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

v3.9.4

Browse Source
This commit is contained in:
Jay D Dee
2019-06-18 13:15:45 -04:00
parent 71d6b97ee8
commit d6e8d7a46e
75 changed files with 1790 additions and 781 deletions
Download Patch File Download Diff File Expand all files Collapse all files

97
algo/x16/x16r-4way.c
View File

@@ -32,8 +32,8 @@
static __thread uint32_t s_ntime = UINT32_MAX;
static __thread char hashOrder[X16R_HASH_FUNC_COUNT + 1] = { 0 };
typedef struct {
union _x16r_4way_context_overlay
{
blake512_4way_context blake;
bmw512_4way_context bmw;
hashState_echo echo;
@@ -50,16 +50,8 @@ typedef struct {
shabal512_4way_context shabal;
sph_whirlpool_context whirlpool;
sha512_4way_context sha512;
} x16r_4way_ctx_holder;
x16r_4way_ctx_holder x16r_4way_ctx __attribute__ ((aligned (64)));
// Cube needs one full init so fast reinits can be done in the hash loop.
void init_x16r_4way_ctx()
{
cubehashInit( &x16r_4way_ctx.cube, 512, 16, 32 );
};
typedef union _x16r_4way_context_overlay x16r_4way_context_overlay;
void x16r_4way_hash( void* output, const void* input )
{
@@ -68,14 +60,14 @@ void x16r_4way_hash( void* output, const void* input )
uint32_t hash2[24] __attribute__ ((aligned (64)));
uint32_t hash3[24] __attribute__ ((aligned (64)));
uint32_t vhash[24*4] __attribute__ ((aligned (64)));
x16r_4way_ctx_holder ctx;
x16r_4way_context_overlay ctx;
void *in0 = (void*) hash0;
void *in1 = (void*) hash1;
void *in2 = (void*) hash2;
void *in3 = (void*) hash3;
int size = 80;
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, input, 640 );
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, input, 640 );
if ( s_ntime == UINT32_MAX )
{
@@ -104,11 +96,11 @@ void x16r_4way_hash( void* output, const void* input )
blake512_4way( &ctx.blake, input, size );
else
{
mm256_interleave_4x64( vhash, in0, in1, in2, in3, size<<3 );
mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
blake512_4way( &ctx.blake, vhash, size );
}
blake512_4way_close( &ctx.blake, vhash );
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
break;
case BMW:
bmw512_4way_init( &ctx.bmw );
@@ -116,11 +108,11 @@ void x16r_4way_hash( void* output, const void* input )
bmw512_4way( &ctx.bmw, input, size );
else
{
mm256_interleave_4x64( vhash, in0, in1, in2, in3, size<<3 );
mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
bmw512_4way( &ctx.bmw, vhash, size );
}
bmw512_4way_close( &ctx.bmw, vhash );
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
break;
case GROESTL:
init_groestl( &ctx.groestl, 64 );
@@ -142,11 +134,11 @@ void x16r_4way_hash( void* output, const void* input )
skein512_4way( &ctx.skein, input, size );
else
{
mm256_interleave_4x64( vhash, in0, in1, in2, in3, size<<3 );
mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
skein512_4way( &ctx.skein, vhash, size );
}
skein512_4way_close( &ctx.skein, vhash );
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
break;
case JH:
jh512_4way_init( &ctx.jh );
@@ -154,11 +146,11 @@ void x16r_4way_hash( void* output, const void* input )
jh512_4way( &ctx.jh, input, size );
else
{
mm256_interleave_4x64( vhash, in0, in1, in2, in3, size<<3 );
mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
jh512_4way( &ctx.jh, vhash, size );
}
jh512_4way_close( &ctx.jh, vhash );
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
break;
case KECCAK:
keccak512_4way_init( &ctx.keccak );
@@ -166,21 +158,21 @@ void x16r_4way_hash( void* output, const void* input )
keccak512_4way( &ctx.keccak, input, size );
else
{
mm256_interleave_4x64( vhash, in0, in1, in2, in3, size<<3 );
mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
keccak512_4way( &ctx.keccak, vhash, size );
}
keccak512_4way_close( &ctx.keccak, vhash );
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
break;
case LUFFA:
mm256_interleave_2x128( vhash, in0, in1, size<<3 );
mm256_intrlv_2x128( vhash, in0, in1, size<<3 );
luffa_2way_init( &ctx.luffa, 512 );
luffa_2way_update_close( &ctx.luffa, vhash, vhash, size );
mm256_deinterleave_2x128( hash0, hash1, vhash, 512 );
mm256_interleave_2x128( vhash, in2, in3, size<<3 );
mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
mm256_intrlv_2x128( vhash, in2, in3, size<<3 );
luffa_2way_init( &ctx.luffa, 512 );
luffa_2way_update_close( &ctx.luffa, vhash, vhash, size);
mm256_deinterleave_2x128( hash2, hash3, vhash, 512 );
mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
