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

v3.7.6

Browse Source
This commit is contained in:
Jay D Dee
2017-12-14 18:28:51 -05:00
parent af1c940919
commit 7a1389998b
31 changed files with 1285 additions and 377 deletions
Download Patch File Download Diff File Expand all files Collapse all files

53
algo/jh/jha-4way.c
View File

@@ -25,7 +25,7 @@ void jha_hash_4way( void *out, const void *input )
uint64_t vhash[8*4] __attribute__ ((aligned (64)));
uint64_t vhash0[8*4] __attribute__ ((aligned (64)));
uint64_t vhash1[8*4] __attribute__ ((aligned (64)));
__m256i mask0, mask1;
__m256i mask, mask0, mask1;
__m256i* vh = (__m256i*)vhash;
__m256i* vh0 = (__m256i*)vhash0;
__m256i* vh1 = (__m256i*)vhash1;
@@ -47,38 +47,37 @@ void jha_hash_4way( void *out, const void *input )
// Heavy & Light Pair Loop
for ( int round = 0; round < 3; round++ )
{
// memset_zero_256( vh0, 20 );
// memset_zero_256( vh1, 20 );
// positive logic, if maski select vhi
// going from bit to mask reverses logic such that if the test bit is set
// zero will be put in mask0, meaning don't take vh0. mask1 is
// inverted so 1 will be put in mask1 meaning take it.
mask0 = mm256_negate_64(
// select next function based on bit 0 of previous hash.
// Specutively execute both functions and use mask to
// select results from correct function for each lane.
// hash = mask : vhash0 ? vhash1
mask = mm256_negate_64(
_mm256_and_si256( vh[0], _mm256_set1_epi64x( 0x1 ) ) );
mask1 = mm256_not( mask0 );
// second version
// mask0 = mask
// mask1 = mm256_not( mask );
// first version
// mask = _mm256_sub_epi64( _mm256_and_si256( vh[0],
// _mm256_set1_epi64x( 0x1 ) ), _mm256_set1_epi64x( 0x1 ) );
// groestl (serial) v skein
// groestl (serial) vs skein
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
init_groestl( &ctx_groestl, 64 );
update_and_final_groestl( &ctx_groestl, (char*)hash0,
(char*)hash0, 512 );
(char*)hash0, 512 );
init_groestl( &ctx_groestl, 64 );
update_and_final_groestl( &ctx_groestl, (char*)hash1,
(char*)hash1, 512 );
(char*)hash1, 512 );
init_groestl( &ctx_groestl, 64 );
update_and_final_groestl( &ctx_groestl, (char*)hash2,
(char*)hash2, 512 );
(char*)hash2, 512 );
init_groestl( &ctx_groestl, 64 );
update_and_final_groestl( &ctx_groestl, (char*)hash3,
(char*)hash3, 512 );
(char*)hash3, 512 );
mm256_interleave_4x64( vhash0, hash0, hash1, hash2, hash3, 512 );
@@ -91,14 +90,20 @@ void jha_hash_4way( void *out, const void *input )
// merge vectored hash
for ( int i = 0; i < 8; i++ )
{
vh[i] = _mm256_or_si256( _mm256_and_si256( vh0[i], mask0 ),
_mm256_and_si256( vh1[i], mask1 ) );
// blend should be faster
vh[i] = _mm256_blendv_epi8( vh0[i], vh1[i], mask );
// second version
// vh[i] = _mm256_or_si256( _mm256_and_si256( vh0[i], mask0 ),
// _mm256_and_si256( vh1[i], mask1 ) );
// first version
/*
vha256[i] = _mm256_maskload_epi64(
vhasha + i*4, mm256_not( mask ) );
vhb256[i] = _mm256_maskload_epi64(
vhashb + i*4, mask );
vh256[i] = _mm256_or_si256( vha256[i], vhb256[i] );
vh0[i] = _mm256_maskload_epi64(
vhash0 + i*4, mm256_not( mask ) );
vh1[i] = _mm256_maskload_epi64(
vhash1 + i*4, mask );
vh[i] = _mm256_or_si256( vh0[i], vh1[i] );
*/
}