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

v3.11.8

Browse Source
This commit is contained in:
Jay D Dee
2020-01-30 03:47:11 -05:00
parent 88f81fda0b
commit 0681ca996d
46 changed files with 2882 additions and 10675 deletions
Download Patch File Download Diff File Expand all files Collapse all files

608
algo/x16/x16r-4way.c
View File

@@ -17,6 +17,7 @@
#include "algo/keccak/keccak-hash-4way.h"
#include "algo/shavite/sph_shavite.h"
#include "algo/luffa/luffa-hash-2way.h"
#include "algo/luffa/luffa_for_sse2.h"
#include "algo/cubehash/cube-hash-2way.h"
#include "algo/cubehash/cubehash_sse2.h"
#include "algo/simd/simd-hash-2way.h"
@@ -32,11 +33,11 @@
#include "algo/echo/echo-hash-4way.h"
#endif
#if defined (X16R_8WAY)
static __thread uint32_t s_ntime = UINT32_MAX;
static __thread char hashOrder[X16R_HASH_FUNC_COUNT + 1] = { 0 };
#if defined (X16R_8WAY)
union _x16r_8way_context_overlay
{
blake512_8way_context blake;
@@ -45,7 +46,8 @@ union _x16r_8way_context_overlay
jh512_8way_context jh;
keccak512_8way_context keccak;
luffa_4way_context luffa;
cube_4way_context cube;
cubehashParam cube;
// cube_4way_context cube;
simd_4way_context simd;
hamsi512_8way_context hamsi;
sph_fugue512_context fugue;
@@ -65,19 +67,21 @@ union _x16r_8way_context_overlay
typedef union _x16r_8way_context_overlay x16r_8way_context_overlay;
static __thread x16r_8way_context_overlay x16r_ctx;
void x16r_8way_hash( void* output, const void* input )
{
uint32_t vhash[24*8] __attribute__ ((aligned (128)));
uint32_t hash0[24] __attribute__ ((aligned (64)));
uint32_t hash1[24] __attribute__ ((aligned (64)));
uint32_t hash2[24] __attribute__ ((aligned (64)));
uint32_t hash3[24] __attribute__ ((aligned (64)));
uint32_t hash4[24] __attribute__ ((aligned (64)));
uint32_t hash5[24] __attribute__ ((aligned (64)));
uint32_t hash6[24] __attribute__ ((aligned (64)));
uint32_t hash7[24] __attribute__ ((aligned (64)));
uint32_t vhash[20*8] __attribute__ ((aligned (128)));
uint32_t hash0[20] __attribute__ ((aligned (64)));
uint32_t hash1[20] __attribute__ ((aligned (64)));
uint32_t hash2[20] __attribute__ ((aligned (64)));
uint32_t hash3[20] __attribute__ ((aligned (64)));
uint32_t hash4[20] __attribute__ ((aligned (64)));
uint32_t hash5[20] __attribute__ ((aligned (64)));
uint32_t hash6[20] __attribute__ ((aligned (64)));
uint32_t hash7[20] __attribute__ ((aligned (64)));
x16r_8way_context_overlay ctx;
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
void *in0 = (void*) hash0;
void *in1 = (void*) hash1;
void *in2 = (void*) hash2;
@@ -143,28 +147,14 @@ void x16r_8way_hash( void* output, const void* input )
groestl512_full( &ctx.groestl, (char*)hash7, (char*)in7, size<<3 );
#endif
break;
case SKEIN:
skein512_8way_init( &ctx.skein );
if ( i == 0 )
skein512_8way_update( &ctx.skein, input, size );
else
{
intrlv_8x64( vhash, in0, in1, in2, in3, in4, in5, in6, in7,
size<<3 );
skein512_8way_update( &ctx.skein, vhash, size );
}
skein512_8way_close( &ctx.skein, vhash );
dintrlv_8x64_512( hash0, hash1, hash2, hash3, hash4, hash5, hash6,
hash7, vhash );
break;
case JH:
jh512_8way_init( &ctx.jh );
if ( i == 0 )
jh512_8way_update( &ctx.jh, input, size );
jh512_8way_update( &ctx.jh, input + (64<<3), 16 );
else
{
intrlv_8x64( vhash, in0, in1, in2, in3, in4, in5, in6, in7,
size<<3 );
jh512_8way_init( &ctx.jh );
jh512_8way_update( &ctx.jh, vhash, size );
}
jh512_8way_close( &ctx.jh, vhash );
@@ -185,21 +175,97 @@ void x16r_8way_hash( void* output, const void* input )
