* 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"
#include "libavutil/opt.h"
#include "libavutil/samplefmt.h"
AVAudioResampleContext *avr;
int64_t pts; ///< timestamp in samples of the first sample in fifo
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;
float min_delta_sec;
int max_comp;
+
+ /* set by filter_frame() to signal an output frame to request_frame() */
+ int got_output;
} ASyncContext;
#define OFFSET(x) offsetof(ASyncContext, x)
#define A AV_OPT_FLAG_AUDIO_PARAM
static const AVOption options[] = {
- { "compensate", "Stretch/squeeze the data to make it match the timestamps", OFFSET(resample), AV_OPT_TYPE_INT, { 0 }, 0, 1, A },
+ { "compensate", "Stretch/squeeze the data to make it match the timestamps", OFFSET(resample), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, A },
{ "min_delta", "Minimum difference between timestamps and audio data "
- "(in seconds) to trigger padding/trimmin the data.", OFFSET(min_delta_sec), AV_OPT_TYPE_FLOAT, { 0.1 }, 0, INT_MAX, A },
- { "max_comp", "Maximum compensation in samples per second.", OFFSET(max_comp), AV_OPT_TYPE_INT, { 500 }, 0, INT_MAX, A },
+ "(in seconds) to trigger padding/trimmin the data.", OFFSET(min_delta_sec), AV_OPT_TYPE_FLOAT, { .dbl = 0.1 }, 0, INT_MAX, A },
+ { "max_comp", "Maximum compensation in samples per second.", OFFSET(max_comp), AV_OPT_TYPE_INT, { .i64 = 500 }, 0, INT_MAX, A },
+ { "first_pts", "Assume the first pts should be this value.", OFFSET(first_pts), AV_OPT_TYPE_INT64, { .i64 = AV_NOPTS_VALUE }, INT64_MIN, INT64_MAX, A },
{ NULL },
};
.version = LIBAVUTIL_VERSION_INT,
};
-static int init(AVFilterContext *ctx, const char *args, void *opaque)
+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->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;
return 0;
}
+/* get amount of data currently buffered, in samples */
+static int64_t get_delay(ASyncContext *s)
+{
+ return avresample_available(s->avr) + avresample_get_delay(s->avr);
+}
+
+static void handle_trimming(AVFilterContext *ctx)
+{
+ ASyncContext *s = ctx->priv;
+
+ if (s->pts < s->first_pts) {
+ int delta = FFMIN(s->first_pts - s->pts, avresample_available(s->avr));
+ av_log(ctx, AV_LOG_VERBOSE, "Trimming %d samples from start\n",
+ delta);
+ avresample_read(s->avr, NULL, delta);
+ s->pts += delta;
+ } else if (s->first_frame)
+ s->pts = s->first_pts;
+}
+
static int request_frame(AVFilterLink *link)
{
AVFilterContext *ctx = link->src;
ASyncContext *s = ctx->priv;
- int ret = ff_request_frame(ctx->inputs[0]);
+ int ret = 0;
int nb_samples;
+ s->got_output = 0;
+ while (ret >= 0 && !s->got_output)
+ ret = ff_request_frame(ctx->inputs[0]);
+
/* flush the fifo */
- if (ret == AVERROR_EOF && (nb_samples = avresample_get_delay(s->avr))) {
- AVFilterBufferRef *buf = ff_get_audio_buffer(link, AV_PERM_WRITE,
- nb_samples);
- if (!buf)
- return AVERROR(ENOMEM);
- avresample_convert(s->avr, (void**)buf->extended_data, buf->linesize[0],
- nb_samples, NULL, 0, 0);
- buf->pts = s->pts;
- ff_filter_samples(link, buf);
- return 0;
+ if (ret == AVERROR_EOF) {
+ if (s->first_pts != AV_NOPTS_VALUE)
+ handle_trimming(ctx);
+
+ if (nb_samples = get_delay(s)) {
+ 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) {
+ av_frame_free(&buf);
+ return (ret < 0) ? ret : AVERROR_EOF;
+ }
+
+ buf->pts = s->pts;
+ return ff_filter_frame(link, buf);
+ }
}
return ret;
}
-static void write_to_fifo(ASyncContext *s, AVFilterBufferRef *buf)
+static int write_to_fifo(ASyncContext *s, AVFrame *buf)
{
- avresample_convert(s->avr, NULL, 0, 0, (void**)buf->extended_data,
