]> git.sesse.net Git - casparcg/blobdiff - modules/ffmpeg/producer/ffmpeg_producer.cpp
git-svn-id: https://casparcg.svn.sourceforge.net/svnroot/casparcg/server/branches...
[casparcg] / modules / ffmpeg / producer / ffmpeg_producer.cpp
index 15717ce2f17f6b9d74b29c2648d0bc5e5010935a..576f7a3cc05ace01b717e287e915274ce62f6441 100644 (file)
 \r
 #include "ffmpeg_producer.h"\r
 \r
+#include "frame_muxer.h"\r
 #include "input.h"\r
+#include "util.h"\r
 #include "audio/audio_decoder.h"\r
 #include "video/video_decoder.h"\r
 \r
+#include <common/env.h>\r
 #include <common/utility/assert.h>\r
-#include <common/utility/timer.h>\r
 #include <common/diagnostics/graph.h>\r
 \r
-#include <core/producer/frame/basic_frame.h>\r
-#include <core/mixer/write_frame.h>\r
-#include <core/producer/frame/audio_transform.h>\r
 #include <core/video_format.h>\r
+#include <core/producer/frame_producer.h>\r
+#include <core/producer/frame/frame_factory.h>\r
+#include <core/producer/frame/basic_frame.h>\r
 \r
-#include <common/env.h>\r
-\r
-#include <tbb/parallel_invoke.h>\r
-\r
+#include <boost/algorithm/string.hpp>\r
+#include <boost/assign.hpp>\r
 #include <boost/timer.hpp>\r
-#include <boost/range/algorithm.hpp>\r
-#include <boost/range/algorithm_ext.hpp>\r
+#include <boost/foreach.hpp>\r
+#include <boost/filesystem.hpp>\r
+#include <boost/range/algorithm/find_if.hpp>\r
+#include <boost/range/algorithm/find.hpp>\r
 \r
-#include <deque>\r
+#include <tbb/parallel_invoke.h>\r
 \r
-namespace caspar {\r
-       \r
+namespace caspar { namespace ffmpeg {\r
+                               \r
 struct ffmpeg_producer : public core::frame_producer\r
 {\r
-       const std::wstring                                              filename_;\r
+       const std::wstring                                                              filename_;\r
        \r
-       const safe_ptr<diagnostics::graph>              graph_;\r
-       boost::timer                                                    frame_timer_;\r
+       const safe_ptr<diagnostics::graph>                              graph_;\r
+       boost::timer                                                                    frame_timer_;\r
+       boost::timer                                                                    video_timer_;\r
+       boost::timer                                                                    audio_timer_;\r
                                        \r
-       const safe_ptr<core::frame_factory>             frame_factory_;\r
+       const safe_ptr<core::frame_factory>                             frame_factory_;\r
+       const core::video_format_desc                                   format_desc_;\r
 \r
-       input                                                                   input_; \r
-       std::unique_ptr<video_decoder>                  video_decoder_;\r
-       std::unique_ptr<audio_decoder>                  audio_decoder_;\r
+       input                                                                                   input_; \r
+       video_decoder                                                                   video_decoder_;\r
+       audio_decoder                                                                   audio_decoder_; \r
+       double                                                                                  fps_;\r
+       frame_muxer                                                                             muxer_;\r
 \r
-       std::deque<std::pair<int, std::vector<int16_t>>> audio_chunks_;\r
-       std::deque<std::pair<int, safe_ptr<core::write_frame>>> video_frames_;\r
+       const int                                                                               start_;\r
+       const bool                                                                              loop_;\r
+       const size_t                                                                    length_;\r
+\r
+       safe_ptr<core::basic_frame>                                             last_frame_;\r
+\r
+       const size_t                                                                    width_;\r
+       const size_t                                                                    height_;\r
+       bool                                                                                    is_progressive_;\r
+       \r
 public:\r
-       explicit ffmpeg_producer(const safe_ptr<core::frame_factory>& frame_factory, const std::wstring& filename, const std::wstring& filter_str, bool loop, int start, int length) \r
+       explicit ffmpeg_producer(const safe_ptr<core::frame_factory>& frame_factory, const std::wstring& filename, const std::wstring& filter, bool loop, int start, size_t length) \r
                : filename_(filename)\r
-               , graph_(diagnostics::create_graph(narrow(print())))\r
+               , graph_(diagnostics::create_graph(""))\r
