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

v3.21.3 Unreleased

Browse Source
This commit is contained in:
Jay D Dee
2023-03-13 03:20:13 -04:00
parent b339450898
commit c6bc9d67fb
49 changed files with 1126 additions and 1111 deletions
Download Patch File Download Diff File Expand all files Collapse all files

131
algo/lyra2/lyra2z-4way.c
View File

@@ -14,44 +14,6 @@ bool lyra2z_16way_thread_init()
return ( lyra2z_16way_matrix = _mm_malloc( 2*LYRA2Z_MATRIX_SIZE, 64 ) );
}
static uint32_t lyra2z_16way_midstate_vars[16*16] __attribute__ ((aligned (64)));
static __m512i lyra2z_16way_block0_hash[8] __attribute__ ((aligned (64)));
static __m512i lyra2z_16way_block_buf[16] __attribute__ ((aligned (64)));
int lyra2z_16way_prehash ( struct work *work )
{
uint32_t phash[8] __attribute__ ((aligned (32))) =
{
0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A,
0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19
};
uint32_t *pdata = work->data;
// Prehash first block
blake256_transform_le( phash, pdata, 512, 0 );
lyra2z_16way_block0_hash[0] = _mm512_set1_epi32( phash[0] );
lyra2z_16way_block0_hash[1] = _mm512_set1_epi32( phash[1] );
lyra2z_16way_block0_hash[2] = _mm512_set1_epi32( phash[2] );
lyra2z_16way_block0_hash[3] = _mm512_set1_epi32( phash[3] );
lyra2z_16way_block0_hash[4] = _mm512_set1_epi32( phash[4] );
lyra2z_16way_block0_hash[5] = _mm512_set1_epi32( phash[5] );
lyra2z_16way_block0_hash[6] = _mm512_set1_epi32( phash[6] );
lyra2z_16way_block0_hash[7] = _mm512_set1_epi32( phash[7] );
// Build vectored second block, interleave 12 of last 16 bytes of data
// excepting the nonce.
lyra2z_16way_block_buf[ 0] = _mm512_set1_epi32( pdata[16] );
lyra2z_16way_block_buf[ 1] = _mm512_set1_epi32( pdata[17] );
lyra2z_16way_block_buf[ 2] = _mm512_set1_epi32( pdata[18] );
// Partialy prehash second block without touching nonces in block_buf[3].
blake256_16way_round0_prehash_le( lyra2z_16way_midstate_vars,
lyra2z_16way_block0_hash, lyra2z_16way_block_buf );
return 1;
}
static void lyra2z_16way_hash( void *state, const void *midstate_vars,
const void *midhash, const void *block )
{
@@ -129,6 +91,11 @@ int scanhash_lyra2z_16way( struct work *work, uint32_t max_nonce,
uint32_t midstate_vars[16*16] __attribute__ ((aligned (64)));
__m512i block0_hash[8] __attribute__ ((aligned (64)));
__m512i block_buf[16] __attribute__ ((aligned (64)));
uint32_t phash[8] __attribute__ ((aligned (64))) =
{
0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A,
0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19
};
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[19];
@@ -140,18 +107,30 @@ int scanhash_lyra2z_16way( struct work *work, uint32_t max_nonce,
if ( bench ) ( (uint32_t*)ptarget )[7] = 0x0000ff;
pthread_rwlock_rdlock( &g_work_lock );
// Prehash first block
blake256_transform_le( phash, pdata, 512, 0 );
memcpy( midstate_vars, lyra2z_16way_midstate_vars, sizeof midstate_vars );
memcpy( block0_hash, lyra2z_16way_block0_hash, sizeof block0_hash );
memcpy( block_buf, lyra2z_16way_block_buf, sizeof block_buf );
block0_hash[0] = _mm512_set1_epi32( phash[0] );
block0_hash[1] = _mm512_set1_epi32( phash[1] );
block0_hash[2] = _mm512_set1_epi32( phash[2] );
block0_hash[3] = _mm512_set1_epi32( phash[3] );
block0_hash[4] = _mm512_set1_epi32( phash[4] );
block0_hash[5] = _mm512_set1_epi32( phash[5] );
block0_hash[6] = _mm512_set1_epi32( phash[6] );
block0_hash[7] = _mm512_set1_epi32( phash[7] );
pthread_rwlock_unlock( &g_work_lock );
// Build vectored second block, interleave last 16 bytes of data using
// unique nonces.
block_buf[ 0] = _mm512_set1_epi32( pdata[16] );
block_buf[ 1] = _mm512_set1_epi32( pdata[17] );
block_buf[ 2] = _mm512_set1_epi32( pdata[18] );
block_buf[ 3] =
