libavcodec/opusdec.c

   1 /*
   2  * Opus decoder
   3  * Copyright (c) 2012 Andrew D'Addesio
   4  * Copyright (c) 2013-2014 Mozilla Corporation
   5  *
   6  * This file is part of Libav.
   7  *
   8  * Libav is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * Libav is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with Libav; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 /**
  24  * @file
  25  * Opus decoder
  26  * @author Andrew D'Addesio, Anton Khirnov
  27  *
  28  * Codec homepage: http://opus-codec.org/
  29  * Specification: http://tools.ietf.org/html/rfc6716
  30  * Ogg Opus specification: https://tools.ietf.org/html/draft-ietf-codec-oggopus-03
  31  *
  32  * Ogg-contained .opus files can be produced with opus-tools:
  33  * http://git.xiph.org/?p=opus-tools.git
  34  */
  35
  36 #include <stdint.h>
  37
  38 #include "libavutil/attributes.h"
  39 #include "libavutil/audio_fifo.h"
  40 #include "libavutil/channel_layout.h"
  41 #include "libavutil/opt.h"
  42
  43 #include "libavresample/avresample.h"
  44
  45 #include "avcodec.h"
  46 #include "bitstream.h"
  47 #include "celp_filters.h"
  48 #include "fft.h"
  49 #include "internal.h"
  50 #include "mathops.h"
  51 #include "opus.h"
  52
  53 static const uint16_t silk_frame_duration_ms[16] = {
  54     10, 20, 40, 60,
  55     10, 20, 40, 60,
  56     10, 20, 40, 60,
  57     10, 20,
  58     10, 20,
  59 };
  60
  61 /* number of samples of silence to feed to the resampler
  62  * at the beginning */
  63 static const int silk_resample_delay[] = {
  64     4, 8, 11, 11, 11
  65 };
  66
  67 static const uint8_t celt_band_end[] = { 13, 17, 17, 19, 21 };
  68
  69 static int get_silk_samplerate(int config)
  70 {
  71     if (config < 4)
  72         return 8000;
  73     else if (config < 8)
  74         return 12000;
  75     return 16000;
  76 }
  77
  78 /**
  79  * Range decoder
  80  */
  81 static int opus_rc_init(OpusRangeCoder *rc, const uint8_t *data, int size)
  82 {
  83     int ret = bitstream_init8(&rc->bc, data, size);
  84     if (ret < 0)
  85         return ret;
  86
  87     rc->range = 128;
  88     rc->value = 127 - bitstream_read(&rc->bc, 7);
  89     rc->total_read_bits = 9;
  90     opus_rc_normalize(rc);
  91
  92     return 0;
  93 }
  94
  95 static void opus_raw_init(OpusRangeCoder *rc, const uint8_t *rightend,
  96                           unsigned int bytes)
  97 {
  98     rc->rb.position = rightend;
  99     rc->rb.bytes    = bytes;
 100     rc->rb.cachelen = 0;
 101     rc->rb.cacheval = 0;
 102 }
 103
 104 static void opus_fade(float *out,
 105                       const float *in1, const float *in2,
 106                       const float *window, int len)
 107 {
 108     int i;
 109     for (i = 0; i < len; i++)
 110         out[i] = in2[i] * window[i] + in1[i] * (1.0 - window[i]);
 111 }
 112
 113 static int opus_flush_resample(OpusStreamContext *s, int nb_samples)
 114 {
 115     int celt_size = av_audio_fifo_size(s->celt_delay);
 116     int ret, i;
 117
 118     ret = avresample_convert(s->avr, (uint8_t**)s->out, s->out_size, nb_samples,
 119                              NULL, 0, 0);
 120     if (ret < 0)
 121         return ret;
 122     else if (ret != nb_samples) {
 123         av_log(s->avctx, AV_LOG_ERROR, "Wrong number of flushed samples: %d\n",
 124                ret);
 125         return AVERROR_BUG;
 126     }
 127
 128     if (celt_size) {
 129         if (celt_size != nb_samples) {
 130             av_log(s->avctx, AV_LOG_ERROR, "Wrong number of CELT delay samples.\n");
 131             return AVERROR_BUG;
 132         }
 133         av_audio_fifo_read(s->celt_delay, (void**)s->celt_output, nb_samples);
 134         for (i = 0; i < s->output_channels; i++) {
 135             s->fdsp->vector_fmac_scalar(s->out[i],
 136                                         s->celt_output[i], 1.0,
 137                                         nb_samples);
 138         }
 139     }
