-/*\r
-* Copyright (c) 2011 Sveriges Television AB <info@casparcg.com>\r
-*\r
-* This file is part of CasparCG (www.casparcg.com).\r
-*\r
-* CasparCG is free software: you can redistribute it and/or modify\r
-* it under the terms of the GNU General Public License as published by\r
-* the Free Software Foundation, either version 3 of the License, or\r
-* (at your option) any later version.\r
-*\r
-* CasparCG is distributed in the hope that it will be useful,\r
-* but WITHOUT ANY WARRANTY; without even the implied warranty of\r
-* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the\r
-* GNU General Public License for more details.\r
-*\r
-* You should have received a copy of the GNU General Public License\r
-* along with CasparCG. If not, see <http://www.gnu.org/licenses/>.\r
-*\r
-* Author: Robert Nagy, ronag89@gmail.com\r
-*/\r
- \r
-#include "../StdAfx.h"\r
-\r
-#include "../ffmpeg_error.h"\r
-\r
-#include "ffmpeg_consumer.h"\r
-\r
-#include "../producer/tbb_avcodec.h"\r
-\r
-#include <core/frame/frame.h>\r
-#include <core/mixer/audio/audio_util.h>\r
-#include <core/consumer/frame_consumer.h>\r
-#include <core/video_format.h>\r
-\r
-#include <common/array.h>\r
-#include <common/env.h>\r
-#include <common/except.h>\r
-#include <common/executor.h>\r
-#include <common/diagnostics/graph.h>\r
-#include <common/lock.h>\r
-#include <common/memory.h>\r
-#include <common/param.h>\r
-#include <common/utf.h>\r
-\r
-#include <boost/algorithm/string.hpp>\r
-#include <boost/timer.hpp>\r
-#include <boost/property_tree/ptree.hpp>\r
-#include <boost/filesystem.hpp>\r
-#include <boost/range/algorithm.hpp>\r
-#include <boost/range/algorithm_ext.hpp>\r
-#include <boost/lexical_cast.hpp>\r
-\r
-#include <tbb/spin_mutex.h>\r
-\r
-#include <numeric>\r
-\r
-#if defined(_MSC_VER)\r
-#pragma warning (push)\r
-#pragma warning (disable : 4244)\r
-#endif\r
-extern "C" \r
-{\r
- #define __STDC_CONSTANT_MACROS\r
- #define __STDC_LIMIT_MACROS\r
- #include <libavformat/avformat.h>\r
- #include <libswscale/swscale.h>\r
- #include <libavutil/opt.h>\r
- #include <libavutil/pixdesc.h>\r
- #include <libavutil/parseutils.h>\r
- #include <libavutil/samplefmt.h>\r
- #include <libswresample/swresample.h>\r
-}\r
-#if defined(_MSC_VER)\r
-#pragma warning (pop)\r
-#endif\r
-\r
-namespace caspar { namespace ffmpeg {\r
- \r
-int av_opt_set(void *obj, const char *name, const char *val, int search_flags)\r
-{\r
- AVClass* av_class = *(AVClass**)obj;\r
-\r
- if((strcmp(name, "pix_fmt") == 0 || strcmp(name, "pixel_format") == 0) && strcmp(av_class->class_name, "AVCodecContext") == 0)\r
- {\r
- AVCodecContext* c = (AVCodecContext*)obj; \r
- auto pix_fmt = av_get_pix_fmt(val);\r
- if(pix_fmt == PIX_FMT_NONE)\r
- return -1; \r
- c->pix_fmt = pix_fmt;\r
- return 0;\r
- }\r
- //if((strcmp(name, "r") == 0 || strcmp(name, "frame_rate") == 0) && strcmp(av_class->class_name, "AVCodecContext") == 0)\r
- //{\r
- // AVCodecContext* c = (AVCodecContext*)obj; \r
-\r
- // if(c->codec_type != AVMEDIA_TYPE_VIDEO)\r
- // return -1;\r
-\r
- // AVRational rate;\r
- // int ret = av_parse_video_rate(&rate, val);\r
- // if(ret < 0)\r
- // return ret;\r
-\r
- // c->time_base.num = rate.den;\r
- // c->time_base.den = rate.num;\r
- // return 0;\r
- //}\r
-\r
- return ::av_opt_set(obj, name, val, search_flags);\r
-}\r
-\r
-struct option\r
-{\r
- std::string name;\r
- std::string value;\r
-\r
- option(std::string name, std::string value)\r
- : name(std::move(name))\r
- , value(std::move(value))\r
- {\r
- }\r
-};\r
- \r
-struct output_format\r
-{\r
- AVOutputFormat* format;\r
- int width;\r
- int height;\r
- CodecID vcodec;\r
- CodecID acodec;\r
- int croptop;\r
- int cropbot;\r
-\r
- output_format(const core::video_format_desc& format_desc, const std::string& filename, std::vector<option>& options)\r
- : format(av_guess_format(nullptr, filename.c_str(), nullptr))\r
- , width(format_desc.width)\r
- , height(format_desc.height)\r
- , vcodec(CODEC_ID_NONE)\r
- , acodec(CODEC_ID_NONE)\r
- , croptop(0)\r
- , cropbot(0)\r
- {\r
- if(boost::iequals(boost::filesystem::path(filename).extension().string(), ".dv"))\r
- set_opt("f", "dv");\r
-\r
- boost::range::remove_erase_if(options, [&](const option& o)\r
- {\r
- return set_opt(o.name, o.value);\r
- });\r
- \r
- if(vcodec == CODEC_ID_NONE)\r
- vcodec = format->video_codec;\r
-\r
- if(acodec == CODEC_ID_NONE)\r
- acodec = format->audio_codec;\r
- \r
- if(vcodec == CODEC_ID_NONE)\r
- vcodec = CODEC_ID_H264;\r
- \r
- if(acodec == CODEC_ID_NONE)\r
- acodec = CODEC_ID_PCM_S16LE;\r
- }\r
- \r
- bool set_opt(const std::string& name, const std::string& value)\r
- {\r
- //if(name == "target")\r
- //{ \r
- // enum { PAL, NTSC, FILM, UNKNOWN } norm = UNKNOWN;\r
- // \r
- // if(name.find("pal-") != std::string::npos)\r
- // norm = PAL;\r
- // else if(name.find("ntsc-") != std::string::npos)\r
- // norm = NTSC;\r
-\r
- // if(norm == UNKNOWN)\r
- // CASPAR_THROW_EXCEPTION(invalid_argument() << arg_name_info("target"));\r
- // \r
- // if (name.find("-dv") != std::string::npos) \r
- // {\r
- // set_opt("f", "dv");\r
- // if(norm == PAL)\r
- // {\r
- // set_opt("s", "720x576");\r
- // }\r
- // else\r
- // {\r
- // set_opt("s", "720x480");\r
- // if(height == 486)\r
- // {\r
- // set_opt("croptop", "2");\r
- // set_opt("cropbot", "4");\r
- // }\r
- // }\r
- // set_opt("s", norm == PAL ? "720x576" : "720x480");\r
- // } \r
-\r
- // return true;\r
- //}\r
- //else \r
- if(name == "f")\r
- {\r
- format = av_guess_format(value.c_str(), nullptr, nullptr);\r
-\r
- if(format == nullptr)\r
- CASPAR_THROW_EXCEPTION(invalid_argument() << arg_name_info("f"));\r
-\r
- return true;\r
- }\r
- else if(name == "vcodec" || name == "v:codec")\r
- {\r
- auto c = avcodec_find_encoder_by_name(value.c_str());\r
- if(c == nullptr)\r
- CASPAR_THROW_EXCEPTION(invalid_argument() << arg_name_info("vcodec"));\r
-\r
- vcodec = avcodec_find_encoder_by_name(value.c_str())->id;\r
- return true;\r
-\r
- }\r
- else if(name == "acodec" || name == "a:codec")\r
- {\r
- auto c = avcodec_find_encoder_by_name(value.c_str());\r
- if(c == nullptr)\r
- CASPAR_THROW_EXCEPTION(invalid_argument() << arg_name_info("acodec"));\r
-\r
- acodec = avcodec_find_encoder_by_name(value.c_str())->id;\r
-\r
- return true;\r
- }\r
- else if(name == "s")\r
- {\r
- if(av_parse_video_size(&width, &height, value.c_str()) < 0)\r
- CASPAR_THROW_EXCEPTION(invalid_argument() << arg_name_info("s"));\r
- \r
- return true;\r
- }\r
- else if(name == "croptop")\r
- {\r
- croptop = boost::lexical_cast<int>(value);\r
-\r
- return true;\r
- }\r
- else if(name == "cropbot")\r
- {\r
- cropbot = boost::lexical_cast<int>(value);\r
-\r
- return true;\r
- }\r
- \r
- return false;\r
- }\r
-};\r
-\r
-typedef std::vector<uint8_t, tbb::cache_aligned_allocator<uint8_t>> byte_vector;\r
-\r
-struct ffmpeg_consumer : boost::noncopyable\r
-{ \r
- const spl::shared_ptr<diagnostics::graph> graph_;\r
- const std::string filename_; \r
- const std::shared_ptr<AVFormatContext> oc_;\r
- const core::video_format_desc format_desc_; \r
-\r
