git.sesse.net Git - casparcg/blob - modules/ffmpeg/producer/audio/audio_decoder.cpp

   1 /*\r
   2 * copyright (c) 2010 Sveriges Television AB <info@casparcg.com>\r
   3 *\r
   4 *  This file is part of CasparCG.\r
   5 *\r
   6 *    CasparCG is free software: you can redistribute it and/or modify\r
   7 *    it under the terms of the GNU General Public License as published by\r
   8 *    the Free Software Foundation, either version 3 of the License, or\r
   9 *    (at your option) any later version.\r
  10 *\r
  11 *    CasparCG is distributed in the hope that it will be useful,\r
  12 *    but WITHOUT ANY WARRANTY; without even the implied warranty of\r
  13 *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the\r
  14 *    GNU General Public License for more details.\r
  15 \r
  16 *    You should have received a copy of the GNU General Public License\r
  17 *    along with CasparCG.  If not, see <http://www.gnu.org/licenses/>.\r
  18 *\r
  19 */\r
  20 #include "../../stdafx.h"\r
  21 \r
  22 #include "audio_decoder.h"\r
  23 \r
  24 #include <tbb/task_group.h>\r
  25 \r
  26 #if defined(_MSC_VER)\r
  27 #pragma warning (push)\r
  28 #pragma warning (disable : 4244)\r
  29 #endif\r
  30 extern "C" \r
  31 {\r
  32         #define __STDC_CONSTANT_MACROS\r
  33         #define __STDC_LIMIT_MACROS\r
  34         #include <libavformat/avformat.h>\r
  35         #include <libavcodec/avcodec.h>\r
  36 }\r
  37 #if defined(_MSC_VER)\r
  38 #pragma warning (pop)\r
  39 #endif\r
  40 \r
  41 namespace caspar {\r
  42         \r
  43 struct audio_decoder::implementation : boost::noncopyable\r
  44 {       \r
  45         std::shared_ptr<AVCodecContext>                                                         codec_context_;         \r
  46         const core::video_format_desc                                                           format_desc_;\r
  47         int                                                                                                                     index_;\r
  48         std::shared_ptr<ReSampleContext>                                                        resampler_;\r
  49 \r
  50         std::vector<int8_t,  tbb::cache_aligned_allocator<int8_t>>      buffer1_;\r
  51         std::vector<int8_t,  tbb::cache_aligned_allocator<int8_t>>      buffer2_;\r
  52         std::queue<std::shared_ptr<AVPacket>>                                           packets_;\r
  53         size_t                                                                                                          sample_count_;\r
  54 public:\r
  55         explicit implementation(const std::shared_ptr<AVFormatContext>& context, const core::video_format_desc& format_desc) \r
  56                 : format_desc_(format_desc)     \r
  57                 , sample_count_(0)\r
  58         {                          \r
  59                 AVCodec* dec;\r
  60                 index_ = av_find_best_stream(context.get(), AVMEDIA_TYPE_AUDIO, -1, -1, &dec, 0);\r
  61 \r
  62                 int errn = avcodec_open(context->streams[index_]->codec, dec);\r
  63                 if(errn < 0)\r
  64                         return;\r
  65                                 \r
  66                 codec_context_.reset(context->streams[index_]->codec, avcodec_close);\r
  67 \r
  68                 if(codec_context_ &&\r
  69                    (codec_context_->sample_rate != static_cast<int>(format_desc_.audio_sample_rate) || \r
  70                     codec_context_->channels    != static_cast<int>(format_desc_.audio_channels)) ||\r
  71                         codec_context_->sample_fmt      != AV_SAMPLE_FMT_S16)\r
  72                 {       \r
  73                         auto resampler = av_audio_resample_init(format_desc_.audio_channels,    codec_context_->channels,\r
  74                                                                                                         format_desc_.audio_sample_rate, codec_context_->sample_rate,\r
  75                                                                                                         AV_SAMPLE_FMT_S16,                              codec_context_->sample_fmt,\r
  76                                                                                                         16, 10, 0, 0.8);\r
  77 \r
  78                         CASPAR_LOG(warning) << L" Invalid audio format.";\r
  79 \r
  80                         if(resampler)\r
  81                                 resampler_.reset(resampler, audio_resample_close);\r
  82                         else\r
  83                                 codec_context_ = nullptr;\r
  84                 }               \r
  85         }\r
  86 \r
  87         void push(const std::shared_ptr<AVPacket>& packet)\r