dintrlv_8x64_512( hash0, hash1, hash2, hash3, hash4, hash5, hash6,
hash7, vhash );
break;
case SKEIN:
if ( i == 0 )
skein512_8way_update( &ctx.skein, input + (64<<3), 16 );
else
{
intrlv_8x64( vhash, in0, in1, in2, in3, in4, in5, in6, in7,
size<<3 );
skein512_8way_init( &ctx.skein );
skein512_8way_update( &ctx.skein, vhash, size );
}
skein512_8way_close( &ctx.skein, vhash );
dintrlv_8x64_512( hash0, hash1, hash2, hash3, hash4, hash5, hash6,
hash7, vhash );
break;
case LUFFA:
intrlv_4x128( vhash, in0, in1, in2, in3, size<<3 );
luffa512_4way_full( &ctx.luffa, vhash, vhash, size );
dintrlv_4x128_512( hash0, hash1, hash2, hash3, vhash );
intrlv_4x128( vhash, in4, in5, in6, in7, size<<3 );
luffa512_4way_full( &ctx.luffa, vhash, vhash, size );
dintrlv_4x128_512( hash4, hash5, hash6, hash7, vhash );
if ( i == 0 )
{
intrlv_4x128( vhash, in0, in1, in2, in3, size<<3 );
luffa_4way_update_close( &ctx.luffa, vhash,
vhash + (16<<2), 16 );
dintrlv_4x128_512( hash0, hash1, hash2, hash3, vhash );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
intrlv_4x128( vhash, in4, in5, in6, in7, size<<3 );
luffa_4way_update_close( &ctx.luffa, vhash,
vhash + (16<<2), 16 );
dintrlv_4x128_512( hash4, hash5, hash6, hash7, vhash );
}
else
{
intrlv_4x128( vhash, in0, in1, in2, in3, size<<3 );
luffa512_4way_full( &ctx.luffa, vhash, vhash, size );
dintrlv_4x128_512( hash0, hash1, hash2, hash3, vhash );
intrlv_4x128( vhash, in4, in5, in6, in7, size<<3 );
luffa512_4way_full( &ctx.luffa, vhash, vhash, size );
dintrlv_4x128_512( hash4, hash5, hash6, hash7, vhash );
}
break;
case CUBEHASH:
intrlv_4x128( vhash, in0, in1, in2, in3, size<<3 );
cube_4way_full( &ctx.cube, vhash, 512, vhash, size );
dintrlv_4x128_512( hash0, hash1, hash2, hash3, vhash );
intrlv_4x128( vhash, in4, in5, in6, in7, size<<3 );
cube_4way_full( &ctx.cube, vhash, 512, vhash, size );
dintrlv_4x128_512( hash4, hash5, hash6, hash7, vhash );
if ( i == 0 )
{
cubehashUpdateDigest( &ctx.cube, (byte*)hash0,
(const byte*)in0 + 64, 16 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
cubehashUpdateDigest( &ctx.cube, (byte*)hash1,
(const byte*)in1 + 64, 16 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
cubehashUpdateDigest( &ctx.cube, (byte*)hash2,
(const byte*)in2 + 64, 16 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
cubehashUpdateDigest( &ctx.cube, (byte*)hash3,
(const byte*)in3 + 64, 16 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
cubehashUpdateDigest( &ctx.cube, (byte*)hash4,
(const byte*)in4 + 64, 16 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
cubehashUpdateDigest( &ctx.cube, (byte*)hash5,
(const byte*)in5 + 64, 16 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
cubehashUpdateDigest( &ctx.cube, (byte*)hash6,
(const byte*)in6 + 64, 16 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
cubehashUpdateDigest( &ctx.cube, (byte*)hash7,
(const byte*)in7 + 64, 16 );
}
else
{
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash0,
(const byte*)in0, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash1,
(const byte*)in1, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash2,
(const byte*)in2, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash3,
(const byte*)in3, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash4,
(const byte*)in4, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash5,
(const byte*)in5, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash6,
(const byte*)in6, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash7,
(const byte*)in7, size );
}
break;
case SHAVITE:
