yading@10: /*
yading@10:  * This file is part of Libav.
yading@10:  *
yading@10:  * Libav is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * Libav is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with Libav; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #include "libavresample/avresample.h"
yading@10: #include "libavutil/audio_fifo.h"
yading@10: #include "libavutil/common.h"
yading@10: #include "libavutil/mathematics.h"
yading@10: #include "libavutil/opt.h"
yading@10: #include "libavutil/samplefmt.h"
yading@10: 
yading@10: #include "audio.h"
yading@10: #include "avfilter.h"
yading@10: #include "internal.h"
yading@10: 
yading@10: typedef struct ASyncContext {
yading@10:     const AVClass *class;
yading@10: 
yading@10:     AVAudioResampleContext *avr;
yading@10:     int64_t pts;            ///< timestamp in samples of the first sample in fifo
yading@10:     int min_delta;          ///< pad/trim min threshold in samples
yading@10:     int first_frame;        ///< 1 until filter_frame() has processed at least 1 frame with a pts != AV_NOPTS_VALUE
yading@10:     int64_t first_pts;      ///< user-specified first expected pts, in samples
yading@10:     int comp;               ///< current resample compensation
yading@10: 
yading@10:     /* options */
yading@10:     int resample;
yading@10:     float min_delta_sec;
yading@10:     int max_comp;
yading@10: 
yading@10:     /* set by filter_frame() to signal an output frame to request_frame() */
yading@10:     int got_output;
yading@10: } ASyncContext;
yading@10: 
yading@10: #define OFFSET(x) offsetof(ASyncContext, x)
yading@10: #define A AV_OPT_FLAG_AUDIO_PARAM
yading@10: #define F AV_OPT_FLAG_FILTERING_PARAM
yading@10: static const AVOption asyncts_options[] = {
yading@10:     { "compensate", "Stretch/squeeze the data to make it match the timestamps", OFFSET(resample),      AV_OPT_TYPE_INT,   { .i64 = 0 },   0, 1,       A|F },
yading@10:     { "min_delta",  "Minimum difference between timestamps and audio data "
yading@10:                     "(in seconds) to trigger padding/trimmin the data.",        OFFSET(min_delta_sec), AV_OPT_TYPE_FLOAT, { .dbl = 0.1 }, 0, INT_MAX, A|F },
yading@10:     { "max_comp",   "Maximum compensation in samples per second.",              OFFSET(max_comp),      AV_OPT_TYPE_INT,   { .i64 = 500 }, 0, INT_MAX, A|F },
yading@10:     { "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|F },
yading@10:     { NULL },
yading@10: };
yading@10: 
yading@10: AVFILTER_DEFINE_CLASS(asyncts);
yading@10: 
yading@10: static int init(AVFilterContext *ctx)
yading@10: {
yading@10:     ASyncContext *s = ctx->priv;
yading@10: 
yading@10:     s->pts         = AV_NOPTS_VALUE;
yading@10:     s->first_frame = 1;
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static void uninit(AVFilterContext *ctx)
yading@10: {
yading@10:     ASyncContext *s = ctx->priv;
yading@10: 
yading@10:     if (s->avr) {
yading@10:         avresample_close(s->avr);
yading@10:         avresample_free(&s->avr);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static int config_props(AVFilterLink *link)
yading@10: {
yading@10:     ASyncContext *s = link->src->priv;
yading@10:     int ret;
yading@10: 
yading@10:     s->min_delta = s->min_delta_sec * link->sample_rate;
yading@10:     link->time_base = (AVRational){1, link->sample_rate};
yading@10: 
yading@10:     s->avr = avresample_alloc_context();
yading@10:     if (!s->avr)
yading@10:         return AVERROR(ENOMEM);
yading@10: 
yading@10:     av_opt_set_int(s->avr,  "in_channel_layout", link->channel_layout, 0);
yading@10:     av_opt_set_int(s->avr, "out_channel_layout", link->channel_layout, 0);
yading@10:     av_opt_set_int(s->avr,  "in_sample_fmt",     link->format,         0);
yading@10:     av_opt_set_int(s->avr, "out_sample_fmt",     link->format,         0);
yading@10:     av_opt_set_int(s->avr,  "in_sample_rate",    link->sample_rate,    0);
yading@10:     av_opt_set_int(s->avr, "out_sample_rate",    link->sample_rate,    0);
yading@10: 
yading@10:     if (s->resample)
yading@10:         av_opt_set_int(s->avr, "force_resampling", 1, 0);
yading@10: 
yading@10:     if ((ret = avresample_open(s->avr)) < 0)
yading@10:         return ret;
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: /* get amount of data currently buffered, in samples */
yading@10: static int64_t get_delay(ASyncContext *s)