break;
case CUBEHASH:
cubehashInit( &ctx.cube, 512, 16, 32 );
@@ -194,7 +186,7 @@ void x16r_4way_hash( void* output, const void* input )
(const byte*)in2, size );
cubehashInit( &ctx.cube, 512, 16, 32 );
cubehashUpdateDigest( &ctx.cube, (byte*) hash3,
(const byte*)in3, size );
(const byte*)in3, size );
break;
case SHAVITE:
sph_shavite512_init( &ctx.shavite );
@@ -211,14 +203,14 @@ void x16r_4way_hash( void* output, const void* input )
sph_shavite512_close( &ctx.shavite, hash3 );
break;
case SIMD:
mm256_interleave_2x128( vhash, in0, in1, size<<3 );
mm256_intrlv_2x128( vhash, in0, in1, size<<3 );
simd_2way_init( &ctx.simd, 512 );
simd_2way_update_close( &ctx.simd, vhash, vhash, size<<3 );
mm256_deinterleave_2x128( hash0, hash1, vhash, 512 );
mm256_interleave_2x128( vhash, in2, in3, size<<3 );
mm256_dintrlv_2x128( hash0, hash1, vhash, 512 );
mm256_intrlv_2x128( vhash, in2, in3, size<<3 );
simd_2way_init( &ctx.simd, 512 );
simd_2way_update_close( &ctx.simd, vhash, vhash, size<<3 );
mm256_deinterleave_2x128( hash2, hash3, vhash, 512 );
mm256_dintrlv_2x128( hash2, hash3, vhash, 512 );
break;
case ECHO:
init_echo( &ctx.echo, 512 );
@@ -235,11 +227,11 @@ void x16r_4way_hash( void* output, const void* input )
(const BitSequence*)in3, size<<3 );
break;
case HAMSI:
mm256_interleave_4x64( vhash, in0, in1, in2, in3, size<<3 );
mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
hamsi512_4way_init( &ctx.hamsi );
hamsi512_4way( &ctx.hamsi, vhash, size );
hamsi512_4way_close( &ctx.hamsi, vhash );
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
break;
case FUGUE:
sph_fugue512_init( &ctx.fugue );
@@ -256,11 +248,11 @@ void x16r_4way_hash( void* output, const void* input )
sph_fugue512_close( &ctx.fugue, hash3 );
break;
case SHABAL:
mm128_interleave_4x32( vhash, in0, in1, in2, in3, size<<3 );
mm128_intrlv_4x32( vhash, in0, in1, in2, in3, size<<3 );
shabal512_4way_init( &ctx.shabal );
shabal512_4way( &ctx.shabal, vhash, size );
shabal512_4way_close( &ctx.shabal, vhash );
mm128_deinterleave_4x32( hash0, hash1, hash2, hash3, vhash, 512 );
mm128_dintrlv_4x32( hash0, hash1, hash2, hash3, vhash, 512 );
break;
case WHIRLPOOL:
sph_whirlpool_init( &ctx.whirlpool );
@@ -277,11 +269,11 @@ void x16r_4way_hash( void* output, const void* input )
sph_whirlpool_close( &ctx.whirlpool, hash3 );
break;
case SHA_512:
mm256_interleave_4x64( vhash, in0, in1, in2, in3, size<<3 );
mm256_intrlv_4x64( vhash, in0, in1, in2, in3, size<<3 );
sha512_4way_init( &ctx.sha512 );
sha512_4way( &ctx.sha512, vhash, size );
sha512_4way_close( &ctx.sha512, vhash );
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
break;
}
size = 64;
@@ -304,15 +296,13 @@ int scanhash_x16r_4way( int thr_id, struct work *work, uint32_t max_nonce,
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
uint32_t *nonces = work->nonces;
int num_found = 0;
uint32_t *noncep = vdata + 73; // 9*8 + 1
__m256i *noncev = (__m256i*)vdata + 9; // aligned
volatile uint8_t *restart = &(work_restart[thr_id].restart);
for ( int k=0; k < 19; k++ )
be32enc( &endiandata[k], pdata[k] );
casti_m256i( endiandata, 0 ) = mm256_bswap_32( casti_m256i( pdata, 0 ) );
casti_m256i( endiandata, 1 ) = mm256_bswap_32( casti_m256i( pdata, 1 ) );
casti_m128i( endiandata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
// if ( s_ntime != pdata[17] )
if ( s_ntime != endiandata[17] )
{
uint32_t ntime = swab32(pdata[17]);
@@ -326,30 +316,27 @@ int scanhash_x16r_4way( int thr_id, struct work *work, uint32_t max_nonce,
ptarget[7] = 0x0cff;
uint64_t *edata = (uint64_t*)endiandata;
mm256_interleave_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
mm256_intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
do
{
be32enc( noncep, n );
be32enc( noncep+2, n+1 );
be32enc( noncep+4, n+2 );
be32enc( noncep+6, n+3 );
*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 ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
for ( int i = 0; i < 4; i++ ) if ( (hash+(i<<3))[7] <= Htarg )
if( fulltest( hash+(i<<3), ptarget ) && !opt_benchmark )
{
pdata[19] = n+i;
nonces[ num_found++ ] = n+i;
work_set_target_ratio( work, hash+(i<<3) );
submit_solution( work, hash+(i<<3), mythr, i );
}
n += 4;
} while ( ( num_found == 0 ) && ( n < max_nonce ) && !(*restart) );
} while ( ( n < max_nonce ) && !(*restart) );
*hashes_done = n - first_nonce + 1;
return num_found;
return 0;
}
#endif