#if defined(__VAES__)
@@ -272,13 +338,17 @@ void x16r_8way_hash( void* output, const void* input )
#endif
break;
case HAMSI:
intrlv_8x64( vhash, in0, in1, in2, in3, in4, in5, in6, in7,
if ( i == 0 )
hamsi512_8way_update( &ctx.hamsi, input + (64<<3), 16 );
else
{
intrlv_8x64( vhash, in0, in1, in2, in3, in4, in5, in6, in7,
size<<3 );
hamsi512_8way_init( &ctx.hamsi );
hamsi512_8way_update( &ctx.hamsi, vhash, size );
hamsi512_8way_close( &ctx.hamsi, vhash );
dintrlv_8x64_512( hash0, hash1, hash2, hash3, hash4, hash5, hash6,
hamsi512_8way_init( &ctx.hamsi );
hamsi512_8way_update( &ctx.hamsi, vhash, size );
}
hamsi512_8way_close( &ctx.hamsi, vhash );
dintrlv_8x64_512( hash0, hash1, hash2, hash3, hash4, hash5, hash6,
hash7, vhash );
break;
case FUGUE:
@@ -309,38 +379,72 @@ void x16r_8way_hash( void* output, const void* input )
break;
case SHABAL:
intrlv_8x32( vhash, in0, in1, in2, in3, in4, in5, in6, in7,
size<<3 );
shabal512_8way_init( &ctx.shabal );
shabal512_8way_update( &ctx.shabal, vhash, size );
size<<3 );
if ( i == 0 )
shabal512_8way_update( &ctx.shabal, vhash + (16<<3), 16 );
else
{
shabal512_8way_init( &ctx.shabal );
shabal512_8way_update( &ctx.shabal, vhash, size );
}
shabal512_8way_close( &ctx.shabal, vhash );
dintrlv_8x32_512( hash0, hash1, hash2, hash3, hash4, hash5, hash6,
hash7, vhash );
break;
case WHIRLPOOL:
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in0, size );
sph_whirlpool_close( &ctx.whirlpool, hash0 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in1, size );
sph_whirlpool_close( &ctx.whirlpool, hash1 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in2, size );
sph_whirlpool_close( &ctx.whirlpool, hash2 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in3, size );
sph_whirlpool_close( &ctx.whirlpool, hash3 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in4, size );
sph_whirlpool_close( &ctx.whirlpool, hash4 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in5, size );
sph_whirlpool_close( &ctx.whirlpool, hash5 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in6, size );
sph_whirlpool_close( &ctx.whirlpool, hash6 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in7, size );
sph_whirlpool_close( &ctx.whirlpool, hash7 );
if ( i == 0 )
{
sph_whirlpool( &ctx.whirlpool, in0 + 64, 16 );
sph_whirlpool_close( &ctx.whirlpool, hash0 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
sph_whirlpool( &ctx.whirlpool, in1 + 64, 16 );
sph_whirlpool_close( &ctx.whirlpool, hash1 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
sph_whirlpool( &ctx.whirlpool, in2 + 64, 16 );
sph_whirlpool_close( &ctx.whirlpool, hash2 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
sph_whirlpool( &ctx.whirlpool, in3 + 64, 16 );
sph_whirlpool_close( &ctx.whirlpool, hash3 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
sph_whirlpool( &ctx.whirlpool, in4 + 64, 16 );
sph_whirlpool_close( &ctx.whirlpool, hash4 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
sph_whirlpool( &ctx.whirlpool, in5 + 64, 16 );
sph_whirlpool_close( &ctx.whirlpool, hash5 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
sph_whirlpool( &ctx.whirlpool, in6 + 64, 16 );
sph_whirlpool_close( &ctx.whirlpool, hash6 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
sph_whirlpool( &ctx.whirlpool, in7 + 64, 16 );
sph_whirlpool_close( &ctx.whirlpool, hash7 );
}
else
{
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in0, size );
sph_whirlpool_close( &ctx.whirlpool, hash0 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in1, size );
