libavfilter/af_asyncts.c

   1 /*
   2  * This file is part of Libav.
   3  *
   4  * Libav is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU Lesser General Public
   6  * License as published by the Free Software Foundation; either
   7  * version 2.1 of the License, or (at your option) any later version.
   8  *
   9  * Libav is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  12  * Lesser General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU Lesser General Public
  15  * License along with Libav; if not, write to the Free Software
  16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  17  */
  18
  19 #include "libavresample/avresample.h"
  20 #include "libavutil/audio_fifo.h"
  21 #include "libavutil/common.h"
  22 #include "libavutil/mathematics.h"
  23 #include "libavutil/opt.h"
  24 #include "libavutil/samplefmt.h"
  25
  26 #include "audio.h"
  27 #include "avfilter.h"
  28 #include "internal.h"
  29
  30 typedef struct ASyncContext {
  31     const AVClass *class;
  32
  33     AVAudioResampleContext *avr;
  34     int64_t pts;            ///< timestamp in samples of the first sample in fifo
  35     int min_delta;          ///< pad/trim min threshold in samples
  36     int first_frame;        ///< 1 until filter_frame() has processed at least 1 frame with a pts != AV_NOPTS_VALUE
  37     int64_t first_pts;      ///< user-specified first expected pts, in samples
  38
  39     /* options */
  40     int resample;
  41     float min_delta_sec;
  42     int max_comp;
  43
  44     /* set by filter_frame() to signal an output frame to request_frame() */
  45     int got_output;
  46 } ASyncContext;
  47
  48 #define OFFSET(x) offsetof(ASyncContext, x)
  49 #define A AV_OPT_FLAG_AUDIO_PARAM
  50 static const AVOption options[] = {
  51     { "compensate", "Stretch/squeeze the data to make it match the timestamps", OFFSET(resample),      AV_OPT_TYPE_INT,   { .i64 = 0 },   0, 1,       A },
  52     { "min_delta",  "Minimum difference between timestamps and audio data "
  53                     "(in seconds) to trigger padding/trimmin the data.",        OFFSET(min_delta_sec), AV_OPT_TYPE_FLOAT, { .dbl = 0.1 }, 0, INT_MAX, A },
  54     { "max_comp",   "Maximum compensation in samples per second.",              OFFSET(max_comp),      AV_OPT_TYPE_INT,   { .i64 = 500 }, 0, INT_MAX, A },
  55     { "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 },
  56     { NULL },
  57 };
  58
  59 static const AVClass async_class = {
  60     .class_name = "asyncts filter",
  61     .item_name  = av_default_item_name,
  62     .option     = options,
  63     .version    = LIBAVUTIL_VERSION_INT,
  64 };
  65
  66 static int init(AVFilterContext *ctx, const char *args)
  67 {
  68     ASyncContext *s = ctx->priv;
  69     int ret;
  70
  71     s->class = &async_class;
  72     av_opt_set_defaults(s);
  73
  74     if ((ret = av_set_options_string(s, args, "=", ":")) < 0) {
  75         av_log(ctx, AV_LOG_ERROR, "Error parsing options string '%s'.\n", args);
  76         return ret;
  77     }
  78     av_opt_free(s);
  79
  80     s->pts         = AV_NOPTS_VALUE;
  81     s->first_frame = 1;
  82
  83     return 0;
  84 }
  85
  86 static void uninit(AVFilterContext *ctx)
  87 {
  88     ASyncContext *s = ctx->priv;
  89
  90     if (s->avr) {
  91         avresample_close(s->avr);
  92         avresample_free(&s->avr);
  93     }
  94 }
  95
  96 static int config_props(AVFilterLink *link)
  97 {
  98     ASyncContext *s = link->src->priv;
  99     int ret;
 100
 101     s->min_delta = s->min_delta_sec * link->sample_rate;
 102     link->time_base = (AVRational){1, link->sample_rate};
 103
 104     s->avr = avresample_alloc_context();
 105     if (!s->avr)
 106         return AVERROR(ENOMEM);
 107
 108     av_opt_set_int(s->avr,  "in_channel_layout", link->channel_layout, 0);