                , frame_factory_(frame_factory)         \r
-               , input_(safe_ptr<diagnostics::graph>(graph_), filename_, loop, start, length)\r
+               , format_desc_(frame_factory->get_video_format_desc())\r
+               , input_(graph_, filename_, loop, start, length)\r
+               , video_decoder_(input_.context(), frame_factory, filter)\r
+               , audio_decoder_(input_.context(), frame_factory->get_video_format_desc())\r
+               , fps_(video_decoder_.fps())\r
+               , muxer_(fps_, frame_factory)\r
+               , start_(start)\r
+               , loop_(loop)\r
+               , length_(length)\r
+               , last_frame_(core::basic_frame::empty())\r
+               , width_(video_decoder_.width())\r
+               , height_(video_decoder_.height())\r
+               , is_progressive_(true)\r
        {\r
                graph_->add_guide("frame-time", 0.5);\r
-               graph_->set_color("frame-time",  diagnostics::color(1.0f, 0.0f, 0.0f));\r
-               graph_->set_color("underflow", diagnostics::color(0.6f, 0.3f, 0.9f));           \r
+               graph_->set_color("frame-time", diagnostics::color(0.1f, 1.0f, 0.1f));\r
+               graph_->set_color("underflow", diagnostics::color(0.6f, 0.3f, 0.9f));   \r
                \r
-               double frame_time = 1.0f/input_.fps();\r
-               double format_frame_time = 1.0/frame_factory->get_video_format_desc().fps;\r
-               if(abs(frame_time - format_frame_time) > 0.0001 && abs(frame_time - format_frame_time/2) > 0.0001)\r
-                       CASPAR_LOG(warning) << print() << L" Invalid framerate detected. This may cause distorted audio during playback. frame-time: " << frame_time;\r
-               \r
-               video_decoder_.reset(input_.get_video_codec_context() ? \r
-                       new video_decoder(input_, frame_factory, narrow(filter_str)) : nullptr);\r
-                       \r
-               audio_decoder_.reset(input_.get_audio_codec_context() ? \r
-                       new audio_decoder(input_, frame_factory->get_video_format_desc()) : nullptr);           \r
-                                       \r
-               // Fill buffers.\r
-               for(size_t n = 0; n < 2; ++n)\r
-                       decode_packets();\r
+               for(int n = 0; n < 3; ++n)\r
+                       frame_factory->create_frame(this, std::max<size_t>(2, video_decoder_.width()), std::max<size_t>(2, video_decoder_.height()));\r
        }\r
-\r
-       virtual safe_ptr<core::basic_frame> receive()\r
+                       \r
+       virtual safe_ptr<core::basic_frame> receive(int hints)\r
        {\r
+               auto frame = core::basic_frame::late();\r
+               \r
                frame_timer_.restart();\r
+               \r
+               for(int n = 0; n < 64 && muxer_.empty(); ++n)\r
+                       decode_frame(hints);\r
+               \r
+               graph_->update_value("frame-time", static_cast<float>(frame_timer_.elapsed()*format_desc_.fps*0.5));\r
 \r
-               auto result = decode_frame();\r
-                               \r
-               graph_->update_value("frame-time", static_cast<float>(frame_timer_.elapsed()*frame_factory_->get_video_format_desc().fps*0.5));\r
-                                       \r
-               return result;\r
+               if(!muxer_.empty())\r
+                       frame = last_frame_ = muxer_.pop();     \r
+               else\r
+               {\r
+                       if(input_.eof())\r
+                               return core::basic_frame::eof();\r
+                       else                    \r
+                               graph_->add_tag("underflow");   \r
+               }\r
+\r
+               graph_->update_text(narrow(print()));\r
+               \r
+               return frame;\r
        }\r
-       \r
-       virtual std::wstring print() const\r
+\r
+       virtual safe_ptr<core::basic_frame> last_frame() const\r
        {\r
-               return L"ffmpeg[" + boost::filesystem::wpath(filename_).filename() + L"]";\r
+               return disable_audio(last_frame_);\r
        }\r
 \r
-       void decode_packets()\r
+       void push_packets()\r
        {\r
-               tbb::parallel_invoke\r
-               (\r
-                       [&]\r
-                       {\r
-                               if(video_decoder_ && video_frames_.size() < 3)\r
-                                       boost::range::push_back(video_frames_, video_decoder_->receive());              \r
-                       }, \r
-                       [&]\r
+               for(int n = 0; n < 16 && ((!muxer_.video_ready() && !video_decoder_.ready()) || (!muxer_.audio_ready() && !audio_decoder_.ready())); ++n) \r
+               {\r