 140
 141     if (s->redundancy_idx) {
 142         for (i = 0; i < s->output_channels; i++)
 143             opus_fade(s->out[i], s->out[i],
 144                       s->redundancy_output[i] + 120 + s->redundancy_idx,
 145                       ff_celt_window2 + s->redundancy_idx, 120 - s->redundancy_idx);
 146         s->redundancy_idx = 0;
 147     }
 148
 149     s->out[0]   += nb_samples;
 150     s->out[1]   += nb_samples;
 151     s->out_size -= nb_samples * sizeof(float);
 152
 153     return 0;
 154 }
 155
 156 static int opus_init_resample(OpusStreamContext *s)
 157 {
 158     float delay[16] = { 0.0 };
 159     uint8_t *delayptr[2] = { (uint8_t*)delay, (uint8_t*)delay };
 160     int ret;
 161
 162     av_opt_set_int(s->avr, "in_sample_rate", s->silk_samplerate, 0);
 163     ret = avresample_open(s->avr);
 164     if (ret < 0) {
 165         av_log(s->avctx, AV_LOG_ERROR, "Error opening the resampler.\n");
 166         return ret;
 167     }
 168
 169     ret = avresample_convert(s->avr, NULL, 0, 0, delayptr, sizeof(delay),
 170                              silk_resample_delay[s->packet.bandwidth]);
 171     if (ret < 0) {
 172         av_log(s->avctx, AV_LOG_ERROR,
 173                "Error feeding initial silence to the resampler.\n");
 174         return ret;
 175     }
 176
 177     return 0;
 178 }
 179
 180 static int opus_decode_redundancy(OpusStreamContext *s, const uint8_t *data, int size)
 181 {
 182     int ret;
 183     enum OpusBandwidth bw = s->packet.bandwidth;
 184
 185     if (s->packet.mode == OPUS_MODE_SILK &&
 186         bw == OPUS_BANDWIDTH_MEDIUMBAND)
 187         bw = OPUS_BANDWIDTH_WIDEBAND;
 188
 189     ret = opus_rc_init(&s->redundancy_rc, data, size);
 190     if (ret < 0)
 191         goto fail;
 192     opus_raw_init(&s->redundancy_rc, data + size, size);
 193
 194     ret = ff_celt_decode_frame(s->celt, &s->redundancy_rc,
 195                                s->redundancy_output,
 196                                s->packet.stereo + 1, 240,
 197                                0, celt_band_end[s->packet.bandwidth]);
 198     if (ret < 0)
 199         goto fail;
 200
 201     return 0;
 202 fail:
 203     av_log(s->avctx, AV_LOG_ERROR, "Error decoding the redundancy frame.\n");
 204     return ret;
 205 }
 206
 207 static int opus_decode_frame(OpusStreamContext *s, const uint8_t *data, int size)
 208 {
 209     int samples    = s->packet.frame_duration;
 210     int redundancy = 0;
 211     int redundancy_size, redundancy_pos;
 212     int ret, i, consumed;
 213     int delayed_samples = s->delayed_samples;
 214
 215     ret = opus_rc_init(&s->rc, data, size);
 216     if (ret < 0)
 217         return ret;
 218
 219     /* decode the silk frame */
 220     if (s->packet.mode == OPUS_MODE_SILK || s->packet.mode == OPUS_MODE_HYBRID) {
 221         if (!avresample_is_open(s->avr)) {
 222             ret = opus_init_resample(s);
 223             if (ret < 0)
 224                 return ret;
 225         }
 226
 227         samples = ff_silk_decode_superframe(s->silk, &s->rc, s->silk_output,
 228                                             FFMIN(s->packet.bandwidth, OPUS_BANDWIDTH_WIDEBAND),
 229                                             s->packet.stereo + 1,
 230                                             silk_frame_duration_ms[s->packet.config]);
 231         if (samples < 0) {
 232             av_log(s->avctx, AV_LOG_ERROR, "Error decoding a SILK frame.\n");
 233             return samples;
 234         }
 235
 236         samples = avresample_convert(s->avr, (uint8_t**)s->out, s->out_size,
 237                                      s->packet.frame_duration,
 238                                      (uint8_t**)s->silk_output,
 239                                      sizeof(s->silk_buf[0]),
 240                                      samples);
 241         if (samples < 0) {
 242             av_log(s->avctx, AV_LOG_ERROR, "Error resampling SILK data.\n");