- monitor::basic_subject event_subject_;\r
- \r
- tbb::spin_mutex exception_mutex_;\r
- std::exception_ptr exception_;\r
- \r
- std::shared_ptr<AVStream> audio_st_;\r
- std::shared_ptr<AVStream> video_st_;\r
- \r
- byte_vector picture_buffer_;\r
- byte_vector audio_buffer_;\r
- std::shared_ptr<SwrContext> swr_;\r
- std::shared_ptr<SwsContext> sws_;\r
-\r
- int64_t frame_number_;\r
-\r
- output_format output_format_;\r
- \r
- executor executor_;\r
-public:\r
- ffmpeg_consumer(const std::string& filename, const core::video_format_desc& format_desc, std::vector<option> options)\r
- : filename_(filename)\r
- , oc_(avformat_alloc_context(), av_free)\r
- , format_desc_(format_desc)\r
- , frame_number_(0)\r
- , output_format_(format_desc, filename, options)\r
- , executor_(print())\r
- {\r
- check_space();\r
-\r
- // TODO: Ask stakeholders about case where file already exists.\r
- boost::filesystem::remove(boost::filesystem::path(env::media_folder() + u16(filename))); // Delete the file if it exists\r
-\r
- graph_->set_color("frame-time", diagnostics::color(0.1f, 1.0f, 0.1f));\r
- graph_->set_text(print());\r
- diagnostics::register_graph(graph_);\r
-\r
- executor_.set_capacity(8);\r
-\r
- oc_->oformat = output_format_.format;\r
- \r
- strcpy_s(oc_->filename, filename_.c_str());\r
- \r
- // Add the audio and video streams using the default format codecs and initialize the codecs.\r
- video_st_ = add_video_stream(options);\r
- audio_st_ = add_audio_stream(options);\r
- \r
- av_dump_format(oc_.get(), 0, filename_.c_str(), 1);\r
- \r
- // Open the output ffmpeg, if needed.\r
- if (!(oc_->oformat->flags & AVFMT_NOFILE)) \r
- THROW_ON_ERROR2(avio_open(&oc_->pb, filename.c_str(), AVIO_FLAG_WRITE), "[ffmpeg_consumer]");\r
- \r
- THROW_ON_ERROR2(avformat_write_header(oc_.get(), nullptr), "[ffmpeg_consumer]");\r
-\r
- if(options.size() > 0)\r
- {\r
- BOOST_FOREACH(auto& option, options)\r
- CASPAR_LOG(warning) << L"Invalid option: -" << u16(option.name) << L" " << u16(option.value);\r
- }\r
-\r
- CASPAR_LOG(info) << print() << L" Successfully Initialized."; \r
- }\r
-\r
- ~ffmpeg_consumer()\r
- { \r
- executor_.wait();\r
- \r
- LOG_ON_ERROR2(av_write_trailer(oc_.get()), "[ffmpeg_consumer]");\r
- \r
- audio_st_.reset();\r
- video_st_.reset();\r
- \r
- if (!(oc_->oformat->flags & AVFMT_NOFILE)) \r
- LOG_ON_ERROR2(avio_close(oc_->pb), "[ffmpeg_consumer]");\r
-\r
- CASPAR_LOG(info) << print() << L" Successfully Uninitialized."; \r
- }\r
- \r
- // frame_consumer\r
-\r
- bool send(core::const_frame& frame)\r
- {\r
- auto exception = lock(exception_mutex_, [&]\r
- {\r
- return exception_;\r
- });\r
-\r
- if(exception != nullptr)\r
- std::rethrow_exception(exception);\r
- \r
- executor_.begin_invoke([=]\r
- { \r
- encode(frame);\r
- });\r
- \r
- return true;\r
- }\r
-\r
- std::wstring print() const\r
- {\r
- return L"ffmpeg[" + u16(filename_) + L"]";\r
- }\r
- \r
- void subscribe(const monitor::observable::observer_ptr& o)\r
- {\r
- event_subject_.subscribe(o);\r
- }\r
-\r
- void unsubscribe(const monitor::observable::observer_ptr& o)\r
- {\r
- event_subject_.unsubscribe(o);\r
- } \r
-\r
-private:\r
- std::shared_ptr<AVStream> add_video_stream(std::vector<option>& options)\r
- { \r
- if(output_format_.vcodec == CODEC_ID_NONE)\r
- return nullptr;\r
-\r
- auto st = av_new_stream(oc_.get(), 0);\r
- if (!st) \r
- CASPAR_THROW_EXCEPTION(caspar_exception() << msg_info("Could not allocate video-stream.") << boost::errinfo_api_function("av_new_stream")); \r
-\r
- auto encoder = avcodec_find_encoder(output_format_.vcodec);\r
- if (!encoder)\r
- CASPAR_THROW_EXCEPTION(caspar_exception() << msg_info("Codec not found."));\r
-\r
- auto c = st->codec;\r
-\r
- avcodec_get_context_defaults3(c, encoder);\r
- \r
- c->codec_id = output_format_.vcodec;\r
- c->codec_type = AVMEDIA_TYPE_VIDEO;\r
- c->width = output_format_.width;\r
- c->height = output_format_.height - output_format_.croptop - output_format_.cropbot;\r
- c->time_base.den = format_desc_.time_scale;\r
- c->time_base.num = format_desc_.duration;\r
- c->gop_size = 25;\r
- c->flags |= format_desc_.field_mode == core::field_mode::progressive ? 0 : (CODEC_FLAG_INTERLACED_ME | CODEC_FLAG_INTERLACED_DCT);\r
- c->pix_fmt = c->pix_fmt != PIX_FMT_NONE ? c->pix_fmt : PIX_FMT_YUV420P;\r
-\r
- if(c->codec_id == CODEC_ID_PRORES)\r
- { \r
- c->bit_rate = output_format_.width < 1280 ? 63*1000000 : 220*1000000;\r
- c->pix_fmt = PIX_FMT_YUV422P10;\r
- }\r
- else if(c->codec_id == CODEC_ID_DNXHD)\r
- {\r
- if(c->width < 1280 || c->height < 720)\r
- CASPAR_THROW_EXCEPTION(caspar_exception() << msg_info("Unsupported video dimensions."));\r
-\r
- c->bit_rate = 220*1000000;\r
- c->pix_fmt = PIX_FMT_YUV422P;\r
- }\r
- else if(c->codec_id == CODEC_ID_DVVIDEO)\r
- {\r
- c->width = c->height == 1280 ? 960 : c->width;\r
- \r
- if(format_desc_.format == core::video_format::ntsc)\r
- {\r
- c->pix_fmt = PIX_FMT_YUV411P;\r
- output_format_.croptop = 2;\r
- output_format_.cropbot = 4;\r
- c->height = output_format_.height - output_format_.croptop - output_format_.cropbot;\r
- }\r
- else if(format_desc_.format == core::video_format::pal)\r
- c->pix_fmt = PIX_FMT_YUV420P;\r
- else // dv50\r
- c->pix_fmt = PIX_FMT_YUV422P;\r
- \r
- if(format_desc_.duration == 1001) \r
- c->width = c->height == 1080 ? 1280 : c->width; \r
- else\r
- c->width = c->height == 1080 ? 1440 : c->width; \r
- }\r
- else if(c->codec_id == CODEC_ID_H264)\r
- { \r
- c->pix_fmt = PIX_FMT_YUV420P; \r
- av_opt_set(c->priv_data, "preset", "ultrafast", 0);\r
- av_opt_set(c->priv_data, "tune", "fastdecode", 0);\r
- av_opt_set(c->priv_data, "crf", "5", 0);\r
- }\r
- else if(c->codec_id == CODEC_ID_QTRLE)\r
- {\r
- c->pix_fmt = PIX_FMT_ARGB;\r
- }\r
- \r
- boost::range::remove_erase_if(options, [&](const option& o)\r
- {\r
- return o.name.at(0) != 'a' && ffmpeg::av_opt_set(c, o.name.c_str(), o.value.c_str(), AV_OPT_SEARCH_CHILDREN) > -1;\r
- });\r
- \r
- if(output_format_.format->flags & AVFMT_GLOBALHEADER)\r
- c->flags |= CODEC_FLAG_GLOBAL_HEADER;\r
- \r
- THROW_ON_ERROR2(tbb_avcodec_open(c, encoder), "[ffmpeg_consumer]");\r
-\r
- return std::shared_ptr<AVStream>(st, [](AVStream* st)\r
- {\r
- LOG_ON_ERROR2(tbb_avcodec_close(st->codec), "[ffmpeg_consumer]");\r
- av_freep(&st->codec);\r
- av_freep(&st);\r
- });\r
- }\r
- \r
- std::shared_ptr<AVStream> add_audio_stream(std::vector<option>& options)\r
- {\r
- if(output_format_.acodec == CODEC_ID_NONE)\r
- return nullptr;\r
-\r
- auto st = av_new_stream(oc_.get(), 1);\r
- if(!st)\r
- CASPAR_THROW_EXCEPTION(caspar_exception() << msg_info("Could not allocate audio-stream") << boost::errinfo_api_function("av_new_stream")); \r
- \r
- auto encoder = avcodec_find_encoder(output_format_.acodec);\r
- if (!encoder)\r
- CASPAR_THROW_EXCEPTION(caspar_exception() << msg_info("codec not found"));\r
- \r
- auto c = st->codec;\r
-\r
- avcodec_get_context_defaults3(c, encoder);\r
-\r
- c->codec_id = output_format_.acodec;\r
- c->codec_type = AVMEDIA_TYPE_AUDIO;\r
- c->sample_rate = 48000;\r
- c->channels = 2;\r