 109     av_opt_set_int(s->avr, "out_channel_layout", link->channel_layout, 0);
 110     av_opt_set_int(s->avr,  "in_sample_fmt",     link->format,         0);
 111     av_opt_set_int(s->avr, "out_sample_fmt",     link->format,         0);
 112     av_opt_set_int(s->avr,  "in_sample_rate",    link->sample_rate,    0);
 113     av_opt_set_int(s->avr, "out_sample_rate",    link->sample_rate,    0);
 114
 115     if (s->resample)
 116         av_opt_set_int(s->avr, "force_resampling", 1, 0);
 117
 118     if ((ret = avresample_open(s->avr)) < 0)
 119         return ret;
 120
 121     return 0;
 122 }
 123
 124 /* get amount of data currently buffered, in samples */
 125 static int64_t get_delay(ASyncContext *s)
 126 {
 127     return avresample_available(s->avr) + avresample_get_delay(s->avr);
 128 }
 129
 130 static void handle_trimming(AVFilterContext *ctx)
 131 {
 132     ASyncContext *s = ctx->priv;
 133
 134     if (s->pts < s->first_pts) {
 135         int delta = FFMIN(s->first_pts - s->pts, avresample_available(s->avr));
 136         av_log(ctx, AV_LOG_VERBOSE, "Trimming %d samples from start\n",
 137                delta);
 138         avresample_read(s->avr, NULL, delta);
 139         s->pts += delta;
 140     } else if (s->first_frame)
 141         s->pts = s->first_pts;
 142 }
 143
 144 static int request_frame(AVFilterLink *link)
 145 {
 146     AVFilterContext *ctx = link->src;
 147     ASyncContext      *s = ctx->priv;
 148     int ret = 0;
 149     int nb_samples;
 150
 151     s->got_output = 0;
 152     while (ret >= 0 && !s->got_output)
 153         ret = ff_request_frame(ctx->inputs[0]);
 154
 155     /* flush the fifo */
 156     if (ret == AVERROR_EOF) {
 157         if (s->first_pts != AV_NOPTS_VALUE)
 158             handle_trimming(ctx);
 159
 160         if (nb_samples = get_delay(s)) {
 161             AVFilterBufferRef *buf = ff_get_audio_buffer(link, AV_PERM_WRITE,
 162                                                          nb_samples);
 163             if (!buf)
 164                 return AVERROR(ENOMEM);
 165             ret = avresample_convert(s->avr, buf->extended_data,
 166                                      buf->linesize[0], nb_samples, NULL, 0, 0);
 167             if (ret <= 0) {
 168                 avfilter_unref_bufferp(&buf);
 169                 return (ret < 0) ? ret : AVERROR_EOF;
 170             }
 171
 172             buf->pts = s->pts;
 173             return ff_filter_frame(link, buf);
 174         }
 175     }
 176
 177     return ret;
 178 }
 179
 180 static int write_to_fifo(ASyncContext *s, AVFilterBufferRef *buf)
 181 {
 182     int ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
 183                                  buf->linesize[0], buf->audio->nb_samples);
 184     avfilter_unref_buffer(buf);
 185     return ret;
 186 }
 187
 188 static int filter_frame(AVFilterLink *inlink, AVFilterBufferRef *buf)
 189 {
 190     AVFilterContext  *ctx = inlink->dst;
 191     ASyncContext       *s = ctx->priv;
 192     AVFilterLink *outlink = ctx->outputs[0];
 193     int nb_channels = av_get_channel_layout_nb_channels(buf->audio->channel_layout);
 194     int64_t pts = (buf->pts == AV_NOPTS_VALUE) ? buf->pts :
 195                   av_rescale_q(buf->pts, inlink->time_base, outlink->time_base);
 196     int out_size, ret;
 197     int64_t delta;
 198
 199     /* buffer data until we get the next timestamp */
 200     if (s->pts == AV_NOPTS_VALUE || pts == AV_NOPTS_VALUE) {
 201         if (pts != AV_NOPTS_VALUE) {
 202             s->pts = pts - get_delay(s);
 203         }
 204         return write_to_fifo(s, buf);
 205     }
 206
 207     if (s->first_pts != AV_NOPTS_VALUE) {
 208         handle_trimming(ctx);
 209         if (!avresample_available(s->avr))
