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

v3.6.4

Browse Source
This commit is contained in:
Jay D Dee
2017-05-02 10:28:19 -04:00
parent 0155cd1bfe
commit e7dbd27636
8 changed files with 535 additions and 116 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
algo/lbry.c
View File

@@ -16,30 +16,6 @@
#define LBRY_WORK_DATA_SIZE 192
#define LBRY_WORK_CMP_SIZE 76 // same as default
/* Move init out of loop, so init once externally, and then use one single memcpy with that bigger memory block */
typedef struct {
#if defined __SHA__
SHA256_CTX sha256;
#else
sph_sha256_context sha256;
#endif
sph_sha512_context sha512;
sph_ripemd160_context ripemd;
} lbryhash_context_holder;
/* no need to copy, because close reinit the context */
static lbryhash_context_holder ctx __attribute__ ((aligned (64)));
void init_lbry_contexts(void *dummy)
{
#if defined __SHA__
SHA256_Init( &ctx.sha256 );
#else
sph_sha256_init( &ctx.sha256 );
#endif
sph_sha512_init( &ctx.sha512 );
sph_ripemd160_init( &ctx.ripemd );
}
void lbry_hash(void* output, const void* input)
{

429
algo/timetravel10.c Normal file
View File

@@ -0,0 +1,429 @@
#include <miner.h>
#include "algo-gate-api.h"
#include <stdlib.h>
#include <stdint.h>
#include <string.h>
#include <stdio.h>
#include "avxdefs.h"
#include "algo/blake/sph_blake.h"
#include "algo/bmw/sph_bmw.h"
#include "algo/jh/sph_jh.h"
#include "algo/keccak/sph_keccak.h"
#include "algo/skein/sph_skein.h"
#include "algo/luffa/sse2/luffa_for_sse2.h"
#include "algo/cubehash/sse2/cubehash_sse2.h"
#include "algo/shavite/sph_shavite.h"
#include "algo/simd/sse2/nist.h"
#ifdef NO_AES_NI
#include "algo/groestl/sph_groestl.h"
#else
#include "algo/groestl/aes_ni/hash-groestl.h"
#endif
// BitCore Genesis Timestamp
#define HASH_FUNC_BASE_TIMESTAMP 1492973331U
#define HASH_FUNC_COUNT 10
#define HASH_FUNC_COUNT_PERMUTATIONS 40320
static __thread uint32_t s_ntime = UINT32_MAX;
static __thread int permutation[HASH_FUNC_COUNT] = { 0 };
inline void tt10_swap( int *a, int *b )
{
int c = *a;
*a = *b;
*b = c;
}
inline void reverse( int *pbegin, int *pend )
{
while ( (pbegin != pend) && (pbegin != --pend) )
{
tt10_swap( pbegin, pend );
pbegin++;
}
}
static void next_permutation( int *pbegin, int *pend )
{
if ( pbegin == pend )
return;
int *i = pbegin;
++i;
if ( i == pend )
return;
i = pend;
--i;
while (1)
{
int *j = i;
--i;
if ( *i < *j )
{
int *k = pend;
while ( !(*i < *--k) ) /* do nothing */ ;
tt10_swap( i, k );
reverse(j, pend);
return; // true
}
if ( i == pbegin )
{
reverse(pbegin, pend);
return; // false
}
// else?
}
}
typedef struct {
sph_blake512_context blake;
sph_bmw512_context bmw;
sph_skein512_context skein;
sph_jh512_context jh;
sph_keccak512_context keccak;
hashState_luffa luffa;
cubehashParam cube;
sph_shavite512_context shavite;
hashState_sd simd;
#ifdef NO_AES_NI
sph_groestl512_context groestl;
#else
hashState_groestl groestl;
#endif
} tt10_ctx_holder;
tt10_ctx_holder tt10_ctx __attribute__ ((aligned (64)));
__thread tt10_ctx_holder tt10_mid __attribute__ ((aligned (64)));
void init_tt10_ctx()
{
sph_blake512_init( &tt10_ctx.blake );
sph_bmw512_init( &tt10_ctx.bmw );
sph_skein512_init( &tt10_ctx.skein );
sph_jh512_init( &tt10_ctx.jh );
sph_keccak512_init( &tt10_ctx.keccak );
init_luffa( &tt10_ctx.luffa, 512 );
cubehashInit( &tt10_ctx.cube, 512, 16, 32 );
sph_shavite512_init( &tt10_ctx.shavite );
init_sd( &tt10_ctx.simd, 512 );
#ifdef NO_AES_NI
sph_groestl512_init( &tt10_ctx.groestl );
#else
init_groestl( &tt10_ctx.groestl, 64 );
#endif
};