- c->sample_fmt = AV_SAMPLE_FMT_S16;\r
- c->time_base.num = 1;\r
- c->time_base.den = c->sample_rate;\r
-\r
- if(output_format_.vcodec == CODEC_ID_FLV1) \r
- c->sample_rate = 44100; \r
-\r
- if(output_format_.format->flags & AVFMT_GLOBALHEADER)\r
- c->flags |= CODEC_FLAG_GLOBAL_HEADER;\r
- \r
- boost::range::remove_erase_if(options, [&](const option& o)\r
- {\r
- return ffmpeg::av_opt_set(c, o.name.c_str(), o.value.c_str(), AV_OPT_SEARCH_CHILDREN) > -1;\r
- });\r
-\r
- THROW_ON_ERROR2(avcodec_open(c, encoder), "[ffmpeg_consumer]");\r
-\r
- return std::shared_ptr<AVStream>(st, [](AVStream* st)\r
- {\r
- LOG_ON_ERROR2(avcodec_close(st->codec), "[ffmpeg_consumer]");;\r
- av_freep(&st->codec);\r
- av_freep(&st);\r
- });\r
- }\r
- \r
- void encode_video_frame(core::const_frame frame)\r
- { \r
- if(!video_st_)\r
- return;\r
- \r
- auto enc = video_st_->codec;\r
- \r
- auto av_frame = convert_video(frame, enc);\r
- av_frame->interlaced_frame = format_desc_.field_mode != core::field_mode::progressive;\r
- av_frame->top_field_first = format_desc_.field_mode == core::field_mode::upper;\r
- av_frame->pts = frame_number_++;\r
-\r
- event_subject_ << monitor::event("frame") % static_cast<int64_t>(frame_number_)\r
- % static_cast<int64_t>(std::numeric_limits<int64_t>::max());\r
-\r
- AVPacket pkt;\r
- av_init_packet(&pkt);\r
- pkt.data = nullptr;\r
- pkt.size = 0;\r
-\r
- int got_packet = 0;\r
- THROW_ON_ERROR2(avcodec_encode_video2(enc, &pkt, av_frame.get(), &got_packet), "[ffmpeg_consumer]");\r
- std::shared_ptr<AVPacket> guard(&pkt, av_free_packet);\r
-\r
- if(!got_packet)\r
- return;\r
- \r
- if (pkt.pts != AV_NOPTS_VALUE)\r
- pkt.pts = av_rescale_q(pkt.pts, enc->time_base, video_st_->time_base);\r
- if (pkt.dts != AV_NOPTS_VALUE)\r
- pkt.dts = av_rescale_q(pkt.dts, enc->time_base, video_st_->time_base);\r
- \r
- pkt.stream_index = video_st_->index;\r
- \r
- THROW_ON_ERROR2(av_interleaved_write_frame(oc_.get(), &pkt), "[ffmpeg_consumer]");\r
- }\r
- \r
- uint64_t get_channel_layout(AVCodecContext* dec)\r
- {\r
- auto layout = (dec->channel_layout && dec->channels == av_get_channel_layout_nb_channels(dec->channel_layout)) ? dec->channel_layout : av_get_default_channel_layout(dec->channels);\r
- return layout;\r
- }\r
- \r
- void encode_audio_frame(core::const_frame frame)\r
- { \r
- if(!audio_st_)\r
- return;\r
- \r
- auto enc = audio_st_->codec;\r
-\r
- boost::push_back(audio_buffer_, convert_audio(frame, enc));\r
- \r
- auto frame_size = enc->frame_size != 0 ? enc->frame_size * enc->channels * av_get_bytes_per_sample(enc->sample_fmt) : static_cast<int>(audio_buffer_.size());\r
- \r
- while(audio_buffer_.size() >= frame_size)\r
- { \r
- std::shared_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);\r
- avcodec_get_frame_defaults(av_frame.get()); \r
- av_frame->nb_samples = frame_size / (enc->channels * av_get_bytes_per_sample(enc->sample_fmt));\r
-\r
- AVPacket pkt;\r
- av_init_packet(&pkt);\r
- pkt.data = nullptr;\r
- pkt.size = 0; \r
- \r
- THROW_ON_ERROR2(avcodec_fill_audio_frame(av_frame.get(), enc->channels, enc->sample_fmt, audio_buffer_.data(), frame_size, 1), "[ffmpeg_consumer]");\r
-\r
- int got_packet = 0;\r
- THROW_ON_ERROR2(avcodec_encode_audio2(enc, &pkt, av_frame.get(), &got_packet), "[ffmpeg_consumer]");\r
- std::shared_ptr<AVPacket> guard(&pkt, av_free_packet);\r
- \r
- audio_buffer_.erase(audio_buffer_.begin(), audio_buffer_.begin() + frame_size);\r
-\r
- if(!got_packet)\r
- return;\r
- \r
- if (pkt.pts != AV_NOPTS_VALUE)\r
- pkt.pts = av_rescale_q(pkt.pts, enc->time_base, audio_st_->time_base);\r
- if (pkt.dts != AV_NOPTS_VALUE)\r
- pkt.dts = av_rescale_q(pkt.dts, enc->time_base, audio_st_->time_base);\r
- if (pkt.duration > 0)\r
- pkt.duration = static_cast<int>(av_rescale_q(pkt.duration, enc->time_base, audio_st_->time_base));\r
- \r
- pkt.stream_index = audio_st_->index;\r
- \r
- THROW_ON_ERROR2(av_interleaved_write_frame(oc_.get(), &pkt), "[ffmpeg_consumer]");\r
- }\r
- } \r
- \r
- std::shared_ptr<AVFrame> convert_video(core::const_frame frame, AVCodecContext* c)\r
- {\r
- if(!sws_) \r
- {\r
- sws_.reset(sws_getContext(format_desc_.width, \r
- format_desc_.height - output_format_.croptop - output_format_.cropbot, \r
- PIX_FMT_BGRA,\r
- c->width,\r
- c->height, \r
- c->pix_fmt, \r
- SWS_BICUBIC, nullptr, nullptr, nullptr), \r
- sws_freeContext);\r
- if (sws_ == nullptr) \r
- CASPAR_THROW_EXCEPTION(caspar_exception() << msg_info("Cannot initialize the conversion context"));\r
- }\r
-\r
- std::shared_ptr<AVFrame> in_frame(avcodec_alloc_frame(), av_free);\r
-\r
- avpicture_fill(reinterpret_cast<AVPicture*>(in_frame.get()), \r
- const_cast<uint8_t*>(frame.image_data().begin()),\r
- PIX_FMT_BGRA, \r
- format_desc_.width,\r
- format_desc_.height - output_format_.croptop - output_format_.cropbot);\r
-\r
- for(int n = 0; n < 4; ++n) \r
- in_frame->data[n] += in_frame->linesize[n] * output_format_.croptop; \r
- \r
- picture_buffer_.resize(avpicture_get_size(c->pix_fmt, c->width, c->height));\r
-\r
- std::shared_ptr<AVFrame> out_frame(avcodec_alloc_frame(), av_free);\r
- \r
- avpicture_fill(reinterpret_cast<AVPicture*>(out_frame.get()),\r
- picture_buffer_.data(), \r
- c->pix_fmt, \r
- c->width, \r
- c->height);\r
-\r
- sws_scale(sws_.get(), \r
- in_frame->data, \r
- in_frame->linesize,\r
- 0, \r
- format_desc_.height - output_format_.cropbot - output_format_.croptop, \r
- out_frame->data, \r
- out_frame->linesize);\r
-\r
- return out_frame;\r
- }\r
- \r
- byte_vector convert_audio(core::const_frame& frame, AVCodecContext* c)\r
- {\r
- if(!swr_) \r
- {\r
- swr_ = std::shared_ptr<SwrContext>(swr_alloc_set_opts(nullptr,\r
- get_channel_layout(c), c->sample_fmt, c->sample_rate,\r
- av_get_default_channel_layout(format_desc_.audio_channels), AV_SAMPLE_FMT_S32, format_desc_.audio_sample_rate,\r
- 0, nullptr), [](SwrContext* p){swr_free(&p);});\r
-\r
- if(!swr_)\r
- CASPAR_THROW_EXCEPTION(bad_alloc());\r
-\r
- THROW_ON_ERROR2(swr_init(swr_.get()), "[audio_decoder]");\r
- }\r
- \r
- byte_vector buffer(48000);\r
-\r
- const uint8_t* in[] = {reinterpret_cast<const uint8_t*>(frame.audio_data().data())};\r
- uint8_t* out[] = {buffer.data()};\r
-\r
- auto channel_samples = swr_convert(swr_.get(), \r
- out, static_cast<int>(buffer.size()) / c->channels / av_get_bytes_per_sample(c->sample_fmt), \r
- in, static_cast<int>(frame.audio_data().size()/format_desc_.audio_channels));\r
-\r
- buffer.resize(channel_samples * c->channels * av_get_bytes_per_sample(c->sample_fmt)); \r
-\r
- return buffer;\r
- }\r
-\r
- void check_space()\r
- {\r
- auto space = boost::filesystem::space(boost::filesystem::path(filename_).parent_path());\r
- if(space.available < 512*1000000)\r
- BOOST_THROW_EXCEPTION(file_write_error() << msg_info("out of space"));\r
- }\r
-\r
- void encode(const core::const_frame& frame)\r
- {\r
- try\r
- {\r
- if(frame_number_ % 25 == 0)\r
- check_space();\r
-\r
- boost::timer frame_timer;\r
-\r
- encode_video_frame(frame);\r
- encode_audio_frame(frame);\r
-\r
- graph_->set_value("frame-time", frame_timer.elapsed()*format_desc_.fps*0.5);\r