 210             return write_to_fifo(s, buf);
 211     }
 212
 213     /* when we have two timestamps, compute how many samples would we have
 214      * to add/remove to get proper sync between data and timestamps */
 215     delta    = pts - s->pts - get_delay(s);
 216     out_size = avresample_available(s->avr);
 217
 218     if (labs(delta) > s->min_delta ||
 219         (s->first_frame && delta && s->first_pts != AV_NOPTS_VALUE)) {
 220         av_log(ctx, AV_LOG_VERBOSE, "Discontinuity - %"PRId64" samples.\n", delta);
 221         out_size = av_clipl_int32((int64_t)out_size + delta);
 222     } else {
 223         if (s->resample) {
 224             int comp = av_clip(delta, -s->max_comp, s->max_comp);
 225             av_log(ctx, AV_LOG_VERBOSE, "Compensating %d samples per second.\n", comp);
 226             avresample_set_compensation(s->avr, comp, inlink->sample_rate);
 227         }
 228         delta = 0;
 229     }
 230
 231     if (out_size > 0) {
 232         AVFilterBufferRef *buf_out = ff_get_audio_buffer(outlink, AV_PERM_WRITE,
 233                                                          out_size);
 234         if (!buf_out) {
 235             ret = AVERROR(ENOMEM);
 236             goto fail;
 237         }
 238
 239         if (s->first_frame && delta > 0) {
 240             int ch;
 241
 242             av_samples_set_silence(buf_out->extended_data, 0, delta,
 243                                    nb_channels, buf->format);
 244
 245             for (ch = 0; ch < nb_channels; ch++)
 246                 buf_out->extended_data[ch] += delta;
 247
 248             avresample_read(s->avr, buf_out->extended_data, out_size);
 249
 250             for (ch = 0; ch < nb_channels; ch++)
 251                 buf_out->extended_data[ch] -= delta;
 252         } else {
 253             avresample_read(s->avr, buf_out->extended_data, out_size);
 254
 255             if (delta > 0) {
 256                 av_samples_set_silence(buf_out->extended_data, out_size - delta,
 257                                        delta, nb_channels, buf->format);
 258             }
 259         }
 260         buf_out->pts = s->pts;
 261         ret = ff_filter_frame(outlink, buf_out);
 262         if (ret < 0)
 263             goto fail;
 264         s->got_output = 1;
 265     } else if (avresample_available(s->avr)) {
 266         av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
 267                "whole buffer.\n");
 268     }
 269
 270     /* drain any remaining buffered data */
 271     avresample_read(s->avr, NULL, avresample_available(s->avr));
 272
 273     s->pts = pts - avresample_get_delay(s->avr);
 274     ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
 275                              buf->linesize[0], buf->audio->nb_samples);
 276
 277     s->first_frame = 0;
 278 fail:
 279     avfilter_unref_buffer(buf);
 280
 281     return ret;
 282 }
 283
 284 static const AVFilterPad avfilter_af_asyncts_inputs[] = {
 285     {
 286         .name           = "default",
 287         .type           = AVMEDIA_TYPE_AUDIO,
 288         .filter_frame   = filter_frame,
 289     },
 290     { NULL }
 291 };
 292
 293 static const AVFilterPad avfilter_af_asyncts_outputs[] = {
 294     {
 295         .name          = "default",
 296         .type          = AVMEDIA_TYPE_AUDIO,
 297         .config_props  = config_props,
 298         .request_frame = request_frame
 299     },
 300     { NULL }
 301 };
 302
 303 AVFilter avfilter_af_asyncts = {
 304     .name        = "asyncts",
 305     .description = NULL_IF_CONFIG_SMALL("Sync audio data to timestamps"),
 306
 307     .init        = init,
 308     .uninit      = uninit,
 309
 310     .priv_size   = sizeof(ASyncContext),
 311
 312     .inputs      = avfilter_af_asyncts_inputs,
 313     .outputs     = avfilter_af_asyncts_outputs,
 314 };