_mm512_set_epi32( n+15, n+14, n+13, n+12, n+11, n+10, n+ 9, n+ 8,
n+ 7, n+ 6, n+ 5, n+ 4, n+ 3, n+ 2, n +1, n );
// Partialy prehash second block without touching nonces in block_buf[3].
blake256_16way_round0_prehash_le( midstate_vars, block0_hash, block_buf );
do {
lyra2z_16way_hash( hash, midstate_vars, block0_hash, block_buf );
@@ -178,44 +157,6 @@ bool lyra2z_8way_thread_init()
return ( lyra2z_8way_matrix = _mm_malloc( LYRA2Z_MATRIX_SIZE, 64 ) );
}
static uint32_t lyra2z_8way_midstate_vars[16*8] __attribute__ ((aligned (64)));
static __m256i lyra2z_8way_block0_hash[8] __attribute__ ((aligned (64)));
static __m256i lyra2z_8way_block_buf[16] __attribute__ ((aligned (64)));
int lyra2z_8way_prehash ( struct work *work )
{
uint32_t phash[8] __attribute__ ((aligned (32))) =
{
0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A,
0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19
};
uint32_t *pdata = work->data;
// Prehash first block
blake256_transform_le( phash, pdata, 512, 0 );
lyra2z_8way_block0_hash[0] = _mm256_set1_epi32( phash[0] );
lyra2z_8way_block0_hash[1] = _mm256_set1_epi32( phash[1] );
lyra2z_8way_block0_hash[2] = _mm256_set1_epi32( phash[2] );
lyra2z_8way_block0_hash[3] = _mm256_set1_epi32( phash[3] );
lyra2z_8way_block0_hash[4] = _mm256_set1_epi32( phash[4] );
lyra2z_8way_block0_hash[5] = _mm256_set1_epi32( phash[5] );
lyra2z_8way_block0_hash[6] = _mm256_set1_epi32( phash[6] );
lyra2z_8way_block0_hash[7] = _mm256_set1_epi32( phash[7] );
// Build vectored second block, interleave last 16 bytes of data using
// unique nonces.
lyra2z_8way_block_buf[ 0] = _mm256_set1_epi32( pdata[16] );
lyra2z_8way_block_buf[ 1] = _mm256_set1_epi32( pdata[17] );
lyra2z_8way_block_buf[ 2] = _mm256_set1_epi32( pdata[18] );
// Partialy prehash second block without touching nonces
blake256_8way_round0_prehash_le( lyra2z_8way_midstate_vars,
lyra2z_8way_block0_hash, lyra2z_8way_block_buf );
return 1;
}
static void lyra2z_8way_hash( void *state, const void *midstate_vars,
const void *midhash, const void *block )
{
@@ -260,6 +201,11 @@ int scanhash_lyra2z_8way( struct work *work, uint32_t max_nonce,
uint32_t midstate_vars[16*8] __attribute__ ((aligned (64)));
__m256i block0_hash[8] __attribute__ ((aligned (64)));
__m256i block_buf[16] __attribute__ ((aligned (64)));
uint32_t phash[8] __attribute__ ((aligned (32))) =
{
0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A,
0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19
};
uint32_t *pdata = work->data;
uint64_t *ptarget = (uint64_t*)work->target;
const uint32_t first_nonce = pdata[19];
@@ -269,14 +215,23 @@ int scanhash_lyra2z_8way( struct work *work, uint32_t max_nonce,
const bool bench = opt_benchmark;
const __m256i eight = m256_const1_32( 8 );
pthread_rwlock_rdlock( &g_work_lock );
// Prehash first block
blake256_transform_le( phash, pdata, 512, 0 );
memcpy( midstate_vars, lyra2z_8way_midstate_vars, sizeof midstate_vars );
memcpy( block0_hash, lyra2z_8way_block0_hash, sizeof block0_hash );
memcpy( block_buf, lyra2z_8way_block_buf, sizeof block_buf );
block0_hash[0] = _mm256_set1_epi32( phash[0] );
block0_hash[1] = _mm256_set1_epi32( phash[1] );
block0_hash[2] = _mm256_set1_epi32( phash[2] );
block0_hash[3] = _mm256_set1_epi32( phash[3] );
block0_hash[4] = _mm256_set1_epi32( phash[4] );
block0_hash[5] = _mm256_set1_epi32( phash[5] );
block0_hash[6] = _mm256_set1_epi32( phash[6] );
block0_hash[7] = _mm256_set1_epi32( phash[7] );
pthread_rwlock_unlock( &g_work_lock );
// Build vectored second block, interleave last 16 bytes of data using
// unique nonces.
block_buf[ 0] = _mm256_set1_epi32( pdata[16] );
block_buf[ 1] = _mm256_set1_epi32( pdata[17] );
block_buf[ 2] = _mm256_set1_epi32( pdata[18] );
block_buf[ 3] =
_mm256_set_epi32( n+ 7, n+ 6, n+ 5, n+ 4, n+ 3, n+ 2, n +1, n );