 243             return samples;
 244         }
 245         s->delayed_samples += s->packet.frame_duration - samples;
 246     } else
 247         ff_silk_flush(s->silk);
 248
 249     // decode redundancy information
 250     consumed = opus_rc_tell(&s->rc);
 251     if (s->packet.mode == OPUS_MODE_HYBRID && consumed + 37 <= size * 8)
 252         redundancy = opus_rc_p2model(&s->rc, 12);
 253     else if (s->packet.mode == OPUS_MODE_SILK && consumed + 17 <= size * 8)
 254         redundancy = 1;
 255
 256     if (redundancy) {
 257         redundancy_pos = opus_rc_p2model(&s->rc, 1);
 258
 259         if (s->packet.mode == OPUS_MODE_HYBRID)
 260             redundancy_size = opus_rc_unimodel(&s->rc, 256) + 2;
 261         else
 262             redundancy_size = size - (consumed + 7) / 8;
 263         size -= redundancy_size;
 264         if (size < 0) {
 265             av_log(s->avctx, AV_LOG_ERROR, "Invalid redundancy frame size.\n");
 266             return AVERROR_INVALIDDATA;
 267         }
 268
 269         if (redundancy_pos) {
 270             ret = opus_decode_redundancy(s, data + size, redundancy_size);
 271             if (ret < 0)
 272                 return ret;
 273             ff_celt_flush(s->celt);
 274         }
 275     }
 276
 277     /* decode the CELT frame */
 278     if (s->packet.mode == OPUS_MODE_CELT || s->packet.mode == OPUS_MODE_HYBRID) {
 279         float *out_tmp[2] = { s->out[0], s->out[1] };
 280         float **dst = (s->packet.mode == OPUS_MODE_CELT) ?
 281                       out_tmp : s->celt_output;
 282         int celt_output_samples = samples;
 283         int delay_samples = av_audio_fifo_size(s->celt_delay);
 284
 285         if (delay_samples) {
 286             if (s->packet.mode == OPUS_MODE_HYBRID) {
 287                 av_audio_fifo_read(s->celt_delay, (void**)s->celt_output, delay_samples);
 288
 289                 for (i = 0; i < s->output_channels; i++) {
 290                     s->fdsp->vector_fmac_scalar(out_tmp[i], s->celt_output[i], 1.0,
 291                                                 delay_samples);
 292                     out_tmp[i] += delay_samples;
 293                 }
 294                 celt_output_samples -= delay_samples;
 295             } else {
 296                 av_log(s->avctx, AV_LOG_WARNING,
 297                        "Spurious CELT delay samples present.\n");
 298                 av_audio_fifo_drain(s->celt_delay, delay_samples);
 299                 if (s->avctx->err_recognition & AV_EF_EXPLODE)
 300                     return AVERROR_BUG;
 301             }
 302         }
 303
 304         opus_raw_init(&s->rc, data + size, size);
 305
 306         ret = ff_celt_decode_frame(s->celt, &s->rc, dst,
 307                                    s->packet.stereo + 1,
 308                                    s->packet.frame_duration,
 309                                    (s->packet.mode == OPUS_MODE_HYBRID) ? 17 : 0,
 310                                    celt_band_end[s->packet.bandwidth]);
 311         if (ret < 0)
 312             return ret;
 313
 314         if (s->packet.mode == OPUS_MODE_HYBRID) {
 315             int celt_delay = s->packet.frame_duration - celt_output_samples;
 316             void *delaybuf[2] = { s->celt_output[0] + celt_output_samples,
 317                                   s->celt_output[1] + celt_output_samples };
 318
 319             for (i = 0; i < s->output_channels; i++) {
 320                 s->fdsp->vector_fmac_scalar(out_tmp[i],
 321                                             s->celt_output[i], 1.0,
 322                                             celt_output_samples);
 323             }
 324
 325             ret = av_audio_fifo_write(s->celt_delay, delaybuf, celt_delay);
 326             if (ret < 0)
 327                 return ret;
 328         }
 329     } else
 330         ff_celt_flush(s->celt);
 331
 332     if (s->redundancy_idx) {
 333         for (i = 0; i < s->output_channels; i++)