- buf->linesize[0], buf->audio->nb_samples);
- avfilter_unref_buffer(buf);
-}
-
-/* get amount of data currently buffered, in samples */
-static int64_t get_delay(ASyncContext *s)
-{
- return avresample_available(s->avr) + avresample_get_delay(s->avr);
+ int ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
+ buf->linesize[0], buf->nb_samples);
+ av_frame_free(&buf);
+ return ret;
}
-static void filter_samples(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;
+ 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);
}
- write_to_fifo(s, buf);
- return;
+ return write_to_fifo(s, buf);
}
- /* now wait for the next timestamp */
- if (pts == AV_NOPTS_VALUE) {
- write_to_fifo(s, buf);
- return;
+ if (s->first_pts != AV_NOPTS_VALUE) {
+ handle_trimming(ctx);
+ if (!avresample_available(s->avr))
+ return write_to_fifo(s, buf);
}
/* when we have two timestamps, compute how many samples would we have
delta = pts - s->pts - get_delay(s);
out_size = avresample_available(s->avr);
- if (labs(delta) > s->min_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 += 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, delta, 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);
- if (!buf_out)
- return;
+ AVFrame *buf_out = ff_get_audio_buffer(outlink, out_size);
+ if (!buf_out) {
+ ret = AVERROR(ENOMEM);
+ goto fail;
+ }
- avresample_read(s->avr, (void**)buf_out->extended_data, out_size);
- buf_out->pts = s->pts;
+ 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;
- if (delta > 0) {
- av_samples_set_silence(buf_out->extended_data, out_size - delta,
- delta, nb_channels, buf->format);
+ av_samples_set_silence(buf_out->extended_data, 0, delta,
+ nb_channels, buf->format);
+
+ 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 < planes; ch++)
+ buf_out->extended_data[ch] -= delta * block_size;
+ } else {
+ avresample_read(s->avr, buf_out->extended_data, out_size);
+
+ if (delta > 0) {
+ av_samples_set_silence(buf_out->extended_data, out_size - delta,
+ delta, nb_channels, buf->format);
+ }
}
- ff_filter_samples(outlink, buf_out);
- } else {
+ buf_out->pts = s->pts;
+ ret = ff_filter_frame(outlink, buf_out);
+ if (ret < 0)
+ goto fail;
+ s->got_output = 1;
+ } else if (avresample_available(s->avr)) {
av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
"whole buffer.\n");
}
/* drain any remaining buffered data */
avresample_read(s->avr, NULL, avresample_available(s->avr));
- s->pts = pts - avresample_get_delay(s->avr);
- avresample_convert(s->avr, NULL, 0, 0, (void**)buf->extended_data,
- buf->linesize[0], buf->audio->nb_samples);
- avfilter_unref_buffer(buf);
+ 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:
+ av_frame_free(&buf);
+
+ return ret;
}
-AVFilter avfilter_af_asyncts = {
+static const AVFilterPad avfilter_af_asyncts_inputs[] = {
+ {
+ .name = "default",
+ .type = AVMEDIA_TYPE_AUDIO,
+ .filter_frame = filter_frame,
+ },
+ { NULL }
+};
+
+static const AVFilterPad avfilter_af_asyncts_outputs[] = {
+ {
+ .name = "default",
+ .type = AVMEDIA_TYPE_AUDIO,
+ .config_props = config_props,
+ .request_frame = request_frame
+ },
+ { NULL }
+};
+
+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 = (const AVFilterPad[]) {{ .name = "default",
- .type = AVMEDIA_TYPE_AUDIO,
- .filter_samples = filter_samples },
- { NULL }},
- .outputs = (const AVFilterPad[]) {{ .name = "default",
- .type = AVMEDIA_TYPE_AUDIO,
- .config_props = config_props,
- .request_frame = request_frame },
- { NULL }},
+ .inputs = avfilter_af_asyncts_inputs,
+ .outputs = avfilter_af_asyncts_outputs,
};
projects / ffmpeg / blobdiff