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 "../../ffmpeg_error.h"\r
  25 \r
  26 #include <core/video_format.h>\r
  27 \r
  28 #include <tbb/cache_aligned_allocator.h>\r
  29 \r
  30 #include <queue>\r
  31 \r
  32 #if defined(_MSC_VER)\r
  33 #pragma warning (push)\r
  34 #pragma warning (disable : 4244)\r
  35 #endif\r
  36 extern "C" \r
  37 {\r
  38         #include <libavformat/avformat.h>\r
  39         #include <libavcodec/avcodec.h>\r
  40 }\r
  41 #if defined(_MSC_VER)\r
  42 #pragma warning (pop)\r
  43 #endif\r
  44 \r
  45 namespace caspar {\r
  46         \r
  47 struct audio_decoder::implementation : boost::noncopyable\r
  48 {       \r
  49         std::shared_ptr<AVCodecContext>                                                         codec_context_;         \r
  50         const core::video_format_desc                                                           format_desc_;\r
  51         int                                                                                                                     index_;\r
  52         std::shared_ptr<ReSampleContext>                                                        resampler_;\r
  53 \r
  54         std::vector<int8_t,  tbb::cache_aligned_allocator<int8_t>>      buffer1_;\r
  55         std::vector<int8_t,  tbb::cache_aligned_allocator<int8_t>>      buffer2_;\r
  56         std::vector<int16_t, tbb::cache_aligned_allocator<int16_t>>     audio_samples_; \r
  57         std::queue<std::shared_ptr<AVPacket>>                                           packets_;\r
  58 \r
  59         int64_t                                                                                                         nb_frames_;\r
  60         double                                                                                                          duration_;\r
  61 public:\r
  62         explicit implementation(const safe_ptr<AVFormatContext>& context, const core::video_format_desc& format_desc) \r
  63                 : format_desc_(format_desc)     \r
  64                 , nb_frames_(0)\r
  65                 , duration_(std::numeric_limits<double>::max())\r
  66         {                               \r
  67                 try\r
  68                 {\r
  69                         AVCodec* dec;\r
  70                         index_ = THROW_ON_ERROR2(av_find_best_stream(context.get(), AVMEDIA_TYPE_AUDIO, -1, -1, &dec, 0), "[audio_decoder]");\r
  71 \r
  72                         THROW_ON_ERROR2(avcodec_open(context->streams[index_]->codec, dec), "[audio_decoder]");\r
  73                         \r
  74                         codec_context_.reset(context->streams[index_]->codec, avcodec_close);\r
  75 \r
  76                         buffer1_.resize(AVCODEC_MAX_AUDIO_FRAME_SIZE*2);\r
  77 \r
  78                         //nb_frames_ = context->streams[index_]->nb_frames;\r
  79                         //if(nb_frames_ == 0)\r
  80                         //      nb_frames_ = context->streams[index_]->duration / (codec_context_->channels*codec_context_->sample_rate);\r
  81                         duration_ = context->streams[index_]->duration / static_cast<double>(codec_context_->sample_rate);\r
  82 \r
  83                         if(codec_context_->sample_rate  != static_cast<int>(format_desc_.audio_sample_rate) || \r
  84                            codec_context_->channels             != static_cast<int>(format_desc_.audio_channels) ||\r
  85                            codec_context_->sample_fmt   != AV_SAMPLE_FMT_S16)\r
  86                         {       \r
  87                                 auto resampler = av_audio_resample_init(format_desc_.audio_channels,    codec_context_->channels,\r
  88                                                                                                                 format_desc_.audio_sample_rate, codec_context_->sample_rate,\r
  89                                                                                                                 AV_SAMPLE_FMT_S16,                              codec_context_->sample_fmt,\r
  90                                                                                                                 16, 10, 0, 0.8);\r
  91 \r
  92                                 buffer2_.resize(AVCODEC_MAX_AUDIO_FRAME_SIZE*2);\r
  93 \r
  94                                 CASPAR_LOG(warning) << L"Invalid audio format. Resampling." <<\r
  95                                                                            L" sample_rate:" << static_cast<int>(codec_context_->sample_rate)  <<\r
  96                                                                            L" audio_channels:" << static_cast<int>(codec_context_->channels)  <<\r
  97                                                                            L" sample_fmt:" << static_cast<int>(codec_context_->sample_fmt);\r
  98 \r
  99                                 if(resampler)\r
 100                                         resampler_.reset(resampler, audio_resample_close);\r