sph_whirlpool_close( &ctx.whirlpool, hash1 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in2, size );
sph_whirlpool_close( &ctx.whirlpool, hash2 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in3, size );
sph_whirlpool_close( &ctx.whirlpool, hash3 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in4, size );
sph_whirlpool_close( &ctx.whirlpool, hash4 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in5, size );
sph_whirlpool_close( &ctx.whirlpool, hash5 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in6, size );
sph_whirlpool_close( &ctx.whirlpool, hash6 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in7, size );
sph_whirlpool_close( &ctx.whirlpool, hash7 );
}
break;
case SHA_512:
sha512_8way_init( &ctx.sha512 );
@@ -355,7 +459,7 @@ void x16r_8way_hash( void* output, const void* input )
sha512_8way_close( &ctx.sha512, vhash );
dintrlv_8x64_512( hash0, hash1, hash2, hash3, hash4, hash5, hash6,
hash7, vhash );
break;
break;
}
size = 64;
}
@@ -373,23 +477,22 @@ void x16r_8way_hash( void* output, const void* input )
int scanhash_x16r_8way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr)
{
uint32_t hash[8*16] __attribute__ ((aligned (128)));
uint32_t vdata[24*8] __attribute__ ((aligned (64)));
uint32_t hash[16*8] __attribute__ ((aligned (128)));
uint32_t vdata[20*8] __attribute__ ((aligned (64)));
uint32_t vdata2[20*8] __attribute__ ((aligned (64)));
uint32_t edata[20] __attribute__ ((aligned (64)));
uint32_t bedata1[2] __attribute__((aligned(64)));
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 8;
uint32_t n = first_nonce;
__m512i *noncev = (__m512i*)vdata + 9; // aligned
int thr_id = mythr->id;
const int thr_id = mythr->id;
volatile uint8_t *restart = &(work_restart[thr_id].restart);
const bool bench = opt_benchmark;
if ( opt_benchmark )
ptarget[7] = 0x0cff;
mm512_bswap32_intrlv80_8x64( vdata, pdata );
if ( bench ) ptarget[7] = 0x0cff;
bedata1[0] = bswap_32( pdata[1] );
bedata1[1] = bswap_32( pdata[2] );
@@ -402,32 +505,84 @@ int scanhash_x16r_8way( struct work *work, uint32_t max_nonce,
applog( LOG_INFO, "hash order %s (%08x)", hashOrder, ntime );
}
// Do midstate prehash on hash functions with block size <= 64 bytes.
const char elem = hashOrder[0];
const uint8_t algo = elem >= 'A' ? elem - 'A' + 10 : elem - '0';
switch ( algo )
{
case JH:
mm512_bswap32_intrlv80_8x64( vdata, pdata );
jh512_8way_init( &x16r_ctx.jh );
jh512_8way_update( &x16r_ctx.jh, vdata, 64 );
break;
case SKEIN:
mm512_bswap32_intrlv80_8x64( vdata, pdata );
skein512_8way_init( &x16r_ctx.skein );
skein512_8way_update( &x16r_ctx.skein, vdata, 64 );
break;
case LUFFA:
mm128_bswap32_80( edata, pdata );
intrlv_4x128( vdata2, edata, edata, edata, edata, 640 );
luffa_4way_init( &x16r_ctx.luffa, 512 );
luffa_4way_update( &x16r_ctx.luffa, vdata2, 64 );
rintrlv_4x128_8x64( vdata, vdata2, vdata2, 640 );
break;
case CUBEHASH:
mm128_bswap32_80( edata, pdata );
cubehashInit( &x16r_ctx.cube, 512, 16, 32 );
cubehashUpdate( &x16r_ctx.cube, (const byte*)edata, 64 );
intrlv_8x64( vdata, edata, edata, edata, edata,
edata, edata, edata, edata, 640 );
break;
case HAMSI:
mm512_bswap32_intrlv80_8x64( vdata, pdata );
hamsi512_8way_init( &x16r_ctx.hamsi );
hamsi512_8way_update( &x16r_ctx.hamsi, vdata, 64 );
break;
case SHABAL:
mm256_bswap32_intrlv80_8x32( vdata2, pdata );
shabal512_8way_init( &x16r_ctx.shabal );
shabal512_8way_update( &x16r_ctx.shabal, vdata2, 64 );
rintrlv_8x32_8x64( vdata, vdata2, 640 );
break;
case WHIRLPOOL:
