Browse Source

update c11 like tribus + 2.2.1 readme

pull/2/head
Tanguy Pruvot 7 years ago
parent
commit
ebf055d585
  1. 4
      README.txt
  2. 376
      compat/ccminer-config.h
  3. 1
      tribus/tribus.cu
  4. 46
      x11/c11.cu

4
README.txt

@ -1,5 +1,5 @@
ccminer 2.2 (August 2017) "Equihash, tribus and optimized skunk" ccminer 2.2.1 (Sept. 2017) "optimized tribus kernel (Maxwell+)"
--------------------------------------------------------------- ---------------------------------------------------------------
*************************************************************** ***************************************************************
@ -277,6 +277,8 @@ so we can more efficiently implement new algorithms using the latest hardware
features. features.
>>> RELEASE HISTORY <<< >>> RELEASE HISTORY <<<
Sep. 01st 2017 v2.2.1
Improve tribus algo on recent cards (up to +10%)
Aug. 13th 2017 v2.2 Aug. 13th 2017 v2.2
New skunk algo, using the heavy streebog algorithm New skunk algo, using the heavy streebog algorithm

376
compat/ccminer-config.h

@ -1,188 +1,188 @@
/* CONFIG ONLY FOR MS VC++ BUILD */ /* CONFIG ONLY FOR MS VC++ BUILD */
/* Define to one of `_getb67', `GETB67', `getb67' for Cray-2 and Cray-YMP /* Define to one of `_getb67', `GETB67', `getb67' for Cray-2 and Cray-YMP
systems. This function is required for `alloca.c' support on those systems. systems. This function is required for `alloca.c' support on those systems.
*/ */
/* #undef CRAY_STACKSEG_END */ /* #undef CRAY_STACKSEG_END */
/* Define to 1 if using `alloca.c'. */ /* Define to 1 if using `alloca.c'. */
/* #undef C_ALLOCA */ /* #undef C_ALLOCA */
/* Define to 1 if you have `alloca', as a function or macro. */ /* Define to 1 if you have `alloca', as a function or macro. */
#define HAVE_ALLOCA 1 #define HAVE_ALLOCA 1
/* Define to 1 if you have <alloca.h> and it should be used (not on Ultrix). /* Define to 1 if you have <alloca.h> and it should be used (not on Ultrix).
*/ */
#define HAVE_ALLOCA_H 1 #define HAVE_ALLOCA_H 1
/* Define to 1 if you have the declaration of `be32dec', and to 0 if you /* Define to 1 if you have the declaration of `be32dec', and to 0 if you
don't. */ don't. */
#define HAVE_DECL_BE32DEC 0 #define HAVE_DECL_BE32DEC 0
/* Define to 1 if you have the declaration of `be32enc', and to 0 if you /* Define to 1 if you have the declaration of `be32enc', and to 0 if you
don't. */ don't. */
#define HAVE_DECL_BE32ENC 0 #define HAVE_DECL_BE32ENC 0
/* Define to 1 if you have the declaration of `le32dec', and to 0 if you /* Define to 1 if you have the declaration of `le32dec', and to 0 if you
don't. */ don't. */
#define HAVE_DECL_LE32DEC 0 #define HAVE_DECL_LE32DEC 0
/* Define to 1 if you have the declaration of `le32enc', and to 0 if you /* Define to 1 if you have the declaration of `le32enc', and to 0 if you
don't. */ don't. */
