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::vector<int8_t, tbb::cache_aligned_allocator<int8_t>>       buffer1_;               // avcodec_decode_audio3 needs 4 byte alignment\r
  49         std::vector<int8_t, tbb::cache_aligned_allocator<int8_t>>       buffer2_;               // avcodec_decode_audio3 needs 4 byte alignment\r
  50         std::vector<int16_t, tbb::cache_aligned_allocator<int16_t>>     audio_samples_;         // avcodec_decode_audio3 needs 4 byte alignment\r
  51         std::queue<std::shared_ptr<AVPacket>>                                           packets_;\r
  52         std::shared_ptr<ReSampleContext>                                                        resampler_;\r
  53 public:\r
  54         explicit implementation(const std::shared_ptr<AVFormatContext>& context, const core::video_format_desc& format_desc) \r
  55                 : format_desc_(format_desc)     \r
  56         {                          \r
  57                 AVCodec* dec;\r
  58                 index_ = av_find_best_stream(context.get(), AVMEDIA_TYPE_AUDIO, -1, -1, &dec, 0);\r
  59 \r
  60                 int errn = avcodec_open(context->streams[index_]->codec, dec);\r
  61                 if(errn < 0)\r
  62                         return;\r
  63                                 \r
  64                 codec_context_.reset(context->streams[index_]->codec, avcodec_close);\r
  65 \r
  66                 if(codec_context_ &&\r
  67                    (codec_context_->sample_rate != static_cast<int>(format_desc_.audio_sample_rate) || \r
  68                     codec_context_->channels    != static_cast<int>(format_desc_.audio_channels)) ||\r
  69                         codec_context_->sample_fmt      != AV_SAMPLE_FMT_S16)\r
  70                 {       \r
  71                         auto resampler = av_audio_resample_init(format_desc_.audio_channels,    codec_context_->channels,
  72                                                                                                         format_desc_.audio_sample_rate, codec_context_->sample_rate,
  73                                                                                                         AV_SAMPLE_FMT_S16,                              codec_context_->sample_fmt,
  74                                                                                                         16, 10, 0, 0.8);\r
  75 \r
  76                         CASPAR_LOG(warning) << L" Invalid audio format.";\r
  77 \r
  78                         if(resampler)\r
  79                                 resampler_.reset(resampler, audio_resample_close);\r
  80                         else\r
  81                                 codec_context_ = nullptr;\r
  82                 }               \r
  83         }\r
  84 \r
  85         void push(const std::shared_ptr<AVPacket>& packet)\r
  86         {                       \r
  87                 if(!codec_context_)\r
  88                         return;\r
  89 \r
  90                 if(packet && packet->stream_index != index_)\r
  91                         return;\r
  92 \r
  93                 packets_.push(packet);\r
  94         }       \r
  95         \r
  96         std::vector<std::vector<int16_t>> poll()\r
  97         {\r
  98                 std::vector<std::vector<int16_t>> result;\r
  99 \r
 100                 if(!codec_context_)\r
 101                         result.push_back(std::vector<int16_t>(format_desc_.audio_samples_per_frame, 0));\r
 102                 else if(!packets_.empty())\r
 103                 {\r
 104                         decode(packets_.front());\r
 105                         packets_.pop();\r
 106 \r
 107                         while(audio_samples_.size() > format_desc_.audio_samples_per_frame)\r
 108                         {\r
 109                                 const auto begin = audio_samples_.begin();\r
 110                                 const auto end   = audio_samples_.begin() + format_desc_.audio_samples_per_frame;\r
 111 \r
 112                                 result.push_back(std::vector<int16_t>(begin, end));\r
 113                                 audio_samples_.erase(begin, end);\r
 114                         }\r
 115                 }\r
 116 \r
 117                 return result;\r
 118         }\r
 119 \r
 120         void decode(const std::shared_ptr<AVPacket>& packet)\r
 121         {                                                                                       \r
 122                 if(!packet) // eof\r
 123                 {\r
 124                         auto truncate = audio_samples_.size() % format_desc_.audio_samples_per_frame;\r
 125                         if(truncate > 0)\r
 126                         {\r
 127                                 audio_samples_.resize(audio_samples_.size() - truncate); \r
 128                                 CASPAR_LOG(info) << L"Truncating " << truncate << L" audio-samples."; \r
 129                         }\r
 130                         avcodec_flush_buffers(codec_context_.get());\r
 131                 }\r
 132                 else\r
 133                 {\r
 134                         buffer1_.resize(AVCODEC_MAX_AUDIO_FRAME_SIZE*2, 0);\r
 135                         int written_bytes = buffer1_.size() - FF_INPUT_BUFFER_PADDING_SIZE;\r
 136                         // TODO: Packet might contain multiple frames\r
 137                         const int errn = avcodec_decode_audio3(codec_context_.get(), reinterpret_cast<int16_t*>(buffer1_.data()), &written_bytes, packet.get());\r
 138                         if(errn < 0)\r
 139                         {       \r
 140                                 BOOST_THROW_EXCEPTION(\r
 141                                         invalid_operation() <<\r
 142                                         boost::errinfo_api_function("avcodec_decode_audio2") <<\r
 143                                         boost::errinfo_errno(AVUNERROR(errn)));\r
 144                         }\r
 145 \r
 146                         buffer1_.resize(written_bytes);\r
 147 \r
 148                         if(resampler_)\r
 149                         {\r
 150                                 buffer2_.resize(AVCODEC_MAX_AUDIO_FRAME_SIZE*2, 0);\r
 151                                 auto ret = audio_resample(resampler_.get(),\r
 152                                                                                   reinterpret_cast<short*>(buffer2_.data()), \r
 153                                                                                   reinterpret_cast<short*>(buffer1_.data()), \r
 154                                                                                   buffer1_.size() / av_get_bytes_per_sample(codec_context_->sample_fmt)); \r
 155                                 buffer2_.resize(ret);\r
 156                                 std::swap(buffer1_, buffer2_);\r
 157                         }\r
 158 \r
 159                         const auto n_samples = buffer1_.size() / av_get_bytes_per_sample(AV_SAMPLE_FMT_S16);\r
 160                         const auto samples = reinterpret_cast<int16_t*>(buffer1_.data());\r
 161 \r
 162                         audio_samples_.insert(audio_samples_.end(), samples, samples + n_samples);      \r
 163                 }\r
 164         }\r
 165 \r
 166         bool ready() const\r
 167         {\r
 168                 return !codec_context_ || !packets_.empty();\r
 169         }\r
 170 };\r
 171 \r
 172 audio_decoder::audio_decoder(const std::shared_ptr<AVFormatContext>& context, const core::video_format_desc& format_desc) : impl_(new implementation(context, format_desc)){}\r
 173 void audio_decoder::push(const std::shared_ptr<AVPacket>& packet){impl_->push(packet);}\r
 174 bool audio_decoder::ready() const{return impl_->ready();}\r
 175 std::vector<std::vector<int16_t>> audio_decoder::poll(){return impl_->poll();}\r
 176 }