Merge pull request #23 from ckolivas/master
Here's my x86_64 and linux optimisations. Hopefully shouldn't break other OSs now.
This commit is contained in:
commit
48a888dbfb
4 changed files with 83 additions and 30 deletions
6
NEWS
6
NEWS
|
@ -1,3 +1,9 @@
|
||||||
|
- Linux x86_64 optimisations - Con Kolivas
|
||||||
|
- Optimise for x86_64 by default by using sse2_64 algo
|
||||||
|
- Detects CPUs and sets number of threads accordingly
|
||||||
|
- Uses CPU affinity for each thread where appropriate
|
||||||
|
- Sets scheduling policy to lowest possible
|
||||||
|
- Minor performance tweaks
|
||||||
|
|
||||||
Version 1.0.1 - May 14, 2011
|
Version 1.0.1 - May 14, 2011
|
||||||
|
|
||||||
|
|
89
cpu-miner.c
89
cpu-miner.c
|
@ -9,6 +9,7 @@
|
||||||
*/
|
*/
|
||||||
|
|
||||||
#include "cpuminer-config.h"
|
#include "cpuminer-config.h"
|
||||||
|
#define _GNU_SOURCE
|
||||||
|
|
||||||
#include <stdio.h>
|
#include <stdio.h>
|
||||||
#include <stdlib.h>
|
#include <stdlib.h>
|
||||||
|
@ -33,6 +34,35 @@
|
||||||
#define DEF_RPC_PASSWORD "rpcpass"
|
#define DEF_RPC_PASSWORD "rpcpass"
|
||||||
#define DEF_RPC_USERPASS DEF_RPC_USERNAME ":" DEF_RPC_PASSWORD
|
#define DEF_RPC_USERPASS DEF_RPC_USERNAME ":" DEF_RPC_PASSWORD
|
||||||
|
|
||||||
|
#ifdef __linux /* Linux specific policy and affinity management */
|
||||||
|
#include <sched.h>
|
||||||
|
static inline void drop_policy(void)
|
||||||
|
{
|
||||||
|
struct sched_param param;
|
||||||
|
|
||||||
|
if (unlikely(sched_setscheduler(0, SCHED_IDLE, ¶m) == -1))
|
||||||
|
sched_setscheduler(0, SCHED_BATCH, ¶m);
|
||||||
|
}
|
||||||
|
|
||||||
|
static inline void affine_to_cpu(int id, int cpu)
|
||||||
|
{
|
||||||
|
cpu_set_t set;
|
||||||
|
|
||||||
|
CPU_ZERO(&set);
|
||||||
|
CPU_SET(cpu, &set);
|
||||||
|
sched_setaffinity(0, sizeof(&set), &set);
|
||||||
|
applog(LOG_INFO, "Binding thread %d to cpu %d", id, cpu);
|
||||||
|
}
|
||||||
|
#else
|
||||||
|
static inline void drop_policy(void)
|
||||||
|
{
|
||||||
|
}
|
||||||
|
|
||||||
|
static inline void affine_to_cpu(int id, int cpu)
|
||||||
|
{
|
||||||
|
}
|
||||||
|
#endif
|
||||||
|
|
||||||
enum workio_commands {
|
enum workio_commands {
|
||||||
WC_GET_WORK,
|
WC_GET_WORK,
|
||||||
WC_SUBMIT_WORK,
|
WC_SUBMIT_WORK,
|
||||||
|
@ -83,8 +113,13 @@ static int opt_fail_pause = 30;
|
||||||
int opt_scantime = 5;
|
int opt_scantime = 5;
|
||||||
static json_t *opt_config;
|
static json_t *opt_config;
|
||||||
static const bool opt_time = true;
|
static const bool opt_time = true;
|
||||||
|
#ifdef WANT_X8664_SSE2
|
||||||
|
static enum sha256_algos opt_algo = ALGO_SSE2_64;
|
||||||
|
#else
|
||||||
static enum sha256_algos opt_algo = ALGO_C;
|
static enum sha256_algos opt_algo = ALGO_C;
|
||||||
static int opt_n_threads = 1;
|
#endif
|
||||||
|
static int opt_n_threads;
|
||||||
|
static int num_processors;
