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

v3.22.0

Browse Source
This commit is contained in:
Jay D Dee
2023-03-21 17:12:51 -04:00
parent cae1ce2ab7
commit 3dd6787531
17 changed files with 305 additions and 1006 deletions
Download Patch File Download Diff File Expand all files Collapse all files

74
algo/blake/decred-4way.c
View File

@@ -1,74 +0,0 @@
#include "decred-gate.h"
#include "blake-hash-4way.h"
#include <string.h>
#include <stdint.h>
#include <memory.h>
#include <unistd.h>
#if defined (DECRED_4WAY)
static __thread blake256_4way_context blake_mid;
void decred_hash_4way( void *state, const void *input )
{
uint32_t vhash[8*4] __attribute__ ((aligned (64)));
// uint32_t hash0[8] __attribute__ ((aligned (32)));
// uint32_t hash1[8] __attribute__ ((aligned (32)));
// uint32_t hash2[8] __attribute__ ((aligned (32)));
// uint32_t hash3[8] __attribute__ ((aligned (32)));
const void *tail = input + ( DECRED_MIDSTATE_LEN << 2 );
int tail_len = 180 - DECRED_MIDSTATE_LEN;
blake256_4way_context ctx __attribute__ ((aligned (64)));
memcpy( &ctx, &blake_mid, sizeof(blake_mid) );
blake256_4way_update( &ctx, tail, tail_len );
blake256_4way_close( &ctx, vhash );
dintrlv_4x32( state, state+32, state+64, state+96, vhash, 256 );
}
int scanhash_decred_4way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t vdata[48*4] __attribute__ ((aligned (64)));
uint32_t hash[8*4] __attribute__ ((aligned (32)));
uint32_t _ALIGN(64) edata[48];
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[DECRED_NONCE_INDEX];
uint32_t n = first_nonce;
const uint32_t HTarget = opt_benchmark ? 0x7f : ptarget[7];
int thr_id = mythr->id; // thr_id arg is deprecated
// copy to buffer guaranteed to be aligned.
memcpy( edata, pdata, 180 );
// use the old way until new way updated for size.
mm128_intrlv_4x32x( vdata, edata, edata, edata, edata, 180*8 );
blake256_4way_init( &blake_mid );
blake256_4way_update( &blake_mid, vdata, DECRED_MIDSTATE_LEN );
uint32_t *noncep = vdata + DECRED_NONCE_INDEX * 4;
do {
* noncep = n;
*(noncep+1) = n+1;
*(noncep+2) = n+2;
*(noncep+3) = n+3;
decred_hash_4way( hash, vdata );
for ( int i = 0; i < 4; i++ )
if ( (hash+(i<<3))[7] <= HTarget )
if ( fulltest( hash+(i<<3), ptarget ) && !opt_benchmark )
{
pdata[DECRED_NONCE_INDEX] = n+i;
submit_solution( work, hash+(i<<3), mythr );
}
n += 4;
} while ( (n < max_nonce) && !work_restart[thr_id].restart );
*hashes_done = n - first_nonce + 1;
return 0;
}
#endif

