386 lines
10 KiB
C
386 lines
10 KiB
C
/*
|
|
* Copyright (c) 2015 John R. Bradley <jrb@turrettech.com>
|
|
*
|
|
* Permission to use, copy, modify, and distribute this software for any
|
|
* purpose with or without fee is hereby granted, provided that the above
|
|
* copyright notice and this permission notice appear in all copies.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
|
|
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
|
|
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
|
|
* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
|
|
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
|
|
* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
|
|
* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
|
|
*/
|
|
|
|
#include "ff-decoder.h"
|
|
|
|
#include <libavutil/time.h>
|
|
#include <assert.h>
|
|
|
|
typedef void *(*ff_decoder_thread_t)(void *opaque_decoder);
|
|
|
|
extern void *ff_audio_decoder_thread(void *opaque_audio_decoder);
|
|
extern void *ff_video_decoder_thread(void *opaque_video_decoder);
|
|
|
|
struct ff_decoder *ff_decoder_init(AVCodecContext *codec_context,
|
|
AVStream *stream, unsigned int packet_queue_size,
|
|
unsigned int frame_queue_size)
|
|
{
|
|
bool success;
|
|
|
|
assert(codec_context != NULL);
|
|
assert(stream != NULL);
|
|
|
|
struct ff_decoder *decoder = av_mallocz(sizeof(struct ff_decoder));
|
|
|
|
if (decoder == NULL)
|
|
goto fail;
|
|
|
|
decoder->codec = codec_context;
|
|
decoder->codec->opaque = decoder;
|
|
decoder->stream = stream;
|
|
decoder->abort = false;
|
|
decoder->finished = false;
|
|
|
|
decoder->packet_queue_size = packet_queue_size;
|
|
if (!packet_queue_init(&decoder->packet_queue))
|
|
goto fail1;
|
|
|
|
decoder->timer_next_wake = (double)av_gettime() / 1000000.0;
|
|
decoder->previous_pts_diff = 40e-3;
|
|
decoder->current_pts_time = av_gettime();
|
|
decoder->start_pts = 0;
|
|
decoder->predicted_pts = 0;
|
|
decoder->first_frame = true;
|
|
|
|
success = ff_timer_init(&decoder->refresh_timer, ff_decoder_refresh,
|
|
decoder);
|
|
if (!success)
|
|
goto fail2;
|
|
|
|
success = ff_circular_queue_init(&decoder->frame_queue,
|
|
sizeof(struct ff_frame), frame_queue_size);
|
|
if (!success)
|
|
goto fail3;
|
|
|
|
return decoder;
|
|
|
|
fail3:
|
|
ff_timer_free(&decoder->refresh_timer);
|
|
fail2:
|
|
packet_queue_free(&decoder->packet_queue);
|
|
fail1:
|
|
av_free(decoder);
|
|
fail:
|
|
return NULL;
|
|
}
|
|
|
|
bool ff_decoder_start(struct ff_decoder *decoder)
|
|
{
|
|
assert(decoder != NULL);
|
|
|
|
ff_decoder_thread_t decoder_thread;
|
|
if (decoder->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
|
|
decoder_thread = ff_audio_decoder_thread;
|
|
} else if (decoder->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
|
|
decoder_thread = ff_video_decoder_thread;
|
|
} else {
|
|
av_log(NULL, AV_LOG_ERROR, "no decoder found for type %d",
|
|
decoder->codec->codec_type);
|
|
return false;
|
|
}
|
|
|
|
ff_decoder_schedule_refresh(decoder, 40);
|
|
|
|
return (pthread_create(&decoder->decoder_thread, NULL,
|
|
decoder_thread, decoder) != 0);
|
|
}
|
|
|
|
void ff_decoder_free(struct ff_decoder *decoder)
|
|
{
|
|
void *decoder_thread_result;
|
|
int i;
|
|
|
|
assert(decoder != NULL);
|
|
|
|
decoder->abort = true;
|
|
|
|
ff_circular_queue_abort(&decoder->frame_queue);
|
|
packet_queue_abort(&decoder->packet_queue);
|
|
|
|
ff_timer_free(&decoder->refresh_timer);
|
|
|
|
pthread_join(decoder->decoder_thread, &decoder_thread_result);
|
|
|
|