- }\r
- catch(...)\r
- { \r
- lock(exception_mutex_, [&]\r
- {\r
- exception_ = std::current_exception();\r
- });\r
- }\r
- }\r
-};\r
-\r
-struct ffmpeg_consumer_proxy : public core::frame_consumer\r
-{\r
- const std::wstring filename_;\r
- const std::vector<option> options_;\r
-\r
- std::unique_ptr<ffmpeg_consumer> consumer_;\r
-\r
-public:\r
-\r
- ffmpeg_consumer_proxy(const std::wstring& filename, const std::vector<option>& options)\r
- : filename_(filename)\r
- , options_(options)\r
- {\r
- }\r
- \r
- virtual void initialize(const core::video_format_desc& format_desc, int)\r
- {\r
- if(consumer_)\r
- BOOST_THROW_EXCEPTION(invalid_operation() << msg_info("Cannot reinitialize ffmpeg-consumer."));\r
-\r
- consumer_.reset(new ffmpeg_consumer(u8(filename_), format_desc, options_));\r
- }\r
- \r
- bool send(core::const_frame frame) override\r
- {\r
- return consumer_->send(frame);\r
- }\r
- \r
- std::wstring print() const override\r
- {\r
- return consumer_ ? consumer_->print() : L"[ffmpeg_consumer]";\r
- }\r
-\r
- std::wstring name() const override\r
- {\r
- return L"file";\r
- }\r
-\r
- boost::property_tree::wptree info() const override\r
- {\r
- boost::property_tree::wptree info;\r
- info.add(L"type", L"file");\r
- info.add(L"filename", filename_);\r
- return info;\r
- }\r
- \r
- bool has_synchronization_clock() const override\r
- {\r
- return false;\r
- }\r
-\r
- int buffer_depth() const override\r
- {\r
- return 1;\r
- }\r
-\r
- int index() const override\r
- {\r
- return 200;\r
- }\r
-\r
- void subscribe(const monitor::observable::observer_ptr& o) override\r
- {\r
- consumer_->subscribe(o);\r
- }\r
-\r
- void unsubscribe(const monitor::observable::observer_ptr& o) override\r
- {\r
- consumer_->unsubscribe(o);\r
- } \r
-}; \r
-spl::shared_ptr<core::frame_consumer> create_consumer(const std::vector<std::wstring>& params)\r
-{\r
- auto str = std::accumulate(params.begin(), params.end(), std::wstring(), [](const std::wstring& lhs, const std::wstring& rhs) {return lhs + L" " + rhs;});\r
- \r
- boost::wregex path_exp(L"(FILE)? (?<PATH>.+\\..+).*", boost::regex::icase);\r
-\r
- boost::wsmatch path;\r
- if(!boost::regex_match(str, path, path_exp))\r
- return core::frame_consumer::empty();\r
- \r
- boost::wregex opt_exp(L"-((?<NAME>[^\\s]+)\\s+(?<VALUE>[^\\s]+))"); \r
- \r
- std::vector<option> options;\r
- for(boost::wsregex_iterator it(str.begin(), str.end(), opt_exp); it != boost::wsregex_iterator(); ++it)\r
- {\r
- auto name = u8(boost::trim_copy(boost::to_lower_copy((*it)["NAME"].str())));\r
- auto value = u8(boost::trim_copy(boost::to_lower_copy((*it)["VALUE"].str())));\r
- \r
- if(value == "h264")\r
- value = "libx264";\r
- else if(value == "dvcpro")\r
- value = "dvvideo";\r
-\r
- options.push_back(option(name, value));\r
- }\r
- \r
- return spl::make_shared<ffmpeg_consumer_proxy>(env::media_folder() + path["PATH"].str(), options);\r
-}\r
-\r
-spl::shared_ptr<core::frame_consumer> create_consumer(const boost::property_tree::wptree& ptree)\r
-{\r
- auto filename = ptree.get<std::wstring>(L"path");\r
- auto codec = ptree.get(L"vcodec", L"libx264");\r
-\r
- std::vector<option> options;\r
- options.push_back(option("vcodec", u8(codec)));\r
- \r
- return spl::make_shared<ffmpeg_consumer_proxy>(env::media_folder() + filename, options);\r
-}\r
-\r
-}}\r
+#include "../StdAfx.h"
+
+#include "ffmpeg_consumer.h"
+
+#include "../ffmpeg_error.h"
+#include "../producer/util/util.h"
+#include "../producer/filter/filter.h"
+#include "../producer/filter/audio_filter.h"
+
+#include <common/except.h>
+#include <common/executor.h>
+#include <common/assert.h>
+#include <common/utf.h>
+#include <common/future.h>
+#include <common/diagnostics/graph.h>
+#include <common/env.h>
+#include <common/scope_exit.h>
+#include <common/ptree.h>
+#include <common/param.h>
+#include <common/semaphore.h>
+
+#include <core/consumer/frame_consumer.h>
+#include <core/frame/frame.h>
+#include <core/frame/audio_channel_layout.h>
+#include <core/video_format.h>
+#include <core/monitor/monitor.h>
+#include <core/help/help_repository.h>
+#include <core/help/help_sink.h>
+
+#include <boost/noncopyable.hpp>
+#include <boost/rational.hpp>
+#include <boost/format.hpp>
+#include <boost/algorithm/string/predicate.hpp>
+#include <boost/property_tree/ptree.hpp>
+
+#pragma warning(push)
+#pragma warning(disable: 4244)
+#pragma warning(disable: 4245)
+#include <boost/crc.hpp>
+#pragma warning(pop)
+
+#include <tbb/atomic.h>
+#include <tbb/concurrent_queue.h>
+#include <tbb/parallel_invoke.h>
+#include <tbb/parallel_for.h>
+
+#include <numeric>
+
+#pragma warning(push)
+#pragma warning(disable: 4244)
+
+extern "C"
+{
+ #define __STDC_CONSTANT_MACROS
+ #define __STDC_LIMIT_MACROS
+ #include <libavformat/avformat.h>
+ #include <libavcodec/avcodec.h>
+ #include <libavutil/avutil.h>
+ #include <libavutil/frame.h>
+ #include <libavutil/opt.h>
+ #include <libavutil/imgutils.h>
+ #include <libavutil/parseutils.h>
+ #include <libavfilter/avfilter.h>
+ #include <libavfilter/buffersink.h>
+ #include <libavfilter/buffersrc.h>
+}
+
+#pragma warning(pop)
+
+namespace caspar { namespace ffmpeg {
+
+void set_pixel_format(AVFilterContext* sink, AVPixelFormat pix_fmt)
+{
+#pragma warning (push)
+#pragma warning (disable : 4245)
+
+ FF(av_opt_set_int_list(
+ sink,
+ "pix_fmts",
+ std::vector<AVPixelFormat>({ pix_fmt, AVPixelFormat::AV_PIX_FMT_NONE }).data(),
+ -1,
+ AV_OPT_SEARCH_CHILDREN));
+
+#pragma warning (pop)
+}
+
+void adjust_video_filter(const AVCodec& codec, const core::video_format_desc& in_format, AVFilterContext* sink, std::string& filter)
+{
+ switch (codec.id)
+ {
+ case AV_CODEC_ID_DVVIDEO:
+ // Crop
+ if (in_format.format == core::video_format::ntsc)
+ filter = u8(append_filter(u16(filter), L"crop=720:480:0:2"));
+
+ // Pixel format selection
+ if (in_format.format == core::video_format::ntsc)
+ set_pixel_format(sink, AVPixelFormat::AV_PIX_FMT_YUV411P);
+ else if (in_format.format == core::video_format::pal)
+ set_pixel_format(sink, AVPixelFormat::AV_PIX_FMT_YUV420P);
+ else
+ set_pixel_format(sink, AVPixelFormat::AV_PIX_FMT_YUV422P);
+
+ // Scale
+ if (in_format.height == 1080)
+ filter = u8(append_filter(u16(filter), in_format.duration == 1001
+ ? L"scale=1280:1080"
+ : L"scale=1440:1080"));
+ else if (in_format.height == 720)
+ filter = u8(append_filter(u16(filter), L"scale=960:720"));
+
+ break;
+ }
+}
+
+void setup_codec_defaults(AVCodecContext& encoder)
+{
+ static const int MEGABIT = 1000000;
+
+ switch (encoder.codec_id)
+ {
+ case AV_CODEC_ID_DNXHD:
+ encoder.bit_rate = 220 * MEGABIT;
+
+ break;
+ case AV_CODEC_ID_PRORES:
+ encoder.bit_rate = encoder.width < 1280
+ ? 63 * MEGABIT
+ : 220 * MEGABIT;
+
+ break;
+ case AV_CODEC_ID_H264:
+ av_opt_set(encoder.priv_data, "preset", "ultrafast", 0);
+ av_opt_set(encoder.priv_data, "tune", "fastdecode", 0);
+ av_opt_set(encoder.priv_data, "crf", "5", 0);
+
+ break;
+ }
+}
+
+bool is_pcm_s24le_not_supported(const AVFormatContext& container)
+{
+ auto name = std::string(container.oformat->name);