|
||||||
static char *rpc_url;
|
static char *rpc_url;
|
||||||
static char *rpc_userpass;
|
static char *rpc_userpass;
|
||||||
static char *rpc_user, *rpc_pass;
|
static char *rpc_user, *rpc_pass;
|
||||||
|
@ -214,12 +249,12 @@ static bool jobj_binary(const json_t *obj, const char *key,
|
||||||
json_t *tmp;
|
json_t *tmp;
|
||||||
|
|
||||||
tmp = json_object_get(obj, key);
|
tmp = json_object_get(obj, key);
|
||||||
if (!tmp) {
|
if (unlikely(!tmp)) {
|
||||||
applog(LOG_ERR, "JSON key '%s' not found", key);
|
applog(LOG_ERR, "JSON key '%s' not found", key);
|
||||||
return false;
|
return false;
|
||||||
}
|
}
|
||||||
hexstr = json_string_value(tmp);
|
hexstr = json_string_value(tmp);
|
||||||
if (!hexstr) {
|
if (unlikely(!hexstr)) {
|
||||||
applog(LOG_ERR, "JSON key '%s' is not a string", key);
|
applog(LOG_ERR, "JSON key '%s' is not a string", key);
|
||||||
return false;
|
return false;
|
||||||
}
|
}
|
||||||
|
@ -231,23 +266,23 @@ static bool jobj_binary(const json_t *obj, const char *key,
|
||||||
|
|
||||||
static bool work_decode(const json_t *val, struct work *work)
|
static bool work_decode(const json_t *val, struct work *work)
|
||||||
{
|
{
|
||||||
if (!jobj_binary(val, "midstate",
|
if (unlikely(!jobj_binary(val, "midstate",
|
||||||
work->midstate, sizeof(work->midstate))) {
|
work->midstate, sizeof(work->midstate)))) {
|
||||||
applog(LOG_ERR, "JSON inval midstate");
|
applog(LOG_ERR, "JSON inval midstate");
|
||||||
goto err_out;
|
goto err_out;
|
||||||
}
|
}
|
||||||
|
|
||||||
if (!jobj_binary(val, "data", work->data, sizeof(work->data))) {
|
if (unlikely(!jobj_binary(val, "data", work->data, sizeof(work->data)))) {
|
||||||
applog(LOG_ERR, "JSON inval data");
|
applog(LOG_ERR, "JSON inval data");
|
||||||
goto err_out;
|
goto err_out;
|
||||||
}
|
}
|
||||||
|
|
||||||
if (!jobj_binary(val, "hash1", work->hash1, sizeof(work->hash1))) {
|
if (unlikely(!jobj_binary(val, "hash1", work->hash1, sizeof(work->hash1)))) {
|
||||||
applog(LOG_ERR, "JSON inval hash1");
|
applog(LOG_ERR, "JSON inval hash1");
|
||||||
goto err_out;
|
goto err_out;
|
||||||
}
|
}
|
||||||
|
|
||||||
if (!jobj_binary(val, "target", work->target, sizeof(work->target))) {
|
if (unlikely(!jobj_binary(val, "target", work->target, sizeof(work->target)))) {
|
||||||
applog(LOG_ERR, "JSON inval target");
|
applog(LOG_ERR, "JSON inval target");
|
||||||
goto err_out;
|
goto err_out;
|
||||||
}
|
}
|
||||||
|
@ -269,7 +304,7 @@ static bool submit_upstream_work(CURL *curl, const struct work *work)
|
||||||
|
|
||||||
/* build hex string */
|
/* build hex string */
|
||||||
hexstr = bin2hex(work->data, sizeof(work->data));
|
hexstr = bin2hex(work->data, sizeof(work->data));
|
||||||
if (!hexstr) {
|
if (unlikely(!hexstr)) {
|
||||||
applog(LOG_ERR, "submit_upstream_work OOM");
|
applog(LOG_ERR, "submit_upstream_work OOM");
