* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
+#include <stdint.h>
+
#include "libavresample/avresample.h"
+#include "libavutil/attributes.h"
#include "libavutil/audio_fifo.h"
#include "libavutil/common.h"
#include "libavutil/mathematics.h"
int min_delta; ///< pad/trim min threshold in samples
int first_frame; ///< 1 until filter_frame() has processed at least 1 frame with a pts != AV_NOPTS_VALUE
int64_t first_pts; ///< user-specified first expected pts, in samples
+ int comp; ///< current resample compensation
/* options */
int resample;
.version = LIBAVUTIL_VERSION_INT,
};
-static int init(AVFilterContext *ctx, const char *args)
+static av_cold int init(AVFilterContext *ctx)
{
ASyncContext *s = ctx->priv;
- int ret;
-
- s->class = &async_class;
- av_opt_set_defaults(s);
-
- if ((ret = av_set_options_string(s, args, "=", ":")) < 0) {
- av_log(ctx, AV_LOG_ERROR, "Error parsing options string '%s'.\n", args);
- return ret;
- }
- av_opt_free(s);
s->pts = AV_NOPTS_VALUE;
s->first_frame = 1;
return 0;
}
-static void uninit(AVFilterContext *ctx)
+static av_cold void uninit(AVFilterContext *ctx)
{
ASyncContext *s = ctx->priv;
handle_trimming(ctx);
if (nb_samples = get_delay(s)) {
- AVFilterBufferRef *buf = ff_get_audio_buffer(link, AV_PERM_WRITE,
- nb_samples);
+ AVFrame *buf = ff_get_audio_buffer(link, nb_samples);
if (!buf)
return AVERROR(ENOMEM);
ret = avresample_convert(s->avr, buf->extended_data,
buf->linesize[0], nb_samples, NULL, 0, 0);
if (ret <= 0) {
- avfilter_unref_bufferp(&buf);
+ av_frame_free(&buf);
return (ret < 0) ? ret : AVERROR_EOF;
}
return ret;
}
-static int write_to_fifo(ASyncContext *s, AVFilterBufferRef *buf)
+static int write_to_fifo(ASyncContext *s, AVFrame *buf)
{
int ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
- buf->linesize[0], buf->audio->nb_samples);
- avfilter_unref_buffer(buf);
+ buf->linesize[0], buf->nb_samples);
+ av_frame_free(&buf);
return ret;
}
-static int filter_frame(AVFilterLink *inlink, AVFilterBufferRef *buf)
+static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
{
AVFilterContext *ctx = inlink->dst;
ASyncContext *s = ctx->priv;
AVFilterLink *outlink = ctx->outputs[0];
- int nb_channels = av_get_channel_layout_nb_channels(buf->audio->channel_layout);
+ int nb_channels = av_get_channel_layout_nb_channels(buf->channel_layout);
int64_t pts = (buf->pts == AV_NOPTS_VALUE) ? buf->pts :
av_rescale_q(buf->pts, inlink->time_base, outlink->time_base);
int out_size, ret;
int64_t delta;
+ int64_t new_pts;
- /* buffer data until we get the first timestamp */
- if (s->pts == AV_NOPTS_VALUE) {
+ /* buffer data until we get the next timestamp */
+ if (s->pts == AV_NOPTS_VALUE || pts == AV_NOPTS_VALUE) {
if (pts != AV_NOPTS_VALUE) {
s->pts = pts - get_delay(s);
}
return write_to_fifo(s, buf);
}
- /* now wait for the next timestamp */
- if (pts == AV_NOPTS_VALUE) {
- return write_to_fifo(s, buf);
- }
-
if (s->first_pts != AV_NOPTS_VALUE) {
handle_trimming(ctx);
if (!avresample_available(s->avr))
delta = pts - s->pts - get_delay(s);
out_size = avresample_available(s->avr);
- if (labs(delta) > s->min_delta || (s->first_frame && delta)) {
+ if (labs(delta) > s->min_delta ||
+ (s->first_frame && delta && s->first_pts != AV_NOPTS_VALUE)) {
av_log(ctx, AV_LOG_VERBOSE, "Discontinuity - %"PRId64" samples.\n", delta);
out_size = av_clipl_int32((int64_t)out_size + delta);
} else {
if (s->resample) {
- int comp = av_clip(delta, -s->max_comp, s->max_comp);
- av_log(ctx, AV_LOG_VERBOSE, "Compensating %d samples per second.\n", comp);
- avresample_set_compensation(s->avr, comp, inlink->sample_rate);
+ // adjust the compensation if delta is non-zero
+ int delay = get_delay(s);
+ int comp = s->comp + av_clip(delta * inlink->sample_rate / delay,
+ -s->max_comp, s->max_comp);
+ if (comp != s->comp) {
+ av_log(ctx, AV_LOG_VERBOSE, "Compensating %d samples per second.\n", comp);
+ if (avresample_set_compensation(s->avr, comp, inlink->sample_rate) == 0) {
+ s->comp = comp;
+ }
+ }
}
+ // adjust PTS to avoid monotonicity errors with input PTS jitter
+ pts -= delta;
delta = 0;
}
if (out_size > 0) {
- AVFilterBufferRef *buf_out = ff_get_audio_buffer(outlink, AV_PERM_WRITE,
- out_size);
+ AVFrame *buf_out = ff_get_audio_buffer(outlink, out_size);
if (!buf_out) {
ret = AVERROR(ENOMEM);
goto fail;
}
if (s->first_frame && delta > 0) {
+ int planar = av_sample_fmt_is_planar(buf_out->format);
+ int planes = planar ? nb_channels : 1;
+ int block_size = av_get_bytes_per_sample(buf_out->format) *
+ (planar ? 1 : nb_channels);
+
int ch;
av_samples_set_silence(buf_out->extended_data, 0, delta,
nb_channels, buf->format);
- for (ch = 0; ch < nb_channels; ch++)
- buf_out->extended_data[ch] += delta;
+ for (ch = 0; ch < planes; ch++)
+ buf_out->extended_data[ch] += delta * block_size;
avresample_read(s->avr, buf_out->extended_data, out_size);
- for (ch = 0; ch < nb_channels; ch++)
- buf_out->extended_data[ch] -= delta;
+ for (ch = 0; ch < planes; ch++)
+ buf_out->extended_data[ch] -= delta * block_size;
} else {
avresample_read(s->avr, buf_out->extended_data, out_size);
/* drain any remaining buffered data */
avresample_read(s->avr, NULL, avresample_available(s->avr));
- s->pts = pts - avresample_get_delay(s->avr);
- ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
- buf->linesize[0], buf->audio->nb_samples);
+ new_pts = pts - avresample_get_delay(s->avr);
+ /* check for s->pts monotonicity */
+ if (new_pts > s->pts) {
+ s->pts = new_pts;
+ ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
+ buf->linesize[0], buf->nb_samples);
+ } else {
+ av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
+ "whole buffer.\n");
+ ret = 0;
+ }
s->first_frame = 0;
fail:
- avfilter_unref_buffer(buf);
+ av_frame_free(&buf);
return ret;
}
{ NULL }
};
-AVFilter avfilter_af_asyncts = {
+AVFilter ff_af_asyncts = {
.name = "asyncts",
.description = NULL_IF_CONFIG_SMALL("Sync audio data to timestamps"),
.uninit = uninit,
.priv_size = sizeof(ASyncContext),
+ .priv_class = &async_class,
.inputs = avfilter_af_asyncts_inputs,
.outputs = avfilter_af_asyncts_outputs,