+
+ if (name == "mp4" || name == "dv")
+ return true;
+
+ return false;
+}
+
+template<typename Out, typename In>
+std::vector<Out> from_terminated_array(const In* array, In terminator)
+{
+ std::vector<Out> result;
+
+ while (array != nullptr && *array != terminator)
+ {
+ In val = *array;
+ Out casted = static_cast<Out>(val);
+
+ result.push_back(casted);
+
+ ++array;
+ }
+
+ return result;
+}
+
+class ffmpeg_consumer
+{
+private:
+ const spl::shared_ptr<diagnostics::graph> graph_;
+ core::monitor::subject subject_;
+ std::string path_;
+ boost::filesystem::path full_path_;
+
+ std::map<std::string, std::string> options_;
+ bool mono_streams_;
+
+ core::video_format_desc in_video_format_;
+ core::audio_channel_layout in_channel_layout_ = core::audio_channel_layout::invalid();
+
+ std::shared_ptr<AVFormatContext> oc_;
+ tbb::atomic<bool> abort_request_;
+
+ std::shared_ptr<AVStream> video_st_;
+ std::vector<std::shared_ptr<AVStream>> audio_sts_;
+
+ std::int64_t video_pts_ = 0;
+ std::int64_t audio_pts_ = 0;
+
+ std::unique_ptr<audio_filter> audio_filter_;
+
+ // TODO: make use of already existent avfilter abstraction for video also
+ AVFilterContext* video_graph_in_;
+ AVFilterContext* video_graph_out_;
+ std::shared_ptr<AVFilterGraph> video_graph_;
+
+ executor video_encoder_executor_;
+ executor audio_encoder_executor_;
+
+ semaphore tokens_ { 0 };
+
+ tbb::atomic<int64_t> current_encoding_delay_;
+
+ executor write_executor_;
+
+public:
+
+ ffmpeg_consumer(
+ std::string path,
+ std::string options,
+ bool mono_streams)
+ : path_(path)
+ , full_path_(path)
+ , mono_streams_(mono_streams)
+ , audio_encoder_executor_(print() + L" audio_encoder")
+ , video_encoder_executor_(print() + L" video_encoder")
+ , write_executor_(print() + L" io")
+ {
+ abort_request_ = false;
+ current_encoding_delay_ = 0;
+
+ for(auto it =
+ boost::sregex_iterator(
+ options.begin(),
+ options.end(),
+ boost::regex("-(?<NAME>[^-\\s]+)(\\s+(?<VALUE>[^\\s]+))?"));
+ it != boost::sregex_iterator();
+ ++it)
+ {
+ options_[(*it)["NAME"].str()] = (*it)["VALUE"].matched ? (*it)["VALUE"].str() : "";
+ }
+
+ if (options_.find("threads") == options_.end())
+ options_["threads"] = "auto";
+
+ tokens_.release(
+ std::max(
+ 1,
+ try_remove_arg<int>(
+ options_,
+ boost::regex("tokens")).get_value_or(2)));
+ }
+
+ ~ffmpeg_consumer()
+ {
+ if(oc_)
+ {
+ video_encoder_executor_.begin_invoke([&] { encode_video(core::const_frame::empty(), nullptr); });
+ audio_encoder_executor_.begin_invoke([&] { encode_audio(core::const_frame::empty(), nullptr); });
+
+ video_encoder_executor_.stop();
+ audio_encoder_executor_.stop();
+ video_encoder_executor_.join();
+ audio_encoder_executor_.join();
+
+ video_graph_.reset();
+ audio_filter_.reset();
+ video_st_.reset();
+ audio_sts_.clear();
+
+ write_packet(nullptr, nullptr);
+
+ write_executor_.stop();
+ write_executor_.join();
+
+ FF(av_write_trailer(oc_.get()));
+
+ if (!(oc_->oformat->flags & AVFMT_NOFILE) && oc_->pb)
+ avio_close(oc_->pb);
+
+ oc_.reset();
+ }
+ }
+
+ void initialize(
+ const core::video_format_desc& format_desc,
+ const core::audio_channel_layout& channel_layout)
+ {
+ try
+ {
+ static boost::regex prot_exp("^.+:.*" );
+
+ if(!boost::regex_match(
+ path_,
+ prot_exp))
+ {
+ if(!full_path_.is_complete())
+ {
+ full_path_ =
+ u8(
+ env::media_folder()) +
+ path_;
+ }
+
+ if(boost::filesystem::exists(full_path_))
+ boost::filesystem::remove(full_path_);
+
+ boost::filesystem::create_directories(full_path_.parent_path());
+ }
+
+ graph_->set_color("frame-time", diagnostics::color(0.1f, 1.0f, 0.1f));
+ graph_->set_color("dropped-frame", diagnostics::color(0.3f, 0.6f, 0.3f));
+ graph_->set_text(print());
+ diagnostics::register_graph(graph_);
+
+ const auto oformat_name =
+ try_remove_arg<std::string>(
+ options_,
+ boost::regex("^f|format$"));
+
+ AVFormatContext* oc;
+
+ FF(avformat_alloc_output_context2(
+ &oc,
+ nullptr,
+ oformat_name && !oformat_name->empty() ? oformat_name->c_str() : nullptr,
+ full_path_.string().c_str()));
+
+ oc_.reset(
+ oc,
+ avformat_free_context);
+
+ CASPAR_VERIFY(oc_->oformat);
+
+ oc_->interrupt_callback.callback = ffmpeg_consumer::interrupt_cb;
+ oc_->interrupt_callback.opaque = this;
+
+ CASPAR_VERIFY(format_desc.format != core::video_format::invalid);
+
+ in_video_format_ = format_desc;
+ in_channel_layout_ = channel_layout;
+
+ CASPAR_VERIFY(oc_->oformat);
+
+ const auto video_codec_name =
+ try_remove_arg<std::string>(
+ options_,
+ boost::regex("^c:v|codec:v|vcodec$"));
+
+ const auto video_codec =
+ video_codec_name
+ ? avcodec_find_encoder_by_name(video_codec_name->c_str())
+ : avcodec_find_encoder(oc_->oformat->video_codec);
+
+ const auto audio_codec_name =
+ try_remove_arg<std::string>(
+ options_,
+ boost::regex("^c:a|codec:a|acodec$"));
+
+ const auto audio_codec =
+ audio_codec_name
+ ? avcodec_find_encoder_by_name(audio_codec_name->c_str())
+ : (is_pcm_s24le_not_supported(*oc_)
+ ? avcodec_find_encoder(oc_->oformat->audio_codec)
+ : avcodec_find_encoder_by_name("pcm_s24le"));
+
+ if (!video_codec)
+ CASPAR_THROW_EXCEPTION(user_error() << msg_info(
+ "Failed to find video codec " + (video_codec_name
+ ? *video_codec_name
+ : "with id " + boost::lexical_cast<std::string>(
+ oc_->oformat->video_codec))));
+ if (!audio_codec)
+ CASPAR_THROW_EXCEPTION(user_error() << msg_info(
+ "Failed to find audio codec " + (audio_codec_name
+ ? *audio_codec_name
+ : "with id " + boost::lexical_cast<std::string>(
+ oc_->oformat->audio_codec))));
+
+ // Filters
+
+ {
+ configure_video_filters(
+ *video_codec,
+ try_remove_arg<std::string>(options_,
+ boost::regex("vf|f:v|filter:v")).get_value_or(""));
+
+ configure_audio_filters(
+ *audio_codec,
+ try_remove_arg<std::string>(options_,
+ boost::regex("af|f:a|filter:a")).get_value_or(""));
+ }
+
+ // Encoders
+
+ {
+ auto video_options = options_;
+ auto audio_options = options_;
+
+ video_st_ = open_encoder(
+ *video_codec,
+ video_options,
+ 0);
+
+ for (int i = 0; i < audio_filter_->get_num_output_pads(); ++i)
+ audio_sts_.push_back(open_encoder(
+ *audio_codec,
+ audio_options,
+ i));
+
+ auto it = options_.begin();
+ while(it != options_.end())
+ {
+ if(video_options.find(it->first) == video_options.end() || audio_options.find(it->first) == audio_options.end())
+ it = options_.erase(it);
+ else
+ ++it;
+ }
+ }
+
+ // Output
+ {
+ AVDictionary* av_opts = nullptr;
+
+ to_dict(
+ &av_opts,
+ std::move(options_));
+
+ CASPAR_SCOPE_EXIT
+ {
+ av_dict_free(&av_opts);
+ };
+
+ if (!(oc_->oformat->flags & AVFMT_NOFILE))
+ {
+ FF(avio_open2(
+ &oc_->pb,
+ full_path_.string().c_str(),
+ AVIO_FLAG_WRITE,
+ &oc_->interrupt_callback,
+ &av_opts));
+ }
+
+ FF(avformat_write_header(
+ oc_.get(),
+ &av_opts));
+
+ options_ = to_map(av_opts);
+ }
+
+ // Dump Info
+
+ av_dump_format(
+ oc_.get(),
+ 0,
+ oc_->filename,
+ 1);
+
+ for (const auto& option : options_)
+ {
+ CASPAR_LOG(warning)
+ << L"Invalid option: -"
+ << u16(option.first)
+ << L" "
+ << u16(option.second);
+ }
+ }
+ catch(...)
