#include "config.h"
+#include <stdatomic.h>
#include <stdint.h>
#if HAVE_PTHREADS
#endif
#include "avcodec.h"
+#include "hwaccel.h"
#include "internal.h"
#include "pthread_internal.h"
#include "thread.h"
+#include "version.h"
#include "libavutil/avassert.h"
#include "libavutil/buffer.h"
#include "libavutil/common.h"
#include "libavutil/cpu.h"
#include "libavutil/frame.h"
+#include "libavutil/internal.h"
#include "libavutil/log.h"
#include "libavutil/mem.h"
+enum {
+ ///< Set when the thread is awaiting a packet.
+ STATE_INPUT_READY,
+ ///< Set before the codec has called ff_thread_finish_setup().
+ STATE_SETTING_UP,
+ /**
+ * Set when the codec calls get_buffer().
+ * State is returned to STATE_SETTING_UP afterwards.
+ */
+ STATE_GET_BUFFER,
+ ///< Set after the codec has called ff_thread_finish_setup().
+ STATE_SETUP_FINISHED,
+};
+
/**
* Context used by codec threads and stored in their AVCodecInternal thread_ctx.
*/
AVCodecContext *avctx; ///< Context used to decode packets passed to this thread.
AVPacket avpkt; ///< Input packet (for decoding) or output (for encoding).
- 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.
int result; ///< The result of the last codec decode/encode() call.
- enum {
- STATE_INPUT_READY, ///< Set when the thread is awaiting a packet.
- STATE_SETTING_UP, ///< Set before the codec has called ff_thread_finish_setup().
- STATE_GET_BUFFER, /**<
- * Set when the codec calls get_buffer().
- * State is returned to STATE_SETTING_UP afterwards.
- */
- STATE_SETUP_FINISHED ///< Set after the codec has called ff_thread_finish_setup().
- } state;
+ atomic_int state;
/**
* Array of frames passed to ff_thread_release_buffer().
AVFrame *requested_frame; ///< AVFrame the codec passed to get_buffer()
int requested_flags; ///< flags passed to get_buffer() for requested_frame
+
+ int die; ///< Set when the thread should exit.
+
+ int hwaccel_serializing;
+ int async_serializing;
} PerThreadContext;
/**
PerThreadContext *prev_thread; ///< The last thread submit_packet() was called on.
pthread_mutex_t buffer_mutex; ///< Mutex used to protect get/release_buffer().
+ /**
+ * This lock is used for ensuring threads run in serial when hwaccel
+ * is used.
+ */
+ pthread_mutex_t hwaccel_mutex;
+ pthread_mutex_t async_mutex;
int next_decoding; ///< The next context to submit a packet to.
int next_finished; ///< The next context to return output from.
* Set for the first N packets, where N is the number of threads.
* While it is set, ff_thread_en/decode_frame won't return any results.
*/
-
- int die; ///< Set when threads should exit.
} FrameThreadContext;
/**
static attribute_align_arg void *frame_worker_thread(void *arg)
{
PerThreadContext *p = arg;
- FrameThreadContext *fctx = p->parent;
AVCodecContext *avctx = p->avctx;
const AVCodec *codec = avctx->codec;
while (1) {
- if (p->state == STATE_INPUT_READY && !fctx->die) {
+ if (atomic_load(&p->state) == STATE_INPUT_READY) {
pthread_mutex_lock(&p->mutex);
- while (p->state == STATE_INPUT_READY && !fctx->die)
+ while (atomic_load(&p->state) == STATE_INPUT_READY) {
+ if (p->die) {
+ pthread_mutex_unlock(&p->mutex);
+ goto die;
+ }
pthread_cond_wait(&p->input_cond, &p->mutex);
+ }
pthread_mutex_unlock(&p->mutex);
}
- if (fctx->die) break;
-
if (!codec->update_thread_context && avctx->thread_safe_callbacks)
ff_thread_finish_setup(avctx);
pthread_mutex_lock(&p->mutex);
+
+ /* If a decoder supports hwaccel, then it must call ff_get_format().
