]> git.sesse.net Git - casparcg/blobdiff - modules/ffmpeg/producer/ffmpeg_producer.cpp
2.0. Updated namespaces.
[casparcg] / modules / ffmpeg / producer / ffmpeg_producer.cpp
index 6cc704a48d2870aa315d34fdc191c428666eb738..480e95b8d0d3739eb4db855e35613adc428ca9fc 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/color/color_producer.h>\r
-\r
-#include <common/env.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 <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
-\r
-#include <tbb/task_group.h>\r
-\r
-#include <deque>\r
-#include <vector>\r
-\r
-namespace caspar {\r
-\r
-struct display_mode\r
-{\r
-       enum type\r
-       {\r
-               simple,\r
-               duplicate,\r
-               half,\r
-               interlace,\r
-               deinterlace,\r
-               deinterlace_half,\r
-               count,\r
-               invalid\r
-       };\r
-\r
-       static std::wstring print(display_mode::type value)\r
-       {\r
-               switch(value)\r
-               {\r
-                       case simple:\r
-                               return L"simple";\r
-                       case duplicate:\r
-                               return L"duplicate";\r
-                       case half:\r
-                               return L"half";\r
-                       case interlace:\r
-                               return L"interlace";\r
-                       case deinterlace:\r
-                               return L"deinterlace";\r
-                       case deinterlace_half:\r
-                               return L"deinterlace_half";\r
-                       default:\r
-                               return L"invalid";\r
-               }\r
-       }\r
-};\r
-\r
-display_mode::type get_display_mode(const core::video_mode::type in_mode, double in_fps, const core::video_mode::type out_mode, double out_fps)\r
-{              \r
-       if(in_mode == core::video_mode::invalid || out_mode == core::video_mode::invalid)\r
-               return display_mode::invalid;\r
-\r
-       static const auto epsilon = 2.0;\r
-\r
-       if(std::abs(in_fps - out_fps) < epsilon)\r
-       {\r
-               if(in_mode != core::video_mode::progressive && out_mode == core::video_mode::progressive)\r
-                       return display_mode::deinterlace_half;\r
-               //else if(in_mode == core::video_mode::progressive && out_mode != core::video_mode::progressive)\r
-               //      simple(); // interlace_duplicate();\r
-               else\r
-                       return display_mode::simple;\r
-       }\r
-       else if(std::abs(in_fps/2.0 - out_fps) < epsilon)\r
-       {\r
-               if(in_mode != core::video_mode::progressive)\r
-                       return display_mode::invalid;\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
-               if(out_mode != core::video_mode::progressive)\r
-                       return display_mode::interlace;\r
-               else\r
-                       return display_mode::half;\r
-       }\r
-       else if(std::abs(in_fps - out_fps/2.0) < epsilon)\r
-       {\r
-               if(out_mode != core::video_mode::progressive)\r
-                       return display_mode::invalid;\r
+#include <tbb/parallel_invoke.h>\r
 \r
-               if(in_mode != core::video_mode::progressive)\r
-                       return display_mode::deinterlace;\r
-               else\r
-                       return display_mode::duplicate;\r
-       }\r
-\r
-       return display_mode::invalid;\r
-}\r
-               \r
+namespace caspar { namespace ffmpeg {\r
+                               \r
 struct ffmpeg_producer : public core::frame_producer\r
 {\r
        const std::wstring                                                              filename_;\r
        \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 core::video_format_desc                                   format_desc_;\r
 \r
        input                                                                                   input_; \r
        video_decoder                                                                   video_decoder_;\r
-       audio_decoder                                                                   audio_decoder_;\r
-\r
-       std::queue<safe_ptr<core::basic_frame>>                 frame_buffer_;\r
-       std::queue<safe_ptr<core::basic_frame>>                 output_buffer_;\r
+       audio_decoder                                                                   audio_decoder_; \r
+       double                                                                                  fps_;\r
+       frame_muxer                                                                             muxer_;\r
 \r
-       const bool                                                                              auto_convert_;\r
-       display_mode::type                                                              display_mode_;\r
+       const int                                                                               start_;\r
+       const bool                                                                              loop_;\r
+       const size_t                                                                    length_;\r
 \r
-       std::deque<safe_ptr<core::write_frame>>                 video_frames_;\r
-       std::deque<std::vector<int16_t>>                                audio_chunks_;\r
-\r
-       tbb::task_group                                                                 tasks_;\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([this]{return print();}))\r
