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

v3.8.8.1

Browse Source
This commit is contained in:
Jay D Dee
2018-05-11 11:52:36 -04:00
parent 9edc650042
commit bfd1c002f9
27 changed files with 539 additions and 403 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
algo/cryptonight/cryptolight.c
View File

@@ -325,7 +325,7 @@ int scanhash_cryptolight(int thr_id, struct work *work,
struct cryptonight_ctx *ctx = (struct cryptonight_ctx*)malloc(sizeof(struct cryptonight_ctx));
#ifndef NO_AES_NI
#if defined(__AES__)
do {
*nonceptr = ++n;
cryptolight_hash_ctx_aes_ni(hash, pdata, 76, ctx);

61
algo/cryptonight/cryptonight-aesni.c
View File

@@ -1,14 +1,11 @@
#if defined(__AES__)
#include <x86intrin.h>
#include <memory.h>
#include "cryptonight.h"
#include "miner.h"
#include "crypto/c_keccak.h"
#include <immintrin.h>
//#include "avxdefs.h"
void aesni_parallel_noxor(uint8_t *long_state, uint8_t *text, uint8_t *ExpandedKey);
void aesni_parallel_xor(uint8_t *text, uint8_t *ExpandedKey, uint8_t *long_state);
void that_fucking_loop(uint8_t a[16], uint8_t b[16], uint8_t *long_state);
static inline void ExpandAESKey256_sub1(__m128i *tmp1, __m128i *tmp2)
{
@@ -25,7 +22,6 @@ static inline void ExpandAESKey256_sub1(__m128i *tmp1, __m128i *tmp2)
static inline void ExpandAESKey256_sub2(__m128i *tmp1, __m128i *tmp3)
{
#ifndef NO_AES_NI
__m128i tmp2, tmp4;
tmp4 = _mm_aeskeygenassist_si128(*tmp1, 0x00);
@@ -37,14 +33,12 @@ static inline void ExpandAESKey256_sub2(__m128i *tmp1, __m128i *tmp3)
tmp4 = _mm_slli_si128(tmp4, 0x04);
*tmp3 = _mm_xor_si128(*tmp3, tmp4);
*tmp3 = _mm_xor_si128(*tmp3, tmp2);
#endif
}
// Special thanks to Intel for helping me
// with ExpandAESKey256() and its subroutines
static inline void ExpandAESKey256(char *keybuf)
{
#ifndef NO_AES_NI
__m128i tmp1, tmp2, tmp3, *keys;
keys = (__m128i *)keybuf;
@@ -91,7 +85,6 @@ static inline void ExpandAESKey256(char *keybuf)
tmp2 = _mm_aeskeygenassist_si128(tmp3, 0x40);
ExpandAESKey256_sub1(&tmp1, &tmp2);
keys[14] = tmp1;
#endif
}
// align to 64 byte cache line
@@ -109,13 +102,19 @@ static __thread cryptonight_ctx ctx;
void cryptonight_hash_aes( void *restrict output, const void *input, int len )
{
#ifndef NO_AES_NI
uint8_t ExpandedKey[256] __attribute__((aligned(64)));
__m128i *longoutput, *expkey, *xmminput;
size_t i, j;
keccak( (const uint8_t*)input, 76, (char*)&ctx.state.hs.b, 200 );
if ( cryptonightV7 && len < 43 )
return;
const uint64_t tweak = cryptonightV7
? *((const uint64_t*) (((const uint8_t*)input) + 35))
^ ctx.state.hs.w[24] : 0;
memcpy( ExpandedKey, ctx.state.hs.b, AES_KEY_SIZE );
ExpandAESKey256( ExpandedKey );
memcpy( ctx.text, ctx.state.init, INIT_SIZE_BYTE );
@@ -214,7 +213,15 @@ void cryptonight_hash_aes( void *restrict output, const void *input, int len )
_mm_store_si128( (__m128i*)c, c_x );
b_x = _mm_xor_si128( b_x, c_x );
nextblock = (uint64_t *)&ctx.long_state[c[0] & 0x1FFFF0];
_mm_store_si128( lsa, b_x );
_mm_store_si128( lsa, b_x );
if ( cryptonightV7 )
{
const uint8_t tmp = ( (const uint8_t*)(lsa) )[11];
const uint8_t index = ( ( (tmp >> 3) & 6 ) | (tmp & 1) ) << 1;
((uint8_t*)(lsa))[11] = tmp ^ ( ( 0x75310 >> index) & 0x30 );
}
b[0] = nextblock[0];
b[1] = nextblock[1];
@@ -227,10 +234,14 @@ void cryptonight_hash_aes( void *restrict output, const void *input, int len )
: "cc" );
b_x = c_x;
nextblock[0] = a[0] + hi;
nextblock[1] = a[1] + lo;
a[0] = b[0] ^ nextblock[0];
a[1] = b[1] ^ nextblock[1];
a[0] += hi;
a[1] += lo;
nextblock[0] = a[0];
nextblock[1] = cryptonightV7 ? a[1] ^ tweak : a[1];
a[0] ^= b[0];
a[1] ^= b[1];
lsa = (__m128i*)&ctx.long_state[ a[0] & 0x1FFFF0 ];
a_x = _mm_load_si128( (__m128i*)a );
c_x = _mm_load_si128( lsa );
@@ -241,6 +252,14 @@ void cryptonight_hash_aes( void *restrict output, const void *input, int len )
b_x = _mm_xor_si128( b_x, c_x );
nextblock = (uint64_t *)&ctx.long_state[c[0] & 0x1FFFF0];
_mm_store_si128( lsa, b_x );
if ( cryptonightV7 )
{
const uint8_t tmp = ( (const uint8_t*)(lsa) )[11];
const uint8_t index = ( ( (tmp >> 3) & 6 ) | (tmp & 1) ) << 1;
((uint8_t*)(lsa))[11] = tmp ^ ( ( 0x75310 >> index) & 0x30 );
}
b[0] = nextblock[0];
b[1] = nextblock[1];
@@ -251,8 +270,12 @@ void cryptonight_hash_aes( void *restrict output, const void *input, int len )
"rm" ( b[0] )
: "cc" );
nextblock[0] = a[0] + hi;
nextblock[1] = a[1] + lo;
a[0] += hi;
a[1] += lo;
nextblock[0] = a[0];
nextblock[1] = cryptonightV7 ? a[1] ^ tweak : a[1];
a[0] ^= b[0];
a[1] ^= b[1];
memcpy( ExpandedKey, &ctx.state.hs.b[32], AES_KEY_SIZE );
ExpandAESKey256( ExpandedKey );
@@ -330,5 +353,5 @@ void cryptonight_hash_aes( void *restrict output, const void *input, int len )
keccakf( (uint64_t*)&ctx.state.hs.w, 24 );
extra_hashes[ctx.state.hs.b[0] & 3](&ctx.state, 200, output);
#endif
}
#endif