171
algo/blake/decred-gate.c
View File

@@ -1,171 +0,0 @@
#include "decred-gate.h"
#include <unistd.h>
#include <memory.h>
#include <string.h>
uint32_t *decred_get_nonceptr( uint32_t *work_data )
{
return &work_data[ DECRED_NONCE_INDEX ];
}
long double decred_calc_network_diff( struct work* work )
{
// sample for diff 43.281 : 1c05ea29
// todo: endian reversed on longpoll could be zr5 specific...
uint32_t nbits = work->data[ DECRED_NBITS_INDEX ];
uint32_t bits = ( nbits & 0xffffff );
int16_t shift = ( swab32(nbits) & 0xff ); // 0x1c = 28
int m;
long double d = (long double)0x0000ffff / (long double)bits;
for ( m = shift; m < 29; m++ )
d *= 256.0;
for ( m = 29; m < shift; m++ )
d /= 256.0;
if ( shift == 28 )
d *= 256.0; // testnet
if ( opt_debug_diff )
applog( LOG_DEBUG, "net diff: %f -> shift %u, bits %08x", (double)d,
shift, bits );
return net_diff;
}
void decred_decode_extradata( struct work* work, uint64_t* net_blocks )
{
// some random extradata to make the work unique
work->data[ DECRED_XNONCE_INDEX ] = (rand()*4);
work->height = work->data[32];
if (!have_longpoll && work->height > *net_blocks + 1)
{
char netinfo[64] = { 0 };
if ( net_diff > 0. )
{
if (net_diff != work->targetdiff)
sprintf(netinfo, ", diff %.3f, target %.1f", net_diff,
work->targetdiff);
else
sprintf(netinfo, ", diff %.3f", net_diff);
}
applog(LOG_BLUE, "%s block %d%s", algo_names[opt_algo], work->height,
netinfo);
*net_blocks = work->height - 1;
}
}
void decred_be_build_stratum_request( char *req, struct work *work,
struct stratum_ctx *sctx )
{
unsigned char *xnonce2str;
uint32_t ntime, nonce;
char ntimestr[9], noncestr[9];
be32enc( &ntime, work->data[ DECRED_NTIME_INDEX ] );
be32enc( &nonce, work->data[ DECRED_NONCE_INDEX ] );
bin2hex( ntimestr, (char*)(&ntime), sizeof(uint32_t) );
bin2hex( noncestr, (char*)(&nonce), sizeof(uint32_t) );
xnonce2str = abin2hex( (char*)( &work->data[ DECRED_XNONCE_INDEX ] ),
sctx->xnonce1_size );
snprintf( req, JSON_BUF_LEN,
"{\"method\": \"mining.submit\", \"params\": [\"%s\", \"%s\", \"%s\", \"%s\", \"%s\"], \"id\":4}",
rpc_user, work->job_id, xnonce2str, ntimestr, noncestr );
free(xnonce2str);
}
#if !defined(min)
#define min(a,b) (a>b ? (b) :(a))
#endif
void decred_build_extraheader( struct work* g_work, struct stratum_ctx* sctx )
{
uchar merkle_root[64] = { 0 };
uint32_t extraheader[32] = { 0 };
int headersize = 0;
uint32_t* extradata = (uint32_t*) sctx->xnonce1;
int i;
// getwork over stratum, getwork merkle + header passed in coinb1
memcpy(merkle_root, sctx->job.coinbase, 32);
headersize = min((int)sctx->job.coinbase_size - 32,
sizeof(extraheader) );
memcpy( extraheader, &sctx->job.coinbase[32], headersize );
// Assemble block header
memset( g_work->data, 0, sizeof(g_work->data) );
g_work->data[0] = le32dec( sctx->job.version );
for ( i = 0; i < 8; i++ )
g_work->data[1 + i] = swab32(
le32dec( (uint32_t *) sctx->job.prevhash + i ) );
for ( i = 0; i < 8; i++ )
g_work->data[9 + i] = swab32( be32dec( (uint32_t *) merkle_root + i ) );
// for ( i = 0; i < 8; i++ ) // prevhash
// g_work->data[1 + i] = swab32( g_work->data[1 + i] );
// for ( i = 0; i < 8; i++ ) // merkle
// g_work->data[9 + i] = swab32( g_work->data[9 + i] );
for ( i = 0; i < headersize/4; i++ ) // header
g_work->data[17 + i] = extraheader[i];
// extradata
for ( i = 0; i < sctx->xnonce1_size/4; i++ )
g_work->data[ DECRED_XNONCE_INDEX + i ] = extradata[i];
for ( i = DECRED_XNONCE_INDEX + sctx->xnonce1_size/4; i < 45; i++ )
g_work->data[i] = 0;
g_work->data[37] = (rand()*4) << 8;
// block header suffix from coinb2 (stake version)
memcpy( &g_work->data[44],
&sctx->job.coinbase[ sctx->job.coinbase_size-4 ], 4 );
sctx->block_height = g_work->data[32];
//applog_hex(work->data, 180);
//applog_hex(&work->data[36], 36);
}
#undef min
bool decred_ready_to_mine( struct work* work, struct stratum_ctx* stratum,
int thr_id )
{
if ( have_stratum && strcmp(stratum->job.job_id, work->job_id) )
// need to regen g_work..
return false;
if ( have_stratum && !work->data[0] && !opt_benchmark )
{
sleep(1);
return false;
}
// extradata: prevent duplicates
work->data[ DECRED_XNONCE_INDEX ] += 1;
work->data[ DECRED_XNONCE_INDEX + 1 ] |= thr_id;
return true;
}
int decred_get_work_data_size() { return DECRED_DATA_SIZE; }
bool register_decred_algo( algo_gate_t* gate )
{
#if defined(DECRED_4WAY)
four_way_not_tested();
gate->scanhash = (void*)&scanhash_decred_4way;
gate->hash = (void*)&decred_hash_4way;
#else
gate->scanhash = (void*)&scanhash_decred;
gate->hash = (void*)&decred_hash;
#endif
gate->optimizations = AVX2_OPT;
// gate->get_nonceptr = (void*)&decred_get_nonceptr;
gate->decode_extra_data = (void*)&decred_decode_extradata;
gate->build_stratum_request = (void*)&decred_be_build_stratum_request;
gate->work_decode = (void*)&std_be_work_decode;
gate->submit_getwork_result = (void*)&std_be_submit_getwork_result;
gate->build_extraheader = (void*)&decred_build_extraheader;
gate->ready_to_mine = (void*)&decred_ready_to_mine;
gate->nbits_index = DECRED_NBITS_INDEX;
gate->ntime_index = DECRED_NTIME_INDEX;
gate->nonce_index = DECRED_NONCE_INDEX;
gate->get_work_data_size = (void*)&decred_get_work_data_size;
gate->work_cmp_size = DECRED_WORK_COMPARE_SIZE;
allow_mininginfo = false;
have_gbt = false;
return true;
}