+ * Since that call must happen before ff_thread_finish_setup(), the
+ * decoder is required to implement update_thread_context() and call
+ * ff_thread_finish_setup() manually. Therefore the above
+ * ff_thread_finish_setup() call did not happen and hwaccel_serializing
+ * cannot be true here. */
+ av_assert0(!p->hwaccel_serializing);
+
+ /* if the previous thread uses hwaccel then we take the lock to ensure
+ * the threads don't run concurrently */
+ if (avctx->hwaccel) {
+ pthread_mutex_lock(&p->parent->hwaccel_mutex);
+ p->hwaccel_serializing = 1;
+ }
+
av_frame_unref(p->frame);
p->got_frame = 0;
p->result = codec->decode(avctx, p->frame, &p->got_frame, &p->avpkt);
av_frame_unref(p->frame);
}
- if (p->state == STATE_SETTING_UP) ff_thread_finish_setup(avctx);
+ if (atomic_load(&p->state) == STATE_SETTING_UP)
+ ff_thread_finish_setup(avctx);
+
+ if (p->hwaccel_serializing) {
+ p->hwaccel_serializing = 0;
+ pthread_mutex_unlock(&p->parent->hwaccel_mutex);
+ }
+
+ if (p->async_serializing) {
+ p->async_serializing = 0;
+ pthread_mutex_unlock(&p->parent->async_mutex);
+ }
- p->state = STATE_INPUT_READY;
+ atomic_store(&p->state, STATE_INPUT_READY);
pthread_mutex_lock(&p->progress_mutex);
pthread_cond_signal(&p->output_cond);
pthread_mutex_unlock(&p->mutex);
}
+die:
return NULL;
}
if (dst != src) {
dst->time_base = src->time_base;
+ dst->framerate = src->framerate;
dst->width = src->width;
dst->height = src->height;
dst->pix_fmt = src->pix_fmt;
dst->bits_per_coded_sample = src->bits_per_coded_sample;
dst->sample_aspect_ratio = src->sample_aspect_ratio;
+#if FF_API_AFD
+FF_DISABLE_DEPRECATION_WARNINGS
dst->dtg_active_format = src->dtg_active_format;
+FF_ENABLE_DEPRECATION_WARNINGS
+#endif /* FF_API_AFD */
dst->profile = src->profile;
dst->level = src->level;
dst->hwaccel = src->hwaccel;
dst->hwaccel_context = src->hwaccel_context;
dst->internal->hwaccel_priv_data = src->internal->hwaccel_priv_data;
+
+ if (!!dst->hw_frames_ctx != !!src->hw_frames_ctx ||
+ (dst->hw_frames_ctx && dst->hw_frames_ctx->data != src->hw_frames_ctx->data)) {
+ av_buffer_unref(&dst->hw_frames_ctx);
+
+ if (src->hw_frames_ctx) {
+ dst->hw_frames_ctx = av_buffer_ref(src->hw_frames_ctx);
+ if (!dst->hw_frames_ctx)
+ return AVERROR(ENOMEM);
+ }
+ }
+
+ dst->hwaccel_flags = src->hwaccel_flags;
}
if (for_user) {
+#if FF_API_CODED_FRAME
+FF_DISABLE_DEPRECATION_WARNINGS
dst->coded_frame = src->coded_frame;
+FF_ENABLE_DEPRECATION_WARNINGS
+#endif
} else {
if (dst->codec->update_thread_context)
err = dst->codec->update_thread_context(dst, src);
dst->draw_horiz_band= src->draw_horiz_band;
dst->get_buffer2 = src->get_buffer2;
-#if FF_API_GET_BUFFER
-FF_DISABLE_DEPRECATION_WARNINGS
- dst->get_buffer = src->get_buffer;
- dst->release_buffer = src->release_buffer;
-FF_ENABLE_DEPRECATION_WARNINGS
-#endif
dst->opaque = src->opaque;
dst->debug = src->debug;
PerThreadContext *prev_thread = fctx->prev_thread;
const AVCodec *codec = p->avctx->codec;
- if (!avpkt->size && !(codec->capabilities & CODEC_CAP_DELAY)) return 0;
+ if (!avpkt->size && !(codec->capabilities & AV_CODEC_CAP_DELAY))
+ return 0;
pthread_mutex_lock(&p->mutex);
if (prev_thread) {
int err;
- if (prev_thread->state == STATE_SETTING_UP) {
+ if (atomic_load(&prev_thread->state) == STATE_SETTING_UP) {
pthread_mutex_lock(&prev_thread->progress_mutex);
- while (prev_thread->state == STATE_SETTING_UP)
+ while (atomic_load(&prev_thread->state) == STATE_SETTING_UP)
pthread_cond_wait(&prev_thread->progress_cond, &prev_thread->progress_mutex);
pthread_mutex_unlock(&prev_thread->progress_mutex);
}
}
}
- av_buffer_unref(&p->avpkt.buf);
- p->avpkt = *avpkt;
- 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);
- }
+ av_packet_unref(&p->avpkt);
+ av_packet_ref(&p->avpkt, avpkt);
- p->state = STATE_SETTING_UP;
+ atomic_store(&p->state, STATE_SETTING_UP);
pthread_cond_signal(&p->input_cond);
pthread_mutex_unlock(&p->mutex);
* and it calls back to the client here.