void timetravel10_hash(void *output, const void *input)
{
uint32_t hash[128] __attribute__ ((aligned (64)));
uint32_t *hashA, *hashB;
tt10_ctx_holder ctx __attribute__ ((aligned (64)));
uint32_t dataLen = 64;
uint32_t *work_data = (uint32_t *)input;
int i;
const int midlen = 64; // bytes
const int tail = 80 - midlen; // 16
memcpy( &ctx, &tt10_ctx, sizeof(tt10_ctx) );
for ( i = 0; i < HASH_FUNC_COUNT; i++ )
{
if (i == 0)
{
dataLen = 80;
hashA = work_data;
}
else
{
dataLen = 64;
hashA = &hash[16 * (i - 1)];
}
hashB = &hash[16 * i];
switch ( permutation[i] )
{
case 0:
if ( i == 0 )
{
memcpy( &ctx.blake, &tt10_mid.blake, sizeof tt10_mid.blake );
sph_blake512( &ctx.blake, input + midlen, tail );
sph_blake512_close( &ctx.blake, hashB );
}
else
{
sph_blake512( &ctx.blake, hashA, dataLen );
sph_blake512_close( &ctx.blake, hashB );
}
break;
case 1:
if ( i == 0 )
{
memcpy( &ctx.bmw, &tt10_mid.bmw, sizeof tt10_mid.bmw );
sph_bmw512( &ctx.bmw, input + midlen, tail );
sph_bmw512_close( &ctx.bmw, hashB );
}
else
{
sph_bmw512( &ctx.bmw, hashA, dataLen );
sph_bmw512_close( &ctx.bmw, hashB );
}
break;
case 2:
#ifdef NO_AES_NI
if ( i == 0 )
{
memcpy( &ctx.groestl, &tt10_mid.groestl, sizeof tt10_mid.groestl );
sph_groestl512( &ctx.groestl, input + midlen, tail );
sph_groestl512_close( &ctx.groestl, hashB );
}
else
{
sph_groestl512( &ctx.groestl, hashA, dataLen );
sph_groestl512_close( &ctx.groestl, hashB );
}
#else
// groestl midstate is slower
// if ( i == 0 )
// {
// memcpy( &ctx.groestl, &tt10_mid.groestl, sizeof tt10_mid.groestl );
// update_and_final_groestl( &ctx.groestl, (char*)hashB,
// (char*)input + midlen, tail*8 );
// }
// else
// {
update_and_final_groestl( &ctx.groestl, (char*)hashB,
(char*)hashA, dataLen*8 );
// }
#endif
break;
case 3:
if ( i == 0 )
{
memcpy( &ctx.skein, &tt10_mid.skein, sizeof tt10_mid.skein );
sph_skein512( &ctx.skein, input + midlen, tail );
sph_skein512_close( &ctx.skein, hashB );
}
else
{
sph_skein512( &ctx.skein, hashA, dataLen );
sph_skein512_close( &ctx.skein, hashB );
}
break;
case 4:
if ( i == 0 )
{
memcpy( &ctx.jh, &tt10_mid.jh, sizeof tt10_mid.jh );
sph_jh512( &ctx.jh, input + midlen, tail );
sph_jh512_close( &ctx.jh, hashB );
}
else
{
sph_jh512( &ctx.jh, hashA, dataLen );
sph_jh512_close( &ctx.jh, hashB);
}
break;
case 5:
if ( i == 0 )
{
memcpy( &ctx.keccak, &tt10_mid.keccak, sizeof tt10_mid.keccak );
sph_keccak512( &ctx.keccak, input + midlen, tail );
sph_keccak512_close( &ctx.keccak, hashB );
}
else
{
sph_keccak512( &ctx.keccak, hashA, dataLen );
sph_keccak512_close( &ctx.keccak, hashB );
}
break;
case 6:
if ( i == 0 )
{
memcpy( &ctx.luffa, &tt10_mid.luffa, sizeof tt10_mid.luffa );
update_and_final_luffa( &ctx.luffa, (BitSequence*)hashB,
(const BitSequence *)input + 64, 16 );
}
else
{
update_and_final_luffa( &ctx.luffa, (BitSequence*)hashB,
(const BitSequence *)hashA, dataLen );
}
break;
case 7:
if ( i == 0 )
{
memcpy( &ctx.cube, &tt10_mid.cube, sizeof tt10_mid.cube );
cubehashUpdateDigest( &ctx.cube, (byte*)hashB,
(const byte*)input + midlen, tail );
}
else
{
cubehashUpdateDigest( &ctx.cube, (byte*)hashB, (const byte*)hashA,
dataLen );
}
break;
case 8:
if ( i == 0 )
{
memcpy( &ctx.shavite, &tt10_mid.shavite, sizeof tt10_mid.shavite );
sph_shavite512( &ctx.shavite, input + midlen, tail*8 );
sph_shavite512_close( &ctx.shavite, hashB );
}
else
{
sph_shavite512( &ctx.shavite, hashA, dataLen );
sph_shavite512_close( &ctx.shavite, hashB );
}
break;
case 9:
if ( i == 0 )
{
memcpy( &ctx.simd, &tt10_mid.simd, sizeof tt10_mid.simd );