#define HAVE_DECL_LE32ENC 0 #define HAVE_DECL_LE32ENC 0
/* Define to 1 if you have the `getopt_long' function. */ /* Define to 1 if you have the `getopt_long' function. */
#define HAVE_GETOPT_LONG 1 #define HAVE_GETOPT_LONG 1
/* Define to 1 if you have the <inttypes.h> header file. */ /* Define to 1 if you have the <inttypes.h> header file. */
#define HAVE_INTTYPES_H 1 #define HAVE_INTTYPES_H 1
/* Define to 1 if you have the `crypto' library (-lcrypto). */ /* Define to 1 if you have the `crypto' library (-lcrypto). */
#define HAVE_LIBCRYPTO 1 #define HAVE_LIBCRYPTO 1
/* Define to 1 if you have a functional curl library. */ /* Define to 1 if you have a functional curl library. */
#define HAVE_LIBCURL 1 #define HAVE_LIBCURL 1
/* Define to 1 if you have the `ssl' library (-lssl). */ /* Define to 1 if you have the `ssl' library (-lssl). */
#define HAVE_LIBSSL 1 #define HAVE_LIBSSL 1
/* Define to 1 if you have the `z' library (-lz). */ /* Define to 1 if you have the `z' library (-lz). */
#define HAVE_LIBZ 1 #define HAVE_LIBZ 1
/* Define to 1 if you have the <memory.h> header file. */ /* Define to 1 if you have the <memory.h> header file. */
#define HAVE_MEMORY_H 1 #define HAVE_MEMORY_H 1
/* Define to 1 if you have the <stdint.h> header file. */ /* Define to 1 if you have the <stdint.h> header file. */
#define HAVE_STDINT_H 1 #define HAVE_STDINT_H 1
/* Define to 1 if you have the <stdlib.h> header file. */ /* Define to 1 if you have the <stdlib.h> header file. */
#define HAVE_STDLIB_H 1 #define HAVE_STDLIB_H 1
/* Define to 1 if you have the <strings.h> header file. */ /* Define to 1 if you have the <strings.h> header file. */
#define HAVE_STRINGS_H 1 #define HAVE_STRINGS_H 1
/* Define to 1 if you have the <string.h> header file. */ /* Define to 1 if you have the <string.h> header file. */
#define HAVE_STRING_H 1 #define HAVE_STRING_H 1
/* Define to 1 if you have the <syslog.h> header file. */ /* Define to 1 if you have the <syslog.h> header file. */
/* #undef HAVE_SYSLOG_H */ /* #undef HAVE_SYSLOG_H */
/* Define to 1 if you have the <sys/endian.h> header file. */ /* Define to 1 if you have the <sys/endian.h> header file. */
/* #undef HAVE_SYS_ENDIAN_H */ /* #undef HAVE_SYS_ENDIAN_H */
/* Define to 1 if you have the <sys/param.h> header file. */ /* Define to 1 if you have the <sys/param.h> header file. */
#define HAVE_SYS_PARAM_H 1 #define HAVE_SYS_PARAM_H 1
/* Define to 1 if you have the <sys/stat.h> header file. */ /* Define to 1 if you have the <sys/stat.h> header file. */
#define HAVE_SYS_STAT_H 1 #define HAVE_SYS_STAT_H 1
/* Define to 1 if you have the <sys/sysctl.h> header file. */ /* Define to 1 if you have the <sys/sysctl.h> header file. */
#define HAVE_SYS_SYSCTL_H 1 #define HAVE_SYS_SYSCTL_H 1
/* Define to 1 if you have the <sys/types.h> header file. */ /* Define to 1 if you have the <sys/types.h> header file. */
