#if HAVE_SCHED_GETAFFINITY
#define _GNU_SOURCE
#include <sched.h>
-#elif HAVE_GETSYSTEMINFO
+#endif
+#if HAVE_GETPROCESSAFFINITYMASK
#include <windows.h>
-#elif HAVE_SYSCTL
+#endif
+#if HAVE_SYSCTL
+#if HAVE_SYS_PARAM_H
+#include <sys/param.h>
+#endif
#include <sys/types.h>
#include <sys/sysctl.h>
#endif
+#if HAVE_SYSCONF
+#include <unistd.h>
+#endif
#include "avcodec.h"
#include "internal.h"
#include "thread.h"
+#include "libavutil/avassert.h"
+#include "libavutil/common.h"
#if HAVE_PTHREADS
#include <pthread.h>
int done;
} ThreadContext;
-/// Max number of frame buffers that can be allocated when using frame threads.
-#define MAX_BUFFERS (32+1)
-
/**
* Context used by codec threads and stored in their AVCodecContext thread_opaque.
*/
AVCodecContext *avctx; ///< Context used to decode packets passed to this thread.
AVPacket avpkt; ///< Input packet (for decoding) or output (for encoding).
- int allocated_buf_size; ///< Size allocated for avpkt.data
+ uint8_t *buf; ///< backup storage for packet data when the input packet is not refcounted
+ int allocated_buf_size; ///< Size allocated for buf
AVFrame frame; ///< Output frame (for decoding) or input (for encoding).
int got_frame; ///< The output of got_picture_ptr from the last avcodec_decode_video() call.
* Array of frames passed to ff_thread_release_buffer().
* Frames are released after all threads referencing them are finished.
*/
- AVFrame released_buffers[MAX_BUFFERS];
- int num_released_buffers;
-
- /**
- * Array of progress values used by ff_thread_get_buffer().
- */
- int progress[MAX_BUFFERS][2];
- uint8_t progress_used[MAX_BUFFERS];
+ AVFrame *released_buffers;
+ int num_released_buffers;
+ int released_buffers_allocated;
AVFrame *requested_frame; ///< AVFrame the codec passed to get_buffer()
+ int requested_flags; ///< flags passed to get_buffer() for requested_frame
} PerThreadContext;
/**
if (!ret) {
nb_cpus = CPU_COUNT(&cpuset);
}
-#elif HAVE_GETSYSTEMINFO
- SYSTEM_INFO sysinfo;
- GetSystemInfo(&sysinfo);
- nb_cpus = sysinfo.dwNumberOfProcessors;
+#elif HAVE_GETPROCESSAFFINITYMASK
+ DWORD_PTR proc_aff, sys_aff;
+ ret = GetProcessAffinityMask(GetCurrentProcess(), &proc_aff, &sys_aff);
+ if (ret)
+ nb_cpus = av_popcount64(proc_aff);
#elif HAVE_SYSCTL && defined(HW_NCPU)
int mib[2] = { CTL_HW, HW_NCPU };
size_t len = sizeof(nb_cpus);
ret = sysctl(mib, 2, &nb_cpus, &len, NULL, 0);
if (ret == -1)
nb_cpus = 0;
+#elif HAVE_SYSCONF && defined(_SC_NPROC_ONLN)
+ nb_cpus = sysconf(_SC_NPROC_ONLN);
+#elif HAVE_SYSCONF && defined(_SC_NPROCESSORS_ONLN)
+ nb_cpus = sysconf(_SC_NPROCESSORS_ONLN);
#endif
av_log(avctx, AV_LOG_DEBUG, "detected %d logical cores\n", nb_cpus);
- return FFMIN(nb_cpus, MAX_AUTO_THREADS);
+ return nb_cpus;
}
if (!thread_count) {
int nb_cpus = get_logical_cpus(avctx);
- // use number of cores + 1 as thread count if there is motre than one
+ // use number of cores + 1 as thread count if there is more than one
if (nb_cpus > 1)
- thread_count = avctx->thread_count = nb_cpus + 1;
