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

v3.9.10

Browse Source
This commit is contained in:
Jay D Dee
2019-11-22 20:29:18 -05:00
parent 86b889e1b0
commit a52c5eccf7
29 changed files with 2015 additions and 1672 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
algo/x16/x16rt-4way.c
View File

@@ -24,7 +24,6 @@
#include "algo/sha/sha-hash-4way.h"
static __thread uint32_t s_ntime = UINT32_MAX;
static __thread bool s_implemented = false;
static __thread char hashOrder[X16R_HASH_FUNC_COUNT + 1] = { 0 };
union _x16rt_4way_context_overlay
@@ -64,26 +63,8 @@ void x16rt_4way_hash( void* output, const void* input )
dintrlv_4x64( hash0, hash1, hash2, hash3, input, 640 );
/*
void *in = (void*) input;
uint32_t *in32 = (uint32_t*) hash0;
uint32_t ntime = in32[17];
if ( s_ntime == UINT32_MAX )
{
uint32_t _ALIGN(64) timeHash[8];
x16rt_getTimeHash(ntime, &timeHash);
x16rt_getAlgoString(&timeHash[0], hashOrder);
}
*/
// Input data is both 64 bit interleaved (input)
// and deinterleaved in inp0-3.
// If First function uses 64 bit data it is not required to interleave inp
// first. It may use the inerleaved data dmost convenient, ie 4way 64 bit.
// All other functions assume data is deinterleaved in hash0-3
// All functions must exit with data deinterleaved in hash0-3.
// Alias in0-3 points to either inp0-3 or hash0-3 according to
// its hashOrder position. Size is also set accordingly.
// and deinterleaved in inp0-3. First function has no need re-interleave.
for ( int i = 0; i < 16; i++ )
{
const char elem = hashOrder[i];
@@ -290,44 +271,31 @@ int scanhash_x16rt_4way( struct work *work, uint32_t max_nonce,
{
uint32_t hash[4*16] __attribute__ ((aligned (64)));
uint32_t vdata[24*4] __attribute__ ((aligned (64)));
uint32_t endiandata[20] __attribute__((aligned(64)));
uint32_t _ALIGN(64) timeHash[4*8];
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t Htarg = ptarget[7];
const uint32_t first_nonce = pdata[19];
uint32_t n = first_nonce;
int thr_id = mythr->id; // thr_id arg is deprecated
int thr_id = mythr->id;
__m256i *noncev = (__m256i*)vdata + 9; // aligned
volatile uint8_t *restart = &(work_restart[thr_id].restart);
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 ( opt_benchmark )
ptarget[7] = 0x0cff;
uint32_t ntime = swab32( pdata[17] );
mm256_bswap32_intrlv80_4x64( vdata, pdata );
uint32_t ntime = bswap_32( pdata[17] );
if ( s_ntime != ntime )
{
x16rt_getTimeHash( ntime, &timeHash );
x16rt_getAlgoString( &timeHash[0], hashOrder );
s_ntime = ntime;
s_implemented = true;
if ( opt_debug && !thr_id )
applog( LOG_INFO, "hash order: %s time: (%08x) time hash: (%08x)",
hashOrder, ntime, timeHash );
}
if ( !s_implemented )
{
applog( LOG_WARNING, "s not implemented");
sleep(1);
return 0;
}
if ( opt_benchmark )
ptarget[7] = 0x0cff;
uint64_t *edata = (uint64_t*)endiandata;
intrlv_4x64( (uint64_t*)vdata, edata, edata, edata, edata, 640 );
do
{