\r
#include "ffmpeg_consumer.h"\r
\r
+#include "../producer/audio/audio_resampler.h"\r
+\r
#include <core/mixer/read_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/concurrency/executor.h>\r
+#include <common/concurrency/future_util.h>\r
#include <common/diagnostics/graph.h>\r
#include <common/env.h>\r
#include <common/utility/string.h>\r
#include <common/utility/param.h>\r
+#include <common/memory/memshfl.h>\r
\r
#include <boost/algorithm/string.hpp>\r
#include <boost/timer.hpp>\r
#include <tbb/cache_aligned_allocator.h>\r
#include <tbb/parallel_invoke.h>\r
\r
+#include <boost/range/algorithm.hpp>\r
+#include <boost/range/algorithm_ext.hpp>\r
+#include <boost/lexical_cast.hpp>\r
+\r
#include <string>\r
\r
#if defined(_MSC_VER)\r
#include <libswscale/swscale.h>\r
#include <libavutil/opt.h>\r
#include <libavutil/pixdesc.h>\r
+ #include <libavutil/parseutils.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
}\r
};\r
\r
-void set_format(AVOutputFormat*& format, const std::string& value)\r
+struct output_format\r
{\r
- format = av_guess_format(value.c_str(), nullptr, nullptr);\r
-\r
- if(format == nullptr)\r
- BOOST_THROW_EXCEPTION(invalid_argument() << arg_name_info("f")); \r
-}\r
-\r
-bool parse_format(AVOutputFormat*& format, std::vector<option>& options)\r
-{ \r
- auto format_it = std::find_if(options.begin(), options.end(), [](const option& o)\r
+ AVOutputFormat* format;\r
+ int width;\r
+ int height;\r
+ CodecID vcodec;\r
+ CodecID acodec;\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
{\r
- return o.name == "f" || o.name == "format";\r
- });\r
-\r
- if(format_it == options.end())\r
- return false;\r
-\r
- set_format(format, format_it->value);\r
-\r
- options.erase(format_it);\r
-\r
- return true;\r
-}\r
-\r
-void set_vcodec(CodecID& vcodec, const std::string& value)\r
-{ \r
- vcodec = avcodec_find_encoder_by_name(value.c_str())->id;\r
- if(vcodec == CODEC_ID_NONE)\r
- BOOST_THROW_EXCEPTION(invalid_argument() << arg_name_info("vcodec"));\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
-bool parse_vcodec(CodecID& vcodec, std::vector<option>& options)\r
-{\r
- auto vcodec_it = std::find_if(options.begin(), options.end(), [](const option& o)\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
- return o.name == "vcodec";\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
+ // BOOST_THROW_EXCEPTION(invalid_argument() << arg_name_info("target"));\r
+ // \r
+ // if (name.find("-dv") != std::string::npos) \r
+ // {\r
+ // set_opt("f", "dv");\r
+ // set_opt("s", norm == PAL ? "720x576" : "720x480");\r
+ // //set_opt("pix_fmt", name.find("-dv50") != std::string::npos ? "yuv422p" : norm == PAL ? "yuv420p" : "yuv411p");\r
+ // //set_opt("ar", "48000");\r
+ // //set_opt("ac", "2");\r
+ // } \r
+ //}\r
+ if(name == "f")\r
+ {\r
+ format = av_guess_format(value.c_str(), nullptr, nullptr);\r
\r
- if(vcodec_it == options.end())\r
- return false;\r
- \r
- set_vcodec(vcodec, vcodec_it->value);\r
+ if(format == nullptr)\r
+ BOOST_THROW_EXCEPTION(invalid_argument() << arg_name_info("f"));\r
\r
- options.erase(vcodec_it);\r
+ return true;\r
+ }\r
+ else if(name == "vcodec")\r
+ {\r
+ auto c = avcodec_find_encoder_by_name(value.c_str());\r
+ if(c == nullptr)\r
+ BOOST_THROW_EXCEPTION(invalid_argument() << arg_name_info("vcodec"));\r
\r
- return true;\r
-}\r
+ vcodec = avcodec_find_encoder_by_name(value.c_str())->id;\r