36
algo/blake/decred-gate.h
View File

@@ -1,36 +0,0 @@
#ifndef __DECRED_GATE_H__
#define __DECRED_GATE_H__
#include "algo-gate-api.h"
#include <stdint.h>
#define DECRED_NBITS_INDEX 29
#define DECRED_NTIME_INDEX 34
#define DECRED_NONCE_INDEX 35
#define DECRED_XNONCE_INDEX 36
#define DECRED_DATA_SIZE 192
#define DECRED_WORK_COMPARE_SIZE 140
#define DECRED_MIDSTATE_LEN 128
#if defined (__AVX2__)
//void blakehash_84way(void *state, const void *input);
//int scanhash_blake_8way( struct work *work, uint32_t max_nonce,
// uint64_t *hashes_done );
#endif
#if defined(__SSE4_2__)
#define DECRED_4WAY
#endif
#if defined (DECRED_4WAY)
void decred_hash_4way(void *state, const void *input);
int scanhash_decred_4way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr );
#endif
void decred_hash( void *state, const void *input );
int scanhash_decred( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr );
#endif

282
algo/blake/decred.c
View File

@@ -1,282 +0,0 @@
#include "decred-gate.h"
#if !defined(DECRED_8WAY) && !defined(DECRED_4WAY)
#include "sph_blake.h"
#include <string.h>
#include <stdint.h>
#include <memory.h>
#include <unistd.h>
/*
#ifndef min
#define min(a,b) (a>b ? b : a)
#endif
#ifndef max
#define max(a,b) (a<b ? b : a)
#endif
*/
/*
#define DECRED_NBITS_INDEX 29
#define DECRED_NTIME_INDEX 34
#define DECRED_NONCE_INDEX 35
#define DECRED_XNONCE_INDEX 36
#define DECRED_DATA_SIZE 192
#define DECRED_WORK_COMPARE_SIZE 140
*/
static __thread sph_blake256_context blake_mid;
static __thread bool ctx_midstate_done = false;
void decred_hash(void *state, const void *input)
{
// #define MIDSTATE_LEN 128
sph_blake256_context ctx __attribute__ ((aligned (64)));
uint8_t *ending = (uint8_t*) input;
ending += DECRED_MIDSTATE_LEN;
if (!ctx_midstate_done) {
sph_blake256_init(&blake_mid);
sph_blake256(&blake_mid, input, DECRED_MIDSTATE_LEN);
ctx_midstate_done = true;
}
memcpy(&ctx, &blake_mid, sizeof(blake_mid));
sph_blake256(&ctx, ending, (180 - DECRED_MIDSTATE_LEN));
sph_blake256_close(&ctx, state);
}
void decred_hash_simple(void *state, const void *input)
{
sph_blake256_context ctx;
sph_blake256_init(&ctx);
sph_blake256(&ctx, input, 180);
sph_blake256_close(&ctx, state);
}
int scanhash_decred( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr )
{
uint32_t _ALIGN(64) endiandata[48];
uint32_t _ALIGN(64) hash32[8];
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
int thr_id = mythr->id; // thr_id arg is deprecated
// #define DCR_NONCE_OFT32 35
const uint32_t first_nonce = pdata[DECRED_NONCE_INDEX];
const uint32_t HTarget = opt_benchmark ? 0x7f : ptarget[7];
uint32_t n = first_nonce;
ctx_midstate_done = false;
#if 1
memcpy(endiandata, pdata, 180);
#else
for (int k=0; k < (180/4); k++)
be32enc(&endiandata[k], pdata[k]);
#endif
do {
//be32enc(&endiandata[DCR_NONCE_OFT32], n);
endiandata[DECRED_NONCE_INDEX] = n;
decred_hash(hash32, endiandata);