 334             opus_fade(s->out[i], s->out[i],
 335                       s->redundancy_output[i] + 120 + s->redundancy_idx,
 336                       ff_celt_window2 + s->redundancy_idx, 120 - s->redundancy_idx);
 337         s->redundancy_idx = 0;
 338     }
 339     if (redundancy) {
 340         if (!redundancy_pos) {
 341             ff_celt_flush(s->celt);
 342             ret = opus_decode_redundancy(s, data + size, redundancy_size);
 343             if (ret < 0)
 344                 return ret;
 345
 346             for (i = 0; i < s->output_channels; i++) {
 347                 opus_fade(s->out[i] + samples - 120 + delayed_samples,
 348                           s->out[i] + samples - 120 + delayed_samples,
 349                           s->redundancy_output[i] + 120,
 350                           ff_celt_window2, 120 - delayed_samples);
 351                 if (delayed_samples)
 352                     s->redundancy_idx = 120 - delayed_samples;
 353             }
 354         } else {
 355             for (i = 0; i < s->output_channels; i++) {
 356                 memcpy(s->out[i] + delayed_samples, s->redundancy_output[i], 120 * sizeof(float));
 357                 opus_fade(s->out[i] + 120 + delayed_samples,
 358                           s->redundancy_output[i] + 120,
 359                           s->out[i] + 120 + delayed_samples,
 360                           ff_celt_window2, 120);
 361             }
 362         }
 363     }
 364
 365     return samples;
 366 }
 367
 368 static int opus_decode_subpacket(OpusStreamContext *s,
 369                                  const uint8_t *buf, int buf_size,
 370                                  float **out, int out_size,
 371                                  int nb_samples)
 372 {
 373     int output_samples = 0;
 374     int flush_needed   = 0;
 375     int i, j, ret;
 376
 377     s->out[0]   = out[0];
 378     s->out[1]   = out[1];
 379     s->out_size = out_size;
 380
 381     /* check if we need to flush the resampler */
 382     if (avresample_is_open(s->avr)) {
 383         if (buf) {
 384             int64_t cur_samplerate;
 385             av_opt_get_int(s->avr, "in_sample_rate", 0, &cur_samplerate);
 386             flush_needed = (s->packet.mode == OPUS_MODE_CELT) || (cur_samplerate != s->silk_samplerate);
 387         } else {
 388             flush_needed = !!s->delayed_samples;
 389         }
 390     }
 391
 392     if (!buf && !flush_needed)
 393         return 0;
 394
 395     /* use dummy output buffers if the channel is not mapped to anything */
 396     if (!s->out[0] ||
 397         (s->output_channels == 2 && !s->out[1])) {
 398         av_fast_malloc(&s->out_dummy, &s->out_dummy_allocated_size, s->out_size);
 399         if (!s->out_dummy)
 400             return AVERROR(ENOMEM);
 401         if (!s->out[0])
 402             s->out[0] = s->out_dummy;
 403         if (!s->out[1])
 404             s->out[1] = s->out_dummy;
 405     }
 406
 407     /* flush the resampler if necessary */
 408     if (flush_needed) {
 409         ret = opus_flush_resample(s, s->delayed_samples);
 410         if (ret < 0) {
 411             av_log(s->avctx, AV_LOG_ERROR, "Error flushing the resampler.\n");
 412             return ret;
 413         }
 414         avresample_close(s->avr);
 415         output_samples += s->delayed_samples;
 416         s->delayed_samples = 0;
 417
 418         if (!buf)
 419             goto finish;
 420     }
 421
 422     /* decode all the frames in the packet */
 423     for (i = 0; i < s->packet.frame_count; i++) {
 424         int size = s->packet.frame_size[i];
 425         int samples = opus_decode_frame(s, buf + s->packet.frame_offset[i], size);
 426
 427         if (samples < 0) {
 428             av_log(s->avctx, AV_LOG_ERROR, "Error decoding an Opus frame.\n");
 429             if (s->avctx->err_recognition & AV_EF_EXPLODE)
 430                 return samples;
 431
 432             for (j = 0; j < s->output_channels; j++)