*/
-FF_DISABLE_DEPRECATION_WARNINGS
- if (!p->avctx->thread_safe_callbacks && (
-#if FF_API_GET_BUFFER
- p->avctx->get_buffer ||
-#endif
- p->avctx->get_buffer2 != avcodec_default_get_buffer2)) {
-FF_ENABLE_DEPRECATION_WARNINGS
- while (p->state != STATE_SETUP_FINISHED && p->state != STATE_INPUT_READY) {
+ if (!p->avctx->thread_safe_callbacks &&
+ p->avctx->get_buffer2 != avcodec_default_get_buffer2) {
+ while (atomic_load(&p->state) != STATE_SETUP_FINISHED &&
+ atomic_load(&p->state) != STATE_INPUT_READY) {
pthread_mutex_lock(&p->progress_mutex);
- while (p->state == STATE_SETTING_UP)
+ while (atomic_load(&p->state) == STATE_SETTING_UP)
pthread_cond_wait(&p->progress_cond, &p->progress_mutex);
- if (p->state == STATE_GET_BUFFER) {
+ if (atomic_load_explicit(&p->state, memory_order_acquire) == STATE_GET_BUFFER) {
p->result = ff_get_buffer(p->avctx, p->requested_frame, p->requested_flags);
- p->state = STATE_SETTING_UP;
+ atomic_store(&p->state, STATE_SETTING_UP);
pthread_cond_signal(&p->progress_cond);
}
pthread_mutex_unlock(&p->progress_mutex);
FrameThreadContext *fctx = avctx->internal->thread_ctx;
int finished = fctx->next_finished;
PerThreadContext *p;
- int err;
+ int err, ret;
+
+ /* release the async lock, permitting blocked hwaccel threads to
+ * go forward while we are in this function */
+ pthread_mutex_unlock(&fctx->async_mutex);
/*
* Submit a packet to the next decoding thread.
p = &fctx->threads[fctx->next_decoding];
err = update_context_from_user(p->avctx, avctx);
- if (err) return err;
+ if (err)
+ goto finish;
err = submit_packet(p, avpkt);
- if (err) return err;
+ if (err)
+ goto finish;
/*
* If we're still receiving the initial packets, don't return a frame.