+ {
+ video_st_.reset();
+ audio_sts_.clear();
+ oc_.reset();
+ throw;
+ }
+ }
+
+ core::monitor::subject& monitor_output()
+ {
+ return subject_;
+ }
+
+ void send(core::const_frame frame)
+ {
+ CASPAR_VERIFY(in_video_format_.format != core::video_format::invalid);
+
+ auto frame_timer = spl::make_shared<caspar::timer>();
+
+ std::shared_ptr<void> token(
+ nullptr,
+ [this, frame, frame_timer](void*)
+ {
+ tokens_.release();
+ current_encoding_delay_ = frame.get_age_millis();
+ graph_->set_value("frame-time", frame_timer->elapsed() * in_video_format_.fps * 0.5);
+ });
+ tokens_.acquire();
+
+ video_encoder_executor_.begin_invoke([=]() mutable
+ {
+ encode_video(
+ frame,
+ token);
+ });
+
+ audio_encoder_executor_.begin_invoke([=]() mutable
+ {
+ encode_audio(
+ frame,
+ token);
+ });
+ }
+
+ bool ready_for_frame() const
+ {
+ return tokens_.permits() > 0;
+ }
+
+ void mark_dropped()
+ {
+ graph_->set_tag(diagnostics::tag_severity::WARNING, "dropped-frame");
+ }
+
+ std::wstring print() const
+ {
+ return L"ffmpeg_consumer[" + u16(path_) + L"]";
+ }
+
+ int64_t presentation_frame_age_millis() const
+ {
+ return current_encoding_delay_;
+ }
+
+private:
+
+ static int interrupt_cb(void* ctx)
+ {
+ CASPAR_ASSERT(ctx);
+ return reinterpret_cast<ffmpeg_consumer*>(ctx)->abort_request_;
+ }
+
+ std::shared_ptr<AVStream> open_encoder(
+ const AVCodec& codec,
+ std::map<std::string,
+ std::string>& options,
+ int stream_number_for_media_type)
+ {
+ auto st =
+ avformat_new_stream(
+ oc_.get(),
+ &codec);
+
+ if (!st)
+ CASPAR_THROW_EXCEPTION(caspar_exception() << msg_info("Could not allocate video-stream.") << boost::errinfo_api_function("avformat_new_stream"));
+
+ auto enc = st->codec;
+
+ CASPAR_VERIFY(enc);
+
+ switch(enc->codec_type)
+ {
+ case AVMEDIA_TYPE_VIDEO:
+ {
+ enc->time_base = video_graph_out_->inputs[0]->time_base;
+ enc->pix_fmt = static_cast<AVPixelFormat>(video_graph_out_->inputs[0]->format);
+ enc->sample_aspect_ratio = st->sample_aspect_ratio = video_graph_out_->inputs[0]->sample_aspect_ratio;
+ enc->width = video_graph_out_->inputs[0]->w;
+ enc->height = video_graph_out_->inputs[0]->h;
+ enc->bit_rate_tolerance = 400 * 1000000;
+
+ break;
+ }
+ case AVMEDIA_TYPE_AUDIO:
+ {
+ enc->time_base = audio_filter_->get_output_pad_info(stream_number_for_media_type).time_base;
+ enc->sample_fmt = static_cast<AVSampleFormat>(audio_filter_->get_output_pad_info(stream_number_for_media_type).format);
+ enc->sample_rate = audio_filter_->get_output_pad_info(stream_number_for_media_type).sample_rate;
+ enc->channel_layout = audio_filter_->get_output_pad_info(stream_number_for_media_type).channel_layout;
+ enc->channels = audio_filter_->get_output_pad_info(stream_number_for_media_type).channels;
+
+ break;
+ }
+ }
+
+ setup_codec_defaults(*enc);
+
+ if(oc_->oformat->flags & AVFMT_GLOBALHEADER)
+ enc->flags |= CODEC_FLAG_GLOBAL_HEADER;
+
+ static const std::array<std::string, 4> char_id_map = {{"v", "a", "d", "s"}};
+
+ const auto char_id = char_id_map.at(enc->codec_type);
+
+ const auto codec_opts =
+ remove_options(
+ options,
+ boost::regex("^(" + char_id + "?[^:]+):" + char_id + "$"));
+
+ AVDictionary* av_codec_opts = nullptr;
+
+ to_dict(
+ &av_codec_opts,
+ options);
+
+ to_dict(
+ &av_codec_opts,
+ codec_opts);
+
+ options.clear();
+
+ FF(avcodec_open2(
+ enc,
+ &codec,
+ av_codec_opts ? &av_codec_opts : nullptr));
+
+ if(av_codec_opts)
+ {
+ auto t =
+ av_dict_get(
+ av_codec_opts,
+ "",
+ nullptr,
+ AV_DICT_IGNORE_SUFFIX);
+
+ while(t)
+ {
+ options[t->key + (codec_opts.find(t->key) != codec_opts.end() ? ":" + char_id : "")] = t->value;
+
+ t = av_dict_get(
+ av_codec_opts,
+ "",
+ t,
+ AV_DICT_IGNORE_SUFFIX);
+ }
+
+ av_dict_free(&av_codec_opts);
+ }
+
+ if(enc->codec_type == AVMEDIA_TYPE_AUDIO && !(codec.capabilities & CODEC_CAP_VARIABLE_FRAME_SIZE))
+ {
+ CASPAR_ASSERT(enc->frame_size > 0);
+ audio_filter_->set_guaranteed_output_num_samples_per_frame(
+ stream_number_for_media_type,
+ enc->frame_size);
+ }
+
+ return std::shared_ptr<AVStream>(st, [this](AVStream* st)
+ {
+ avcodec_close(st->codec);
+ });
+ }
+
+ void configure_video_filters(
+ const AVCodec& codec,
+ std::string filtergraph)
+ {
+ video_graph_.reset(
+ avfilter_graph_alloc(),
+ [](AVFilterGraph* p)
+ {
+ avfilter_graph_free(&p);
+ });
+
+ video_graph_->nb_threads = boost::thread::hardware_concurrency()/2;
+ video_graph_->thread_type = AVFILTER_THREAD_SLICE;
+
+ const auto sample_aspect_ratio =
+ boost::rational<int>(
+ in_video_format_.square_width,
+ in_video_format_.square_height) /
+ boost::rational<int>(
+ in_video_format_.width,
+ in_video_format_.height);
+
+ const auto vsrc_options = (boost::format("video_size=%1%x%2%:pix_fmt=%3%:time_base=%4%/%5%:pixel_aspect=%6%/%7%:frame_rate=%8%/%9%")
+ % in_video_format_.width % in_video_format_.height
+ % AVPixelFormat::AV_PIX_FMT_BGRA
+ % in_video_format_.duration % in_video_format_.time_scale
+ % sample_aspect_ratio.numerator() % sample_aspect_ratio.denominator()
+ % in_video_format_.time_scale % in_video_format_.duration).str();
+
+ AVFilterContext* filt_vsrc = nullptr;
+ FF(avfilter_graph_create_filter(
+ &filt_vsrc,
+ avfilter_get_by_name("buffer"),
+ "ffmpeg_consumer_buffer",
+ vsrc_options.c_str(),
+ nullptr,
+ video_graph_.get()));
+
+ AVFilterContext* filt_vsink = nullptr;
+ FF(avfilter_graph_create_filter(
+ &filt_vsink,
+ avfilter_get_by_name("buffersink"),
+ "ffmpeg_consumer_buffersink",
+ nullptr,
+ nullptr,
+ video_graph_.get()));
+
+#pragma warning (push)
+#pragma warning (disable : 4245)
+
+ FF(av_opt_set_int_list(
+ filt_vsink,
+ "pix_fmts",
+ codec.pix_fmts,
+ -1,
+ AV_OPT_SEARCH_CHILDREN));
+
+#pragma warning (pop)
+
+ adjust_video_filter(codec, in_video_format_, filt_vsink, filtergraph);
+
+ configure_filtergraph(
+ *video_graph_,
+ filtergraph,
+ *filt_vsrc,
+ *filt_vsink);
+
+ video_graph_in_ = filt_vsrc;
+ video_graph_out_ = filt_vsink;
+
+ CASPAR_LOG(info)
+ << u16(std::string("\n")
+ + avfilter_graph_dump(
+ video_graph_.get(),
+ nullptr));
+ }
+
+ void configure_audio_filters(
+ const AVCodec& codec,
+ std::string filtergraph)
+ {
+ int num_output_pads = 1;
+
+ if (mono_streams_)
+ {
+ num_output_pads = in_channel_layout_.num_channels;
+ }
+
+ if (num_output_pads > 1)
+ {
+ std::string splitfilter = "[a:0]channelsplit=channel_layout=";
+
+ splitfilter += (boost::format("0x%|1$x|") % create_channel_layout_bitmask(in_channel_layout_.num_channels)).str();
+
+ for (int i = 0; i < num_output_pads; ++i)
+ splitfilter += "[aout:" + boost::lexical_cast<std::string>(i) + "]";
+
+ filtergraph = u8(append_filter(u16(filtergraph), u16(splitfilter)));
+ }
+
+ std::vector<audio_output_pad> output_pads(
+ num_output_pads,
+ audio_output_pad(
+ from_terminated_array<int>( codec.supported_samplerates, 0),
+ from_terminated_array<AVSampleFormat>( codec.sample_fmts, AVSampleFormat::AV_SAMPLE_FMT_NONE),
+ from_terminated_array<uint64_t>( codec.channel_layouts, 0ull)));
+
+ audio_filter_.reset(new audio_filter(
+ { audio_input_pad(
+ boost::rational<int>(1, in_video_format_.audio_sample_rate),
+ in_video_format_.audio_sample_rate,
+ AVSampleFormat::AV_SAMPLE_FMT_S32,
+ create_channel_layout_bitmask(in_channel_layout_.num_channels)) },
+ output_pads,
+ filtergraph));
+ }
+
+ void configure_filtergraph(
+ AVFilterGraph& graph,
+ const std::string& filtergraph,
+ AVFilterContext& source_ctx,
+ AVFilterContext& sink_ctx)
+ {
+ AVFilterInOut* outputs = nullptr;
+ AVFilterInOut* inputs = nullptr;
+
+ if(!filtergraph.empty())
+ {
+ outputs = avfilter_inout_alloc();
+ inputs = avfilter_inout_alloc();
+
+ try
+ {
+ CASPAR_VERIFY(outputs && inputs);
+
+ outputs->name = av_strdup("in");
+ outputs->filter_ctx = &source_ctx;
+ outputs->pad_idx = 0;
+ outputs->next = nullptr;
+
+ inputs->name = av_strdup("out");
+ inputs->filter_ctx = &sink_ctx;
+ inputs->pad_idx = 0;
+ inputs->next = nullptr;
+ }
+ catch (...)