if (hash32[7] <= HTarget && fulltest(hash32, ptarget))
{
pdata[DECRED_NONCE_INDEX] = n;
submit_solution( work, hash32, mythr );
}
n++;
} while (n < max_nonce && !work_restart[thr_id].restart);
*hashes_done = n - first_nonce + 1;
pdata[DECRED_NONCE_INDEX] = n;
return 0;
}
/*
uint32_t *decred_get_nonceptr( uint32_t *work_data )
{
return &work_data[ DECRED_NONCE_INDEX ];
}
double decred_calc_network_diff( struct work* work )
{
// sample for diff 43.281 : 1c05ea29
// todo: endian reversed on longpoll could be zr5 specific...
uint32_t nbits = work->data[ DECRED_NBITS_INDEX ];
uint32_t bits = ( nbits & 0xffffff );
int16_t shift = ( swab32(nbits) & 0xff ); // 0x1c = 28
int m;
double d = (double)0x0000ffff / (double)bits;
for ( m = shift; m < 29; m++ )
d *= 256.0;
for ( m = 29; m < shift; m++ )
d /= 256.0;
if ( shift == 28 )
d *= 256.0; // testnet
if ( opt_debug_diff )
applog( LOG_DEBUG, "net diff: %f -> shift %u, bits %08x", d,
shift, bits );
return net_diff;
}
void decred_decode_extradata( struct work* work, uint64_t* net_blocks )
{
// some random extradata to make the work unique
work->data[ DECRED_XNONCE_INDEX ] = (rand()*4);
work->height = work->data[32];
if (!have_longpoll && work->height > *net_blocks + 1)
{
char netinfo[64] = { 0 };
if (net_diff > 0.)
{
if (net_diff != work->targetdiff)
sprintf(netinfo, ", diff %.3f, target %.1f", net_diff,
work->targetdiff);
else
sprintf(netinfo, ", diff %.3f", net_diff);
}
applog(LOG_BLUE, "%s block %d%s", algo_names[opt_algo], work->height,
netinfo);
*net_blocks = work->height - 1;
}
}
void decred_be_build_stratum_request( char *req, struct work *work,
struct stratum_ctx *sctx )
{
unsigned char *xnonce2str;
uint32_t ntime, nonce;
char ntimestr[9], noncestr[9];
be32enc( &ntime, work->data[ DECRED_NTIME_INDEX ] );
be32enc( &nonce, work->data[ DECRED_NONCE_INDEX ] );
bin2hex( ntimestr, (char*)(&ntime), sizeof(uint32_t) );
bin2hex( noncestr, (char*)(&nonce), sizeof(uint32_t) );
xnonce2str = abin2hex( (char*)( &work->data[ DECRED_XNONCE_INDEX ] ),
sctx->xnonce1_size );
snprintf( req, JSON_BUF_LEN,
"{\"method\": \"mining.submit\", \"params\": [\"%s\", \"%s\", \"%s\", \"%s\", \"%s\"], \"id\":4}",
rpc_user, work->job_id, xnonce2str, ntimestr, noncestr );
free(xnonce2str);
}
*/
/*
// data shared between gen_merkle_root and build_extraheader.
__thread uint32_t decred_extraheader[32] = { 0 };
__thread int decred_headersize = 0;
void decred_gen_merkle_root( char* merkle_root, struct stratum_ctx* sctx )
{
// getwork over stratum, getwork merkle + header passed in coinb1
memcpy(merkle_root, sctx->job.coinbase, 32);
decred_headersize = min((int)sctx->job.coinbase_size - 32,
sizeof(decred_extraheader) );
memcpy( decred_extraheader, &sctx->job.coinbase[32], decred_headersize);
}
*/
/*
#define min(a,b) (a>b ? (b) :(a))
void decred_build_extraheader( struct work* g_work, struct stratum_ctx* sctx )