|
||||||
goto out;
|
goto out;
|
||||||
}
|
}
|
||||||
|
@ -284,7 +319,7 @@ static bool submit_upstream_work(CURL *curl, const struct work *work)
|
||||||
|
|
||||||
/* issue JSON-RPC request */
|
/* issue JSON-RPC request */
|
||||||
val = json_rpc_call(curl, rpc_url, rpc_userpass, s, false, false);
|
val = json_rpc_call(curl, rpc_url, rpc_userpass, s, false, false);
|
||||||
if (!val) {
|
if (unlikely(!val)) {
|
||||||
applog(LOG_ERR, "submit_upstream_work json_rpc_call failed");
|
applog(LOG_ERR, "submit_upstream_work json_rpc_call failed");
|
||||||
goto out;
|
goto out;
|
||||||
}
|
}
|
||||||
|
@ -351,7 +386,7 @@ static bool workio_get_work(struct workio_cmd *wc, CURL *curl)
|
||||||
|
|
||||||
/* obtain new work from bitcoin via JSON-RPC */
|
/* obtain new work from bitcoin via JSON-RPC */
|
||||||
while (!get_upstream_work(curl, ret_work)) {
|
while (!get_upstream_work(curl, ret_work)) {
|
||||||
if ((opt_retries >= 0) && (++failures > opt_retries)) {
|
if (unlikely((opt_retries >= 0) && (++failures > opt_retries))) {
|
||||||
applog(LOG_ERR, "json_rpc_call failed, terminating workio thread");
|
applog(LOG_ERR, "json_rpc_call failed, terminating workio thread");
|
||||||
free(ret_work);
|
free(ret_work);
|
||||||
return false;
|
return false;
|
||||||
|
@ -376,7 +411,7 @@ static bool workio_submit_work(struct workio_cmd *wc, CURL *curl)
|
||||||
|
|
||||||
/* submit solution to bitcoin via JSON-RPC */
|
/* submit solution to bitcoin via JSON-RPC */
|
||||||
while (!submit_upstream_work(curl, wc->u.work)) {
|
while (!submit_upstream_work(curl, wc->u.work)) {
|
||||||
if ((opt_retries >= 0) && (++failures > opt_retries)) {
|
if (unlikely((opt_retries >= 0) && (++failures > opt_retries))) {
|
||||||
applog(LOG_ERR, "...terminating workio thread");
|
applog(LOG_ERR, "...terminating workio thread");
|
||||||
return false;
|
return false;
|
||||||
}
|
}
|
||||||
|
@ -397,7 +432,7 @@ static void *workio_thread(void *userdata)
|
||||||
bool ok = true;
|
bool ok = true;
|
||||||
|
|
||||||
curl = curl_easy_init();
|
curl = curl_easy_init();
|
||||||
if (!curl) {
|
if (unlikely(!curl)) {
|
||||||
applog(LOG_ERR, "CURL initialization failed");
|
applog(LOG_ERR, "CURL initialization failed");
|
||||||
return NULL;
|
return NULL;
|
||||||
}
|
}
|
||||||
|
@ -514,6 +549,17 @@ static void *miner_thread(void *userdata)
|
||||||
int thr_id = mythr->id;
|
int thr_id = mythr->id;
|
||||||
uint32_t max_nonce = 0xffffff;
|
uint32_t max_nonce = 0xffffff;
|
||||||
|
|
||||||
|
/* Set worker threads to nice 19 and then preferentially to SCHED_IDLE
|
||||||
|
* and if that fails, then SCHED_BATCH. No need for this to be an
|
||||||
|
* error if it fails */
|
||||||
|
setpriority(PRIO_PROCESS, 0, 19);
|
||||||
|
drop_policy();
|
||||||
|
|
||||||
|
/* Cpu affinity only makes sense if the number of threads is a multiple
|
||||||
|