+ thread_count = avctx->thread_count = FFMIN(nb_cpus + 1, MAX_AUTO_THREADS);
+ else
+ thread_count = avctx->thread_count = 1;
}
if (thread_count <= 1) {
PerThreadContext *p = arg;
FrameThreadContext *fctx = p->parent;
AVCodecContext *avctx = p->avctx;
- AVCodec *codec = avctx->codec;
+ const AVCodec *codec = avctx->codec;
while (1) {
if (p->state == STATE_INPUT_READY && !fctx->die) {
p->got_frame = 0;
p->result = codec->decode(avctx, &p->frame, &p->got_frame, &p->avpkt);
+ /* many decoders assign whole AVFrames, thus overwriting extended_data;
+ * make sure it's set correctly */
+ p->frame.extended_data = p->frame.data;
+
if (p->state == STATE_SETTING_UP) ff_thread_finish_setup(avctx);
p->state = STATE_INPUT_READY;
int err = 0;
if (dst != src) {
- dst->sub_id = src->sub_id;
dst->time_base = src->time_base;
dst->width = src->width;
dst->height = src->height;
dst->has_b_frames = src->has_b_frames;
dst->idct_algo = src->idct_algo;
- dst->slice_count = src->slice_count;
dst->bits_per_coded_sample = src->bits_per_coded_sample;
dst->sample_aspect_ratio = src->sample_aspect_ratio;
dst->colorspace = src->colorspace;
dst->color_range = src->color_range;
dst->chroma_sample_location = src->chroma_sample_location;
+
+ dst->hwaccel = src->hwaccel;
+ dst->hwaccel_context = src->hwaccel_context;
}
if (for_user) {
*
* @param dst The destination context.
* @param src The source context.
+ * @return 0 on success, negative error code on failure
*/
-static void update_context_from_user(AVCodecContext *dst, AVCodecContext *src)
+static int update_context_from_user(AVCodecContext *dst, AVCodecContext *src)
{
#define copy_fields(s, e) memcpy(&dst->s, &src->s, (char*)&dst->e - (char*)&dst->s);
dst->flags = src->flags;
dst->draw_horiz_band= src->draw_horiz_band;
+ dst->get_buffer2 = src->get_buffer2;
+#if FF_API_GET_BUFFER
dst->get_buffer = src->get_buffer;
dst->release_buffer = src->release_buffer;
+#endif
dst->opaque = src->opaque;
- dst->dsp_mask = src->dsp_mask;
dst->debug = src->debug;
dst->debug_mv = src->debug_mv;
dst->slice_flags = src->slice_flags;
dst->flags2 = src->flags2;
- copy_fields(skip_loop_filter, bidir_refine);
+ copy_fields(skip_loop_filter, subtitle_header);
dst->frame_number = src->frame_number;
dst->reordered_opaque = src->reordered_opaque;
-#undef copy_fields
-}
-
-static void free_progress(AVFrame *f)
-{
- PerThreadContext *p = f->owner->thread_opaque;
- int *progress = f->thread_opaque;
- p->progress_used[(progress - p->progress[0]) / 2] = 0;
+ if (src->slice_count && src->slice_offset) {
+ if (dst->slice_count < src->slice_count) {
+ int *tmp = av_realloc(dst->slice_offset, src->slice_count *
+ sizeof(*dst->slice_offset));
+ if (!tmp) {
+ av_free(dst->slice_offset);
+ return AVERROR(ENOMEM);
+ }
+ dst->slice_offset = tmp;
+ }
+ memcpy(dst->slice_offset, src->slice_offset,
+ src->slice_count * sizeof(*dst->slice_offset));
+ }
+ dst->slice_count = src->slice_count;
+ return 0;
+#undef copy_fields
}
/// Releases the buffers that this decoding thread was the last user of.