{
uchar merkle_root[64] = { 0 };
uint32_t extraheader[32] = { 0 };
int headersize = 0;
uint32_t* extradata = (uint32_t*) sctx->xnonce1;
size_t t;
int i;
// getwork over stratum, getwork merkle + header passed in coinb1
memcpy(merkle_root, sctx->job.coinbase, 32);
headersize = min((int)sctx->job.coinbase_size - 32,
sizeof(extraheader) );
memcpy( extraheader, &sctx->job.coinbase[32], headersize );
// Increment extranonce2
for ( t = 0; t < sctx->xnonce2_size && !( ++sctx->job.xnonce2[t] ); t++ );
// Assemble block header
memset( g_work->data, 0, sizeof(g_work->data) );
g_work->data[0] = le32dec( sctx->job.version );
for ( i = 0; i < 8; i++ )
g_work->data[1 + i] = swab32(
le32dec( (uint32_t *) sctx->job.prevhash + i ) );
for ( i = 0; i < 8; i++ )
g_work->data[9 + i] = swab32( be32dec( (uint32_t *) merkle_root + i ) );
// for ( i = 0; i < 8; i++ ) // prevhash
// g_work->data[1 + i] = swab32( g_work->data[1 + i] );
// for ( i = 0; i < 8; i++ ) // merkle
// g_work->data[9 + i] = swab32( g_work->data[9 + i] );
for ( i = 0; i < headersize/4; i++ ) // header
g_work->data[17 + i] = extraheader[i];
// extradata
for ( i = 0; i < sctx->xnonce1_size/4; i++ )
g_work->data[ DECRED_XNONCE_INDEX + i ] = extradata[i];
for ( i = DECRED_XNONCE_INDEX + sctx->xnonce1_size/4; i < 45; i++ )
g_work->data[i] = 0;
g_work->data[37] = (rand()*4) << 8;
// block header suffix from coinb2 (stake version)
memcpy( &g_work->data[44],
&sctx->job.coinbase[ sctx->job.coinbase_size-4 ], 4 );
sctx->bloc_height = g_work->data[32];
//applog_hex(work->data, 180);
//applog_hex(&work->data[36], 36);
}
#undef min
bool decred_ready_to_mine( struct work* work, struct stratum_ctx* stratum,
int thr_id )
{
if ( have_stratum && strcmp(stratum->job.job_id, work->job_id) )
// need to regen g_work..
return false;
if ( have_stratum && !work->data[0] && !opt_benchmark )
{
sleep(1);
return false;
}
// extradata: prevent duplicates
work->data[ DECRED_XNONCE_INDEX ] += 1;
work->data[ DECRED_XNONCE_INDEX + 1 ] |= thr_id;
return true;
}
bool register_decred_algo( algo_gate_t* gate )
{
gate->optimizations = SSE2_OPT;
gate->scanhash = (void*)&scanhash_decred;
gate->hash = (void*)&decred_hash;
gate->get_nonceptr = (void*)&decred_get_nonceptr;
gate->decode_extra_data = (void*)&decred_decode_extradata;
gate->build_stratum_request = (void*)&decred_be_build_stratum_request;
gate->work_decode = (void*)&std_be_work_decode;
gate->submit_getwork_result = (void*)&std_be_submit_getwork_result;
gate->build_extraheader = (void*)&decred_build_extraheader;
gate->ready_to_mine = (void*)&decred_ready_to_mine;
gate->nbits_index = DECRED_NBITS_INDEX;
gate->ntime_index = DECRED_NTIME_INDEX;
gate->nonce_index = DECRED_NONCE_INDEX;
gate->work_data_size = DECRED_DATA_SIZE;
gate->work_cmp_size = DECRED_WORK_COMPARE_SIZE;
allow_mininginfo = false;
have_gbt = false;
return true;
}
*/
#endif