#define HAVE_SYS_TYPES_H 1 #define HAVE_SYS_TYPES_H 1
/* Define to 1 if you have the <unistd.h> header file. */ /* Define to 1 if you have the <unistd.h> header file. */
#define HAVE_UNISTD_H 1 #define HAVE_UNISTD_H 1
/* Defined if libcurl supports AsynchDNS */ /* Defined if libcurl supports AsynchDNS */
/* #undef LIBCURL_FEATURE_ASYNCHDNS */ /* #undef LIBCURL_FEATURE_ASYNCHDNS */
/* Defined if libcurl supports IDN */ /* Defined if libcurl supports IDN */
#define LIBCURL_FEATURE_IDN 1 #define LIBCURL_FEATURE_IDN 1
/* Defined if libcurl supports IPv6 */ /* Defined if libcurl supports IPv6 */
#define LIBCURL_FEATURE_IPV6 1 #define LIBCURL_FEATURE_IPV6 1
/* Defined if libcurl supports KRB4 */ /* Defined if libcurl supports KRB4 */
/* #undef LIBCURL_FEATURE_KRB4 */ /* #undef LIBCURL_FEATURE_KRB4 */
/* Defined if libcurl supports libz */ /* Defined if libcurl supports libz */
#define LIBCURL_FEATURE_LIBZ 1 #define LIBCURL_FEATURE_LIBZ 1
/* Defined if libcurl supports NTLM */ /* Defined if libcurl supports NTLM */
#define LIBCURL_FEATURE_NTLM 1 #define LIBCURL_FEATURE_NTLM 1
/* Defined if libcurl supports SSL */ /* Defined if libcurl supports SSL */
#define LIBCURL_FEATURE_SSL 1 #define LIBCURL_FEATURE_SSL 1
/* Defined if libcurl supports SSPI */ /* Defined if libcurl supports SSPI */
/* #undef LIBCURL_FEATURE_SSPI */ /* #undef LIBCURL_FEATURE_SSPI */
/* Defined if libcurl supports DICT */ /* Defined if libcurl supports DICT */
/* #undef LIBCURL_PROTOCOL_DICT */ /* #undef LIBCURL_PROTOCOL_DICT */
/* Defined if libcurl supports FILE */ /* Defined if libcurl supports FILE */
#define LIBCURL_PROTOCOL_FILE 1 #define LIBCURL_PROTOCOL_FILE 1
/* Defined if libcurl supports FTP */ /* Defined if libcurl supports FTP */
#define LIBCURL_PROTOCOL_FTP 1 #define LIBCURL_PROTOCOL_FTP 1
/* Defined if libcurl supports FTPS */ /* Defined if libcurl supports FTPS */
#define LIBCURL_PROTOCOL_FTPS 1 #define LIBCURL_PROTOCOL_FTPS 1
/* Defined if libcurl supports HTTP */ /* Defined if libcurl supports HTTP */
#define LIBCURL_PROTOCOL_HTTP 1 #define LIBCURL_PROTOCOL_HTTP 1
/* Defined if libcurl supports HTTPS */ /* Defined if libcurl supports HTTPS */
#define LIBCURL_PROTOCOL_HTTPS 1 #define LIBCURL_PROTOCOL_HTTPS 1
/* Defined if libcurl supports IMAP */ /* Defined if libcurl supports IMAP */
/* #undef LIBCURL_PROTOCOL_IMAP */ /* #undef LIBCURL_PROTOCOL_IMAP */
/* Defined if libcurl supports LDAP */ /* Defined if libcurl supports LDAP */
/* #undef LIBCURL_PROTOCOL_LDAP */ /* #undef LIBCURL_PROTOCOL_LDAP */
/* Defined if libcurl supports POP3 */ /* Defined if libcurl supports POP3 */
/* #undef LIBCURL_PROTOCOL_POP3 */ /* #undef LIBCURL_PROTOCOL_POP3 */
/* Defined if libcurl supports RTSP */ /* Defined if libcurl supports RTSP */
/* #undef LIBCURL_PROTOCOL_RTSP */ /* #undef LIBCURL_PROTOCOL_RTSP */
/* Defined if libcurl supports SMTP */ /* Defined if libcurl supports SMTP */
