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::vector<int16_t, tbb::cache_aligned_allocator<int16_t>>     audio_samples_; \r
  53         std::queue<std::shared_ptr<AVPacket>>                                           packets_;\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         {                          \r
  58                 AVCodec* dec;\r
  59                 index_ = av_find_best_stream(context.get(), AVMEDIA_TYPE_AUDIO, -1, -1, &dec, 0);\r
  60 \r
  61                 int errn = avcodec_open(context->streams[index_]->codec, dec);\r
  62                 if(errn < 0)\r
  63                         return;\r
  64                                 \r
  65                 codec_context_.reset(context->streams[index_]->codec, avcodec_close);\r
  66 \r
  67                 if(codec_context_ &&\r
  68                    (codec_context_->sample_rate != static_cast<int>(format_desc_.audio_sample_rate) || \r
  69                     codec_context_->channels    != static_cast<int>(format_desc_.audio_channels)) ||\r
  70                         codec_context_->sample_fmt      != AV_SAMPLE_FMT_S16)\r
  71                 {       \r
  72                         auto resampler = av_audio_resample_init(format_desc_.audio_channels,    codec_context_->channels,\r
  73                                                                                                         format_desc_.audio_sample_rate, codec_context_->sample_rate,\r
  74                                                                                                         AV_SAMPLE_FMT_S16,                              codec_context_->sample_fmt,\r
  75                                                                                                         16, 10, 0, 0.8);\r
  76 \r
  77                         CASPAR_LOG(warning) << L" Invalid audio format.";\r
  78 \r
  79                         if(resampler)\r
  80                                 resampler_.reset(resampler, audio_resample_close);\r
  81                         else\r
  82                                 codec_context_ = nullptr;\r
  83                 }               \r
  84         }\r
  85 \r
  86         void push(const std::shared_ptr<AVPacket>& packet)\r
  87         {                       \r
  88                 if(!codec_context_)\r
  89                         return;\r
  90 \r
  91                 if(packet && packet->stream_index != index_)\r
  92                         return;\r
  93 \r
  94                 packets_.push(packet);\r
  95         }       \r
  96         \r
  97         std::vector<std::vector<int16_t>> poll()\r
  98         {\r
  99                 std::vector<std::vector<int16_t>> result;\r
 100 \r
 101                 if(!codec_context_)\r
 102                         result.push_back(std::vector<int16_t>(format_desc_.audio_samples_per_frame, 0));\r
 103                 else if(!packets_.empty())\r
 104                 {                       \r
 105                         if(packets_.front())            \r
 106                         {\r
 107                                 AVPacket pkt;\r
 108                                 av_init_packet(&pkt);\r
 109                                 pkt.data = packets_.front()->data;\r
 110                                 pkt.size = packets_.front()->size;\r
 111 \r
 112                                 for(int n = 0; n < 64 && pkt.size > 0; ++n)\r
 113                                         decode(pkt);\r
 114                         }\r
 115                         else\r
 116                                 flush();\r
 117 \r
 118                         packets_.pop();\r
 119 \r
 120                         while(audio_samples_.size() > format_desc_.audio_samples_per_frame)\r
 121                         {\r
 122                                 const auto begin = audio_samples_.begin();\r
 123                                 const auto end   = audio_samples_.begin() + format_desc_.audio_samples_per_frame;\r
 124 \r
 125                                 result.push_back(std::vector<int16_t>(begin, end));\r
 126                                 audio_samples_.erase(begin, end);\r
 127                         }\r
 128                 }\r
 129 \r
 130                 return result;\r
 131         }\r
 132 \r
 133         void decode(AVPacket& pkt)\r
 134         {               \r
 135                 buffer1_.resize(AVCODEC_MAX_AUDIO_FRAME_SIZE*2, 0);\r
 136                 int written_bytes = buffer1_.size() - FF_INPUT_BUFFER_PADDING_SIZE;\r
 137 \r
 138                 const int ret = avcodec_decode_audio3(codec_context_.get(), reinterpret_cast<int16_t*>(buffer1_.data()), &written_bytes, &pkt);\r
 139                 if(ret < 0)\r
 140                 {       \r
 141                         BOOST_THROW_EXCEPTION(\r
 142                                 invalid_operation() <<\r
 143                                 boost::errinfo_api_function("avcodec_decode_audio2") <<\r
 144                                 boost::errinfo_errno(AVUNERROR(ret)));\r
 145                 }\r
 146 \r
 147                 pkt.size -= ret;\r
 148                 pkt.data += ret;\r
 149                         \r
 150                 buffer1_.resize(written_bytes);\r
 151 \r
 152                 if(resampler_)\r
 153                 {\r
 154                         buffer2_.resize(AVCODEC_MAX_AUDIO_FRAME_SIZE*2, 0);\r
 155                         auto ret = audio_resample(resampler_.get(),\r
 156                                                                                 reinterpret_cast<short*>(buffer2_.data()), \r
 157                                                                                 reinterpret_cast<short*>(buffer1_.data()), \r
 158                                                                                 buffer1_.size() / (av_get_bytes_per_sample(codec_context_->sample_fmt) * codec_context_->channels)); \r
 159                         buffer2_.resize(ret * av_get_bytes_per_sample(AV_SAMPLE_FMT_S16) * format_desc_.audio_channels);\r
 160                         std::swap(buffer1_, buffer2_);\r
 161                 }\r
 162 \r
 163                 const auto n_samples = buffer1_.size() / av_get_bytes_per_sample(AV_SAMPLE_FMT_S16);\r
 164                 const auto samples = reinterpret_cast<int16_t*>(buffer1_.data());\r
 165 \r
 166                 audio_samples_.insert(audio_samples_.end(), samples, samples + n_samples);\r
 167         }\r
 168 \r
 169         void flush()\r
 170         {\r
 171                 auto truncate = audio_samples_.size() % format_desc_.audio_samples_per_frame;\r
 172                 if(truncate > 0)\r
 173                 {\r
 174                         audio_samples_.resize(audio_samples_.size() - truncate); \r
 175                         CASPAR_LOG(info) << L"Truncating " << truncate << L" audio-samples."; \r
 176                 }\r
 177                 avcodec_flush_buffers(codec_context_.get());\r
 178         }\r
 179 \r
 180         bool ready() const\r
 181         {\r
 182                 return !codec_context_ || !packets_.empty();\r
 183         }\r
 184 };\r
 185 \r
 186 audio_decoder::audio_decoder(const std::shared_ptr<AVFormatContext>& context, const core::video_format_desc& format_desc) : impl_(new implementation(context, format_desc)){}\r
 187 void audio_decoder::push(const std::shared_ptr<AVPacket>& packet){impl_->push(packet);}\r
 188 bool audio_decoder::ready() const{return impl_->ready();}\r
 189 std::vector<std::vector<int16_t>> audio_decoder::poll(){return impl_->poll();}\r
 190 }