 433                 memset(s->out[j], 0, s->packet.frame_duration * sizeof(float));
 434             samples = s->packet.frame_duration;
 435         }
 436         output_samples += samples;
 437
 438         for (j = 0; j < s->output_channels; j++)
 439             s->out[j] += samples;
 440         s->out_size -= samples * sizeof(float);
 441     }
 442
 443 finish:
 444     s->out[0] = s->out[1] = NULL;
 445     s->out_size = 0;
 446
 447     return output_samples;
 448 }
 449
 450 static int opus_decode_packet(AVCodecContext *avctx, void *data,
 451                               int *got_frame_ptr, AVPacket *avpkt)
 452 {
 453     OpusContext *c      = avctx->priv_data;
 454     AVFrame *frame      = data;
 455     const uint8_t *buf  = avpkt->data;
 456     int buf_size        = avpkt->size;
 457     int coded_samples   = 0;
 458     int decoded_samples = INT_MAX;
 459     int delayed_samples = 0;
 460     int i, ret;
 461
 462     /* calculate the number of delayed samples */
 463     for (i = 0; i < c->nb_streams; i++) {
 464         delayed_samples = FFMAX(delayed_samples,
 465                                 c->streams[i].delayed_samples + av_audio_fifo_size(c->sync_buffers[i]));
 466     }
 467
 468     /* decode the header of the first sub-packet to find out the sample count */
 469     if (buf) {
 470         OpusPacket *pkt = &c->streams[0].packet;
 471         ret = ff_opus_parse_packet(pkt, buf, buf_size, c->nb_streams > 1);
 472         if (ret < 0) {
 473             av_log(avctx, AV_LOG_ERROR, "Error parsing the packet header.\n");
 474             return ret;
 475         }
 476         coded_samples += pkt->frame_count * pkt->frame_duration;
 477         c->streams[0].silk_samplerate = get_silk_samplerate(pkt->config);
 478     }
 479
 480     frame->nb_samples = coded_samples + delayed_samples;
 481
 482     /* no input or buffered data => nothing to do */
 483     if (!frame->nb_samples) {
 484         *got_frame_ptr = 0;
 485         return 0;
 486     }
 487
 488     /* setup the data buffers */
 489     ret = ff_get_buffer(avctx, frame, 0);
 490     if (ret < 0) {
 491         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
 492         return ret;
 493     }
 494     frame->nb_samples = 0;
 495
 496     memset(c->out, 0, c->nb_streams * 2 * sizeof(*c->out));
 497     for (i = 0; i < avctx->channels; i++) {
 498         ChannelMap *map = &c->channel_maps[i];
 499         if (!map->copy)
 500             c->out[2 * map->stream_idx + map->channel_idx] = (float*)frame->extended_data[i];
 501     }
 502
 503     /* read the data from the sync buffers */
 504     for (i = 0; i < c->nb_streams; i++) {
 505         float          **out = c->out + 2 * i;
 506         int sync_size = av_audio_fifo_size(c->sync_buffers[i]);
 507
 508         float sync_dummy[32];
 509         int out_dummy = (!out[0]) | ((!out[1]) << 1);
 510
 511         if (!out[0])
 512             out[0] = sync_dummy;
 513         if (!out[1])
 514             out[1] = sync_dummy;
 515         if (out_dummy && sync_size > FF_ARRAY_ELEMS(sync_dummy))
 516             return AVERROR_BUG;
 517
 518         ret = av_audio_fifo_read(c->sync_buffers[i], (void**)out, sync_size);
 519         if (ret < 0)
 520             return ret;
 521
 522         if (out_dummy & 1)
 523             out[0] = NULL;
 524         else
 525             out[0] += ret;
 526         if (out_dummy & 2)
 527             out[1] = NULL;
 528         else
 529             out[1] += ret;
 530
 531         c->out_size[i] = frame->linesize[0] - ret * sizeof(float);
 532     }
 533
 534     /* decode each sub-packet */
 535     for (i = 0; i < c->nb_streams; i++) {
 536         OpusStreamContext *s = &c->streams[i];
 537
 538         if (i && buf) {
 539             ret = ff_opus_parse_packet(&s->packet, buf, buf_size, i != c->nb_streams - 1);
 540             if (ret < 0) {
 541                 av_log(avctx, AV_LOG_ERROR, "Error parsing the packet header.\n");