for (i = 0; i < decoder->frame_queue.capacity; i++) {
|
|
void *item = decoder->frame_queue.slots[i];
|
|
struct ff_frame *frame = (struct ff_frame *)item;
|
|
|
|
ff_callbacks_frame_free(frame, decoder->callbacks);
|
|
|
|
if (frame != NULL) {
|
|
if (frame->frame != NULL)
|
|
av_frame_unref(frame->frame);
|
|
if (frame->clock != NULL)
|
|
ff_clock_release(&frame->clock);
|
|
av_free(frame);
|
|
}
|
|
}
|
|
|
|
packet_queue_free(&decoder->packet_queue);
|
|
ff_circular_queue_free(&decoder->frame_queue);
|
|
|
|
avcodec_close(decoder->codec);
|
|
|
|
av_free(decoder);
|
|
}
|
|
|
|
void ff_decoder_schedule_refresh(struct ff_decoder *decoder, int delay)
|
|
{
|
|
ff_timer_schedule(&decoder->refresh_timer, 1000*delay);
|
|
}
|
|
|
|
double ff_decoder_clock(void *opaque)
|
|
{
|
|
struct ff_decoder *decoder = opaque;
|
|
double delta = (av_gettime() - decoder->current_pts_time) / 1000000.0;
|
|
return decoder->current_pts + delta;
|
|
}
|
|
|
|
static double get_sync_adjusted_pts_diff(struct ff_clock *clock,
|
|
double pts, double pts_diff)
|
|
{
|
|
double new_pts_diff = pts_diff;
|
|
double sync_time = ff_get_sync_clock(clock);
|
|
double diff = pts - sync_time;
|
|
double sync_threshold;
|
|
|
|
sync_threshold = (pts_diff > AV_SYNC_THRESHOLD)
|
|
? pts_diff : AV_SYNC_THRESHOLD;
|
|
|
|
if (fabs(diff) < AV_NOSYNC_THRESHOLD) {
|
|
if (diff <= -sync_threshold) {
|
|
new_pts_diff = 0;
|
|
|
|
} else if (diff >= sync_threshold) {
|
|
new_pts_diff = 2 * pts_diff;
|
|
}
|
|
}
|
|
|
|
return new_pts_diff;
|
|
}
|
|
|
|
void ff_decoder_refresh(void *opaque)
|
|
{
|
|
struct ff_decoder *decoder = (struct ff_decoder *)opaque;
|
|
|
|
struct ff_frame *frame;
|
|
|
|
if (decoder->stream) {
|
|
if (decoder->frame_queue.size == 0) {
|
|
if (!decoder->eof || !decoder->finished) {
|
|
// We expected a frame, but there were none
|
|
// available
|
|
|
|
// Schedule another call as soon as possible
|
|
ff_decoder_schedule_refresh(decoder, 1);
|
|
} else {
|
|
ff_callbacks_frame(decoder->callbacks, NULL);
|
|
decoder->refresh_timer.abort = true;
|
|
// no more refreshes, we are at the eof
|
|
av_log(NULL, AV_LOG_INFO,
|
|
"refresh timer stopping; eof");
|
|
return;
|
|
}
|
|
} else {
|
|
double pts_diff;
|
|
double delay_until_next_wake;
|
|
bool late_first_frame = false;
|
|
|
|
frame = ff_circular_queue_peek_read(
|
|
&decoder->frame_queue);
|
|
|
|
// Get frame clock and start it if needed
|
|
ff_clock_t *clock = ff_clock_move(&frame->clock);
|
|
if (!ff_clock_start(clock, decoder->natural_sync_clock,
|
|
&decoder->refresh_timer.abort)) {
|
|
ff_clock_release(&clock);
|
|
|
|
// Our clock was never started and deleted or
|
|
// aborted
|
|
|
|
if (decoder->refresh_timer.abort) {
|
|
av_log(NULL, AV_LOG_INFO,
|
|
"refresh timer aborted");
|
|
return;
|
|
}
|
|
|
|
// Drop this frame? The only way this can happen
|
|
// is if one stream finishes before another and
|
|
// the input is looping or canceled. Until we
|
|
// get another clock we will unable to continue
|
|
|
|
ff_decoder_schedule_refresh(decoder, 100);
|
|
|
|
// Drop this frame as we have no way of timing
|
|
// it
|
|
ff_circular_queue_advance_read(
|
|
&decoder->frame_queue);
|
|
return;
|
|
}
|
|
|
|
decoder->current_pts = frame->pts;
|
|
decoder->current_pts_time = av_gettime();
|
|
|
|
// the amount of time until we need to display this
|
|
// frame
|
|
pts_diff = frame->pts - decoder->previous_pts;
|
|
|
|
// if the first frame is a very large value, we've most
|
|
// likely started mid-stream, and the initial diff
|
|
// should be ignored.