46
algo/cryptonight/cryptonight-common.c
View File

@@ -7,11 +7,11 @@
#include "cpuminer-config.h"
#include "algo-gate-api.h"
#ifndef NO_AES_NI
#if defined(__AES__)
#include "algo/groestl/aes_ni/hash-groestl256.h"
#endif
#else
#include "crypto/c_groestl.h"
#endif
#include "crypto/c_blake256.h"
#include "crypto/c_jh.h"
#include "crypto/c_skein.h"
@@ -30,12 +30,12 @@ void do_blake_hash(const void* input, size_t len, char* output) {
}
void do_groestl_hash(const void* input, size_t len, char* output) {
#ifdef NO_AES_NI
groestl(input, len * 8, (uint8_t*)output);
#else
#if defined(__AES__)
hashState_groestl256 ctx;
init_groestl256( &ctx, 32 );
update_and_final_groestl256( &ctx, output, input, len * 8 );
#else
groestl(input, len * 8, (uint8_t*)output);
#endif
}
@@ -52,23 +52,24 @@ void (* const extra_hashes[4])( const void *, size_t, char *) =
void cryptonight_hash( void *restrict output, const void *input, int len )
{
#ifdef NO_AES_NI
cryptonight_hash_ctx ( output, input, len );
#else
#if defined(__AES__)
cryptonight_hash_aes( output, input, len );
#else
cryptonight_hash_ctx ( output, input, len );
#endif
}
void cryptonight_hash_suw( void *restrict output, const void *input )
{
#ifdef NO_AES_NI
cryptonight_hash_ctx ( output, input, 76 );
#else
#if defined(__AES__)
cryptonight_hash_aes( output, input, 76 );
#else
cryptonight_hash_ctx ( output, input, 76 );
#endif
}
bool cryptonightV7 = false;
int scanhash_cryptonight( int thr_id, struct work *work, uint32_t max_nonce,
uint64_t *hashes_done )
{
@@ -80,6 +81,11 @@ int scanhash_cryptonight( int thr_id, struct work *work, uint32_t max_nonce,
const uint32_t first_nonce = n + 1;
const uint32_t Htarg = ptarget[7];
uint32_t hash[32 / 4] __attribute__((aligned(32)));
// if ( ( cryptonightV7 && ( *(uint8_t*)pdata < 7 ) )
// || ( !cryptonightV7 && ( *(uint8_t*)pdata == 7 ) ) )
// applog(LOG_WARNING,"Cryptonight variant mismatch, shares may be rejected.");
do
{
*nonceptr = ++n;
@@ -87,6 +93,7 @@ int scanhash_cryptonight( int thr_id, struct work *work, uint32_t max_nonce,
if (unlikely( hash[7] < Htarg ))
{
*hashes_done = n - first_nonce + 1;
// work_set_target_ratio( work, hash );
return true;
}
} while (likely((n <= max_nonce && !work_restart[thr_id].restart)));
@@ -97,6 +104,7 @@ int scanhash_cryptonight( int thr_id, struct work *work, uint32_t max_nonce,
bool register_cryptonight_algo( algo_gate_t* gate )
{
cryptonightV7 = false;
register_json_rpc2( gate );
gate->optimizations = SSE2_OPT | AES_OPT;
gate->scanhash = (void*)&scanhash_cryptonight;
@@ -106,3 +114,15 @@ bool register_cryptonight_algo( algo_gate_t* gate )
return true;
};
bool register_cryptonightv7_algo( algo_gate_t* gate )
{
cryptonightV7 = true;
register_json_rpc2( gate );
gate->optimizations = SSE2_OPT | AES_OPT;
gate->scanhash = (void*)&scanhash_cryptonight;
gate->hash = (void*)&cryptonight_hash;
gate->hash_suw = (void*)&cryptonight_hash_suw;
gate->get_max64 = (void*)&get_max64_0x40LL;
return true;
};