if (fctx->next_decoding >= (avctx->thread_count-1)) fctx->delaying = 0;
*got_picture_ptr=0;
- if (avpkt->size)
- return avpkt->size;
+ if (avpkt->size) {
+ ret = avpkt->size;
+ goto finish;
+ }
}
/*
do {
p = &fctx->threads[finished++];
- if (p->state != STATE_INPUT_READY) {
+ if (atomic_load(&p->state) != STATE_INPUT_READY) {
pthread_mutex_lock(&p->progress_mutex);
- while (p->state != STATE_INPUT_READY)
+ while (atomic_load_explicit(&p->state, memory_order_relaxed) != STATE_INPUT_READY)
pthread_cond_wait(&p->output_cond, &p->progress_mutex);
pthread_mutex_unlock(&p->progress_mutex);
}
fctx->next_finished = finished;
/* return the size of the consumed packet if no error occurred */
- return (p->result >= 0) ? avpkt->size : p->result;
+ ret = (p->result >= 0) ? avpkt->size : p->result;
+finish:
+ pthread_mutex_lock(&fctx->async_mutex);
+ if (err < 0)
+ return err;
+ return ret;
}
void ff_thread_report_progress(ThreadFrame *f, int n, int field)
{
PerThreadContext *p;
- int *progress = f->progress ? (int*)f->progress->data : NULL;
+ atomic_int *progress = f->progress ? (atomic_int*)f->progress->data : NULL;
- if (!progress || progress[field] >= n) return;
+ if (!progress ||
+ atomic_load_explicit(&progress[field], memory_order_relaxed) >= n)
+ return;
p = f->owner->internal->thread_ctx;
av_log(f->owner, AV_LOG_DEBUG, "%p finished %d field %d\n", progress, n, field);
pthread_mutex_lock(&p->progress_mutex);
- progress[field] = n;
+
+ atomic_store_explicit(&progress[field], n, memory_order_release);
+
pthread_cond_broadcast(&p->progress_cond);
pthread_mutex_unlock(&p->progress_mutex);
}
void ff_thread_await_progress(ThreadFrame *f, int n, int field)
{
PerThreadContext *p;
- int *progress = f->progress ? (int*)f->progress->data : NULL;
+ atomic_int *progress = f->progress ? (atomic_int*)f->progress->data : NULL;
- if (!progress || progress[field] >= n) return;
+ if (!progress ||
+ atomic_load_explicit(&progress[field], memory_order_acquire) >= n)
+ return;
p = f->owner->internal->thread_ctx;
av_log(f->owner, AV_LOG_DEBUG, "thread awaiting %d field %d from %p\n", n, field, progress);
pthread_mutex_lock(&p->progress_mutex);
- while (progress[field] < n)
+ while (atomic_load_explicit(&progress[field], memory_order_relaxed) < n)
pthread_cond_wait(&p->progress_cond, &p->progress_mutex);
pthread_mutex_unlock(&p->progress_mutex);
}
if (!(avctx->active_thread_type&FF_THREAD_FRAME)) return;
+ if (avctx->hwaccel && !p->hwaccel_serializing) {
+ pthread_mutex_lock(&p->parent->hwaccel_mutex);
+ p->hwaccel_serializing = 1;
+ }
+
+ /* this assumes that no hwaccel calls happen before ff_thread_finish_setup() */
+ if (avctx->hwaccel &&
+ !(avctx->hwaccel->caps_internal & HWACCEL_CAP_ASYNC_SAFE)) {
+ p->async_serializing = 1;
+ pthread_mutex_lock(&p->parent->async_mutex);
+ }
+
pthread_mutex_lock(&p->progress_mutex);
- p->state = STATE_SETUP_FINISHED;
+
+ atomic_store(&p->state, STATE_SETUP_FINISHED);
+
pthread_cond_broadcast(&p->progress_cond);
pthread_mutex_unlock(&p->progress_mutex);
}
{
int i;
+ pthread_mutex_unlock(&fctx->async_mutex);
+
for (i = 0; i < thread_count; i++) {
PerThreadContext *p = &fctx->threads[i];
- if (p->state != STATE_INPUT_READY) {
+ if (atomic_load(&p->state) != STATE_INPUT_READY) {
pthread_mutex_lock(&p->progress_mutex);
- while (p->state != STATE_INPUT_READY)
+ while (atomic_load(&p->state) != STATE_INPUT_READY)
pthread_cond_wait(&p->output_cond, &p->progress_mutex);
pthread_mutex_unlock(&p->progress_mutex);
}
}
+
+ pthread_mutex_lock(&fctx->async_mutex);
}
void ff_frame_thread_free(AVCodecContext *avctx, int thread_count)
if (fctx->prev_thread && fctx->prev_thread != fctx->threads)