 101                                 else\r
 102                                         codec_context_ = nullptr;\r
 103                         }               \r
 104                 }\r
 105                 catch(...)\r
 106                 {\r
 107                         index_ = THROW_ON_ERROR2(av_find_best_stream(context.get(), AVMEDIA_TYPE_VIDEO, -1, -1, nullptr, 0), "[audio_decoder]");\r
 108 \r
 109                         CASPAR_LOG_CURRENT_EXCEPTION();\r
 110                         CASPAR_LOG(warning) << "[audio_decoder] Failed to open audio-stream. Running without audio.";                   \r
 111                 }\r
 112         }\r
 113 \r
 114         void push(const std::shared_ptr<AVPacket>& packet)\r
 115         {                       \r
 116                 if(packet && packet->stream_index != index_)\r
 117                         return;\r
 118 \r
 119                 packets_.push(packet);\r
 120         }       \r
 121         \r
 122         std::vector<std::shared_ptr<std::vector<int16_t>>> poll()\r
 123         {\r
 124                 std::vector<std::shared_ptr<std::vector<int16_t>>> result;\r
 125 \r
 126                 if(packets_.empty())\r
 127                         return result;\r
 128 \r
 129                 if(!codec_context_)\r
 130                         return empty_poll();\r
 131                 \r
 132                 auto packet = packets_.front();\r
 133 \r
 134                 if(packet)              \r
 135                 {\r
 136                         result.push_back(decode(*packet));\r
 137                         if(packet->size == 0)                                   \r
 138                                 packets_.pop();\r
 139                 }\r
 140                 else                    \r
 141                 {       \r
 142                         avcodec_flush_buffers(codec_context_.get());\r
 143                         result.push_back(nullptr);\r
 144                         packets_.pop();\r
 145                 }               \r
 146 \r
 147                 return result;\r
 148         }\r
 149 \r
 150         std::vector<std::shared_ptr<std::vector<int16_t>>> empty_poll()\r
 151         {\r
 152                 auto packet = packets_.front();\r
 153                 packets_.pop();\r
 154 \r
 155                 if(!packet)                     \r
 156                         return boost::assign::list_of(nullptr);\r
 157                 \r
 158                 return boost::assign::list_of(std::make_shared<std::vector<int16_t>>(format_desc_.audio_samples_per_frame, 0)); \r
 159         }\r
 160 \r
 161         std::shared_ptr<std::vector<int16_t>> decode(AVPacket& pkt)\r
 162         {               \r
 163                 int written_bytes = buffer1_.size() - FF_INPUT_BUFFER_PADDING_SIZE;\r
 164 \r
 165                 int ret = THROW_ON_ERROR2(avcodec_decode_audio3(codec_context_.get(), reinterpret_cast<int16_t*>(buffer1_.data()), &written_bytes, &pkt), "[audio_decoder]");\r
 166 \r
 167                 // There might be several frames in one packet.\r
 168                 pkt.size -= ret;\r
 169                 pkt.data += ret;\r
 170                         \r
 171                 if(resampler_)\r
 172                 {\r
 173                         auto ret = audio_resample(resampler_.get(),\r
 174                                                                                 reinterpret_cast<short*>(buffer2_.data()), \r
 175                                                                                 reinterpret_cast<short*>(buffer1_.data()), \r
 176                                                                                 written_bytes / (av_get_bytes_per_sample(codec_context_->sample_fmt) * codec_context_->channels)); \r
 177                         written_bytes = ret * av_get_bytes_per_sample(AV_SAMPLE_FMT_S16) * format_desc_.audio_channels;\r
 178                         std::swap(buffer1_, buffer2_);\r
 179                 }\r
 180 \r
 181                 const auto n_samples = written_bytes / av_get_bytes_per_sample(AV_SAMPLE_FMT_S16);\r
 182                 const auto samples = reinterpret_cast<int16_t*>(buffer1_.data());\r
 183 \r
 184                 return std::make_shared<std::vector<int16_t>>(samples, samples + n_samples);\r
 185         }\r
 186 \r
 187         bool ready() const\r
 188         {\r
 189                 return !packets_.empty();\r
 190         }\r
 191 };\r
 192 \r
 193 audio_decoder::audio_decoder(const safe_ptr<AVFormatContext>& context, const core::video_format_desc& format_desc) : impl_(new implementation(context, format_desc)){}\r
 194 void audio_decoder::push(const std::shared_ptr<AVPacket>& packet){impl_->push(packet);}\r
 195 bool audio_decoder::ready() const{return impl_->ready();}\r
 196 std::vector<std::shared_ptr<std::vector<int16_t>>> audio_decoder::poll(){return impl_->poll();}\r
 197 int64_t audio_decoder::nb_frames() const{return impl_->nb_frames_;}\r
 198 double audio_decoder::duration() const {return impl_->duration_;}\r
 199 }