* of the number of CPUs */
|
||||||
|
if (!(opt_n_threads % num_processors))
|
||||||
|
affine_to_cpu(mythr->id, mythr->id % num_processors);
|
||||||
|
|
||||||
while (1) {
|
while (1) {
|
||||||
struct work work __attribute__((aligned(128)));
|
struct work work __attribute__((aligned(128)));
|
||||||
unsigned long hashes_done;
|
unsigned long hashes_done;
|
||||||
|
@ -522,7 +568,7 @@ static void *miner_thread(void *userdata)
|
||||||
bool rc;
|
bool rc;
|
||||||
|
|
||||||
/* obtain new work from internal workio thread */
|
/* obtain new work from internal workio thread */
|
||||||
if (!get_work(mythr, &work)) {
|
if (unlikely(!get_work(mythr, &work))) {
|
||||||
applog(LOG_ERR, "work retrieval failed, exiting "
|
applog(LOG_ERR, "work retrieval failed, exiting "
|
||||||
"mining thread %d", mythr->id);
|
"mining thread %d", mythr->id);
|
||||||
goto out;
|
goto out;
|
||||||
|
@ -658,7 +704,7 @@ static void *longpoll_thread(void *userdata)
|
||||||
applog(LOG_INFO, "Long-polling activated for %s", lp_url);
|
applog(LOG_INFO, "Long-polling activated for %s", lp_url);
|
||||||
|
|
||||||
curl = curl_easy_init();
|
curl = curl_easy_init();
|
||||||
if (!curl) {
|
if (unlikely(!curl)) {
|
||||||
applog(LOG_ERR, "CURL initialization failed");
|
applog(LOG_ERR, "CURL initialization failed");
|
||||||
goto out;
|
goto out;
|
||||||
}
|
}
|
||||||
|
@ -668,7 +714,7 @@ static void *longpoll_thread(void *userdata)
|
||||||
|
|
||||||
val = json_rpc_call(curl, lp_url, rpc_userpass, rpc_req,
|
val = json_rpc_call(curl, lp_url, rpc_userpass, rpc_req,
|
||||||
false, true);
|
false, true);
|
||||||
if (val) {
|
if (likely(val)) {
|
||||||
failures = 0;
|
failures = 0;
|
||||||
json_decref(val);
|
json_decref(val);
|
||||||
|
|
||||||
|
@ -809,6 +855,9 @@ static void parse_arg (int key, char *arg)
|
||||||
default:
|
default:
|
||||||
show_usage();
|
show_usage();
|
||||||
}
|
}
|
||||||
|
num_processors = sysconf(_SC_NPROCESSORS_ONLN);
|
||||||
|
if (!opt_n_threads)
|
||||||
|
opt_n_threads = num_processors;
|
||||||
}
|
}
|
||||||
|
|
||||||
static void parse_config(void)
|
static void parse_config(void)
|
||||||
|
@ -886,10 +935,6 @@ int main (int argc, char *argv[])
|
||||||
openlog("cpuminer", LOG_PID, LOG_USER);
|
openlog("cpuminer", LOG_PID, LOG_USER);
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
/* set our priority to the highest (aka "nicest, least intrusive") */
|
|
||||||
if (setpriority(PRIO_PROCESS, 0, 19))
|
|
||||||
perror("setpriority");
|
|
||||||
|
|
||||||
work_restart = calloc(opt_n_threads, sizeof(*work_restart));
|
work_restart = calloc(opt_n_threads, sizeof(*work_restart));
|
||||||
if (!work_restart)
|
if (!work_restart)
|
||||||
return 1;
|
return 1;
|
||||||
|
@ -922,7 +967,7 @@ int main (int argc, char *argv[])
|
||||||
return 1;
|
return 1;
|
||||||
|
|
||||||
/* start longpoll thread */