+ return true;\r
\r
-void set_pix_fmt(AVCodecContext* vcodec, const std::string& value)\r
-{\r
- auto pix_fmt = av_get_pix_fmt(value.c_str());\r
- if(pix_fmt == PIX_FMT_NONE)\r
- BOOST_THROW_EXCEPTION(invalid_argument() << arg_name_info("pix_fmt"));\r
+ }\r
+ else if(name == "acodec")\r
+ {\r
+ auto c = avcodec_find_encoder_by_name(value.c_str());\r
+ if(c == nullptr)\r
+ BOOST_THROW_EXCEPTION(invalid_argument() << arg_name_info("acodec"));\r
\r
- vcodec->pix_fmt = pix_fmt;\r
-}\r
+ acodec = avcodec_find_encoder_by_name(value.c_str())->id;\r
\r
-bool parse_pix_fmt(AVCodecContext* vcodec, std::vector<option>& options)\r
-{\r
- auto pix_fmt_it = std::find_if(options.begin(), options.end(), [](const option& o)\r
- {\r
- return o.name == "pix_fmt" || o.name == "pixel_format";\r
- });\r
- \r
- if(pix_fmt_it == options.end())\r
- return false;\r
+ return true;\r
+ }\r
+ else if(name == "s")\r
+ {\r
+ if(av_parse_video_size(&width, &height, value.c_str()) < 0)\r
+ BOOST_THROW_EXCEPTION(invalid_argument() << arg_name_info("s"));\r
+ \r
+ return true;\r
+ }\r
\r
- set_pix_fmt(vcodec, pix_fmt_it->value);\r
- \r
- options.erase(pix_fmt_it);\r
+ return false;\r
+ }\r
+};\r
\r
- return true;\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 core::video_format_desc format_desc_;\r
\r
const safe_ptr<diagnostics::graph> graph_;\r
- boost::timer frame_timer_;\r
- boost::timer write_timer_;\r
\r
- executor executor_;\r
- executor file_write_executor_;\r
-\r
- // Audio\r
- std::shared_ptr<AVStream> audio_st_;\r
+ executor encode_executor_;\r
\r
- // Video\r
+ std::shared_ptr<AVStream> audio_st_;\r
std::shared_ptr<AVStream> video_st_;\r
-\r
- std::vector<uint8_t> video_outbuf_;\r
- std::vector<uint8_t> picture_buf_;\r
+ \r
+ byte_vector audio_outbuf_;\r
+ byte_vector audio_buf_;\r
+ byte_vector video_outbuf_;\r
+ byte_vector key_picture_buf_;\r
+ byte_vector picture_buf_;\r
+ std::shared_ptr<audio_resampler> swr_;\r
std::shared_ptr<SwsContext> sws_;\r
\r
- int64_t frame_number_;\r
+ int64_t in_frame_number_;\r
+ int64_t out_frame_number_;\r
+\r
+ output_format output_format_;\r
+ bool key_only_;\r
\r
public:\r
- ffmpeg_consumer(const std::string& filename, const core::video_format_desc& format_desc, std::vector<option> options)\r
+ ffmpeg_consumer(const std::string& filename, const core::video_format_desc& format_desc, std::vector<option> options, bool key_only)\r
: filename_(filename)\r
, video_outbuf_(1920*1080*8)\r
+ , audio_outbuf_(10000)\r
, oc_(avformat_alloc_context(), av_free)\r
, format_desc_(format_desc)\r
- , executor_(print())\r
- , file_write_executor_(print() + L"/output")\r
- , frame_number_(0)\r
+ , encode_executor_(print())\r
+ , in_frame_number_(0)\r
+ , out_frame_number_(0)\r
+ , output_format_(format_desc, filename, options)\r
+ , key_only_(key_only)\r
{\r
// TODO: Ask stakeholders about case where file already exists.\r
boost::filesystem2::remove(boost::filesystem2::wpath(env::media_folder() + widen(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_color("write-time", diagnostics::color(0.5f, 0.5f, 0.1f));\r
+ graph_->set_color("dropped-frame", diagnostics::color(0.3f, 0.6f, 0.3f));\r
graph_->set_text(print());\r
diagnostics::register_graph(graph_);\r
\r
- executor_.set_capacity(8);\r
- file_write_executor_.set_capacity(8);\r
+ encode_executor_.set_capacity(8);\r
\r
- oc_->oformat = av_guess_format(nullptr, filename_.c_str(), nullptr);\r