mm128_bswap32_80( edata, pdata );
sph_whirlpool_init( &x16r_ctx.whirlpool );
sph_whirlpool( &x16r_ctx.whirlpool, edata, 64 );
intrlv_8x64( vdata, edata, edata, edata, edata,
edata, edata, edata, edata, 640 );
break;
default:
mm512_bswap32_intrlv80_8x64( vdata, pdata );
}
*noncev = mm512_intrlv_blend_32( _mm512_set_epi32(
n+7, 0, n+6, 0, n+5, 0, n+4, 0,
n+3, 0, n+2, 0, n+1, 0, n, 0 ), *noncev );
do
{
*noncev = mm512_intrlv_blend_32( mm512_bswap_32(
_mm512_set_epi32( n+7, 0, n+6, 0, n+5, 0, n+4, 0,
n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );
x16r_8way_hash( hash, vdata );
pdata[19] = n;
for ( int i = 0; i < 8; i++ )
if ( unlikely( (hash+(i<<3))[7] <= Htarg ) )
if( likely( fulltest( hash+(i<<3), ptarget ) && !opt_benchmark ) )
if ( unlikely( valid_hash( hash + (i<<3), ptarget ) && !bench ) )
{
pdata[19] = n+i;
pdata[19] = bswap_32( n+i );
submit_lane_solution( work, hash+(i<<3), mythr, i );
}
*noncev = _mm512_add_epi32( *noncev,
m512_const1_64( 0x0000000800000000 ) );
n += 8;
} while ( likely( ( n < last_nonce ) && !(*restart) ) );
pdata[19] = n;
*hashes_done = n - first_nonce;
return 0;
}
#elif defined (X16R_4WAY)
static __thread uint32_t s_ntime = UINT32_MAX;
static __thread char hashOrder[X16R_HASH_FUNC_COUNT + 1] = { 0 };
union _x16r_4way_context_overlay
{
blake512_4way_context blake;
@@ -438,6 +593,7 @@ union _x16r_4way_context_overlay
jh512_4way_context jh;
keccak512_4way_context keccak;
luffa_2way_context luffa;
hashState_luffa luffa1;
cubehashParam cube;
sph_shavite512_context shavite;
simd_2way_context simd;
@@ -449,14 +605,17 @@ union _x16r_4way_context_overlay
} __attribute__ ((aligned (64)));
typedef union _x16r_4way_context_overlay x16r_4way_context_overlay;
static __thread x16r_4way_context_overlay x16r_ctx;
void x16r_4way_hash( void* output, const void* input )
{
uint32_t vhash[24*4] __attribute__ ((aligned (128)));
uint32_t hash0[24] __attribute__ ((aligned (64)));
uint32_t hash1[24] __attribute__ ((aligned (64)));
uint32_t hash2[24] __attribute__ ((aligned (64)));
uint32_t hash3[24] __attribute__ ((aligned (64)));
uint32_t vhash[20*4] __attribute__ ((aligned (128)));
uint32_t hash0[20] __attribute__ ((aligned (64)));
uint32_t hash1[20] __attribute__ ((aligned (64)));
uint32_t hash2[20] __attribute__ ((aligned (64)));
uint32_t hash3[20] __attribute__ ((aligned (64)));
x16r_4way_context_overlay ctx;
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
void *in0 = (void*) hash0;
void *in1 = (void*) hash1;
void *in2 = (void*) hash2;
@@ -500,25 +659,13 @@ void x16r_4way_hash( void* output, const void* input )
groestl512_full( &ctx.groestl, (char*)hash2, (char*)in2, size<<3 );
groestl512_full( &ctx.groestl, (char*)hash3, (char*)in3, size<<3 );
break;
case SKEIN:
skein512_4way_init( &ctx.skein );
if ( i == 0 )
skein512_4way_update( &ctx.skein, input, size );
else
{
intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
skein512_4way_update( &ctx.skein, vhash, size );
}
skein512_4way_close( &ctx.skein, vhash );
dintrlv_4x64_512( hash0, hash1, hash2, hash3, vhash );
break;
case JH:
jh512_4way_init( &ctx.jh );
if ( i == 0 )
jh512_4way_update( &ctx.jh, input, size );
jh512_4way_update( &ctx.jh, input + (64<<2), 16 );
else
{
intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
jh512_4way_init( &ctx.jh );
jh512_4way_update( &ctx.jh, vhash, size );
}
jh512_4way_close( &ctx.jh, vhash );
@@ -536,27 +683,68 @@ void x16r_4way_hash( void* output, const void* input )
keccak512_4way_close( &ctx.keccak, vhash );
dintrlv_4x64_512( hash0, hash1, hash2, hash3, vhash );