84
algo/cryptonight/cryptonight.c
View File

@@ -20,8 +20,8 @@
#include "crypto/c_jh.h"
#include "crypto/c_skein.h"
#include "crypto/int-util.h"
#include "crypto/hash-ops.h"
//#include "cryptonight.h"
//#include "crypto/hash-ops.h"
#include "cryptonight.h"
#if USE_INT128
@@ -51,6 +51,7 @@ typedef __uint128_t uint128_t;
#define INIT_SIZE_BLK 8
#define INIT_SIZE_BYTE (INIT_SIZE_BLK * AES_BLOCK_SIZE)
/*
#pragma pack(push, 1)
union cn_slow_hash_state {
union hash_state hs;
@@ -78,6 +79,7 @@ static void do_skein_hash(const void* input, size_t len, char* output) {
int r = skein_hash(8 * HASH_SIZE, input, 8 * len, (uint8_t*)output);
assert(likely(SKEIN_SUCCESS == r));
}
*/
extern int aesb_single_round(const uint8_t *in, uint8_t*out, const uint8_t *expandedKey);
extern int aesb_pseudo_round_mut(uint8_t *val, uint8_t *expandedKey);
@@ -120,9 +122,11 @@ static uint64_t mul128(uint64_t multiplier, uint64_t multiplicand, uint64_t* pro
extern uint64_t mul128(uint64_t multiplier, uint64_t multiplicand, uint64_t* product_hi);
#endif
/*
static void (* const extra_hashes[4])(const void *, size_t, char *) = {
do_blake_hash, do_groestl_hash, do_jh_hash, do_skein_hash
};
*/
static inline size_t e2i(const uint8_t* a) {
#if !LITE
@@ -132,14 +136,16 @@ static inline size_t e2i(const uint8_t* a) {
#endif
}
static inline void mul_sum_xor_dst(const uint8_t* a, uint8_t* c, uint8_t* dst) {
static inline void mul_sum_xor_dst( const uint8_t* a, uint8_t* c, uint8_t* dst,
const uint64_t tweak )
{
uint64_t hi, lo = mul128(((uint64_t*) a)[0], ((uint64_t*) dst)[0], &hi) + ((uint64_t*) c)[1];
hi += ((uint64_t*) c)[0];
((uint64_t*) c)[0] = ((uint64_t*) dst)[0] ^ hi;
((uint64_t*) c)[1] = ((uint64_t*) dst)[1] ^ lo;
((uint64_t*) dst)[0] = hi;
((uint64_t*) dst)[1] = lo;
((uint64_t*) dst)[1] = cryptonightV7 ? lo ^ tweak : lo;
}
static inline void xor_blocks(uint8_t* a, const uint8_t* b) {
@@ -174,8 +180,16 @@ static __thread cryptonight_ctx ctx;
void cryptonight_hash_ctx(void* output, const void* input, int len)
{
hash_process(&ctx.state.hs, (const uint8_t*) input, len);
ctx.aes_ctx = (oaes_ctx*) oaes_alloc();
// hash_process(&ctx.state.hs, (const uint8_t*) input, len);
keccak( (const uint8_t*)input, 76, (char*)&ctx.state.hs.b, 200 );
if ( cryptonightV7 && len < 43 )
return;
const uint64_t tweak = cryptonightV7
? *((const uint64_t*) (((const uint8_t*)input) + 35))
^ ctx.state.hs.w[24] : 0;
ctx.aes_ctx = (oaes_ctx*) oaes_alloc();