45
algo/ripemd/lbry-gate.c
View File

@@ -4,24 +4,6 @@
#include <string.h>
#include <stdio.h>
long double lbry_calc_network_diff( struct work *work )
{
// sample for diff 43.281 : 1c05ea29
// todo: endian reversed on longpoll could be zr5 specific...
uint32_t nbits = swab32( work->data[ LBRY_NBITS_INDEX ] );
uint32_t bits = (nbits & 0xffffff);
int16_t shift = (swab32(nbits) & 0xff); // 0x1c = 28
long double d = (long double)0x0000ffff / (long double)bits;
for (int m=shift; m < 29; m++) d *= 256.0;
for (int m=29; m < shift; m++) d /= 256.0;
if (opt_debug_diff)
applog(LOG_DEBUG, "net diff: %f -> shift %u, bits %08x", d, shift, bits);
return d;
}
// std_le should work but it doesn't
void lbry_le_build_stratum_request( char *req, struct work *work,
struct stratum_ctx *sctx )
@@ -41,31 +23,6 @@ void lbry_le_build_stratum_request( char *req, struct work *work,
free(xnonce2str);
}
/*
void lbry_build_block_header( struct work* g_work, uint32_t version,
uint32_t *prevhash, uint32_t *merkle_root,
uint32_t ntime, uint32_t nbits )
{
int i;
memset( g_work->data, 0, sizeof(g_work->data) );
g_work->data[0] = version;
if ( have_stratum )
for ( i = 0; i < 8; i++ )
g_work->data[1 + i] = le32dec( prevhash + i );
else
for (i = 0; i < 8; i++)
g_work->data[ 8-i ] = le32dec( prevhash + i );
for ( i = 0; i < 8; i++ )
g_work->data[9 + i] = be32dec( merkle_root + i );
g_work->data[ LBRY_NTIME_INDEX ] = ntime;
g_work->data[ LBRY_NBITS_INDEX ] = nbits;
g_work->data[28] = 0x80000000;
}
*/
void lbry_build_extraheader( struct work* g_work, struct stratum_ctx* sctx )
{
unsigned char merkle_root[64] = { 0 };
@@ -112,9 +69,7 @@ bool register_lbry_algo( algo_gate_t* gate )
gate->hash = (void*)&lbry_hash;
gate->optimizations = AVX2_OPT | AVX512_OPT | SHA_OPT;
#endif
gate->calc_network_diff = (void*)&lbry_calc_network_diff;
gate->build_stratum_request = (void*)&lbry_le_build_stratum_request;
// gate->build_block_header = (void*)&build_block_header;
gate->build_extraheader = (void*)&lbry_build_extraheader;
gate->ntime_index = LBRY_NTIME_INDEX;
gate->nbits_index = LBRY_NBITS_INDEX;

2
algo/x17/x17-4way.c
View File

@@ -257,6 +257,7 @@ int scanhash_x17_8way( struct work *work, uint32_t max_nonce,
const __m512i eight = m512_const1_64( 8 );
const bool bench = opt_benchmark;
// convert LE32 to LE64
edata[0] = mm128_swap64_32( casti_m128i( pdata, 0 ) );
edata[1] = mm128_swap64_32( casti_m128i( pdata, 1 ) );
edata[2] = mm128_swap64_32( casti_m128i( pdata, 2 ) );
@@ -470,6 +471,7 @@ int scanhash_x17_4way( struct work *work, uint32_t max_nonce,
const __m256i four = m256_const1_64( 4 );
const bool bench = opt_benchmark;
// convert LE32 to LE64
edata[0] = mm128_swap64_32( casti_m128i( pdata, 0 ) );
edata[1] = mm128_swap64_32( casti_m128i( pdata, 1 ) );
edata[2] = mm128_swap64_32( casti_m128i( pdata, 2 ) );