/* #undef LIBCURL_PROTOCOL_SMTP */ /* #undef LIBCURL_PROTOCOL_SMTP */
/* Defined if libcurl supports TELNET */ /* Defined if libcurl supports TELNET */
/* #undef LIBCURL_PROTOCOL_TELNET */ /* #undef LIBCURL_PROTOCOL_TELNET */
/* Defined if libcurl supports TFTP */ /* Defined if libcurl supports TFTP */
/* #undef LIBCURL_PROTOCOL_TFTP */ /* #undef LIBCURL_PROTOCOL_TFTP */
/* Define to 1 if your C compiler doesn't accept -c and -o together. */ /* Define to 1 if your C compiler doesn't accept -c and -o together. */
/* #undef NO_MINUS_C_MINUS_O */ /* #undef NO_MINUS_C_MINUS_O */
/* Name of package */ /* Name of package */
#define PACKAGE "ccminer" #define PACKAGE "ccminer"
/* Define to the address where bug reports for this package should be sent. */ /* Define to the address where bug reports for this package should be sent. */
#define PACKAGE_BUGREPORT "" #define PACKAGE_BUGREPORT ""
/* Define to the full name of this package. */ /* Define to the full name of this package. */
#define PACKAGE_NAME "ccminer" #define PACKAGE_NAME "ccminer"
/* Define to the home page for this package. */ /* Define to the home page for this package. */
#define PACKAGE_URL "http://github.com/tpruvot/ccminer" #define PACKAGE_URL "http://github.com/tpruvot/ccminer"
/* Define to the version of this package. */ /* Define to the version of this package. */
#define PACKAGE_VERSION "2.2" #define PACKAGE_VERSION "2.2.1"
/* If using the C implementation of alloca, define if you know the /* If using the C implementation of alloca, define if you know the
direction of stack growth for your system; otherwise it will be direction of stack growth for your system; otherwise it will be
automatically deduced at runtime. automatically deduced at runtime.
STACK_DIRECTION > 0 => grows toward higher addresses STACK_DIRECTION > 0 => grows toward higher addresses
STACK_DIRECTION < 0 => grows toward lower addresses STACK_DIRECTION < 0 => grows toward lower addresses
STACK_DIRECTION = 0 => direction of growth unknown */ STACK_DIRECTION = 0 => direction of growth unknown */
/* #undef STACK_DIRECTION */ /* #undef STACK_DIRECTION */
/* Define to 1 if you have the ANSI C header files. */ /* Define to 1 if you have the ANSI C header files. */
#define STDC_HEADERS 1 #define STDC_HEADERS 1
/* Define curl_free() as free() if our version of curl lacks curl_free. */ /* Define curl_free() as free() if our version of curl lacks curl_free. */
/* #undef curl_free */ /* #undef curl_free */
/* Define to `unsigned int' if <sys/types.h> does not define. */ /* Define to `unsigned int' if <sys/types.h> does not define. */
//#define size_t unsigned int //#define size_t unsigned int
#if !defined(HAVE_STRUCT_TIMESPEC) && _MSC_VER >= 1900 #if !defined(HAVE_STRUCT_TIMESPEC) && _MSC_VER >= 1900
#define HAVE_STRUCT_TIMESPEC #define HAVE_STRUCT_TIMESPEC
#endif #endif