+                       std::shared_ptr<AVPacket> pkt;\r
+                       if(input_.try_pop(pkt))\r
                        {\r
-                               if(audio_decoder_ && audio_chunks_.size() < 3)\r
-                                       boost::range::push_back(audio_chunks_, audio_decoder_->receive());                              \r
+                               video_decoder_.push(pkt);\r
+                               audio_decoder_.push(pkt);\r
                        }\r
-               );\r
-               \r
-               // If video is on first frame, sync with audio\r
-               if(audio_decoder_ && video_decoder_ && !video_frames_.empty() && !audio_chunks_.empty() &&\r
-                  video_frames_.front().first == 0 && audio_chunks_.front().first != 0)\r
-               {\r
-                       audio_decoder_->restart(); // Notify decoder to wait for eof which was sent with video eof.\r
-                       audio_chunks_ = audio_decoder_->receive();              \r
                }\r
-               \r
-               CASPAR_ASSERT(!(video_decoder_ && audio_decoder_ && !video_frames_.empty() && !audio_chunks_.empty()) ||\r
-                                     video_frames_.front().first == audio_chunks_.front().first);\r
        }\r
 \r
-       // FIXME: Don't re-interlace when going from 50i to 50p, maybe do this inside decoder?\r
-       safe_ptr<core::basic_frame> get_video_frame(std::vector<int16_t>&& audio_chunk)\r
+       void decode_frame(int hints)\r
        {\r
-               auto frame = std::move(video_frames_.front().second);   \r
-               auto frame_number = video_frames_.front().first;\r
-               video_frames_.pop_front();\r
-                               \r
-               frame->audio_data() = std::move(audio_chunk);\r
-               if(frame->audio_data().empty())\r
-                       frame->get_audio_transform().set_has_audio(false);      \r
-\r
-               if(!video_frames_.empty()) // interlace if we have double frames\r
+               push_packets();\r
+               \r
+               tbb::parallel_invoke(\r
+               [&]\r
                {\r
-                       if(video_frames_.front().first == frame_number)\r
-                       {\r
-                               auto frame2 = std::move(video_frames_.front().second);  \r
-                               video_frames_.pop_front();\r
-                               frame2->get_audio_transform().set_has_audio(false);     \r
+                       if(muxer_.video_ready())\r
+                               return;\r
 \r
-                               return core::basic_frame::interlace(frame, frame2, frame_factory_->get_video_format_desc().mode);\r
+                       auto video_frames = video_decoder_.poll();\r
+                       BOOST_FOREACH(auto& video, video_frames)        \r
+                       {\r
+                               is_progressive_ = video ? video->interlaced_frame == 0 : is_progressive_;\r
+                               muxer_.push(video, hints);      \r
                        }\r
-               }\r
+               },\r
+               [&]\r
+               {\r
+                       if(muxer_.audio_ready())\r
+                               return;\r
+                                       \r
+                       auto audio_samples = audio_decoder_.poll();\r
+                       BOOST_FOREACH(auto& audio, audio_samples)\r
+                               muxer_.push(audio);                             \r
+               });\r
 \r
-               return frame;\r
+               muxer_.commit();\r
        }\r
 \r
-       safe_ptr<core::basic_frame> decode_frame()\r
+       virtual int64_t nb_frames() const \r
        {\r
-               decode_packets();\r
+               if(loop_)\r
+                       return std::numeric_limits<int64_t>::max();\r
 \r
-               if(video_decoder_ && audio_decoder_ && !video_frames_.empty() && !audio_chunks_.empty())\r
-               {\r
-                       auto audio_chunk = std::move(audio_chunks_.front().second);\r
-                       audio_chunks_.pop_front();\r
-                                               \r
-                       return get_video_frame(std::move(audio_chunk));\r
-               }\r
-               else if(video_decoder_ && !audio_decoder_ && !video_frames_.empty())\r
-               {                                               \r
-                       return get_video_frame(std::vector<int16_t>());\r
-               }\r
-               else if(audio_decoder_ && !video_decoder_ && !audio_chunks_.empty())\r
-               {\r
-                       auto frame = frame_factory_->create_frame(this, 1, 1);\r
-                       std::fill(frame->image_data().begin(), frame->image_data().end(), 0);\r
-                               \r