|
|
if (decoder->first_frame) {
|
|
late_first_frame = pts_diff >= 1.0;
|
|
decoder->first_frame = false;
|
|
}
|
|
|
|
if (pts_diff <= 0 || late_first_frame) {
|
|
// if diff is invalid, use previous
|
|
pts_diff = decoder->previous_pts_diff;
|
|
}
|
|
|
|
// save for next time
|
|
decoder->previous_pts_diff = pts_diff;
|
|
decoder->previous_pts = frame->pts;
|
|
|
|
// if not synced against natural clock
|
|
if (clock->sync_type
|
|
!= decoder->natural_sync_clock) {
|
|
pts_diff = get_sync_adjusted_pts_diff(clock,
|
|
frame->pts, pts_diff);
|
|
}
|
|
|
|
decoder->timer_next_wake += pts_diff;
|
|
|
|
// compute the amount of time until next refresh
|
|
delay_until_next_wake = decoder->timer_next_wake -
|
|
(av_gettime() / 1000000.0L);
|
|
if (delay_until_next_wake < 0.010L) {
|
|
delay_until_next_wake = 0.010L;
|
|
}
|
|
|
|
if (delay_until_next_wake > pts_diff)
|
|
delay_until_next_wake = pts_diff;
|
|
|
|
ff_clock_release(&clock);
|
|
ff_callbacks_frame(decoder->callbacks, frame);
|
|
|
|
ff_decoder_schedule_refresh(decoder,
|
|
(int)(delay_until_next_wake * 1000
|
|
+ 0.5L));
|
|
|
|
av_frame_free(&frame->frame);
|
|
|
|
ff_circular_queue_advance_read(&decoder->frame_queue);
|
|
}
|
|
} else {
|
|
ff_decoder_schedule_refresh(decoder, 100);
|
|
}
|
|
}
|
|
|
|
bool ff_decoder_full(struct ff_decoder *decoder)
|
|
{
|
|
if (decoder == NULL)
|
|
return false;
|
|
|
|
return (decoder->packet_queue.total_size > decoder->packet_queue_size);
|
|
}
|
|
|
|
bool ff_decoder_accept(struct ff_decoder *decoder, struct ff_packet *packet)
|
|
{
|
|
if (decoder && packet->base.stream_index == decoder->stream->index) {
|
|
packet_queue_put(&decoder->packet_queue, packet);
|
|
return true;
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
double ff_decoder_get_best_effort_pts(struct ff_decoder *decoder,
|
|
AVFrame *frame)
|
|
{
|
|
// this is how long each frame is added to the amount of repeated frames
|
|
// according to the codec
|
|
double estimated_frame_delay;
|
|
int64_t best_effort_pts;
|
|
double d_pts;
|
|
|
|
// This function is ffmpeg only, replace with frame->pkt_pts
|
|
// if you are trying to compile for libav as a temporary
|
|
// measure
|
|
best_effort_pts = av_frame_get_best_effort_timestamp(frame);
|
|
|
|
if (best_effort_pts != AV_NOPTS_VALUE) {
|
|
// Fix the first pts if less than start_pts
|
|
if (best_effort_pts < decoder->start_pts) {
|
|
if (decoder->first_frame) {
|
|
best_effort_pts = decoder->start_pts;
|
|
} else {
|
|
av_log(NULL, AV_LOG_WARNING, "multiple pts < "
|
|
"start_pts; setting start pts "
|
|
"to 0");
|
|
decoder->start_pts = 0;
|
|
}
|
|
}
|
|
|
|
best_effort_pts -= decoder->start_pts;
|
|
|
|
// Since the best effort pts came from the stream we use his
|
|
// time base
|
|
d_pts = best_effort_pts * av_q2d(decoder->stream->time_base);
|
|
decoder->predicted_pts = d_pts;
|
|
} else {
|
|
d_pts = decoder->predicted_pts;
|
|
}
|
|
|
|
// Update our predicted pts to include the repeated picture count
|
|
// Our predicted pts clock is based on the codecs time base
|
|
estimated_frame_delay = av_frame_get_pkt_duration(frame)
|
|
* av_q2d(decoder->codec->time_base);
|
|
// Add repeat frame delay
|
|
estimated_frame_delay += frame->repeat_pict
|
|
/ (1.0L / estimated_frame_delay);
|
|
|
|
decoder->predicted_pts += estimated_frame_delay;
|
|
|
|
return d_pts;
|
|
}
|
|
|
|
bool ff_decoder_set_frame_drop_state(struct ff_decoder *decoder,
|
|
int64_t start_time, int64_t pts)
|
|
{
|
|
if (pts != AV_NOPTS_VALUE) {
|
|
int64_t rescaled_pts = av_rescale_q(pts,
|
|
decoder->stream->time_base, AV_TIME_BASE_Q);
|
|
int64_t master_clock = av_gettime() -
|
|
start_time;
|
|
|
|
int64_t diff = master_clock - rescaled_pts;
|
|
|
|
if (diff > (AV_TIME_BASE / 2)) {
|
|
decoder->codec->skip_frame = decoder->frame_drop;
|
|
decoder->codec->skip_idct = decoder->frame_drop;
|
|
decoder->codec->skip_loop_filter = decoder->frame_drop;
|
|
return true;
|
|
} else {
|
|
decoder->codec->skip_frame = AVDISCARD_DEFAULT;
|
|
decoder->codec->skip_idct = AVDISCARD_DEFAULT;
|
|
decoder->codec->skip_loop_filter = AVDISCARD_DEFAULT;
|
|
return false;
|
|
}
|
|
}
|
|
|
|
return false;
|
|
}
|