yading@10: {
yading@10:     return avresample_available(s->avr) + avresample_get_delay(s->avr);
yading@10: }
yading@10: 
yading@10: static void handle_trimming(AVFilterContext *ctx)
yading@10: {
yading@10:     ASyncContext *s = ctx->priv;
yading@10: 
yading@10:     if (s->pts < s->first_pts) {
yading@10:         int delta = FFMIN(s->first_pts - s->pts, avresample_available(s->avr));
yading@10:         av_log(ctx, AV_LOG_VERBOSE, "Trimming %d samples from start\n",
yading@10:                delta);
yading@10:         avresample_read(s->avr, NULL, delta);
yading@10:         s->pts += delta;
yading@10:     } else if (s->first_frame)
yading@10:         s->pts = s->first_pts;
yading@10: }
yading@10: 
yading@10: static int request_frame(AVFilterLink *link)
yading@10: {
yading@10:     AVFilterContext *ctx = link->src;
yading@10:     ASyncContext      *s = ctx->priv;
yading@10:     int ret = 0;
yading@10:     int nb_samples;
yading@10: 
yading@10:     s->got_output = 0;
yading@10:     while (ret >= 0 && !s->got_output)
yading@10:         ret = ff_request_frame(ctx->inputs[0]);
yading@10: 
yading@10:     /* flush the fifo */
yading@10:     if (ret == AVERROR_EOF) {
yading@10:         if (s->first_pts != AV_NOPTS_VALUE)
yading@10:             handle_trimming(ctx);
yading@10: 
yading@10:         if (nb_samples = get_delay(s)) {
yading@10:             AVFrame *buf = ff_get_audio_buffer(link, nb_samples);
yading@10:             if (!buf)
yading@10:                 return AVERROR(ENOMEM);
yading@10:             ret = avresample_convert(s->avr, buf->extended_data,
yading@10:                                      buf->linesize[0], nb_samples, NULL, 0, 0);
yading@10:             if (ret <= 0) {
yading@10:                 av_frame_free(&buf);
yading@10:                 return (ret < 0) ? ret : AVERROR_EOF;
yading@10:             }
yading@10: 
yading@10:             buf->pts = s->pts;
yading@10:             return ff_filter_frame(link, buf);
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     return ret;
yading@10: }
yading@10: 
yading@10: static int write_to_fifo(ASyncContext *s, AVFrame *buf)
yading@10: {
yading@10:     int ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
yading@10:                                  buf->linesize[0], buf->nb_samples);
yading@10:     av_frame_free(&buf);
yading@10:     return ret;
yading@10: }
yading@10: 
yading@10: static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
yading@10: {
yading@10:     AVFilterContext  *ctx = inlink->dst;
yading@10:     ASyncContext       *s = ctx->priv;
yading@10:     AVFilterLink *outlink = ctx->outputs[0];
yading@10:     int nb_channels = av_get_channel_layout_nb_channels(buf->channel_layout);
yading@10:     int64_t pts = (buf->pts == AV_NOPTS_VALUE) ? buf->pts :
yading@10:                   av_rescale_q(buf->pts, inlink->time_base, outlink->time_base);
yading@10:     int out_size, ret;
yading@10:     int64_t delta;
yading@10:     int64_t new_pts;
yading@10: 
yading@10:     /* buffer data until we get the next timestamp */
yading@10:     if (s->pts == AV_NOPTS_VALUE || pts == AV_NOPTS_VALUE) {
yading@10:         if (pts != AV_NOPTS_VALUE) {
yading@10:             s->pts = pts - get_delay(s);
yading@10:         }
yading@10:         return write_to_fifo(s, buf);
yading@10:     }
yading@10: 
yading@10:     if (s->first_pts != AV_NOPTS_VALUE) {
yading@10:         handle_trimming(ctx);
yading@10:         if (!avresample_available(s->avr))
yading@10:             return write_to_fifo(s, buf);
yading@10:     }
yading@10: 
yading@10:     /* when we have two timestamps, compute how many samples would we have
yading@10:      * to add/remove to get proper sync between data and timestamps */
yading@10:     delta    = pts - s->pts - get_delay(s);
yading@10:     out_size = avresample_available(s->avr);
yading@10: 
yading@10:     if (labs(delta) > s->min_delta ||
yading@10:         (s->first_frame && delta && s->first_pts != AV_NOPTS_VALUE)) {
yading@10:         av_log(ctx, AV_LOG_VERBOSE, "Discontinuity - %"PRId64" samples.\n", delta);
yading@10:         out_size = av_clipl_int32((int64_t)out_size + delta);
yading@10:     } else {
yading@10:         if (s->resample) {
yading@10:             // adjust the compensation if delta is non-zero
yading@10:             int delay = get_delay(s);
