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

174
algo/ripemd/lbry-4way.c
View File

@@ -40,9 +40,9 @@ void lbry_8way_hash( void* output, const void* input )
sha256_8way_close( &ctx_sha256, vhashA );
// reinterleave to do sha512 4-way 64 bit twice.
mm256_deinterleave_8x32( h0, h1, h2, h3, h4, h5, h6, h7, vhashA, 256 );
mm256_interleave_4x64( vhashA, h0, h1, h2, h3, 256 );
mm256_interleave_4x64( vhashB, h4, h5, h6, h7, 256 );
mm256_dintrlv_8x32( h0, h1, h2, h3, h4, h5, h6, h7, vhashA, 256 );
mm256_intrlv_4x64( vhashA, h0, h1, h2, h3, 256 );
mm256_intrlv_4x64( vhashB, h4, h5, h6, h7, 256 );
sha512_4way_init( &ctx_sha512 );
sha512_4way( &ctx_sha512, vhashA, 32 );
@@ -53,9 +53,9 @@ void lbry_8way_hash( void* output, const void* input )
sha512_4way_close( &ctx_sha512, vhashB );
// back to 8-way 32 bit
mm256_deinterleave_4x64( h0, h1, h2, h3, vhashA, 512 );
mm256_deinterleave_4x64( h4, h5, h6, h7, vhashB, 512 );
mm256_interleave_8x32( vhashA, h0, h1, h2, h3, h4, h5, h6, h7, 512 );
mm256_dintrlv_4x64( h0, h1, h2, h3, vhashA, 512 );
mm256_dintrlv_4x64( h4, h5, h6, h7, vhashB, 512 );
mm256_intrlv_8x32( vhashA, h0, h1, h2, h3, h4, h5, h6, h7, 512 );
ripemd160_8way_init( &ctx_ripemd );
ripemd160_8way( &ctx_ripemd, vhashA, 32 );
@@ -72,27 +72,24 @@ void lbry_8way_hash( void* output, const void* input )
sha256_8way_init( &ctx_sha256 );
sha256_8way( &ctx_sha256, vhashA, 32 );
sha256_8way_close( &ctx_sha256, vhashA );
mm256_deinterleave_8x32( output, output+ 32, output+ 64, output+ 96,
output+128, output+160, output+192, output+224,
vhashA, 256 );
sha256_8way_close( &ctx_sha256, output );
}
int scanhash_lbry_8way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done)
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t hash[8*8] __attribute__ ((aligned (64)));
uint32_t vdata[32*8] __attribute__ ((aligned (64)));
uint32_t lane_hash[8] __attribute__ ((aligned (32)));
uint32_t *hash7 = &(hash[7<<3]);
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t n = pdata[27];
const uint32_t first_nonce = pdata[27];
const uint32_t Htarg = ptarget[7];
uint32_t edata[32] __attribute__ ((aligned (64)));
uint32_t *nonces = work->nonces;
int num_found = 0;
uint32_t *noncep = vdata + 216; // 27*8
__m256i *noncev = (__m256i*)vdata + 27; // aligned
/* int */ thr_id = mythr->id; // thr_id arg is deprecated
uint64_t htmax[] = { 0, 0xF, 0xFF,
0xFFF, 0xFFFF, 0x10000000 };
@@ -100,9 +97,12 @@ int scanhash_lbry_8way( int thr_id, struct work *work, uint32_t max_nonce,
0xFFFFF000, 0xFFFF0000, 0 };
// we need bigendian data...
swab32_array( edata, pdata, 32 );
mm256_interleave_8x32( vdata, edata, edata, edata, edata,
edata, edata, edata, edata, 1024 );
casti_m256i( edata, 0 ) = mm256_bswap_32( casti_m256i( pdata, 0 ) );
casti_m256i( edata, 1 ) = mm256_bswap_32( casti_m256i( pdata, 1 ) );
casti_m256i( edata, 2 ) = mm256_bswap_32( casti_m256i( pdata, 2 ) );
casti_m256i( edata, 3 ) = mm256_bswap_32( casti_m256i( pdata, 3 ) );
mm256_intrlv_8x32( vdata, edata, edata, edata, edata,
edata, edata, edata, edata, 1024 );
sha256_8way_init( &sha256_8w_mid );
sha256_8way( &sha256_8w_mid, vdata, LBRY_MIDSTATE );
@@ -111,136 +111,26 @@ int scanhash_lbry_8way( int thr_id, struct work *work, uint32_t max_nonce,
uint32_t mask = masks[m];
do
{
be32enc( noncep, n );
be32enc( noncep+1, n+1 );
be32enc( noncep+2, n+2 );
be32enc( noncep+3, n+3 );
be32enc( noncep+4, n+4 );
be32enc( noncep+5, n+5 );
be32enc( noncep+6, n+6 );
be32enc( noncep+7, n+7 );
*noncev = mm256_bswap_32( _mm256_set_epi32(
n+7,n+6,n+5,n+4,n+3,n+2,n+1,n ) );
lbry_8way_hash( hash, vdata );
for ( int i = 0; i < 8; i++ )