break;
case SKEIN:
if ( i == 0 )
skein512_4way_update( &ctx.skein, input + (64<<2), 16 );
else
{
intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
skein512_4way_init( &ctx.skein );
skein512_4way_update( &ctx.skein, vhash, size );
}
skein512_4way_close( &ctx.skein, vhash );
dintrlv_4x64_512( hash0, hash1, hash2, hash3, vhash );
break;
case LUFFA:
intrlv_2x128( vhash, in0, in1, size<<3 );
luffa512_2way_full( &ctx.luffa, vhash, vhash, size );
dintrlv_2x128_512( hash0, hash1, vhash );
intrlv_2x128( vhash, in2, in3, size<<3 );
luffa512_2way_full( &ctx.luffa, vhash, vhash, size );
dintrlv_2x128_512( hash2, hash3, vhash );
if ( i == 0 )
{
intrlv_2x128( vhash, in0, in1, size<<3 );
luffa512_2way_full( &ctx.luffa, vhash, vhash + (16<<1), 16 );
dintrlv_2x128_512( hash0, hash1, vhash );
intrlv_2x128( vhash, in2, in3, size<<3 );
luffa512_2way_full( &ctx.luffa, vhash, vhash + (16<<1), 16 );
dintrlv_2x128_512( hash2, hash3, vhash );
}
else
{
intrlv_2x128( vhash, in0, in1, size<<3 );
luffa512_2way_full( &ctx.luffa, vhash, vhash, size );
dintrlv_2x128_512( hash0, hash1, vhash );
intrlv_2x128( vhash, in2, in3, size<<3 );
luffa512_2way_full( &ctx.luffa, vhash, vhash, size );
dintrlv_2x128_512( hash2, hash3, vhash );
}
break;
case CUBEHASH:
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash0,
(const byte*)in0, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash1,
(const byte*)in1, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash2,
(const byte*)in2, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash3,
(const byte*)in3, size );
if ( i == 0 )
{
cubehashUpdateDigest( &ctx.cube, (byte*)hash0,
(const byte*)in0 + 64, 16 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
cubehashUpdateDigest( &ctx.cube, (byte*) hash1,
(const byte*)in1 + 64, 16 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
cubehashUpdateDigest( &ctx.cube, (byte*) hash2,
(const byte*)in2 + 64, 16 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
cubehashUpdateDigest( &ctx.cube, (byte*) hash3,
(const byte*)in3 + 64, 16 );
}
else
{
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash0,
(const byte*)in0, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash1,
(const byte*)in1, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash2,
(const byte*)in2, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash3,
(const byte*)in3, size );
}
break;
case SHAVITE:
sph_shavite512_init( &ctx.shavite );
@@ -591,11 +779,16 @@ void x16r_4way_hash( void* output, const void* input )
(const BitSequence *)in3, size );
break;
case HAMSI:
intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
hamsi512_4way_init( &ctx.hamsi );
hamsi512_4way_update( &ctx.hamsi, vhash, size );
hamsi512_4way_close( &ctx.hamsi, vhash );
dintrlv_4x64_512( hash0, hash1, hash2, hash3, vhash );
if ( i == 0 )
hamsi512_4way_update( &ctx.hamsi, input + (64<<2), 16 );
else
{
intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
hamsi512_4way_init( &ctx.hamsi );
hamsi512_4way_update( &ctx.hamsi, vhash, size );
}
hamsi512_4way_close( &ctx.hamsi, vhash );
dintrlv_4x64_512( hash0, hash1, hash2, hash3, vhash );
break;
case FUGUE:
sph_fugue512_init( &ctx.fugue );
@@ -613,31 +806,59 @@ void x16r_4way_hash( void* output, const void* input )
break;
case SHABAL:
intrlv_4x32( vhash, in0, in1, in2, in3, size<<3 );
shabal512_4way_init( &ctx.shabal );
shabal512_4way_update( &ctx.shabal, vhash, size );
if ( i == 0 )
shabal512_4way_update( &ctx.shabal, vhash + (16<<2), 16 );
else
{