+ {
+ avfilter_inout_free(&outputs);
+ avfilter_inout_free(&inputs);
+ throw;
+ }
+
+ FF(avfilter_graph_parse(
+ &graph,
+ filtergraph.c_str(),
+ inputs,
+ outputs,
+ nullptr));
+ }
+ else
+ {
+ FF(avfilter_link(
+ &source_ctx,
+ 0,
+ &sink_ctx,
+ 0));
+ }
+
+ FF(avfilter_graph_config(
+ &graph,
+ nullptr));
+ }
+
+ void encode_video(core::const_frame frame_ptr, std::shared_ptr<void> token)
+ {
+ if(!video_st_)
+ return;
+
+ auto enc = video_st_->codec;
+
+ if(frame_ptr != core::const_frame::empty())
+ {
+ auto src_av_frame = create_frame();
+
+ const auto sample_aspect_ratio =
+ boost::rational<int>(
+ in_video_format_.square_width,
+ in_video_format_.square_height) /
+ boost::rational<int>(
+ in_video_format_.width,
+ in_video_format_.height);
+
+ src_av_frame->format = AVPixelFormat::AV_PIX_FMT_BGRA;
+ src_av_frame->width = in_video_format_.width;
+ src_av_frame->height = in_video_format_.height;
+ src_av_frame->sample_aspect_ratio.num = sample_aspect_ratio.numerator();
+ src_av_frame->sample_aspect_ratio.den = sample_aspect_ratio.denominator();
+ src_av_frame->pts = video_pts_;
+
+ video_pts_ += 1;
+
+ subject_
+ << core::monitor::message("/frame") % video_pts_
+ << core::monitor::message("/path") % path_
+ << core::monitor::message("/fps") % in_video_format_.fps;
+
+ FF(av_image_fill_arrays(
+ src_av_frame->data,
+ src_av_frame->linesize,
+ frame_ptr.image_data().begin(),
+ static_cast<AVPixelFormat>(src_av_frame->format),
+ in_video_format_.width,
+ in_video_format_.height,
+ 1));
+
+ FF(av_buffersrc_add_frame(
+ video_graph_in_,
+ src_av_frame.get()));
+ }
+
+ int ret = 0;
+
+ while(ret >= 0)
+ {
+ auto filt_frame = create_frame();
+
+ ret = av_buffersink_get_frame(
+ video_graph_out_,
+ filt_frame.get());
+
+ video_encoder_executor_.begin_invoke([=]
+ {
+ if(ret == AVERROR_EOF)
+ {
+ if(enc->codec->capabilities & CODEC_CAP_DELAY)
+ {
+ while(encode_av_frame(
+ *video_st_,
+ avcodec_encode_video2,
+ nullptr, token))
+ {
+ boost::this_thread::yield(); // TODO:
+ }
+ }
+ }
+ else if(ret != AVERROR(EAGAIN))
+ {
+ FF_RET(ret, "av_buffersink_get_frame");
+
+ if (filt_frame->interlaced_frame)
+ {
+ if (enc->codec->id == AV_CODEC_ID_MJPEG)
+ enc->field_order = filt_frame->top_field_first ? AV_FIELD_TT : AV_FIELD_BB;
+ else
+ enc->field_order = filt_frame->top_field_first ? AV_FIELD_TB : AV_FIELD_BT;
+ }
+ else
+ enc->field_order = AV_FIELD_PROGRESSIVE;
+
+ filt_frame->quality = enc->global_quality;
+
+ if (!enc->me_threshold)
+ filt_frame->pict_type = AV_PICTURE_TYPE_NONE;
+
+ encode_av_frame(
+ *video_st_,
+ avcodec_encode_video2,
+ filt_frame,
+ token);
+
+ boost::this_thread::yield(); // TODO:
+ }
+ });
+ }
+ }
+
+ void encode_audio(core::const_frame frame_ptr, std::shared_ptr<void> token)
+ {
+ if(audio_sts_.empty())
+ return;
+
+ if(frame_ptr != core::const_frame::empty())
+ {
+ auto src_av_frame = create_frame();
+
+ src_av_frame->channels = in_channel_layout_.num_channels;
+ src_av_frame->channel_layout = create_channel_layout_bitmask(in_channel_layout_.num_channels);
+ src_av_frame->sample_rate = in_video_format_.audio_sample_rate;
+ src_av_frame->nb_samples = static_cast<int>(frame_ptr.audio_data().size()) / src_av_frame->channels;
+ src_av_frame->format = AV_SAMPLE_FMT_S32;
+ src_av_frame->pts = audio_pts_;
+
+ audio_pts_ += src_av_frame->nb_samples;
+
+ FF(av_samples_fill_arrays(
+ src_av_frame->extended_data,
+ src_av_frame->linesize,
+ reinterpret_cast<const std::uint8_t*>(&*frame_ptr.audio_data().begin()),
+ src_av_frame->channels,
+ src_av_frame->nb_samples,
+ static_cast<AVSampleFormat>(src_av_frame->format),
+ 16));
+
+ audio_filter_->push(0, src_av_frame);
+ }
+
+ for (int pad_id = 0; pad_id < audio_filter_->get_num_output_pads(); ++pad_id)
+ {
+ for (auto filt_frame : audio_filter_->poll_all(pad_id))
+ {
+ audio_encoder_executor_.begin_invoke([=]
+ {
+ encode_av_frame(
+ *audio_sts_.at(pad_id),
+ avcodec_encode_audio2,
+ filt_frame,
+ token);
+
+ boost::this_thread::yield(); // TODO:
+ });
+ }
+ }
+
+ bool eof = frame_ptr == core::const_frame::empty();
+
+ if (eof)
+ {
+ audio_encoder_executor_.begin_invoke([=]
+ {
+ for (int pad_id = 0; pad_id < audio_filter_->get_num_output_pads(); ++pad_id)
+ {
+ auto enc = audio_sts_.at(pad_id)->codec;
+
+ if (enc->codec->capabilities & CODEC_CAP_DELAY)
+ {
+ while (encode_av_frame(
+ *audio_sts_.at(pad_id),
+ avcodec_encode_audio2,
+ nullptr,
+ token))
+ {
+ boost::this_thread::yield(); // TODO:
+ }
+ }
+ }
+ });
+ }
+ }
+
+ template<typename F>
+ bool encode_av_frame(
+ AVStream& st,
+ const F& func,
+ const std::shared_ptr<AVFrame>& src_av_frame,
+ std::shared_ptr<void> token)
+ {
+ AVPacket pkt = {};
+ av_init_packet(&pkt);
+
+ int got_packet = 0;
+
+ FF(func(
+ st.codec,
+ &pkt,
+ src_av_frame.get(),
+ &got_packet));
+
+ if(!got_packet || pkt.size <= 0)
+ return false;
+
+ pkt.stream_index = st.index;
+
+ if (pkt.pts != AV_NOPTS_VALUE)
+ {
+ pkt.pts =
+ av_rescale_q(
+ pkt.pts,
+ st.codec->time_base,
+ st.time_base);
+ }
+
+ if (pkt.dts != AV_NOPTS_VALUE)
+ {
+ pkt.dts =
+ av_rescale_q(
+ pkt.dts,
+ st.codec->time_base,
+ st.time_base);
+ }
+
+ pkt.duration =
+ static_cast<int>(
+ av_rescale_q(
+ pkt.duration,
+ st.codec->time_base, st.time_base));
+
+ write_packet(
+ std::shared_ptr<AVPacket>(
+ new AVPacket(pkt),
+ [](AVPacket* p)
+ {
+ av_free_packet(p);
+ delete p;
+ }), token);
+
+ return true;
+ }
+
+ void write_packet(
+ const std::shared_ptr<AVPacket>& pkt_ptr,
+ std::shared_ptr<void> token)
+ {
+ write_executor_.begin_invoke([this, pkt_ptr, token]() mutable
+ {
+ FF(av_interleaved_write_frame(
+ oc_.get(),
+ pkt_ptr.get()));
+ });
+ }
+
+ template<typename T>
+ static boost::optional<T> try_remove_arg(
+ std::map<std::string, std::string>& options,
+ const boost::regex& expr)
+ {
+ for(auto it = options.begin(); it != options.end(); ++it)
+ {
+ if(boost::regex_search(it->first, expr))
+ {
+ auto arg = it->second;
+ options.erase(it);
+ return boost::lexical_cast<T>(arg);
+ }
+ }
+
+ return boost::optional<T>();
+ }
+
+ static std::map<std::string, std::string> remove_options(
+ std::map<std::string, std::string>& options,
+ const boost::regex& expr)
+ {
+ std::map<std::string, std::string> result;
+
+ auto it = options.begin();
+ while(it != options.end())
+ {
+ boost::smatch what;
+ if(boost::regex_search(it->first, what, expr))
+ {
+ result[
+ what.size() > 0 && what[1].matched
+ ? what[1].str()
+ : it->first] = it->second;
+ it = options.erase(it);
+ }
+ else
+ ++it;
+ }
+
+ return result;
+ }
+
+ static void to_dict(AVDictionary** dest, const std::map<std::string, std::string>& c)
+ {
+ for (const auto& entry : c)
+ {
+ av_dict_set(
+ dest,
+ entry.first.c_str(),
+ entry.second.c_str(), 0);
+ }
+ }
+