update_context_from_thread(fctx->threads->avctx, fctx->prev_thread->avctx, 0);
- fctx->die = 1;
-
for (i = 0; i < thread_count; i++) {
PerThreadContext *p = &fctx->threads[i];
pthread_mutex_lock(&p->mutex);
+ p->die = 1;
pthread_cond_signal(&p->input_cond);
pthread_mutex_unlock(&p->mutex);
pthread_cond_destroy(&p->input_cond);
pthread_cond_destroy(&p->progress_cond);
pthread_cond_destroy(&p->output_cond);
- av_buffer_unref(&p->avpkt.buf);
- av_freep(&p->buf);
+ av_packet_unref(&p->avpkt);
av_freep(&p->released_buffers);
if (i) {
av_freep(&p->avctx->slice_offset);
}
+ av_buffer_unref(&p->avctx->hw_frames_ctx);
+
av_freep(&p->avctx->internal);
av_freep(&p->avctx);
}
av_freep(&fctx->threads);
pthread_mutex_destroy(&fctx->buffer_mutex);
+ pthread_mutex_destroy(&fctx->hwaccel_mutex);
+
+ pthread_mutex_unlock(&fctx->async_mutex);
+ pthread_mutex_destroy(&fctx->async_mutex);
+
av_freep(&avctx->internal->thread_ctx);
}
}
avctx->internal->thread_ctx = fctx = av_mallocz(sizeof(FrameThreadContext));
+ if (!fctx)
+ return AVERROR(ENOMEM);
fctx->threads = av_mallocz(sizeof(PerThreadContext) * thread_count);
+ if (!fctx->threads) {
+ av_freep(&avctx->internal->thread_ctx);
+ return AVERROR(ENOMEM);
+ }
+
pthread_mutex_init(&fctx->buffer_mutex, NULL);
+ pthread_mutex_init(&fctx->hwaccel_mutex, NULL);
+
+ pthread_mutex_init(&fctx->async_mutex, NULL);
+ pthread_mutex_lock(&fctx->async_mutex);
+
fctx->delaying = 1;
for (i = 0; i < thread_count; i++) {
p->frame = av_frame_alloc();
if (!p->frame) {
+ av_freep(©);
err = AVERROR(ENOMEM);
goto error;
}
}
*copy->internal = *src->internal;
copy->internal->thread_ctx = p;
- copy->internal->pkt = &p->avpkt;
+ copy->internal->last_pkt_props = &p->avpkt;
if (!i) {
src = copy;
if (!(avctx->active_thread_type & FF_THREAD_FRAME))
return ff_get_buffer(avctx, f->f, flags);
- if (p->state != STATE_SETTING_UP &&
+ if (atomic_load(&p->state) != STATE_SETTING_UP &&
(avctx->codec->update_thread_context || !avctx->thread_safe_callbacks)) {
av_log(avctx, AV_LOG_ERROR, "get_buffer() cannot be called after ff_thread_finish_setup()\n");
return -1;
}
if (avctx->internal->allocate_progress) {
- int *progress;
- f->progress = av_buffer_alloc(2 * sizeof(int));
+ atomic_int *progress;
+ f->progress = av_buffer_alloc(2 * sizeof(*progress));
if (!f->progress) {
return AVERROR(ENOMEM);
}
- progress = (int*)f->progress->data;
+ progress = (atomic_int*)f->progress->data;
- progress[0] = progress[1] = -1;
+ atomic_init(&progress[0], -1);
+ atomic_init(&progress[1], -1);
}
pthread_mutex_lock(&p->parent->buffer_mutex);
-FF_DISABLE_DEPRECATION_WARNINGS
- if (avctx->thread_safe_callbacks || (
-#if FF_API_GET_BUFFER
- !avctx->get_buffer &&
-#endif
- avctx->get_buffer2 == avcodec_default_get_buffer2)) {
-FF_ENABLE_DEPRECATION_WARNINGS
+ if (avctx->thread_safe_callbacks ||
+ avctx->get_buffer2 == avcodec_default_get_buffer2) {
err = ff_get_buffer(avctx, f->f, flags);
} else {
p->requested_frame = f->f;
p->requested_flags = flags;
- p->state = STATE_GET_BUFFER;
+ atomic_store_explicit(&p->state, STATE_GET_BUFFER, memory_order_release);
pthread_mutex_lock(&p->progress_mutex);
pthread_cond_signal(&p->progress_cond);
- while (p->state != STATE_SETTING_UP)
+ while (atomic_load(&p->state) != STATE_SETTING_UP)
pthread_cond_wait(&p->progress_cond, &p->progress_mutex);
err = p->result;
PerThreadContext *p = avctx->internal->thread_ctx;
FrameThreadContext *fctx;
AVFrame *dst, *tmp;
-FF_DISABLE_DEPRECATION_WARNINGS
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);
-FF_ENABLE_DEPRECATION_WARNINGS
+ avctx->get_buffer2 == avcodec_default_get_buffer2;
if (!f->f || !f->f->buf[0])
return;