- if (!oc_->oformat)\r
- BOOST_THROW_EXCEPTION(caspar_exception() << msg_info("Could not find suitable output format."));\r
- \r
- parse_format(oc_->oformat, options);\r
-\r
- auto vcodec = oc_->oformat->video_codec;\r
- parse_vcodec(vcodec, options);\r
- \r
+ oc_->oformat = output_format_.format;\r
+ \r
THROW_ON_ERROR2(av_set_parameters(oc_.get(), nullptr), "[ffmpeg_consumer]");\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(vcodec, options);\r
- audio_st_ = add_audio_stream();\r
+ // Add the audio and video streams using the default format codecs and initialize the codecs.\r
+ auto options2 = options;\r
+ video_st_ = add_video_stream(options2);\r
+\r
+ if (!key_only)\r
+ audio_st_ = add_audio_stream(options);\r
\r
dump_format(oc_.get(), 0, filename_.c_str(), 1);\r
\r
\r
THROW_ON_ERROR2(av_write_header(oc_.get()), "[ffmpeg_consumer]");\r
\r
+ if(options.size() > 0)\r
+ {\r
+ BOOST_FOREACH(auto& option, options)\r
+ CASPAR_LOG(warning) << L"Invalid option: -" << widen(option.name) << L" " << widen(option.value);\r
+ }\r
+\r
CASPAR_LOG(info) << print() << L" Successfully Initialized."; \r
}\r
\r
~ffmpeg_consumer()\r
{ \r
- executor_.stop();\r
- executor_.join();\r
+ encode_executor_.stop_execute_rest();\r
+ encode_executor_.join();\r
\r
- file_write_executor_.stop();\r
- file_write_executor_.join();\r
+ // Flush\r
+ LOG_ON_ERROR2(av_interleaved_write_frame(oc_.get(), nullptr), "[ffmpeg_consumer]");\r
\r
LOG_ON_ERROR2(av_write_trailer(oc_.get()), "[ffmpeg_consumer]");\r
\r
- audio_st_.reset();\r
+ if (!key_only_)\r
+ audio_st_.reset();\r
video_st_.reset();\r
\r
if (!(oc_->oformat->flags & AVFMT_NOFILE)) \r
return L"ffmpeg[" + widen(filename_) + L"]";\r
}\r
\r
- std::shared_ptr<AVStream> add_video_stream(CodecID vcodec, std::vector<option>& options)\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
- BOOST_THROW_EXCEPTION(caspar_exception() << msg_info("Could not allocate video-stream") << boost::errinfo_api_function("av_new_stream")); \r
+ BOOST_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(vcodec);\r
+ auto encoder = avcodec_find_encoder(output_format_.vcodec);\r
if (!encoder)\r
- BOOST_THROW_EXCEPTION(caspar_exception() << msg_info("codec not found"));\r
+ BOOST_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 = vcodec;\r
+ c->codec_id = output_format_.vcodec;\r
c->codec_type = AVMEDIA_TYPE_VIDEO;\r
- c->width = format_desc_.width;\r
- c->height = format_desc_.height;\r
+ c->width = output_format_.width;\r
+ c->height = output_format_.height;\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
+ if(c->pix_fmt == PIX_FMT_NONE)\r
+ c->pix_fmt = PIX_FMT_YUV420P;\r
\r
if(c->codec_id == CODEC_ID_PRORES)\r
{ \r
- c->bit_rate = format_desc_.width < 1280 ? 63*1000000 : 220*1000000;\r
+ c->bit_rate = c->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(format_desc_.width < 1280 || format_desc_.height < 720)\r
- BOOST_THROW_EXCEPTION(caspar_exception() << msg_info("unsupported dimension"));\r
+ if(c->width < 1280 || c->height < 720)\r
+ BOOST_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->bit_rate = format_desc_.width < 1280 ? 50*1000000 : 100*1000000;\r
- c->width = format_desc_.height == 1280 ? 960 : c->width;\r
+ c->width = c->height == 1280 ? 960 : c->width;\r
\r
- //if(format_desc_.format == core::video_format::ntsc)\r
- // c->pix_fmt = PIX_FMT_YUV411P;\r