+ static std::map<std::string, std::string> to_map(AVDictionary* dict)
+ {
+ std::map<std::string, std::string> result;
+
+ for(auto t = dict
+ ? av_dict_get(
+ dict,
+ "",
+ nullptr,
+ AV_DICT_IGNORE_SUFFIX)
+ : nullptr;
+ t;
+ t = av_dict_get(
+ dict,
+ "",
+ t,
+ AV_DICT_IGNORE_SUFFIX))
+ {
+ result[t->key] = t->value;
+ }
+
+ return result;
+ }
+};
+
+int crc16(const std::string& str)
+{
+ boost::crc_16_type result;
+
+ result.process_bytes(str.data(), str.length());
+
+ return result.checksum();
+}
+
+struct ffmpeg_consumer_proxy : public core::frame_consumer
+{
+ const std::string path_;
+ const std::string options_;
+ const bool separate_key_;
+ const bool mono_streams_;
+ const bool compatibility_mode_;
+ int consumer_index_offset_;
+
+ std::unique_ptr<ffmpeg_consumer> consumer_;
+ std::unique_ptr<ffmpeg_consumer> key_only_consumer_;
+
+public:
+
+ ffmpeg_consumer_proxy(const std::string& path, const std::string& options, bool separate_key, bool mono_streams, bool compatibility_mode)
+ : path_(path)
+ , options_(options)
+ , separate_key_(separate_key)
+ , mono_streams_(mono_streams)
+ , compatibility_mode_(compatibility_mode)
+ , consumer_index_offset_(crc16(path))
+ {
+ }
+
+ void initialize(const core::video_format_desc& format_desc, const core::audio_channel_layout& channel_layout, int) override
+ {
+ if (consumer_)
+ CASPAR_THROW_EXCEPTION(invalid_operation() << msg_info("Cannot reinitialize ffmpeg-consumer."));
+
+ consumer_.reset(new ffmpeg_consumer(path_, options_, mono_streams_));
+ consumer_->initialize(format_desc, channel_layout);
+
+ if (separate_key_)
+ {
+ boost::filesystem::path fill_file(path_);
+ auto without_extension = u16(fill_file.parent_path().string() + "/" + fill_file.stem().string());
+ auto key_file = without_extension + L"_A" + u16(fill_file.extension().string());
+
+ key_only_consumer_.reset(new ffmpeg_consumer(u8(key_file), options_, mono_streams_));
+ key_only_consumer_->initialize(format_desc, channel_layout);
+ }
+ }
+
+ int64_t presentation_frame_age_millis() const override
+ {
+ return consumer_ ? static_cast<int64_t>(consumer_->presentation_frame_age_millis()) : 0;
+ }
+
+ std::future<bool> send(core::const_frame frame) override
+ {
+ bool ready_for_frame = consumer_->ready_for_frame();
+
+ if (ready_for_frame && separate_key_)
+ ready_for_frame = ready_for_frame && key_only_consumer_->ready_for_frame();
+
+ if (ready_for_frame)
+ {
+ consumer_->send(frame);
+
+ if (separate_key_)
+ key_only_consumer_->send(frame.key_only());
+ }
+ else
+ {
+ consumer_->mark_dropped();
+
+ if (separate_key_)
+ key_only_consumer_->mark_dropped();
+ }
+
+ return make_ready_future(true);
+ }
+
+ std::wstring print() const override
+ {
+ return consumer_ ? consumer_->print() : L"[ffmpeg_consumer]";
+ }
+
+ std::wstring name() const override
+ {
+ return L"ffmpeg";
+ }
+
+ boost::property_tree::wptree info() const override
+ {
+ boost::property_tree::wptree info;
+
+ info.add(L"type", L"ffmpeg");
+ info.add(L"path", u16(path_));
+ info.add(L"separate_key", separate_key_);
+ info.add(L"mono_streams", mono_streams_);
+
+ return info;
+ }
+
+ bool has_synchronization_clock() const override
+ {
+ return false;
+ }
+
+ int buffer_depth() const override
+ {
+ return -1;
+ }
+
+ int index() const override
+ {
+ return compatibility_mode_ ? 200 : 100000 + consumer_index_offset_;
+ }
+
+ core::monitor::subject& monitor_output() override
+ {
+ return consumer_->monitor_output();
+ }
+};
+
+void describe_ffmpeg_consumer(core::help_sink& sink, const core::help_repository& repo)
+{
+ sink.short_description(L"For streaming/recording the contents of a channel using FFmpeg.");
+ sink.syntax(L"FILE,STREAM [filename:string],[url:string] {-[ffmpeg_param1:string] [value1:string] {-[ffmpeg_param2:string] [value2:string] {...}}} {[separate_key:SEPARATE_KEY]} {[mono_streams:MONO_STREAMS]}");
+ sink.para()->text(L"For recording or streaming the contents of a channel using FFmpeg");
+ sink.definitions()
+ ->item(L"filename", L"The filename under the media folder including the extension (decides which kind of container format that will be used).")
+ ->item(L"url", L"If the filename is given in the form of an URL a network stream will be created instead of a file on disk.")
+ ->item(L"ffmpeg_paramX", L"A parameter supported by FFmpeg. For example vcodec or acodec etc.")
+ ->item(L"separate_key", L"If defined will create two files simultaneously -- One for fill and one for key (_A will be appended).")
+ ->item(L"mono_streams", L"If defined every audio channel will be written to its own audio stream.");
+ sink.para()->text(L"Examples:");
+ sink.example(L">> ADD 1 FILE output.mov -vcodec dnxhd");
+ sink.example(L">> ADD 1 FILE output.mov -vcodec prores");
+ sink.example(L">> ADD 1 FILE output.mov -vcodec dvvideo");
+ sink.example(L">> ADD 1 FILE output.mov -vcodec libx264 -preset ultrafast -tune fastdecode -crf 25");
+ sink.example(L">> ADD 1 FILE output.mov -vcodec dnxhd SEPARATE_KEY", L"for creating output.mov with fill and output_A.mov with key/alpha");
+ sink.example(L">> ADD 1 FILE output.mxf -vcodec dnxhd MONO_STREAMS", L"for creating output.mxf with every audio channel encoded in its own mono stream.");
+ sink.example(L">> ADD 1 STREAM udp://<client_ip_address>:9250 -format mpegts -vcodec libx264 -crf 25 -tune zerolatency -preset ultrafast",
+ L"for streaming over UDP instead of creating a local file.");
+}
+
+spl::shared_ptr<core::frame_consumer> create_ffmpeg_consumer(
+ const std::vector<std::wstring>& params, core::interaction_sink*, std::vector<spl::shared_ptr<core::video_channel>> channels)
+{
+ if (params.size() < 1 || (!boost::iequals(params.at(0), L"STREAM") && !boost::iequals(params.at(0), L"FILE")))
+ return core::frame_consumer::empty();
+
+ auto params2 = params;
+ bool separate_key = get_and_consume_flag(L"SEPARATE_KEY", params2);
+ bool mono_streams = get_and_consume_flag(L"MONO_STREAMS", params2);
+ auto compatibility_mode = boost::iequals(params.at(0), L"FILE");
+ auto path = u8(params2.size() > 1 ? params2.at(1) : L"");
+ auto args = u8(boost::join(params2, L" "));
+
+ return spl::make_shared<ffmpeg_consumer_proxy>(path, args, separate_key, mono_streams, compatibility_mode);
+}
+
+spl::shared_ptr<core::frame_consumer> create_preconfigured_ffmpeg_consumer(
+ const boost::property_tree::wptree& ptree, core::interaction_sink*, std::vector<spl::shared_ptr<core::video_channel>> channels)
+{
+ return spl::make_shared<ffmpeg_consumer_proxy>(
+ u8(ptree_get<std::wstring>(ptree, L"path")),
+ u8(ptree.get<std::wstring>(L"args", L"")),
+ ptree.get<bool>(L"separate-key", false),
+ ptree.get<bool>(L"mono-streams", false),
+ false);
+}
+
+}}
projects / casparcg / blobdiff