__builtin_prefetch( ctx.text, 0, 3 );
__builtin_prefetch( ctx.text + 64, 0, 3 );
@@ -211,23 +225,44 @@ void cryptonight_hash_ctx(void* output, const void* input, int len)
xor_blocks_dst(&ctx.state.k[0], &ctx.state.k[32], ctx.a);
xor_blocks_dst(&ctx.state.k[16], &ctx.state.k[48], ctx.b);
for (i = 0; likely(i < ITER / 4); ++i) {
/* Dependency chain: address -> read value ------+
* written value <-+ hard function (AES or MUL) <+
* next address <-+
*/
/* Iteration 1 */
j = e2i(ctx.a);
aesb_single_round(&ctx.long_state[j], ctx.c, ctx.a);
xor_blocks_dst(ctx.c, ctx.b, &ctx.long_state[j]);
/* Iteration 2 */
mul_sum_xor_dst(ctx.c, ctx.a, &ctx.long_state[e2i(ctx.c)]);
/* Iteration 3 */
j = e2i(ctx.a);
aesb_single_round(&ctx.long_state[j], ctx.b, ctx.a);
xor_blocks_dst(ctx.b, ctx.c, &ctx.long_state[j]);
/* Iteration 4 */
mul_sum_xor_dst(ctx.b, ctx.a, &ctx.long_state[e2i(ctx.b)]);
for (i = 0; likely(i < ITER / 4); ++i)
{
/* Dependency chain: address -> read value ------+
* written value <-+ hard function (AES or MUL) <+
* next address <-+
*/
/* Iteration 1 */
j = e2i(ctx.a);
aesb_single_round(&ctx.long_state[j], ctx.c, ctx.a);
xor_blocks_dst(ctx.c, ctx.b, &ctx.long_state[j]);
if ( cryptonightV7 )
{
uint8_t *lsa = (uint8_t*)&ctx.long_state[((uint64_t *)(ctx.a))[0] & 0x1FFFF0];
const uint8_t tmp = lsa[11];
const uint8_t index = ( ( (tmp >> 3) & 6 ) | (tmp & 1) ) << 1;
lsa[11] = tmp ^ ( ( 0x75310 >> index) & 0x30 );
}
/* Iteration 2 */
mul_sum_xor_dst(ctx.c, ctx.a, &ctx.long_state[e2i(ctx.c)], tweak );
/* Iteration 3 */
j = e2i(ctx.a);
aesb_single_round(&ctx.long_state[j], ctx.b, ctx.a);
xor_blocks_dst(ctx.b, ctx.c, &ctx.long_state[j]);
if ( cryptonightV7 )
{
uint8_t *lsa = (uint8_t*)&ctx.long_state[((uint64_t *)(ctx.a))[0] & 0x1FFFF0];
const uint8_t tmp = lsa[11];
const uint8_t index = ( ( (tmp >> 3) & 6 ) | (tmp & 1) ) << 1;
lsa[11] = tmp ^ ( ( 0x75310 >> index) & 0x30 );
}
/* Iteration 4 */
mul_sum_xor_dst(ctx.b, ctx.a, &ctx.long_state[e2i(ctx.b)], tweak );
}
__builtin_prefetch( ctx.text, 0, 3 );
@@ -266,7 +301,8 @@ void cryptonight_hash_ctx(void* output, const void* input, int len)
aesb_pseudo_round_mut(&ctx.text[7 * AES_BLOCK_SIZE], ctx.aes_ctx->key->exp_data);
}
memcpy(ctx.state.init, ctx.text, INIT_SIZE_BYTE);
hash_permutation(&ctx.state.hs);
// hash_permutation(&ctx.state.hs);
keccakf( (uint64_t*)&ctx.state.hs.w, 24 );
/*memcpy(hash, &state, 32);*/
extra_hashes[ctx.state.hs.b[0] & 3](&ctx.state, 200, output);
oaes_free((OAES_CTX **) &ctx.aes_ctx);

