From c49b88b93bca53c04f18d78c27dbf1dc6daea909 Mon Sep 17 00:00:00 2001 From: Anton Khirnov Date: Mon, 27 Jul 2015 11:13:53 +0200 Subject: [PATCH] opusdec: properly handle mismatching configurations in multichannel streams The substreams can have different resampling delays, so an additional level of buffering is needed to synchronize them. Bug-Id: 876 --- libavcodec/opus.h | 10 +++++ libavcodec/opusdec.c | 103 +++++++++++++++++++++++++++++++++++++------ 2 files changed, 99 insertions(+), 14 deletions(-) diff --git a/libavcodec/opus.h b/libavcodec/opus.h index c2fac063bc..accb998943 100644 --- a/libavcodec/opus.h +++ b/libavcodec/opus.h @@ -164,6 +164,16 @@ typedef struct ChannelMap { typedef struct OpusContext { OpusStreamContext *streams; + + /* current output buffers for each streams */ + float **out; + int *out_size; + /* Buffers for synchronizing the streams when they have different + * resampling delays */ + AVAudioFifo **sync_buffers; + /* number of decoded samples for each stream */ + int *decoded_samples; + int nb_streams; int nb_stereo_streams; diff --git a/libavcodec/opusdec.c b/libavcodec/opusdec.c index 771922e973..80a80b4e52 100644 --- a/libavcodec/opusdec.c +++ b/libavcodec/opusdec.c @@ -367,12 +367,17 @@ static int opus_decode_frame(OpusStreamContext *s, const uint8_t *data, int size static int opus_decode_subpacket(OpusStreamContext *s, const uint8_t *buf, int buf_size, + float **out, int out_size, int nb_samples) { int output_samples = 0; int flush_needed = 0; int i, j, ret; + s->out[0] = out[0]; + s->out[1] = out[1]; + s->out_size = out_size; + /* check if we need to flush the resampler */ if (avresample_is_open(s->avr)) { if (buf) { @@ -450,9 +455,16 @@ static int opus_decode_packet(AVCodecContext *avctx, void *data, const uint8_t *buf = avpkt->data; int buf_size = avpkt->size; int coded_samples = 0; - int decoded_samples = 0; + int decoded_samples = INT_MAX; + int delayed_samples = 0; int i, ret; + /* calculate the number of delayed samples */ + for (i = 0; i < c->nb_streams; i++) { + delayed_samples = FFMAX(delayed_samples, + c->streams[i].delayed_samples + av_audio_fifo_size(c->sync_buffers[i])); + } + /* decode the header of the first sub-packet to find out the sample count */ if (buf) { OpusPacket *pkt = &c->streams[0].packet; @@ -465,7 +477,7 @@ static int opus_decode_packet(AVCodecContext *avctx, void *data, c->streams[0].silk_samplerate = get_silk_samplerate(pkt->config); } - frame->nb_samples = coded_samples + c->streams[0].delayed_samples; + frame->nb_samples = coded_samples + delayed_samples; /* no input or buffered data => nothing to do */ if (!frame->nb_samples) { @@ -481,14 +493,43 @@ static int opus_decode_packet(AVCodecContext *avctx, void *data, } frame->nb_samples = 0; + memset(c->out, 0, c->nb_streams * 2 * sizeof(*c->out)); for (i = 0; i < avctx->channels; i++) { ChannelMap *map = &c->channel_maps[i]; if (!map->copy) - c->streams[map->stream_idx].out[map->channel_idx] = (float*)frame->extended_data[i]; + c->out[2 * map->stream_idx + map->channel_idx] = (float*)frame->extended_data[i]; } - for (i = 0; i < c->nb_streams; i++) - c->streams[i].out_size = frame->linesize[0]; + /* read the data from the sync buffers */ + for (i = 0; i < c->nb_streams; i++) { + float **out = c->out + 2 * i; + int sync_size = av_audio_fifo_size(c->sync_buffers[i]); + + float sync_dummy[32]; + int out_dummy = (!out[0]) | ((!out[1]) << 1); + + if (!out[0]) + out[0] = sync_dummy; + if (!out[1]) + out[1] = sync_dummy; + if (out_dummy && sync_size > FF_ARRAY_ELEMS(sync_dummy)) + return AVERROR_BUG; + + ret = av_audio_fifo_read(c->sync_buffers[i], (void**)out, sync_size); + if (ret < 0) + return ret; + + if (out_dummy & 1) + out[0] = NULL; + else + out[0] += ret; + if (out_dummy & 2) + out[1] = NULL; + else + out[1] += ret; + + c->out_size[i] = frame->linesize[0] - ret * sizeof(float); + } /* decode each sub-packet */ for (i = 0; i < c->nb_streams; i++) { @@ -509,20 +550,31 @@ static int opus_decode_packet(AVCodecContext *avctx, void *data, s->silk_samplerate = get_silk_samplerate(s->packet.config); } - ret = opus_decode_subpacket(&c->streams[i], buf, - s->packet.data_size, coded_samples); + ret = opus_decode_subpacket(&c->streams[i], buf, s->packet.data_size, + c->out + 2 * i, c->out_size[i], coded_samples); if (ret < 0) return ret; - if (decoded_samples && ret != decoded_samples) { - av_log(avctx, AV_LOG_ERROR, "Different numbers of decoded samples " - "in a multi-channel stream\n"); - return AVERROR_INVALIDDATA; - } - decoded_samples = ret; + c->decoded_samples[i] = ret; + decoded_samples = FFMIN(decoded_samples, ret); + buf += s->packet.packet_size; buf_size -= s->packet.packet_size; } + /* buffer the extra samples */ + for (i = 0; i < c->nb_streams; i++) { + int buffer_samples = c->decoded_samples[i] - decoded_samples; + if (buffer_samples) { + float *buf[2] = { c->out[2 * i + 0] ? c->out[2 * i + 0] : (float*)frame->extended_data[0], + c->out[2 * i + 1] ? c->out[2 * i + 1] : (float*)frame->extended_data[0] }; + buf[0] += buffer_samples; + buf[1] += buffer_samples; + ret = av_audio_fifo_write(c->sync_buffers[i], (void**)buf, buffer_samples); + if (ret < 0) + return ret; + } + } + for (i = 0; i < avctx->channels; i++) { ChannelMap *map = &c->channel_maps[i]; @@ -563,6 +615,8 @@ static av_cold void opus_decode_flush(AVCodecContext *ctx) av_audio_fifo_drain(s->celt_delay, av_audio_fifo_size(s->celt_delay)); avresample_close(s->avr); + av_audio_fifo_drain(c->sync_buffers[i], av_audio_fifo_size(c->sync_buffers[i])); + ff_silk_flush(s->silk); ff_celt_flush(s->celt); } @@ -587,6 +641,16 @@ static av_cold int opus_decode_close(AVCodecContext *avctx) } av_freep(&c->streams); + + if (c->sync_buffers) { + for (i = 0; i < c->nb_streams; i++) + av_audio_fifo_free(c->sync_buffers[i]); + } + av_freep(&c->sync_buffers); + av_freep(&c->decoded_samples); + av_freep(&c->out); + av_freep(&c->out_size); + c->nb_streams = 0; av_freep(&c->channel_maps); @@ -611,7 +675,11 @@ static av_cold int opus_decode_init(AVCodecContext *avctx) /* allocate and init each independent decoder */ c->streams = av_mallocz_array(c->nb_streams, sizeof(*c->streams)); - if (!c->streams) { + c->out = av_mallocz_array(c->nb_streams, 2 * sizeof(*c->out)); + c->out_size = av_mallocz_array(c->nb_streams, sizeof(*c->out_size)); + c->sync_buffers = av_mallocz_array(c->nb_streams, sizeof(*c->sync_buffers)); + c->decoded_samples = av_mallocz_array(c->nb_streams, sizeof(*c->decoded_samples)); + if (!c->streams || !c->sync_buffers || !c->decoded_samples || !c->out || !c->out_size) { c->nb_streams = 0; ret = AVERROR(ENOMEM); goto fail; @@ -658,6 +726,13 @@ static av_cold int opus_decode_init(AVCodecContext *avctx) ret = AVERROR(ENOMEM); goto fail; } + + c->sync_buffers[i] = av_audio_fifo_alloc(avctx->sample_fmt, + s->output_channels, 32); + if (!c->sync_buffers[i]) { + ret = AVERROR(ENOMEM); + goto fail; + } } return 0;