|
/* start longpoll thread */
|
||||||
if (pthread_create(&thr->pth, NULL, longpoll_thread, thr)) {
|
if (unlikely(pthread_create(&thr->pth, NULL, longpoll_thread, thr))) {
|
||||||
applog(LOG_ERR, "longpoll thread create failed");
|
applog(LOG_ERR, "longpoll thread create failed");
|
||||||
return 1;
|
return 1;
|
||||||
}
|
}
|
||||||
|
@ -938,7 +983,7 @@ int main (int argc, char *argv[])
|
||||||
if (!thr->q)
|
if (!thr->q)
|
||||||
return 1;
|
return 1;
|
||||||
|
|
||||||
if (pthread_create(&thr->pth, NULL, miner_thread, thr)) {
|
if (unlikely(pthread_create(&thr->pth, NULL, miner_thread, thr))) {
|
||||||
applog(LOG_ERR, "thread %d create failed", i);
|
applog(LOG_ERR, "thread %d create failed", i);
|
||||||
return 1;
|
return 1;
|
||||||
}
|
}
|
||||||
|
|
8
miner.h
8
miner.h
|
@ -87,12 +87,14 @@ enum {
|
||||||
};
|
};
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
|
#undef unlikely
|
||||||
|
#undef likely
|
||||||
#if defined(__GNUC__) && (__GNUC__ > 2) && defined(__OPTIMIZE__)
|
#if defined(__GNUC__) && (__GNUC__ > 2) && defined(__OPTIMIZE__)
|
||||||
#undef unlikely
|
#define unlikely(expr) (__builtin_expect(!!(expr), 0))
|
||||||
#define unlikely(expr) (__builtin_expect((expr), 0))
|
#define likely(expr) (__builtin_expect(!!(expr), 1))
|
||||||
#else
|
#else
|
||||||
#undef unlikely
|
|
||||||
#define unlikely(expr) (expr)
|
#define unlikely(expr) (expr)
|
||||||
|
#define likely(expr) (expr)
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
#if defined(__i386__)
|
#if defined(__i386__)
|
||||||
|
|
|
@ -100,13 +100,13 @@ int scanhash_sse2_64(int thr_id, const unsigned char *pmidstate,
|
||||||
|
|
||||||
for (j = 0; j < 4; j++) {
|
for (j = 0; j < 4; j++) {
|
||||||
mi.m = m_4hash[7];
|
mi.m = m_4hash[7];
|
||||||
if (mi.i[j] == 0)
|
if (unlikely(mi.i[j] == 0))
|
||||||
break;
|
break;
|
||||||
}
|
}
|
||||||
|
|
||||||
/* If j = true, we found a hit...so check it */
|
/* If j = true, we found a hit...so check it */
|
||||||
/* Use the C version for a check... */
|
/* Use the C version for a check... */
|
||||||
if (j != 4) {
|
if (unlikely(j != 4)) {
|
||||||
for (i = 0; i < 8; i++) {
|
for (i = 0; i < 8; i++) {
|
||||||
mi.m = m_4hash[i];
|
mi.m = m_4hash[i];
|
||||||
*(uint32_t *)&(phash)[i*4] = mi.i[j];
|
*(uint32_t *)&(phash)[i*4] = mi.i[j];
|
||||||
|
@ -121,12 +121,12 @@ int scanhash_sse2_64(int thr_id, const unsigned char *pmidstate,
|
||||||
|
|
||||||
nonce += 4;
|
nonce += 4;
|
||||||
|
|
||||||
if ((nonce >= max_nonce) || work_restart[thr_id].restart)
|
if (unlikely((nonce >= max_nonce) || work_restart[thr_id].restart))
|
||||||
{
|
{
|
||||||
*nHashesDone = nonce;
|
*nHashesDone = nonce;
|
||||||
return -1;
|
return -1;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
#endif /* WANT_X8664_SSE2 */
|
#endif /* WANT_X8664_SSE2 */
|
||||||
|
|
Loading…
Reference in a new issue