2
algo/cryptonight/cryptonight.h
View File

@@ -45,5 +45,7 @@ int scanhash_cryptonight( int thr_id, struct work *work, uint32_t max_nonce,
void cryptonight_hash_aes( void *restrict output, const void *input, int len );
extern bool cryptonightV7;
#endif

2
algo/lyra2/lyra2h-4way.c
View File

@@ -67,7 +67,7 @@ int scanhash_lyra2h_4way( int thr_id, struct work *work, uint32_t max_nonce,
if ( opt_benchmark )
ptarget[7] = 0x0000ff;
for ( int i=0; i < 19; i++ )
for ( int i=0; i < 20; i++ )
be32enc( &edata[i], pdata[i] );
mm_interleave_4x32( vdata, edata, edata, edata, edata, 640 );

2
algo/lyra2/lyra2z-4way.c
View File

@@ -67,7 +67,7 @@ int scanhash_lyra2z_4way( int thr_id, struct work *work, uint32_t max_nonce,
if ( opt_benchmark )
ptarget[7] = 0x0000ff;
for ( int i=0; i < 19; i++ )
for ( int i=0; i < 20; i++ )
be32enc( &edata[i], pdata[i] );
mm_interleave_4x32( vdata, edata, edata, edata, edata, 640 );

2
algo/sha/sha256t-4way.c
View File

@@ -60,7 +60,7 @@ int scanhash_sha256t_8way( int thr_id, struct work *work,
0xFFFF0000,
0 };
for ( int k = 0; k < 19; k++ )
for ( int k = 0; k < 20; k++ )
be32enc( &edata[k], pdata[k] );
mm256_interleave_8x32( vdata, edata, edata, edata, edata,

6
algo/x11/x11evo-gate.c
View File

@@ -31,11 +31,13 @@ int nextPerm( uint8_t n[], uint32_t count )
return 0;
for ( i = count - 1; i>0 && n[i - 1] >= n[i]; i-- );
tail = i;
tail = i;
if ( tail > 0 )
{
for ( j = count - 1; j>tail && n[j] <= n[tail - 1]; j-- );
evo_swap( &n[tail - 1], &n[j] );
evo_swap( &n[tail - 1], &n[j] );
}
for ( i = tail, j = count - 1; i<j; i++, j-- )
evo_swap( &n[i], &n[j] );