1
tribus/tribus.cu

@ -146,6 +146,7 @@ extern "C" int scanhash_tribus(int thr_id, struct work *work, uint32_t max_nonce
gpu_increment_reject(thr_id); gpu_increment_reject(thr_id);
if (!opt_quiet) if (!opt_quiet)
gpulog(LOG_WARNING, thr_id, "result for %08x does not validate on CPU!", work->nonces[0]); gpulog(LOG_WARNING, thr_id, "result for %08x does not validate on CPU!", work->nonces[0]);
cudaMemset(d_resNonce[thr_id], 0xFF, 2 * sizeof(uint32_t));
pdata[19] = work->nonces[0] + 1; pdata[19] = work->nonces[0] + 1;
continue; continue;
} }

46
x11/c11.cu

@ -18,10 +18,13 @@ extern "C"
#include "cuda_helper.h" #include "cuda_helper.h"
#include "cuda_x11.h" #include "cuda_x11.h"
void tribus_echo512_final(int thr_id, uint32_t threads, uint32_t *d_hash, uint32_t *d_resNonce, const uint64_t target);
#include <stdio.h> #include <stdio.h>
#include <memory.h> #include <memory.h>
static uint32_t *d_hash[MAX_GPUS]; static uint32_t *d_hash[MAX_GPUS];
static uint32_t *d_resNonce[MAX_GPUS];
// Flax/Chaincoin C11 CPU Hash // Flax/Chaincoin C11 CPU Hash
extern "C" void c11hash(void *output, const void *input) extern "C" void c11hash(void *output, const void *input)
@ -103,6 +106,7 @@ extern "C" void c11hash(void *output, const void *input)
#endif #endif
static bool init[MAX_GPUS] = { 0 }; static bool init[MAX_GPUS] = { 0 };
static bool use_compat_kernels[MAX_GPUS] = { 0 };
extern "C" int scanhash_c11(int thr_id, struct work* work, uint32_t max_nonce, unsigned long *hashes_done) extern "C" int scanhash_c11(int thr_id, struct work* work, uint32_t max_nonce, unsigned long *hashes_done)
{ {
@ -118,7 +122,8 @@ extern "C" int scanhash_c11(int thr_id, struct work* work, uint32_t max_nonce, u
if (!init[thr_id]) if (!init[thr_id])
{ {
cudaSetDevice(device_map[thr_id]); int dev_id = device_map[thr_id];
cudaSetDevice(dev_id);
if (opt_cudaschedule == -1 && gpu_threads == 1) { if (opt_cudaschedule == -1 && gpu_threads == 1) {
cudaDeviceReset(); cudaDeviceReset();
// reduce cpu usage // reduce cpu usage
@ -127,6 +132,9 @@ extern "C" int scanhash_c11(int thr_id, struct work* work, uint32_t max_nonce, u
} }
gpulog(LOG_INFO, thr_id, "Intensity set to %g, %u cuda threads", throughput2intensity(throughput), throughput); gpulog(LOG_INFO, thr_id, "Intensity set to %g, %u cuda threads", throughput2intensity(throughput), throughput);
cuda_get_arch(thr_id);
use_compat_kernels[thr_id] = (cuda_arch[dev_id] < 500);
quark_blake512_cpu_init(thr_id, throughput); quark_blake512_cpu_init(thr_id, throughput);
quark_bmw512_cpu_init(thr_id, throughput); quark_bmw512_cpu_init(thr_id, throughput);
quark_groestl512_cpu_init(thr_id, throughput); quark_groestl512_cpu_init(thr_id, throughput);
@ -135,11 +143,13 @@ extern "C" int scanhash_c11(int thr_id, struct work* work, uint32_t max_nonce, u
quark_jh512_cpu_init(thr_id, throughput); quark_jh512_cpu_init(thr_id, throughput);
x11_luffaCubehash512_cpu_init(thr_id, throughput); x11_luffaCubehash512_cpu_init(thr_id, throughput);
x11_shavite512_cpu_init(thr_id, throughput); x11_shavite512_cpu_init(thr_id, throughput);
x11_echo512_cpu_init(thr_id, throughput); if (use_compat_kernels[thr_id])
x11_echo512_cpu_init(thr_id, throughput);
if (x11_simd512_cpu_init(thr_id, throughput) != 0) { if (x11_simd512_cpu_init(thr_id, throughput) != 0) {
return 0; return 0;
} }
CUDA_CALL_OR_RET_X(cudaMalloc(&d_hash[thr_id], 64 * throughput), 0); // why 64 ? CUDA_CALL_OR_RET_X(cudaMalloc(&d_hash[thr_id], 64 * throughput), 0);
CUDA_SAFE_CALL(cudaMalloc(&d_resNonce[thr_id], 2 * sizeof(uint32_t)));
cuda_check_cpu_init(thr_id, throughput); cuda_check_cpu_init(thr_id, throughput);