 542                 return ret;
 543             }
 544             if (coded_samples != s->packet.frame_count * s->packet.frame_duration) {
 545                 av_log(avctx, AV_LOG_ERROR,
 546                        "Mismatching coded sample count in substream %d.\n", i);
 547                 return AVERROR_INVALIDDATA;
 548             }
 549
 550             s->silk_samplerate = get_silk_samplerate(s->packet.config);
 551         }
 552
 553         ret = opus_decode_subpacket(&c->streams[i], buf, s->packet.data_size,
 554                                     c->out + 2 * i, c->out_size[i], coded_samples);
 555         if (ret < 0)
 556             return ret;
 557         c->decoded_samples[i] = ret;
 558         decoded_samples       = FFMIN(decoded_samples, ret);
 559
 560         buf      += s->packet.packet_size;
 561         buf_size -= s->packet.packet_size;
 562     }
 563
 564     /* buffer the extra samples */
 565     for (i = 0; i < c->nb_streams; i++) {
 566         int buffer_samples = c->decoded_samples[i] - decoded_samples;
 567         if (buffer_samples) {
 568             float *buf[2] = { c->out[2 * i + 0] ? c->out[2 * i + 0] : (float*)frame->extended_data[0],
 569                               c->out[2 * i + 1] ? c->out[2 * i + 1] : (float*)frame->extended_data[0] };
 570             buf[0] += decoded_samples;
 571             buf[1] += decoded_samples;
 572             ret = av_audio_fifo_write(c->sync_buffers[i], (void**)buf, buffer_samples);
 573             if (ret < 0)
 574                 return ret;
 575         }
 576     }
 577
 578     for (i = 0; i < avctx->channels; i++) {
 579         ChannelMap *map = &c->channel_maps[i];
 580
 581         /* handle copied channels */
 582         if (map->copy) {
 583             memcpy(frame->extended_data[i],
 584                    frame->extended_data[map->copy_idx],
 585                    frame->linesize[0]);
 586         } else if (map->silence) {
 587             memset(frame->extended_data[i], 0, frame->linesize[0]);
 588         }
 589
 590         if (c->gain_i && decoded_samples > 0) {
 591             c->fdsp.vector_fmul_scalar((float*)frame->extended_data[i],
 592                                        (float*)frame->extended_data[i],
 593                                        c->gain, FFALIGN(decoded_samples, 8));
 594         }
 595     }
 596
 597     frame->nb_samples = decoded_samples;
 598     *got_frame_ptr    = !!decoded_samples;
 599
 600     return avpkt->size;
 601 }
 602
 603 static av_cold void opus_decode_flush(AVCodecContext *ctx)
 604 {
 605     OpusContext *c = ctx->priv_data;
 606     int i;
 607
 608     for (i = 0; i < c->nb_streams; i++) {
 609         OpusStreamContext *s = &c->streams[i];
 610
 611         memset(&s->packet, 0, sizeof(s->packet));
 612         s->delayed_samples = 0;
 613
 614         if (s->celt_delay)
 615             av_audio_fifo_drain(s->celt_delay, av_audio_fifo_size(s->celt_delay));
 616         avresample_close(s->avr);
 617
 618         av_audio_fifo_drain(c->sync_buffers[i], av_audio_fifo_size(c->sync_buffers[i]));
 619
 620         ff_silk_flush(s->silk);
 621         ff_celt_flush(s->celt);
 622     }
 623 }
 624
 625 static av_cold int opus_decode_close(AVCodecContext *avctx)
 626 {
 627     OpusContext *c = avctx->priv_data;
 628     int i;
 629
 630     for (i = 0; i < c->nb_streams; i++) {
 631         OpusStreamContext *s = &c->streams[i];
 632
 633         ff_silk_free(&s->silk);
 634         ff_celt_free(&s->celt);
 635
 636         av_freep(&s->out_dummy);
 637         s->out_dummy_allocated_size = 0;
 638
 639         av_audio_fifo_free(s->celt_delay);
 640         avresample_free(&s->avr);
 641     }
 642
 643     av_freep(&c->streams);
 644
 645     if (c->sync_buffers) {
 646         for (i = 0; i < c->nb_streams; i++)
 647             av_audio_fifo_free(c->sync_buffers[i]);
 648     }
 649     av_freep(&c->sync_buffers);