-                       frame->audio_data() = std::move(audio_chunks_.front().second);\r
-                       audio_chunks_.pop_front();\r
+               // This function estimates nb_frames until input has read all packets for one loop, at which point the count should be accurate.\r
 \r
-                       return frame;\r
-               }\r
-               else if(!input_.is_running() || (!video_decoder_ && !audio_decoder_))\r
-               {\r
-                       return core::basic_frame::eof();\r
-               }\r
-               else\r
+               int64_t nb_frames = input_.nb_frames();\r
+               if(input_.nb_loops() < 1) // input still hasn't counted all frames\r
                {\r
-                       graph_->add_tag("underflow");\r
-                       return core::basic_frame::late();\r
+                       int64_t video_nb_frames = video_decoder_.nb_frames();\r
+                       int64_t audio_nb_frames = audio_decoder_.nb_frames();\r
+\r
+                       nb_frames = std::min(static_cast<int64_t>(length_), std::max(nb_frames, std::max(video_nb_frames, audio_nb_frames)));\r
                }\r
+\r
+               nb_frames = muxer_.calc_nb_frames(nb_frames);\r
+\r
+               // TODO: Might need to scale nb_frames av frame_muxer transformations.\r
+\r
+               return nb_frames - start_;\r
+       }\r
+                               \r
+       virtual std::wstring print() const\r
+       {\r
+               return L"ffmpeg[" + boost::filesystem::wpath(filename_).filename() + L"|" \r
+                                                 + boost::lexical_cast<std::wstring>(width_) + L"x" + boost::lexical_cast<std::wstring>(height_)\r
+                                                 + (is_progressive_ ? L"p" : L"i")  + boost::lexical_cast<std::wstring>(is_progressive_ ? fps_ : 2.0 * fps_)\r
+                                                 + L"]";\r
        }\r
 };\r
 \r
-safe_ptr<core::frame_producer> create_ffmpeg_producer(const safe_ptr<core::frame_factory>& frame_factory, const std::vector<std::wstring>& params)\r
+safe_ptr<core::frame_producer> create_producer(const safe_ptr<core::frame_factory>& frame_factory, const std::vector<std::wstring>& params)\r
 {              \r
        static const std::vector<std::wstring> extensions = boost::assign::list_of\r
-               (L"mpg")(L"mpeg")(L"avi")(L"mov")(L"qt")(L"webm")(L"dv")(L"mp4")(L"f4v")(L"flv")(L"mkv")(L"mka")(L"wmv")(L"wma")(L"ogg")(L"divx")(L"xvid")(L"wav")(L"mp3")(L"m2v");\r
+               (L"mpg")(L"mpeg")(L"m2v")(L"m4v")(L"mp3")(L"mp4")(L"mpga")\r
+               (L"avi")\r
+               (L"mov")\r
+               (L"qt")\r
+               (L"webm")\r
+               (L"dv")         \r
+               (L"f4v")(L"flv")\r
+               (L"mkv")(L"mka")\r
+               (L"wmv")(L"wma")(L"wav")\r
+               (L"rm")(L"ram")\r
+               (L"ogg")(L"ogv")(L"oga")(L"ogx")\r
+               (L"divx")(L"xvid");\r
+\r
        std::wstring filename = env::media_folder() + L"\\" + params[0];\r
        \r
        auto ext = boost::find_if(extensions, [&](const std::wstring& ex)\r
@@ -209,29 +235,16 @@ safe_ptr<core::frame_producer> create_ffmpeg_producer(const safe_ptr<core::frame
        if(ext == extensions.end())\r
                return core::frame_producer::empty();\r
 \r
-       std::wstring path = filename + L"." + *ext;\r
-       bool loop = boost::find(params, L"LOOP") != params.end();\r
-\r
-       int start = -1;\r
-       int length = -1;\r
-       \r
-       auto seek_it = std::find(params.begin(), params.end(), L"SEEK");\r
-       if(seek_it != params.end())\r
-       {\r
-               if(++seek_it != params.end())\r
-                       start = boost::lexical_cast<int>(*seek_it);\r
-       }\r
+       auto path               = filename + L"." + *ext;\r
+       auto loop               = boost::range::find(params, L"LOOP") != params.end();\r
+       auto start              = core::get_param(L"SEEK", params, 0);\r
+       auto length             = core::get_param(L"LENGTH", params, std::numeric_limits<size_t>::max());\r
+       auto filter_str = core::get_param<std::wstring>(L"FILTER", params, L"");        \r
+               \r
+       boost::replace_all(filter_str, L"DEINTERLACE", L"YADIF=0:-1");\r
+       boost::replace_all(filter_str, L"DEINTERLACE_BOB", L"YADIF=1:-1");\r
        \r
-       std::wstring filter_str = L"";\r
-\r
-       auto filter_it = std::find(params.begin(), params.end(), L"FILTER");\r
-       if(filter_it != params.end())\r
-       {\r
-               if(++filter_it != params.end())\r
-                       filter_str = *filter_it;\r
-       }\r
-\r
        return make_safe<ffmpeg_producer>(frame_factory, path, filter_str, loop, start, length);\r
 }\r
 \r
-}
\ No newline at end of file
+}}
\ No newline at end of file