update_final_sd( &ctx.simd, (BitSequence *)hashB,
(const BitSequence *)input + midlen, tail*8 );
}
else
{
update_final_sd( &ctx.simd, (BitSequence *)hashB,
(const BitSequence *)hashA, dataLen*8 );
}
break;
default:
break;
}
}
memcpy(output, &hash[16 * (HASH_FUNC_COUNT - 1)], 32);
}
int scanhash_timetravel10( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done )
{
uint32_t _ALIGN(64) hash[8];
uint32_t _ALIGN(64) endiandata[20];
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 nonce = first_nonce;
volatile uint8_t *restart = &(work_restart[thr_id].restart);
int i;
if (opt_benchmark)
ptarget[7] = 0x0cff;
for (int k=0; k < 19; k++)
be32enc(&endiandata[k], pdata[k]);
const uint32_t timestamp = endiandata[17];
if ( timestamp != s_ntime )
{
const int steps = ( timestamp - HASH_FUNC_BASE_TIMESTAMP )
% HASH_FUNC_COUNT_PERMUTATIONS;
for ( i = 0; i < HASH_FUNC_COUNT; i++ )
permutation[i] = i;
for ( i = 0; i < steps; i++ )
next_permutation( permutation, permutation + HASH_FUNC_COUNT );
s_ntime = timestamp;
// do midstate precalc for first function
switch ( permutation[0] )
{
case 0:
memcpy( &tt10_mid.blake, &tt10_ctx.blake, sizeof(tt10_mid.blake) );
sph_blake512( &tt10_mid.blake, endiandata, 64 );
break;
case 1:
memcpy( &tt10_mid.bmw, &tt10_ctx.bmw, sizeof(tt10_mid.bmw) );
sph_bmw512( &tt10_mid.bmw, endiandata, 64 );
break;
case 2:
#ifdef NO_AES_NI
memcpy( &tt10_mid.groestl, &tt10_ctx.groestl, sizeof(tt10_mid.groestl ) );
sph_groestl512( &tt10_mid.groestl, endiandata, 64 );
#else
// groestl midstate is slower
// memcpy( &tt10_mid.groestl, &tt10_ctx.groestl, sizeof(tt10_mid.groestl ) );
// update_groestl( &tt10_mid.groestl, (char*)endiandata, 64*8 );
#endif
break;
case 3:
memcpy( &tt10_mid.skein, &tt10_ctx.skein, sizeof(tt10_mid.skein ) );
sph_skein512( &tt10_mid.skein, endiandata, 64 );
break;
case 4:
memcpy( &tt10_mid.jh, &tt10_ctx.jh, sizeof(tt10_mid.jh ) );
sph_jh512( &tt10_mid.jh, endiandata, 64 );
break;
case 5:
memcpy( &tt10_mid.keccak, &tt10_ctx.keccak, sizeof(tt10_mid.keccak ) );
sph_keccak512( &tt10_mid.keccak, endiandata, 64 );
break;
case 6:
memcpy( &tt10_mid.luffa, &tt10_ctx.luffa, sizeof(tt10_mid.luffa ) );
update_luffa( &tt10_mid.luffa, (const BitSequence*)endiandata, 64 );
break;
case 7:
memcpy( &tt10_mid.cube, &tt10_ctx.cube, sizeof(tt10_mid.cube ) );
cubehashUpdate( &tt10_mid.cube, (const byte*)endiandata, 64 );
break;
case 8:
memcpy( &tt10_mid.shavite, &tt10_ctx.shavite, sizeof(tt10_mid.shavite ) );
sph_shavite512( &tt10_mid.shavite, endiandata, 64 );
break;
case 9:
memcpy( &tt10_mid.simd, &tt10_ctx.simd, sizeof(tt10_mid.simd ) );
update_sd( &tt10_mid.simd, (const BitSequence *)endiandata, 512 );
break;
default:
break;
}
}
do {
be32enc( &endiandata[19], nonce );
timetravel10_hash( hash, endiandata );
if ( hash[7] <= Htarg && fulltest( hash, ptarget) )
{
work_set_target_ratio( work, hash );
pdata[19] = nonce;
*hashes_done = pdata[19] - first_nonce;
return 1;
}
nonce++;
} while (nonce < max_nonce && !(*restart));
pdata[19] = nonce;
*hashes_done = pdata[19] - first_nonce + 1;
return 0;
}
void timetravel10_set_target( struct work* work, double job_diff )
{
work_set_target( work, job_diff / (256.0 * opt_diff_factor) );
}
bool register_timetravel10_algo( algo_gate_t* gate )
{
gate->optimizations = SSE2_OPT | AES_OPT | AVX_OPT | AVX2_OPT;
init_tt10_ctx();
gate->scanhash = (void*)&scanhash_timetravel10;
gate->hash = (void*)&timetravel10_hash;
gate->set_target = (void*)&timetravel10_set_target;
gate->get_max64 = (void*)&get_max64_0xffffLL;
return true;
};