AVFrame *f;
pthread_mutex_lock(&fctx->buffer_mutex);
+
+ // fix extended data in case the caller screwed it up
+ av_assert0(p->avctx->codec_type == AVMEDIA_TYPE_VIDEO);
f = &p->released_buffers[--p->num_released_buffers];
- free_progress(f);
- f->thread_opaque = NULL;
+ f->extended_data = f->data;
+ av_frame_unref(f);
- f->owner->release_buffer(f->owner, f);
pthread_mutex_unlock(&fctx->buffer_mutex);
}
}
{
FrameThreadContext *fctx = p->parent;
PerThreadContext *prev_thread = fctx->prev_thread;
- AVCodec *codec = p->avctx->codec;
- uint8_t *buf = p->avpkt.data;
+ const AVCodec *codec = p->avctx->codec;
if (!avpkt->size && !(codec->capabilities & CODEC_CAP_DELAY)) return 0;
}
}
- av_fast_malloc(&buf, &p->allocated_buf_size, avpkt->size + FF_INPUT_BUFFER_PADDING_SIZE);
+ av_buffer_unref(&p->avpkt.buf);
p->avpkt = *avpkt;
- p->avpkt.data = buf;
- memcpy(buf, avpkt->data, avpkt->size);
- memset(buf + avpkt->size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
+ if (avpkt->buf)
+ p->avpkt.buf = av_buffer_ref(avpkt->buf);
+ else {
+ av_fast_malloc(&p->buf, &p->allocated_buf_size, avpkt->size + FF_INPUT_BUFFER_PADDING_SIZE);
+ p->avpkt.data = p->buf;
+ memcpy(p->buf, avpkt->data, avpkt->size);
+ memset(p->buf + avpkt->size, 0, FF_INPUT_BUFFER_PADDING_SIZE);
+ }
p->state = STATE_SETTING_UP;
pthread_cond_signal(&p->input_cond);
* and it calls back to the client here.
*/
- if (!p->avctx->thread_safe_callbacks &&
- p->avctx->get_buffer != avcodec_default_get_buffer) {
+ if (!p->avctx->thread_safe_callbacks && (
+#if FF_API_GET_BUFFER
+ p->avctx->get_buffer ||
+#endif
+ p->avctx->get_buffer2 != avcodec_default_get_buffer2)) {
while (p->state != STATE_SETUP_FINISHED && p->state != STATE_INPUT_READY) {
pthread_mutex_lock(&p->progress_mutex);
while (p->state == STATE_SETTING_UP)
pthread_cond_wait(&p->progress_cond, &p->progress_mutex);
if (p->state == STATE_GET_BUFFER) {
- p->result = p->avctx->get_buffer(p->avctx, p->requested_frame);
+ p->result = ff_get_buffer(p->avctx, p->requested_frame, p->requested_flags);
p->state = STATE_SETTING_UP;
pthread_cond_signal(&p->progress_cond);
}
*/
p = &fctx->threads[fctx->next_decoding];
- update_context_from_user(p->avctx, avctx);
+ err = update_context_from_user(p->avctx, avctx);
+ if (err) return err;
err = submit_packet(p, avpkt);
if (err) return err;
* If we're still receiving the initial packets, don't return a frame.
*/
- if (fctx->delaying && avpkt->size) {
+ if (fctx->delaying) {
if (fctx->next_decoding >= (avctx->thread_count-1)) fctx->delaying = 0;
*got_picture_ptr=0;
- return avpkt->size;
+ if (avpkt->size)
+ return avpkt->size;
}
/*
pthread_mutex_unlock(&p->progress_mutex);
}
- *picture = p->frame;
+ av_frame_move_ref(picture, &p->frame);
*got_picture_ptr = p->got_frame;
picture->pkt_dts = p->avpkt.dts;
return (p->result >= 0) ? avpkt->size : p->result;
}
-void ff_thread_report_progress(AVFrame *f, int n, int field)
+void ff_thread_report_progress(ThreadFrame *f, int n, int field)
{
PerThreadContext *p;
- int *progress = f->thread_opaque;
+ int *progress = f->progress ? (int*)f->progress->data : NULL;
if (!progress || progress[field] >= n) return;
pthread_mutex_unlock(&p->progress_mutex);
}
-void ff_thread_await_progress(AVFrame *f, int n, int field)
+void ff_thread_await_progress(ThreadFrame *f, int n, int field)
{
PerThreadContext *p;