shabal512_4way_init( &ctx.shabal );
shabal512_4way_update( &ctx.shabal, vhash, size );
}
shabal512_4way_close( &ctx.shabal, vhash );
dintrlv_4x32_512( hash0, hash1, hash2, hash3, vhash );
break;
case WHIRLPOOL:
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in0, size );
sph_whirlpool_close( &ctx.whirlpool, hash0 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in1, size );
sph_whirlpool_close( &ctx.whirlpool, hash1 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in2, size );
sph_whirlpool_close( &ctx.whirlpool, hash2 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in3, size );
sph_whirlpool_close( &ctx.whirlpool, hash3 );
if ( i == 0 )
{
sph_whirlpool( &ctx.whirlpool, in0 + 64, 16 );
sph_whirlpool_close( &ctx.whirlpool, hash0 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
sph_whirlpool( &ctx.whirlpool, in1 + 64, 16 );
sph_whirlpool_close( &ctx.whirlpool, hash1 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
sph_whirlpool( &ctx.whirlpool, in2 + 64, 16 );
sph_whirlpool_close( &ctx.whirlpool, hash2 );
memcpy( &ctx, &x16r_ctx, sizeof(ctx) );
sph_whirlpool( &ctx.whirlpool, in3 + 64, 16 );
sph_whirlpool_close( &ctx.whirlpool, hash3 );
}
else
{
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in0, size );
sph_whirlpool_close( &ctx.whirlpool, hash0 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in1, size );
sph_whirlpool_close( &ctx.whirlpool, hash1 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in2, size );
sph_whirlpool_close( &ctx.whirlpool, hash2 );
sph_whirlpool_init( &ctx.whirlpool );
sph_whirlpool( &ctx.whirlpool, in3, size );
sph_whirlpool_close( &ctx.whirlpool, hash3 );
}
break;
case SHA_512:
intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
sha512_4way_init( &ctx.sha512 );
sha512_4way_update( &ctx.sha512, vhash, size );
sha512_4way_close( &ctx.sha512, vhash );
dintrlv_4x64_512( hash0, hash1, hash2, hash3, vhash );
sha512_4way_init( &ctx.sha512 );
if ( i == 0 )
sha512_4way_update( &ctx.sha512, input, size );
else
{
intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
sha512_4way_init( &ctx.sha512 );
sha512_4way_update( &ctx.sha512, vhash, size );
}
sha512_4way_close( &ctx.sha512, vhash );
dintrlv_4x64_512( hash0, hash1, hash2, hash3, vhash );
break;
}
size = 64;
@@ -651,23 +872,22 @@ void x16r_4way_hash( void* output, const void* input )
int scanhash_x16r_4way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr)
{
uint32_t hash[4*16] __attribute__ ((aligned (64)));
uint32_t vdata[24*4] __attribute__ ((aligned (64)));
uint32_t hash[16*4] __attribute__ ((aligned (64)));
uint32_t vdata[20*4] __attribute__ ((aligned (64)));
uint32_t vdata2[20*4] __attribute__ ((aligned (64)));
uint32_t edata[20] __attribute__ ((aligned (64)));
uint32_t bedata1[2] __attribute__((aligned(64)));
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
const uint32_t last_nonce = max_nonce - 4;
uint32_t n = first_nonce;
__m256i *noncev = (__m256i*)vdata + 9; // aligned
int thr_id = mythr->id;
const int thr_id = mythr->id;
const bool bench = opt_benchmark;
volatile uint8_t *restart = &(work_restart[thr_id].restart);
if ( opt_benchmark )
ptarget[7] = 0x0cff;
mm256_bswap32_intrlv80_4x64( vdata, pdata );
if ( bench ) ptarget[7] = 0x0cff;
bedata1[0] = bswap_32( pdata[1] );
bedata1[1] = bswap_32( pdata[2] );
@@ -680,24 +900,72 @@ int scanhash_x16r_4way( struct work *work, uint32_t max_nonce,
applog( LOG_INFO, "hash order %s (%08x)", hashOrder, ntime );
}
// Do midstate prehash on hash functions with block size <= 64 bytes.
