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

v3.9.5

Browse Source
This commit is contained in:
Jay D Dee
2019-06-26 14:16:01 -04:00
parent d6e8d7a46e
commit 0d48d573ce
174 changed files with 1352 additions and 1556 deletions
Download Patch File Download Diff File Expand all files Collapse all files

65
algo/x11/timetravel10-4way.c
View File

@@ -93,18 +93,18 @@ void timetravel10_4way_hash(void *output, const void *input)
blake512_4way( &ctx.blake, vhashA, dataLen );
blake512_4way_close( &ctx.blake, vhashB );
if ( i == 9 )
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3,
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
vhashB, dataLen<<3 );
break;
case 1:
bmw512_4way( &ctx.bmw, vhashA, dataLen );
bmw512_4way_close( &ctx.bmw, vhashB );
if ( i == 9 )
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3,
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
vhashB, dataLen<<3 );
break;
case 2:
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3,
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
vhashA, dataLen<<3 );
update_and_final_groestl( &ctx.groestl, (char*)hash0,
(char*)hash0, dataLen<<3 );
@@ -118,46 +118,46 @@ void timetravel10_4way_hash(void *output, const void *input)
update_and_final_groestl( &ctx.groestl, (char*)hash3,
(char*)hash3, dataLen<<3 );
if ( i != 9 )
mm256_interleave_4x64( vhashB,
mm256_intrlv_4x64( vhashB,
hash0, hash1, hash2, hash3, dataLen<<3 );
break;
case 3:
skein512_4way( &ctx.skein, vhashA, dataLen );
skein512_4way_close( &ctx.skein, vhashB );
if ( i == 9 )
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3,
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
vhashB, dataLen<<3 );
break;
case 4:
jh512_4way( &ctx.jh, vhashA, dataLen );
jh512_4way_close( &ctx.jh, vhashB );
if ( i == 9 )
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3,
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
vhashB, dataLen<<3 );
break;
case 5:
keccak512_4way( &ctx.keccak, vhashA, dataLen );
keccak512_4way_close( &ctx.keccak, vhashB );
if ( i == 9 )
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3,
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
vhashB, dataLen<<3 );
break;
case 6:
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3,
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
vhashA, dataLen<<3 );
mm256_interleave_2x128( vhashA, hash0, hash1, dataLen<<3 );
mm256_intrlv_2x128( vhashA, hash0, hash1, dataLen<<3 );
luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, dataLen );
mm256_deinterleave_2x128( hash0, hash1, vhashA, dataLen<<3 );
mm256_interleave_2x128( vhashA, hash2, hash3, dataLen<<3 );
mm256_dintrlv_2x128( hash0, hash1, vhashA, dataLen<<3 );
mm256_intrlv_2x128( vhashA, hash2, hash3, dataLen<<3 );
luffa_2way_init( &ctx.luffa, 512 );
luffa_2way_update_close( &ctx.luffa, vhashA, vhashA, dataLen );
mm256_deinterleave_2x128( hash2, hash3, vhashA, dataLen<<3 );
mm256_dintrlv_2x128( hash2, hash3, vhashA, dataLen<<3 );
if ( i != 9 )
mm256_interleave_4x64( vhashB,
mm256_intrlv_4x64( vhashB,
hash0, hash1, hash2, hash3, dataLen<<3 );
break;
case 7:
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3,
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
vhashA, dataLen<<3 );
cubehashUpdateDigest( &ctx.cube, (byte*)hash0,
(const byte*)hash0, dataLen );
@@ -171,11 +171,11 @@ void timetravel10_4way_hash(void *output, const void *input)
cubehashUpdateDigest( &ctx.cube, (byte*)hash3,
(const byte*)hash3, dataLen );
if ( i != 9 )
mm256_interleave_4x64( vhashB,
mm256_intrlv_4x64( vhashB,
hash0, hash1, hash2, hash3, dataLen<<3 );
break;
case 8:
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3,
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
vhashA, dataLen<<3 );
sph_shavite512( &ctx.shavite, hash0, dataLen );
sph_shavite512_close( &ctx.shavite, hash0 );
@@ -189,21 +189,21 @@ void timetravel10_4way_hash(void *output, const void *input)
sph_shavite512( &ctx.shavite, hash3, dataLen );
sph_shavite512_close( &ctx.shavite, hash3 );
if ( i != 9 )
mm256_interleave_4x64( vhashB,
mm256_intrlv_4x64( vhashB,
hash0, hash1, hash2, hash3, dataLen<<3 );
break;
case 9:
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3,
mm256_dintrlv_4x64( hash0, hash1, hash2, hash3,
vhashA, dataLen<<3 );
mm256_interleave_2x128( vhashA, hash0, hash1, dataLen<<3 );
mm256_intrlv_2x128( vhashA, hash0, hash1, dataLen<<3 );
simd_2way_update_close( &ctx.simd, vhashA, vhashA, dataLen<<3 );
mm256_deinterleave_2x128( hash0, hash1, vhashA, dataLen<<3 );
mm256_interleave_2x128( vhashA, hash2, hash3, dataLen<<3 );
mm256_dintrlv_2x128( hash0, hash1, vhashA, dataLen<<3 );
mm256_intrlv_2x128( vhashA, hash2, hash3, dataLen<<3 );
simd_2way_init( &ctx.simd, 512 );
simd_2way_update_close( &ctx.simd, vhashA, vhashA, dataLen<<3 );
mm256_deinterleave_2x128( hash2, hash3, vhashA, dataLen<<3 );
mm256_dintrlv_2x128( hash2, hash3, vhashA, dataLen<<3 );
if ( i != 9 )
mm256_interleave_4x64( vhashB,
mm256_intrlv_4x64( vhashB,
hash0, hash1, hash2, hash3, dataLen<<3 );
break;
default:
@@ -218,8 +218,8 @@ void timetravel10_4way_hash(void *output, const void *input)
memcpy( output+96, hash3, 32 );
}
int scanhash_timetravel10_4way( int thr_id, struct work *work,
uint32_t max_nonce, uint64_t *hashes_done )
int scanhash_timetravel10_4way( struct work *work,
uint32_t max_nonce, uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t hash[4*8] __attribute__ ((aligned (64)));
uint32_t vdata[24*4] __attribute__ ((aligned (64)));
@@ -228,9 +228,8 @@ int scanhash_timetravel10_4way( int thr_id, struct work *work,
uint32_t *ptarget = work->target;
uint32_t n = pdata[19];
const uint32_t first_nonce = pdata[19];
uint32_t *nonces = work->nonces;
int num_found = 0;
uint32_t *noncep = vdata + 73; // 9*8 + 1
int thr_id = mythr->id; // thr_id arg is deprecated
const uint32_t Htarg = ptarget[7];
volatile uint8_t *restart = &(work_restart[thr_id].restart);
int i;
@@ -254,7 +253,7 @@ int scanhash_timetravel10_4way( int thr_id, struct work *work,
}
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
{
@@ -267,16 +266,16 @@ int scanhash_timetravel10_4way( int thr_id, struct work *work,
pdata[19] = n;
for ( int i = 0; i < 4; i++ )
if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
if ( (hash+(i<<3))[7] <= Htarg && 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_lane_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