 650     av_freep(&c->decoded_samples);
 651     av_freep(&c->out);
 652     av_freep(&c->out_size);
 653
 654     c->nb_streams = 0;
 655
 656     av_freep(&c->channel_maps);
 657
 658     return 0;
 659 }
 660
 661 static av_cold int opus_decode_init(AVCodecContext *avctx)
 662 {
 663     OpusContext *c = avctx->priv_data;
 664     int ret, i, j;
 665
 666     avctx->sample_fmt  = AV_SAMPLE_FMT_FLTP;
 667     avctx->sample_rate = 48000;
 668
 669     avpriv_float_dsp_init(&c->fdsp, 0);
 670
 671     /* find out the channel configuration */
 672     ret = ff_opus_parse_extradata(avctx, c);
 673     if (ret < 0)
 674         return ret;
 675
 676     /* allocate and init each independent decoder */
 677     c->streams = av_mallocz_array(c->nb_streams, sizeof(*c->streams));
 678     c->out             = av_mallocz_array(c->nb_streams, 2 * sizeof(*c->out));
 679     c->out_size        = av_mallocz_array(c->nb_streams, sizeof(*c->out_size));
 680     c->sync_buffers    = av_mallocz_array(c->nb_streams, sizeof(*c->sync_buffers));
 681     c->decoded_samples = av_mallocz_array(c->nb_streams, sizeof(*c->decoded_samples));
 682     if (!c->streams || !c->sync_buffers || !c->decoded_samples || !c->out || !c->out_size) {
 683         c->nb_streams = 0;
 684         ret = AVERROR(ENOMEM);
 685         goto fail;
 686     }
 687
 688     for (i = 0; i < c->nb_streams; i++) {
 689         OpusStreamContext *s = &c->streams[i];
 690         uint64_t layout;
 691
 692         s->output_channels = (i < c->nb_stereo_streams) ? 2 : 1;
 693
 694         s->avctx = avctx;
 695
 696         for (j = 0; j < s->output_channels; j++) {
 697             s->silk_output[j]       = s->silk_buf[j];
 698             s->celt_output[j]       = s->celt_buf[j];
 699             s->redundancy_output[j] = s->redundancy_buf[j];
 700         }
 701
 702         s->fdsp = &c->fdsp;
 703
 704         s->avr = avresample_alloc_context();
 705         if (!s->avr)
 706             goto fail;
 707
 708         layout = (s->output_channels == 1) ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
 709         av_opt_set_int(s->avr, "in_sample_fmt",      avctx->sample_fmt,  0);
 710         av_opt_set_int(s->avr, "out_sample_fmt",     avctx->sample_fmt,  0);
 711         av_opt_set_int(s->avr, "in_channel_layout",  layout,             0);
 712         av_opt_set_int(s->avr, "out_channel_layout", layout,             0);
 713         av_opt_set_int(s->avr, "out_sample_rate",    avctx->sample_rate, 0);
 714
 715         ret = ff_silk_init(avctx, &s->silk, s->output_channels);
 716         if (ret < 0)
 717             goto fail;
 718
 719         ret = ff_celt_init(avctx, &s->celt, s->output_channels);
 720         if (ret < 0)
 721             goto fail;
 722
 723         s->celt_delay = av_audio_fifo_alloc(avctx->sample_fmt,
 724                                             s->output_channels, 1024);
 725         if (!s->celt_delay) {
 726             ret = AVERROR(ENOMEM);
 727             goto fail;
 728         }
 729
 730         c->sync_buffers[i] = av_audio_fifo_alloc(avctx->sample_fmt,
 731                                                  s->output_channels, 32);
 732         if (!c->sync_buffers[i]) {
 733             ret = AVERROR(ENOMEM);
 734             goto fail;
 735         }
 736     }
 737
 738     return 0;
 739 fail:
 740     opus_decode_close(avctx);
 741     return ret;
 742 }
 743
 744 AVCodec ff_opus_decoder = {
 745     .name            = "opus",
 746     .long_name       = NULL_IF_CONFIG_SMALL("Opus"),
 747     .type            = AVMEDIA_TYPE_AUDIO,
 748     .id              = AV_CODEC_ID_OPUS,
 749     .priv_data_size  = sizeof(OpusContext),
 750     .init            = opus_decode_init,
 751     .close           = opus_decode_close,
 752     .decode          = opus_decode_packet,
 753     .flush           = opus_decode_flush,
 754     .capabilities    = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY,
 755 };