligned (32)));
uint32_t hash2[4] __attribute__ ((aligned (32)));
uint32_t hash3[4] __attribute__ ((aligned (32)));
blake256_4way_context ctx;

blake256_4way_init( &ctx );
blake256_4way( &ctx, input, 16 );
blake256_4way_close( &ctx, vhash );

mm_deinterleave_4x32( hash0, hash1, hash2, hash3, vhash, 256 );

memcpy( state, hash0, 32 );
memcpy( state+32, hash1, 32 );
memcpy( state+64, hash1, 32 );
memcpy( state+96, hash1, 32 );
uint32_t vhash[8*4] __attribute__ ((aligned (64)));
blake256r14_4way_context ctx;
memcpy( &ctx, &blake_ctx, sizeof ctx );
blake256r14_4way( &ctx, input + (64<<2), 16 );
blake256r14_4way_close( &ctx, vhash );
mm_deinterleave_4x32( state, state+32, state+64, state+96, vhash, 256 );
}

int scanhash_blake_4way( int thr_id, struct work *work, uint32_t max_nonce,
Expand All @@ -36,21 +27,24 @@ int scanhash_blake_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[19];
// uint32_t HTarget = ptarget[7];
uint32_t HTarget = ptarget[7];
uint32_t _ALIGN(32) edata[20];
uint32_t n = first_nonce;
uint32_t *nonces = work->nonces;
bool *found = work->nfound;
int num_found = 0;

// if (opt_benchmark)
// HTarget = 0x7f;
if (opt_benchmark)
HTarget = 0x7f;

// we need big endian data...
swab32_array( edata, pdata, 20 );

mm_interleave_4x32( vdata, edata, edata, edata, edata, 640 );

blake256r14_4way_init( &blake_ctx );
blake256r14_4way( &blake_ctx, vdata, 64 );

uint32_t *noncep = vdata + 76; // 19*4
do {
found[0] = found[1] = found[2] = found[3] = false;
Expand All @@ -61,45 +55,36 @@ int scanhash_blake_4way( int thr_id, struct work *work, uint32_t max_nonce,

blakehash_4way( hash, vdata );

if ( hash[7] == 0 )
if ( hash[7] <= HTarget && fulltest( hash, ptarget ) )
{
if ( fulltest( hash, ptarget ) )
{
found[0] = true;
num_found++;
nonces[0] = n;
pdata[19] = n;
}
found[0] = true;
num_found++;
nonces[0] = n;
pdata[19] = n;
work_set_target_ratio( work, hash );
}
if ( (hash+8)[7] == 0 )
if ( (hash+8)[7] <= HTarget && fulltest( hash+8, ptarget ) )
{
if ( fulltest( hash+8, ptarget ) )
{
found[1] = true;
num_found++;
nonces[1] = n+1;
}
found[1] = true;
num_found++;
nonces[1] = n+1;
work_set_target_ratio( work, hash+8 );
}
if ( (hash+16)[7] == 0 )
if ( (hash+16)[7] <= HTarget && fulltest( hash+16, ptarget ) )
{
if ( fulltest( hash+8, ptarget ) )
{
found[2] = true;
num_found++;
nonces[2] = n+2;
}
found[2] = true;
num_found++;
nonces[2] = n+2;
work_set_target_ratio( work, hash+16 );
}
if ( (hash+24)[7] == 0 )
if ( (hash+24)[7] <= HTarget && fulltest( hash+24, ptarget ) )
{
if ( fulltest( hash+8, ptarget ) )
{
found[3] = true;
num_found++;
nonces[3] = n+3;
}
found[3] = true;
num_found++;
nonces[3] = n+3;
work_set_target_ratio( work, hash+24 );
}
n += 4;
*hashes_done = n - first_nonce + 1;
n += 4;

} while ( (num_found == 0) && (n < max_nonce)
&& !work_restart[thr_id].restart );
Expand Down
Loading

0 comments on commit a90d75b

Please sign in to comment.