yading@10:             int comp = s->comp + av_clip(delta * inlink->sample_rate / delay,
yading@10:                                          -s->max_comp, s->max_comp);
yading@10:             if (comp != s->comp) {
yading@10:                 av_log(ctx, AV_LOG_VERBOSE, "Compensating %d samples per second.\n", comp);
yading@10:                 if (avresample_set_compensation(s->avr, comp, inlink->sample_rate) == 0) {
yading@10:                     s->comp = comp;
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:         // adjust PTS to avoid monotonicity errors with input PTS jitter
yading@10:         pts -= delta;
yading@10:         delta = 0;
yading@10:     }
yading@10: 
yading@10:     if (out_size > 0) {
yading@10:         AVFrame *buf_out = ff_get_audio_buffer(outlink, out_size);
yading@10:         if (!buf_out) {
yading@10:             ret = AVERROR(ENOMEM);
yading@10:             goto fail;
yading@10:         }
yading@10: 
yading@10:         if (s->first_frame && delta > 0) {
yading@10:             int ch;
yading@10: 
yading@10:             av_samples_set_silence(buf_out->extended_data, 0, delta,
yading@10:                                    nb_channels, buf->format);
yading@10: 
yading@10:             for (ch = 0; ch < nb_channels; ch++)
yading@10:                 buf_out->extended_data[ch] += delta;
yading@10: 
yading@10:             avresample_read(s->avr, buf_out->extended_data, out_size);
yading@10: 
yading@10:             for (ch = 0; ch < nb_channels; ch++)
yading@10:                 buf_out->extended_data[ch] -= delta;
yading@10:         } else {
yading@10:             avresample_read(s->avr, buf_out->extended_data, out_size);
yading@10: 
yading@10:             if (delta > 0) {
yading@10:                 av_samples_set_silence(buf_out->extended_data, out_size - delta,
yading@10:                                        delta, nb_channels, buf->format);
yading@10:             }
yading@10:         }
yading@10:         buf_out->pts = s->pts;
yading@10:         ret = ff_filter_frame(outlink, buf_out);
yading@10:         if (ret < 0)
yading@10:             goto fail;
yading@10:         s->got_output = 1;
yading@10:     } else if (avresample_available(s->avr)) {
yading@10:         av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
yading@10:                "whole buffer.\n");
yading@10:     }
yading@10: 
yading@10:     /* drain any remaining buffered data */
yading@10:     avresample_read(s->avr, NULL, avresample_available(s->avr));
yading@10: 
yading@10:     new_pts = pts - avresample_get_delay(s->avr);
yading@10:     /* check for s->pts monotonicity */
yading@10:     if (new_pts > s->pts) {
yading@10:         s->pts = new_pts;
yading@10:         ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
yading@10:                                  buf->linesize[0], buf->nb_samples);
yading@10:     } else {
yading@10:         av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
yading@10:                "whole buffer.\n");
yading@10:         ret = 0;
yading@10:     }
yading@10: 
yading@10:     s->first_frame = 0;
yading@10: fail:
yading@10:     av_frame_free(&buf);
yading@10: 
yading@10:     return ret;
yading@10: }
yading@10: 
yading@10: static const AVFilterPad avfilter_af_asyncts_inputs[] = {
yading@10:     {
yading@10:         .name           = "default",
yading@10:         .type           = AVMEDIA_TYPE_AUDIO,
yading@10:         .filter_frame   = filter_frame
yading@10:     },
yading@10:     { NULL }
yading@10: };
yading@10: 
yading@10: static const AVFilterPad avfilter_af_asyncts_outputs[] = {
yading@10:     {
yading@10:         .name          = "default",
yading@10:         .type          = AVMEDIA_TYPE_AUDIO,
yading@10:         .config_props  = config_props,
yading@10:         .request_frame = request_frame
yading@10:     },
yading@10:     { NULL }
yading@10: };
yading@10: 
yading@10: AVFilter avfilter_af_asyncts = {
yading@10:     .name        = "asyncts",
yading@10:     .description = NULL_IF_CONFIG_SMALL("Sync audio data to timestamps"),
yading@10: 
yading@10:     .init        = init,
yading@10:     .uninit      = uninit,
yading@10: 
yading@10:     .priv_size   = sizeof(ASyncContext),
yading@10:     .priv_class  = &asyncts_class,
yading@10: 
yading@10:     .inputs      = avfilter_af_asyncts_inputs,
yading@10:     .outputs     = avfilter_af_asyncts_outputs,
yading@10: };