const char elem = hashOrder[0];
const uint8_t algo = elem >= 'A' ? elem - 'A' + 10 : elem - '0';
switch ( algo )
{
case JH:
mm256_bswap32_intrlv80_4x64( vdata, pdata );
jh512_4way_init( &x16r_ctx.jh );
jh512_4way_update( &x16r_ctx.jh, vdata, 64 );
break;
case SKEIN:
mm256_bswap32_intrlv80_4x64( vdata, pdata );
skein512_4way_init( &x16r_ctx.skein );
skein512_4way_update( &x16r_ctx.skein, vdata, 64 );
break;
case LUFFA:
mm128_bswap32_80( edata, pdata );
intrlv_2x128( vdata2, edata, edata, 640 );
luffa_2way_init( &x16r_ctx.luffa, 512 );
luffa_2way_update( &x16r_ctx.luffa, vdata2, 64 );
rintrlv_2x128_4x64( vdata, vdata2, vdata2, 512 );
break;
case CUBEHASH:
mm128_bswap32_80( edata, pdata );
cubehashInit( &x16r_ctx.cube, 512, 16, 32 );
cubehashUpdate( &x16r_ctx.cube, (const byte*)edata, 64 );
intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
break;
case HAMSI:
mm256_bswap32_intrlv80_4x64( vdata, pdata );
hamsi512_4way_init( &x16r_ctx.hamsi );
hamsi512_4way_update( &x16r_ctx.hamsi, vdata, 64 );
break;
case SHABAL:
mm128_bswap32_intrlv80_4x32( vdata2, pdata );
shabal512_4way_init( &x16r_ctx.shabal );
shabal512_4way_update( &x16r_ctx.shabal, vdata2, 64 );
rintrlv_4x32_4x64( vdata, vdata2, 640 );
break;
case WHIRLPOOL:
mm128_bswap32_80( edata, pdata );
sph_whirlpool_init( &x16r_ctx.whirlpool );
sph_whirlpool( &x16r_ctx.whirlpool, edata, 64 );
intrlv_4x64( vdata, edata, edata, edata, edata, 640 );
break;
default:
mm256_bswap32_intrlv80_4x64( vdata, pdata );
}
*noncev = mm256_intrlv_blend_32(
_mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ), *noncev );
do
{
*noncev = mm256_intrlv_blend_32( mm256_bswap_32(
_mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );
x16r_4way_hash( hash, vdata );
pdata[19] = n;
for ( int i = 0; i < 4; i++ )
if ( unlikely( (hash+(i<<3))[7] <= Htarg ) )
if( likely( fulltest( hash+(i<<3), ptarget ) && !opt_benchmark ) )
if ( unlikely( valid_hash( hash + (i<<3), ptarget ) && !bench ) )
{
pdata[19] = n+i;
pdata[19] = bswap_32( n+i );
submit_lane_solution( work, hash+(i<<3), mythr, i );
}
*noncev = _mm256_add_epi32( *noncev,
m256_const1_64( 0x0000000400000000 ) );
n += 4;
} while ( likely( ( n < last_nonce ) && !(*restart) ) );
pdata[19] = n;
*hashes_done = n - first_nonce;
return 0;
}