- int *progress = f->thread_opaque;
+ int *progress = f->progress ? (int*)f->progress->data : NULL;
if (!progress || progress[field] >= n) return;
static void frame_thread_free(AVCodecContext *avctx, int thread_count)
{
FrameThreadContext *fctx = avctx->thread_opaque;
- AVCodec *codec = avctx->codec;
+ const AVCodec *codec = avctx->codec;
int i;
park_frame_worker_threads(fctx, thread_count);
avctx->codec = NULL;
release_delayed_buffers(p);
+ av_frame_unref(&p->frame);
}
for (i = 0; i < thread_count; i++) {
PerThreadContext *p = &fctx->threads[i];
- avcodec_default_free_buffers(p->avctx);
-
pthread_mutex_destroy(&p->mutex);
pthread_mutex_destroy(&p->progress_mutex);
pthread_cond_destroy(&p->input_cond);
pthread_cond_destroy(&p->progress_cond);
pthread_cond_destroy(&p->output_cond);
- av_freep(&p->avpkt.data);
+ av_buffer_unref(&p->avpkt.buf);
+ av_freep(&p->buf);
+ av_freep(&p->released_buffers);
if (i) {
av_freep(&p->avctx->priv_data);
av_freep(&p->avctx->internal);
+ av_freep(&p->avctx->slice_offset);
}
av_freep(&p->avctx);
static int frame_thread_init(AVCodecContext *avctx)
{
int thread_count = avctx->thread_count;
- AVCodec *codec = avctx->codec;
+ const AVCodec *codec = avctx->codec;
AVCodecContext *src = avctx;
FrameThreadContext *fctx;
int i, err = 0;
if (!thread_count) {
int nb_cpus = get_logical_cpus(avctx);
- // use number of cores + 1 as thread count if there is motre than one
+ // use number of cores + 1 as thread count if there is more than one
if (nb_cpus > 1)
- thread_count = avctx->thread_count = nb_cpus + 1;
+ thread_count = avctx->thread_count = FFMIN(nb_cpus + 1, MAX_AUTO_THREADS);
+ else
+ thread_count = avctx->thread_count = 1;
}
if (thread_count <= 1) {
err = AVERROR(ENOMEM);
goto error;
}
- *(copy->internal) = *(src->internal);
+ *copy->internal = *src->internal;
copy->internal->is_copy = 1;
if (codec->init_thread_copy)
void ff_thread_flush(AVCodecContext *avctx)
{
+ int i;
FrameThreadContext *fctx = avctx->thread_opaque;
if (!avctx->thread_opaque) return;
fctx->next_decoding = fctx->next_finished = 0;
fctx->delaying = 1;
fctx->prev_thread = NULL;
-}
-
-static int *allocate_progress(PerThreadContext *p)
-{
- int i;
-
- for (i = 0; i < MAX_BUFFERS; i++)
- if (!p->progress_used[i]) break;
+ for (i = 0; i < avctx->thread_count; i++) {
+ PerThreadContext *p = &fctx->threads[i];
+ // Make sure decode flush calls with size=0 won't return old frames
+ p->got_frame = 0;
+ av_frame_unref(&p->frame);
- if (i == MAX_BUFFERS) {
- av_log(p->avctx, AV_LOG_ERROR, "allocate_progress() overflow\n");
- return NULL;
+ release_delayed_buffers(p);
}
-
- p->progress_used[i] = 1;
-
- return p->progress[i];
}
-int ff_thread_get_buffer(AVCodecContext *avctx, AVFrame *f)
+int ff_thread_get_buffer(AVCodecContext *avctx, ThreadFrame *f, int flags)
{
PerThreadContext *p = avctx->thread_opaque;
- int *progress, err;
+ int err;
f->owner = avctx;
- if (!(avctx->active_thread_type&FF_THREAD_FRAME)) {
- f->thread_opaque = NULL;
- return avctx->get_buffer(avctx, f);
- }
+ if (!(avctx->active_thread_type & FF_THREAD_FRAME))
+ return ff_get_buffer(avctx, f->f, flags);
if (p->state != STATE_SETTING_UP &&
(avctx->codec->update_thread_context || !avctx->thread_safe_callbacks)) {
return -1;
}
- pthread_mutex_lock(&p->parent->buffer_mutex);
- f->thread_opaque = progress = allocate_progress(p);