                , frame_factory_(frame_factory)         \r
                , format_desc_(frame_factory->get_video_format_desc())\r
-               , input_(safe_ptr<diagnostics::graph>(graph_), filename_, loop, start, length)\r
-               , video_decoder_(input_.stream(AVMEDIA_TYPE_VIDEO), frame_factory)\r
-               , audio_decoder_(input_.stream(AVMEDIA_TYPE_AUDIO), frame_factory->get_video_format_desc())\r
-               , auto_convert_(env::properties().get("configuration.ffmpeg.auto-mode", false))\r
-               , display_mode_(display_mode::invalid)\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
-               for(int n = 0; n < 128 && frame_buffer_.size() < 4; ++n)\r
-                       decode_frame();\r
-       }\r
+               // Do some pre-work in order to not block rendering thread for initialization and allocations.\r
 \r
-       ~ffmpeg_producer()\r
-       {\r
-               tasks_.cancel();\r
-               tasks_.wait();\r
+               push_packets();\r
+               auto video_frames = video_decoder_.poll();\r
+               if(!video_frames.empty())\r
+               {\r
+                       auto& video_frame = video_frames.front();\r
+                       auto  desc                = get_pixel_format_desc(static_cast<PixelFormat>(video_frame->format), video_frame->width, video_frame->height);\r
+                       if(desc.pix_fmt == core::pixel_format::invalid)\r
+                               get_pixel_format_desc(PIX_FMT_BGRA, video_frame->width, video_frame->height);\r
+                       \r
+                       for(int n = 0; n < 3; ++n)\r
+                               frame_factory->create_frame(this, desc);\r
+               }\r
+               BOOST_FOREACH(auto& video, video_frames)                        \r
+                       muxer_.push(video, 0);  \r
        }\r
-\r
-       virtual safe_ptr<core::basic_frame> receive()\r
+                       \r
+       virtual safe_ptr<core::basic_frame> receive(int hints)\r
        {\r
-               if(output_buffer_.empty())\r
-               {       \r
-                       tasks_.wait();\r
-\r
-                       output_buffer_ = std::move(frame_buffer_);\r
-\r
-                       tasks_.run([=]\r
-                       {\r
-                               frame_timer_.restart();\r
-\r
-                               for(int n = 0; n < 64 && frame_buffer_.empty(); ++n)\r
-                                       decode_frame();\r
-\r
-                               graph_->update_value("frame-time", static_cast<float>(frame_timer_.elapsed()*format_desc_.fps*0.5));\r
-                       });\r
-               }\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
-               if(output_buffer_.empty())\r
-               {\r
-                       if(input_.eof())\r
-                               frame = core::basic_frame::eof();\r
-                       else\r
-                               graph_->add_tag("underflow");                   \r
-               }\r
+               if(!muxer_.empty())\r
+                       frame = last_frame_ = muxer_.pop();     \r
                else\r
                {\r
-                       frame = output_buffer_.front();\r
-                       output_buffer_.pop();\r
+                       if(input_.eof())\r
+                               return core::basic_frame::eof();\r
+                       else                    \r
+                               graph_->add_tag("underflow");   \r
                }\r
                \r
                return frame;\r
        }\r
 \r
-       void decode_frame()\r
+       virtual safe_ptr<core::basic_frame> last_frame() const\r
+       {\r
+               return disable_audio(last_frame_);\r
+       }\r
+\r
+       void push_packets()\r
        {\r
-               for(int n = 0; n < 32 && ((video_frames_.size() < 2 && !video_decoder_.ready()) ||      (audio_chunks_.size() < 2 && !audio_decoder_.ready())); ++n) \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
@@ -223,160 +157,86 @@ public:
                                audio_decoder_.push(pkt);\r
                        }\r
                }\r
+       }\r
+\r
+       void decode_frame(int hints)\r
+       {\r
+               push_packets();\r
                \r
                tbb::parallel_invoke(\r
-               [=]\r
+               [&]\r
                {\r
-                       if(video_frames_.size() < 2)\r
-                               boost::range::push_back(video_frames_, video_decoder_.poll());\r
+                       if(muxer_.video_ready())\r
+                               return;\r
+\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(audio_chunks_.size() < 2)\r
-                               boost::range::push_back(audio_chunks_, audio_decoder_.poll());\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
-               if(video_frames_.empty() || audio_chunks_.empty())\r
-                       return;\r
-\r
-               if(auto_convert_)\r
-                       auto_convert();\r
-               else\r
-                       simple();\r
-       }\r
-\r
-       void auto_convert()\r
-       {\r
-               auto current_display_mode = get_display_mode(video_decoder_.mode(), video_decoder_.fps(),  format_desc_.mode,  format_desc_.fps);               \r
-               if(current_display_mode != display_mode_)\r
-               {\r
-                       display_mode_ = current_display_mode;\r
-                       CASPAR_LOG(info) << print() << " display_mode: " << display_mode::print(display_mode_) << \r
