xenia-project/FFmpeg
1
0
Fork 0
mirror of https://github.com/xenia-project/FFmpeg.git synced 2024-11-24 03:59:43 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

avfilter/pthread: use slice threading from avutil

Browse Source 
Benchmark (with 2 cpus):
./ffmpeg -f rawvideo -s 1280x720 -t 1000 -i /dev/zero \
    -filter_threads $threads -vf transpose=clock -f null null
threads=2:
  old: 31.129s 31.446s 31.574s
  new: 29.602s 29.636s 29.656s
threads=3 (nb_threads = nb_cpus + 1 is bad choice at this situation):
  old: 40.132s 40.279s 40.279s
  new: 39.308s 39.570s 39.693s
threads=4:
  old: 31.306s 31.366s 31.654s
  new: 30.231s 30.360s 30.451s

Signed-off-by: Muhammad Faiz <mfcc64@gmail.com>
This commit is contained in:
Muhammad Faiz 2017-07-12 07:16:33 +07:00
parent b505f15b15
commit 22b72de04b
1 changed files with 12 additions and 116 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

128
libavfilter/pthread.c
View File

@ -27,6 +27,7 @@
#include "libavutil/cpu.h"
#include "libavutil/mem.h"
#include "libavutil/thread.h"
#include "libavutil/slicethread.h"
#include "avfilter.h"
#include "internal.h"
@ -34,85 +35,26 @@
typedef struct ThreadContext {
AVFilterGraph *graph;
int nb_threads;
pthread_t *workers;
AVSliceThread *thread;
avfilter_action_func *func;
/* per-execute parameters */
AVFilterContext *ctx;
void *arg;
int *rets;
int nb_jobs;
pthread_cond_t last_job_cond;
pthread_cond_t current_job_cond;
pthread_mutex_t current_job_lock;
int current_job;
unsigned int current_execute;
int done;
} ThreadContext;
static void* attribute_align_arg worker(void *v)
static void worker_func(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads)
{
ThreadContext *c = v;
int our_job = c->nb_jobs;
int nb_threads = c->nb_threads;
unsigned int last_execute = 0;
int ret, self_id;
pthread_mutex_lock(&c->current_job_lock);
self_id = c->current_job++;
for (;;) {
while (our_job >= c->nb_jobs) {
if (c->current_job == nb_threads + c->nb_jobs)
pthread_cond_signal(&c->last_job_cond);
while (last_execute == c->current_execute && !c->done)
pthread_cond_wait(&c->current_job_cond, &c->current_job_lock);
last_execute = c->current_execute;
our_job = self_id;
if (c->done) {
pthread_mutex_unlock(&c->current_job_lock);
return NULL;
}
}
pthread_mutex_unlock(&c->current_job_lock);
ret = c->func(c->ctx, c->arg, our_job, c->nb_jobs);
if (c->rets)
c->rets[our_job % c->nb_jobs] = ret;
pthread_mutex_lock(&c->current_job_lock);
our_job = c->current_job++;
}
ThreadContext *c = priv;
int ret = c->func(c->ctx, c->arg, jobnr, nb_jobs);
if (c->rets)
c->rets[jobnr] = ret;
}
static void slice_thread_uninit(ThreadContext *c)
{
int i;
pthread_mutex_lock(&c->current_job_lock);
c->done = 1;
pthread_cond_broadcast(&c->current_job_cond);
pthread_mutex_unlock(&c->current_job_lock);
for (i = 0; i < c->nb_threads; i++)
pthread_join(c->workers[i], NULL);
pthread_mutex_destroy(&c->current_job_lock);
pthread_cond_destroy(&c->current_job_cond);
pthread_cond_destroy(&c->last_job_cond);
av_freep(&c->workers);
}
static void slice_thread_park_workers(ThreadContext *c)
{
while (c->current_job != c->nb_threads + c->nb_jobs)
pthread_cond_wait(&c->last_job_cond, &c->current_job_lock);
pthread_mutex_unlock(&c->current_job_lock);
avpriv_slicethread_free(&c->thread);
}
static int thread_execute(AVFilterContext *ctx, avfilter_action_func *func,
@ -122,67 +64,21 @@ static int thread_execute(AVFilterContext *ctx, avfilter_action_func *func,
if (nb_jobs <= 0)
return 0;
pthread_mutex_lock(&c->current_job_lock);
c->current_job = c->nb_threads;
c->nb_jobs = nb_jobs;
c->ctx = ctx;
c->arg = arg;
c->func = func;
c->rets = ret;
c->current_execute++;
pthread_cond_broadcast(&c->current_job_cond);
slice_thread_park_workers(c);
avpriv_slicethread_execute(c->thread, nb_jobs, 0);
return 0;
}
static int thread_init_internal(ThreadContext *c, int nb_threads)
{
int i, ret;
if (!nb_threads) {
int nb_cpus = av_cpu_count();
// use number of cores + 1 as thread count if there is more than one
if (nb_cpus > 1)
nb_threads = nb_cpus + 1;
else
nb_threads = 1;
}
nb_threads = avpriv_slicethread_create(&c->thread, c, worker_func, NULL, nb_threads);
if (nb_threads <= 1)
return 1;
c->nb_threads = nb_threads;
c->workers = av_mallocz_array(sizeof(*c->workers), nb_threads);
if (!c->workers)
return AVERROR(ENOMEM);
c->current_job = 0;
c->nb_jobs = 0;
c->done = 0;
pthread_cond_init(&c->current_job_cond, NULL);
pthread_cond_init(&c->last_job_cond, NULL);
pthread_mutex_init(&c->current_job_lock, NULL);
pthread_mutex_lock(&c->current_job_lock);
for (i = 0; i < nb_threads; i++) {
ret = pthread_create(&c->workers[i], NULL, worker, c);
if (ret) {
pthread_mutex_unlock(&c->current_job_lock);
c->nb_threads = i;
slice_thread_uninit(c);
return AVERROR(ret);
}
}
slice_thread_park_workers(c);
return c->nb_threads;
avpriv_slicethread_free(&c->thread);
return FFMAX(nb_threads, 1);
}
int ff_graph_thread_init(AVFilterGraph *graph)