  88         {                       \r
  89                 if(!codec_context_)\r
  90                         return;\r
  91 \r
  92                 if(packet && packet->stream_index != index_)\r
  93                         return;\r
  94 \r
  95                 packets_.push(packet);\r
  96         }       \r
  97         \r
  98         std::vector<std::vector<int16_t>> poll()\r
  99         {\r
 100                 std::vector<std::vector<int16_t>> result;\r
 101 \r
 102                 if(!codec_context_)\r
 103                         result.push_back(std::vector<int16_t>(format_desc_.audio_samples_per_frame, 0));\r
 104                 else if(!packets_.empty())\r
 105                 {                       \r
 106                         if(packets_.front())            \r
 107                         {\r
 108                                 AVPacket pkt;\r
 109                                 av_init_packet(&pkt);\r
 110                                 pkt.data = packets_.front()->data;\r
 111                                 pkt.size = packets_.front()->size;\r
 112 \r
 113                                 for(int n = 0; n < 64 && pkt.size > 0; ++n)\r
 114                                         decode(pkt, result);\r
 115                         }\r
 116                         else\r
 117                                 avcodec_flush_buffers(codec_context_.get());\r
 118 \r
 119                         packets_.pop();\r
 120                 }\r
 121 \r
 122                 return result;\r
 123         }\r
 124 \r
 125         void decode(AVPacket& pkt, std::vector<std::vector<int16_t>>& audio_chunks)\r
 126         {               \r
 127                 buffer1_.resize(AVCODEC_MAX_AUDIO_FRAME_SIZE*2, 0);\r
 128                 int written_bytes = buffer1_.size() - FF_INPUT_BUFFER_PADDING_SIZE;\r
 129 \r
 130                 const int ret = avcodec_decode_audio3(codec_context_.get(), reinterpret_cast<int16_t*>(buffer1_.data()), &written_bytes, &pkt);\r
 131                 if(ret < 0)\r
 132                 {       \r
 133                         BOOST_THROW_EXCEPTION(\r
 134                                 invalid_operation() <<\r
 135                                 boost::errinfo_api_function("avcodec_decode_audio2") <<\r
 136                                 boost::errinfo_errno(AVUNERROR(ret)));\r
 137                 }\r
 138 \r
 139                 // There might be several frames in one packet.\r
 140                 pkt.size -= ret;\r
 141                 pkt.data += ret;\r
 142                         \r
 143                 buffer1_.resize(written_bytes);\r
 144 \r
 145                 if(resampler_)\r
 146                 {\r
 147                         buffer2_.resize(AVCODEC_MAX_AUDIO_FRAME_SIZE*2, 0);\r
 148                         auto ret = audio_resample(resampler_.get(),\r
 149                                                                                 reinterpret_cast<short*>(buffer2_.data()), \r
 150                                                                                 reinterpret_cast<short*>(buffer1_.data()), \r
 151                                                                                 buffer1_.size() / (av_get_bytes_per_sample(codec_context_->sample_fmt) * codec_context_->channels)); \r
 152                         buffer2_.resize(ret * av_get_bytes_per_sample(AV_SAMPLE_FMT_S16) * format_desc_.audio_channels);\r
 153                         std::swap(buffer1_, buffer2_);\r
 154                 }\r
 155 \r
 156                 const auto n_samples = buffer1_.size() / av_get_bytes_per_sample(AV_SAMPLE_FMT_S16);\r
 157                 const auto samples = reinterpret_cast<int16_t*>(buffer1_.data());\r
 158 \r
 159                 sample_count_ = std::max(sample_count_, n_samples);\r
 160 \r
 161                 if(sample_count_ == 0 || n_samples == sample_count_)\r
 162                         audio_chunks.push_back(std::vector<int16_t>(samples, samples + n_samples));\r
 163                 else\r
 164                         CASPAR_LOG(warning) << L" audio_decoder: Discarding " << n_samples << L" samples.";\r
 165         }\r
 166 \r
 167         bool ready() const\r
 168         {\r
 169                 return !codec_context_ || !packets_.empty();\r
 170         }\r
 171 };\r
 172 \r
 173 audio_decoder::audio_decoder(const std::shared_ptr<AVFormatContext>& context, const core::video_format_desc& format_desc) : impl_(new implementation(context, format_desc)){}\r
 174 void audio_decoder::push(const std::shared_ptr<AVPacket>& packet){impl_->push(packet);}\r
 175 bool audio_decoder::ready() const{return impl_->ready();}\r
 176 std::vector<std::vector<int16_t>> audio_decoder::poll(){return impl_->poll();}\r
 177 }