+ if (avctx->internal->allocate_progress) {
+ int *progress;
+ f->progress = av_buffer_alloc(2 * sizeof(int));
+ if (!f->progress) {
+ return AVERROR(ENOMEM);
+ }
+ progress = (int*)f->progress->data;
- if (!progress) {
- pthread_mutex_unlock(&p->parent->buffer_mutex);
- return -1;
+ progress[0] = progress[1] = -1;
}
- progress[0] =
- progress[1] = -1;
-
- if (avctx->thread_safe_callbacks ||
- avctx->get_buffer == avcodec_default_get_buffer) {
- err = avctx->get_buffer(avctx, f);
+ pthread_mutex_lock(&p->parent->buffer_mutex);
+ if (avctx->thread_safe_callbacks || (
+#if FF_API_GET_BUFFER
+ !avctx->get_buffer &&
+#endif
+ avctx->get_buffer2 == avcodec_default_get_buffer2)) {
+ err = ff_get_buffer(avctx, f->f, flags);
} else {
- p->requested_frame = f;
+ p->requested_frame = f->f;
+ p->requested_flags = flags;
p->state = STATE_GET_BUFFER;
pthread_mutex_lock(&p->progress_mutex);
pthread_cond_signal(&p->progress_cond);
ff_thread_finish_setup(avctx);
}
+ if (err)
+ av_buffer_unref(&f->progress);
+
pthread_mutex_unlock(&p->parent->buffer_mutex);
return err;
}
-void ff_thread_release_buffer(AVCodecContext *avctx, AVFrame *f)
+void ff_thread_release_buffer(AVCodecContext *avctx, ThreadFrame *f)
{
PerThreadContext *p = avctx->thread_opaque;
FrameThreadContext *fctx;
+ AVFrame *dst, *tmp;
+ int can_direct_free = !(avctx->active_thread_type & FF_THREAD_FRAME) ||
+ avctx->thread_safe_callbacks ||
+ (
+#if FF_API_GET_BUFFER
+ !avctx->get_buffer &&
+#endif
+ avctx->get_buffer2 == avcodec_default_get_buffer2);
- if (!(avctx->active_thread_type&FF_THREAD_FRAME)) {
- avctx->release_buffer(avctx, f);
+ if (!f->f->data[0])
return;
- }
- if (p->num_released_buffers >= MAX_BUFFERS) {
- av_log(p->avctx, AV_LOG_ERROR, "too many thread_release_buffer calls!\n");
+ if (avctx->debug & FF_DEBUG_BUFFERS)
+ av_log(avctx, AV_LOG_DEBUG, "thread_release_buffer called on pic %p\n", f);
+
+ av_buffer_unref(&f->progress);
+ f->owner = NULL;
+
+ if (can_direct_free) {
+ av_frame_unref(f->f);
return;
}
- if(avctx->debug & FF_DEBUG_BUFFERS)
- av_log(avctx, AV_LOG_DEBUG, "thread_release_buffer called on pic %p\n", f);
-
fctx = p->parent;
pthread_mutex_lock(&fctx->buffer_mutex);
- p->released_buffers[p->num_released_buffers++] = *f;
+
+ if (p->num_released_buffers + 1 >= INT_MAX / sizeof(*p->released_buffers))
+ goto fail;
+ tmp = av_fast_realloc(p->released_buffers, &p->released_buffers_allocated,
+ (p->num_released_buffers + 1) *
+ sizeof(*p->released_buffers));
+ if (!tmp)
+ goto fail;
+ p->released_buffers = tmp;
+
+ dst = &p->released_buffers[p->num_released_buffers];
+ av_frame_move_ref(dst, f->f);
+
+ p->num_released_buffers++;
+
+fail:
pthread_mutex_unlock(&fctx->buffer_mutex);
- memset(f->data, 0, sizeof(f->data));
}
/**
} else if (avctx->codec->capabilities & CODEC_CAP_SLICE_THREADS &&
avctx->thread_type & FF_THREAD_SLICE) {
avctx->active_thread_type = FF_THREAD_SLICE;
+ } else if (!(avctx->codec->capabilities & CODEC_CAP_AUTO_THREADS)) {
+ avctx->thread_count = 1;
+ avctx->active_thread_type = 0;
}
+
+ if (avctx->thread_count > MAX_AUTO_THREADS)
+ av_log(avctx, AV_LOG_WARNING,
+ "Application has requested %d threads. Using a thread count greater than %d is not recommended.\n",
+ avctx->thread_count, MAX_AUTO_THREADS);
}
int ff_thread_init(AVCodecContext *avctx)
projects / ffmpeg / blobdiff