-                               L" in: " << core::video_mode::print(video_decoder_.mode()) << L" " << video_decoder_.fps() << " fps" <<\r
-                               L" out: " << core::video_mode::print(format_desc_.mode) << L" " << format_desc_.fps << " fps";\r
-               }\r
-\r
-               switch(display_mode_)\r
-               {\r
-               case display_mode::simple:\r
-                       return simple();\r
-               case display_mode::duplicate:\r
-                       return duplicate();\r
-               case display_mode::half:\r
-                       return half();\r
-               case display_mode::interlace:\r
-                       return interlace();\r
-               case display_mode::deinterlace:\r
-                       return deinterlace();\r
-               case display_mode::deinterlace_half:\r
-                       return deinterlace_half();\r
-               default:\r
-                       BOOST_THROW_EXCEPTION(invalid_operation());\r
-               }\r
+               muxer_.commit();\r
        }\r
 \r
-       void simple()\r
+       virtual int64_t nb_frames() const \r
        {\r
-               CASPAR_ASSERT(!video_frames_.empty());\r
-               CASPAR_ASSERT(!audio_chunks_.empty());\r
-\r
-               auto frame1 = video_frames_.front();\r
-               video_frames_.pop_front();\r
-\r
-               frame1->audio_data() = audio_chunks_.front();\r
-               audio_chunks_.pop_front();\r
-\r
-               frame_buffer_.push(frame1);\r
-       }\r
-\r
-       void duplicate()\r
-       {               \r
-               CASPAR_ASSERT(!video_frames_.empty());\r
-               CASPAR_ASSERT(!audio_chunks_.empty());\r
-\r
-               auto frame = video_frames_.front();\r
-               video_frames_.pop_front();\r
-\r
-               auto frame1 = make_safe<core::write_frame>(*frame); // make a copy\r
-               frame1->audio_data() = audio_chunks_.front();\r
-               audio_chunks_.pop_front();\r
+               if(loop_)\r
+                       return std::numeric_limits<int64_t>::max();\r
 \r
-               auto frame2 = frame;\r
-               frame2->audio_data() = audio_chunks_.front();\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
-               frame_buffer_.push(frame1);\r
-               frame_buffer_.push(frame2);\r
-       }\r
-\r
-       void half()\r
-       {       \r
-               CASPAR_ASSERT(!video_frames_.empty());\r
-               CASPAR_ASSERT(!audio_chunks_.empty());\r
-\r
-               if(video_frames_.size() < 2 && !input_.eof())\r
-                       return;\r
-               \r
-               if(video_frames_.size() < 2)\r
-                       video_frames_.push_back(create_color_frame(this, frame_factory_, L"#00000000"));\r
-\r
-               CASPAR_ASSERT(video_frames_.size() == 2);\r
-                               \r
-               auto frame1 =video_frames_.front();\r
-               video_frames_.pop_front();\r
-               frame1->audio_data() = audio_chunks_.front();\r
-               audio_chunks_.pop_front();\r
-                               \r
-               video_frames_.pop_front(); // Throw away\r
-\r
-               frame_buffer_.push(frame1);\r
-       }\r
-       \r
-       void interlace()\r
-       {               \r
-               CASPAR_ASSERT(!video_frames_.empty());\r
-               CASPAR_ASSERT(!audio_chunks_.empty());\r
-\r
-               if(video_frames_.size() < 2 && !input_.eof())\r
-                       return;\r
-\r
-               if(video_frames_.size() < 2)\r
-                       video_frames_.push_back(create_color_frame(this, frame_factory_, L"#00000000"));\r
-               \r
-               CASPAR_ASSERT(video_frames_.size() == 2);\r
+               int64_t nb_frames = input_.nb_frames();\r
+               if(input_.nb_loops() < 1) // input still hasn't counted all frames\r
+               {\r
+                       int64_t video_nb_frames = video_decoder_.nb_frames();\r
+                       int64_t audio_nb_frames = audio_decoder_.nb_frames();\r
 \r
-               auto frame1 = video_frames_.front();\r
-               video_frames_.pop_front();\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
-               frame1->audio_data() = audio_chunks_.front();\r
-               audio_chunks_.pop_front();\r
-                               \r
-               auto frame2 = video_frames_.front();\r
-               video_frames_.pop_front();\r
+               nb_frames = muxer_.calc_nb_frames(nb_frames);\r
 \r
-               frame_buffer_.push(core::basic_frame::interlace(frame1, frame2, format_desc_.mode));            \r
-       }\r
-       \r
-       void deinterlace()\r
-       {\r
-               BOOST_THROW_EXCEPTION(not_implemented() << msg_info("deinterlace"));\r
-       }\r
+               // TODO: Might need to scale nb_frames av frame_muxer transformations.\r
 \r
-       void deinterlace_half()\r
-       {\r
-               BOOST_THROW_EXCEPTION(not_implemented() << msg_info("deinterlace_half"));\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
+               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
@@ -387,20 +247,16 @@ 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
+       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
-       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
-\r
-       return make_safe<ffmpeg_producer>(frame_factory, path, L"", loop, start, length);\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