if ( !( (hash+(i<<3))[7] & mask ) && fulltest( hash+(i<<3), ptarget ) )
for ( int i = 0; i < 8; i++ ) if ( !( hash7[ i ] & mask ) )
{
pdata[27] = n+i;
nonces[ num_found++ ] = n+i;
work_set_target_ratio( work, hash+(i<<3) );
// deinterleave hash for lane
mm256_extract_lane_8x32( lane_hash, hash, i, 256 );
if ( fulltest( lane_hash, ptarget ) && !opt_benchmark )
{
pdata[27] = n + i;
submit_solution( work, lane_hash, mythr, i );
}
}
n+=8;
} while ( ( num_found == 0 ) && ( n < max_nonce )
&& !work_restart[thr_id].restart );
n += 8;
} while ( (n < max_nonce-10) && !work_restart[thr_id].restart );
break;
}
*hashes_done = n - first_nonce;
return num_found;
}
#elif defined(LBRY_4WAY)
static __thread sha256_4way_context sha256_mid;
void lbry_4way_hash( void* output, const void* input )
{
sha256_4way_context ctx_sha256 __attribute__ ((aligned (64)));
sha512_4way_context ctx_sha512;
ripemd160_4way_context ctx_ripemd;
uint32_t _ALIGN(64) vhashA[16<<2];
uint32_t _ALIGN(64) vhashB[16<<2];
uint32_t _ALIGN(64) vhashC[16<<2];
memcpy( &ctx_sha256, &sha256_mid, sizeof(ctx_sha256) );
sha256_4way( &ctx_sha256, input + (LBRY_MIDSTATE<<2), LBRY_TAIL );
sha256_4way_close( &ctx_sha256, vhashA );
sha256_4way_init( &ctx_sha256 );
sha256_4way( &ctx_sha256, vhashA, 32 );
sha256_4way_close( &ctx_sha256, vhashA );
// sha512 64 bit data, 64 byte output
mm256_reinterleave_4x64( vhashB, vhashA, 256 );
sha512_4way_init( &ctx_sha512 );
sha512_4way( &ctx_sha512, vhashB, 32 );
sha512_4way_close( &ctx_sha512, vhashB );
mm256_reinterleave_4x32( vhashA, vhashB, 512 );
ripemd160_4way_init( &ctx_ripemd );
ripemd160_4way( &ctx_ripemd, vhashA, 32 );
ripemd160_4way_close( &ctx_ripemd, vhashB );
ripemd160_4way_init( &ctx_ripemd );
ripemd160_4way( &ctx_ripemd, vhashA+(8<<2), 32 );
ripemd160_4way_close( &ctx_ripemd, vhashC );
sha256_4way_init( &ctx_sha256 );
sha256_4way( &ctx_sha256, vhashB, 20 );
sha256_4way( &ctx_sha256, vhashC, 20 );
sha256_4way_close( &ctx_sha256, vhashA );
sha256_4way_init( &ctx_sha256 );
sha256_4way( &ctx_sha256, vhashA, 32 );
sha256_4way_close( &ctx_sha256, vhashA );
mm128_deinterleave_4x32( output, output+32, output+64, output+96,
vhashA, 256 );
}
int scanhash_lbry_4way( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done)
{
uint32_t hash[4*8] __attribute__ ((aligned (64)));
uint32_t vdata[32*4] __attribute__ ((aligned (64)));
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
uint32_t n = pdata[27];
const uint32_t first_nonce = pdata[27];
const uint32_t Htarg = ptarget[7];
uint32_t edata[32] __attribute__ ((aligned (64)));
uint32_t *nonces = work->nonces;
int num_found = 0;
uint32_t *noncep = vdata + 108; // 27*4
uint64_t htmax[] = { 0, 0xF, 0xFF,
0xFFF, 0xFFFF, 0x10000000 };
uint32_t masks[] = { 0xFFFFFFFF, 0xFFFFFFF0, 0xFFFFFF00,
0xFFFFF000, 0xFFFF0000, 0 };
// we need bigendian data...
swab32_array( edata, pdata, 32 );
mm128_interleave_4x32( vdata, edata, edata, edata, edata, 1024 );
sha256_4way_init( &sha256_mid );
sha256_4way( &sha256_mid, vdata, LBRY_MIDSTATE );
for ( int m = 0; m < sizeof(masks); m++ ) if ( Htarg <= htmax[m] )
{
uint32_t mask = masks[m];
do
{
be32enc( noncep, n );
be32enc( noncep+1, n+1 );
be32enc( noncep+2, n+2 );
be32enc( noncep+3, n+3 );
lbry_4way_hash( hash, vdata );
for ( int i = 0; i < 4; i++ )
if ( !( (hash+(i<<3))[7] & mask ) && fulltest( hash+(i<<3), ptarget ) )
{
pdata[27] = n+i;
nonces[ num_found++ ] = n+i;
work_set_target_ratio( work, hash+(i<<3) );
}
n+=4;
} while ( ( num_found == 0 ) && ( n < max_nonce )
&& !work_restart[thr_id].restart );
break;
}
*hashes_done = n - first_nonce;
return num_found;
*hashes_done = n - first_nonce + 1;
return 0;
}
#endif