@ -151,7 +161,10 @@ extern "C" int scanhash_c11(int thr_id, struct work* work, uint32_t max_nonce, u
be32enc(&endiandata[k], pdata[k]); be32enc(&endiandata[k], pdata[k]);
quark_blake512_cpu_setBlock_80(thr_id, endiandata); quark_blake512_cpu_setBlock_80(thr_id, endiandata);
cuda_check_cpu_setTarget(ptarget); if (use_compat_kernels[thr_id])
cuda_check_cpu_setTarget(ptarget);
else
cudaMemset(d_resNonce[thr_id], 0xFF, 2 * sizeof(uint32_t));
do { do {
int order = 0; int order = 0;
@ -175,24 +188,32 @@ extern "C" int scanhash_c11(int thr_id, struct work* work, uint32_t max_nonce, u
TRACE("shavite:"); TRACE("shavite:");
x11_simd512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++); x11_simd512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
TRACE("simd :"); TRACE("simd :");
x11_echo512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
TRACE("echo => "); if (use_compat_kernels[thr_id]) {
x11_echo512_cpu_hash_64(thr_id, throughput, pdata[19], NULL, d_hash[thr_id], order++);
work->nonces[0] = cuda_check_hash(thr_id, throughput, pdata[19], d_hash[thr_id]);
work->nonces[1] = UINT32_MAX;
} else {
tribus_echo512_final(thr_id, throughput, d_hash[thr_id], d_resNonce[thr_id], AS_U64(&ptarget[6]));
cudaMemcpy(&work->nonces[0], d_resNonce[thr_id], 2 * sizeof(uint32_t), cudaMemcpyDeviceToHost);
}
*hashes_done = pdata[19] - first_nonce + throughput; *hashes_done = pdata[19] - first_nonce + throughput;
work->nonces[0] = cuda_check_hash(thr_id, throughput, pdata[19], d_hash[thr_id]);
if (work->nonces[0] != UINT32_MAX) if (work->nonces[0] != UINT32_MAX)
{ {
const uint32_t Htarg = ptarget[7];
uint32_t _ALIGN(64) vhash[8]; uint32_t _ALIGN(64) vhash[8];
const uint32_t Htarg = ptarget[7];
const uint32_t startNounce = pdata[19];
if (!use_compat_kernels[thr_id]) work->nonces[0] += startNounce;
be32enc(&endiandata[19], work->nonces[0]); be32enc(&endiandata[19], work->nonces[0]);
c11hash(vhash, endiandata); c11hash(vhash, endiandata);
if (vhash[7] <= Htarg && fulltest(vhash, ptarget)) { if (vhash[7] <= Htarg && fulltest(vhash, ptarget)) {
work->valid_nonces = 1; work->valid_nonces = 1;
work_set_target_ratio(work, vhash); work_set_target_ratio(work, vhash);
work->nonces[1] = cuda_check_hash_suppl(thr_id, throughput, pdata[19], d_hash[thr_id], 1); if (work->nonces[1] != UINT32_MAX) {
if (work->nonces[1] != 0) { work->nonces[1] += startNounce;
be32enc(&endiandata[19], work->nonces[1]); be32enc(&endiandata[19], work->nonces[1]);
c11hash(vhash, endiandata); c11hash(vhash, endiandata);
bn_set_target_ratio(work, vhash, 1); bn_set_target_ratio(work, vhash, 1);
@ -206,7 +227,8 @@ extern "C" int scanhash_c11(int thr_id, struct work* work, uint32_t max_nonce, u
else if (vhash[7] > Htarg) { else if (vhash[7] > Htarg) {
gpu_increment_reject(thr_id); gpu_increment_reject(thr_id);
if (!opt_quiet) if (!opt_quiet)
gpulog(LOG_WARNING, thr_id, "result for %08x does not validate on CPU!", work->nonces[0]); gpulog(LOG_WARNING, thr_id, "result for %08x does not validate on CPU!", work->nonces[0]);
cudaMemset(d_resNonce[thr_id], 0xFF, 2 * sizeof(uint32_t));
pdata[19] = work->nonces[0] + 1; pdata[19] = work->nonces[0] + 1;
continue; continue;
} }
@ -234,6 +256,8 @@ extern "C" void free_c11(int thr_id)
cudaThreadSynchronize(); cudaThreadSynchronize();
cudaFree(d_hash[thr_id]); cudaFree(d_hash[thr_id]);
cudaFree(d_resNonce[thr_id]);
quark_blake512_cpu_free(thr_id); quark_blake512_cpu_free(thr_id);
quark_groestl512_cpu_free(thr_id); quark_groestl512_cpu_free(thr_id);
x11_simd512_cpu_free(thr_id); x11_simd512_cpu_free(thr_id);

Loading…
Cancel
Save