12
algo/x17/x16r-4way.c
View File

@@ -68,13 +68,7 @@ void x16r_4way_hash( void* output, const void* input )
uint32_t hash2[24] __attribute__ ((aligned (64)));
uint32_t hash3[24] __attribute__ ((aligned (64)));
uint32_t vhash[24*4] __attribute__ ((aligned (64)));
// uint32_t inp0[24] __attribute__ ((aligned (64)));
// uint32_t inp1[24] __attribute__ ((aligned (64)));
// uint32_t inp2[24] __attribute__ ((aligned (64)));
// uint32_t inp3[24] __attribute__ ((aligned (64)));
x16r_4way_ctx_holder ctx;
void *in0 = (void*) hash0;
void *in1 = (void*) hash1;
void *in2 = (void*) hash2;
@@ -290,10 +284,6 @@ void x16r_4way_hash( void* output, const void* input )
mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash, 512 );
break;
}
// in0 = (void*) hash0;
// in1 = (void*) hash1;
// in2 = (void*) hash2;
// in3 = (void*) hash3;
size = 64;
}
memcpy( output, hash0, 32 );
@@ -325,7 +315,7 @@ int scanhash_x16r_4way( int thr_id, struct work *work, uint32_t max_nonce,
if ( s_ntime != endiandata[17] )
{
uint32_t ntime = swab32(pdata[17]);
x16_r_s_getAlgoString( (const char*) (&endiandata[1]), hashOrder );
x16_r_s_getAlgoString( (const uint8_t*) (&endiandata[1]), hashOrder );
s_ntime = ntime;
if ( opt_debug && !thr_id )
applog( LOG_DEBUG, "hash order %s (%08x)", hashOrder, ntime );

7
algo/x17/x16r-gate.c
View File

@@ -1,6 +1,6 @@
#include "x16r-gate.h"
void x16r_getAlgoString( const char* prevblock, char *output )
void x16r_getAlgoString( const uint8_t* prevblock, char *output )
{
char *sptr = output;
for ( int j = 0; j < X16R_HASH_FUNC_COUNT; j++ )
@@ -16,14 +16,13 @@ void x16r_getAlgoString( const char* prevblock, char *output )
*sptr = '\0';
}
void x16s_getAlgoString( const char* prevblock, char *output )
void x16s_getAlgoString( const uint8_t* prevblock, char *output )
{
uint8_t* data = (uint8_t*)prevblock;
strcpy( output, "0123456789ABCDEF" );
for ( int i = 0; i < 16; i++ )
{
uint8_t b = (15 - i) >> 1; // 16 ascii hex chars, reversed
uint8_t algoDigit = (i & 1) ? data[b] & 0xF : data[b] >> 4;
uint8_t algoDigit = (i & 1) ? prevblock[b] & 0xF : prevblock[b] >> 4;
int offset = algoDigit;
// insert the nth character at the front
char oldVal = output[offset];

6
algo/x17/x16r-gate.h
View File

@@ -29,9 +29,9 @@ enum x16r_Algo {
X16R_HASH_FUNC_COUNT
};
bool (*x16_r_s_getAlgoString) ( const char*, char* );
void x16r_getAlgoString( const char* prevblock, char *output );
void x16s_getAlgoString( const char* prevblock, char *output );
void (*x16_r_s_getAlgoString) ( const uint8_t*, char* );
void x16r_getAlgoString( const uint8_t* prevblock, char *output );
void x16s_getAlgoString( const uint8_t* prevblock, char *output );
bool register_x16r_algo( algo_gate_t* gate );
bool register_x16s_algo( algo_gate_t* gate );

2
algo/x17/x16r.c
View File

@@ -205,7 +205,7 @@ int scanhash_x16r( int thr_id, struct work *work, uint32_t max_nonce,
if ( s_ntime != pdata[17] )
{
uint32_t ntime = swab32(pdata[17]);
x16_r_s_getAlgoString( (const char*) (&endiandata[1]), hashOrder );
x16_r_s_getAlgoString( (const uint8_t*) (&endiandata[1]), hashOrder );
s_ntime = ntime;
if ( opt_debug && !thr_id )
applog( LOG_DEBUG, "hash order %s (%08x)", hashOrder, ntime );