- //else if(format_desc_.format == core::video_format::pal)\r
- // c->pix_fmt = PIX_FMT_YUV420P;\r
-\r
- c->pix_fmt = PIX_FMT_YUV411P;\r
-\r
- if(c->bit_rate >= 50*1000000) // dv50\r
+ if(format_desc_.format == core::video_format::ntsc)\r
+ c->pix_fmt = PIX_FMT_YUV411P;\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 = format_desc_.height == 1080 ? 1280 : c->width; \r
+ c->width = c->height == 1080 ? 1280 : c->width; \r
else\r
- c->width = format_desc_.height == 1080 ? 1440 : c->width; \r
+ c->width = c->height == 1080 ? 1440 : c->width; \r
}\r
else if(c->codec_id == CODEC_ID_H264)\r
{ \r
{\r
c->pix_fmt = PIX_FMT_ARGB;\r
}\r
- else\r
- {\r
- BOOST_THROW_EXCEPTION(invalid_argument() << msg_info("Unsupported output parameters."));\r
- }\r
-\r
- parse_pix_fmt(c, options);\r
- \r
+ \r
c->max_b_frames = 0; // b-frames not supported.\r
-\r
- BOOST_FOREACH(auto& option, options)\r
- THROW_ON_ERROR2(av_opt_set(c, option.name.c_str(), option.value.c_str(), AV_OPT_SEARCH_CHILDREN), "[ffmpeg_consumer]");\r
\r
- if(oc_->oformat->flags & AVFMT_GLOBALHEADER)\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
+ ffmpeg::av_opt_set(c->priv_data, 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
c->thread_count = boost::thread::hardware_concurrency();\r
- THROW_ON_ERROR2(avcodec_open(c, encoder), "[ffmpeg_consumer]");\r
+ if(avcodec_open(c, encoder) < 0)\r
+ {\r
+ c->thread_count = 1;\r
+ THROW_ON_ERROR2(avcodec_open(c, encoder), "[ffmpeg_consumer]");\r
+ }\r
\r
return std::shared_ptr<AVStream>(st, [](AVStream* st)\r
{\r
av_freep(&st);\r
});\r
}\r
- \r
- std::shared_ptr<AVStream> add_audio_stream()\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
BOOST_THROW_EXCEPTION(caspar_exception() << msg_info("Could not allocate audio-stream") << boost::errinfo_api_function("av_new_stream")); \r
-\r
- st->codec->codec_id = CODEC_ID_PCM_S16LE;\r
- st->codec->codec_type = AVMEDIA_TYPE_AUDIO;\r
- st->codec->sample_rate = 48000;\r
- st->codec->channels = 2;\r
- st->codec->sample_fmt = SAMPLE_FMT_S16;\r
\r
- if(oc_->oformat->flags & AVFMT_GLOBALHEADER)\r
- st->codec->flags |= CODEC_FLAG_GLOBAL_HEADER;\r
- \r
- auto codec = avcodec_find_encoder(st->codec->codec_id);\r
- if (!codec)\r
+ auto encoder = avcodec_find_encoder(output_format_.acodec);\r
+ if (!encoder)\r
BOOST_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 = SAMPLE_FMT_S16;\r
+\r
+ if(output_format_.vcodec == CODEC_ID_FLV1) \r
+ c->sample_rate = 44100; \r
\r
- THROW_ON_ERROR2(avcodec_open(st->codec, codec), "[ffmpeg_consumer]");\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
});\r
}\r
\r
- std::shared_ptr<AVFrame> convert_video_frame(const safe_ptr<core::read_frame>& frame, AVCodecContext* c)\r
+ std::shared_ptr<AVFrame> convert_video(core::read_frame& frame, AVCodecContext* c)\r
{\r
if(!sws_) \r
{\r
BOOST_THROW_EXCEPTION(caspar_exception() << msg_info("Cannot initialize the conversion context"));\r
}\r
\r
- std::shared_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);\r
- avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), const_cast<uint8_t*>(frame->image_data().begin()), PIX_FMT_BGRA, format_desc_.width, format_desc_.height);\r
- \r
- std::shared_ptr<AVFrame> local_av_frame(avcodec_alloc_frame(), av_free);\r
- picture_buf_.resize(avpicture_get_size(c->pix_fmt, format_desc_.width, format_desc_.height));\r
- avpicture_fill(reinterpret_cast<AVPicture*>(local_av_frame.get()), picture_buf_.data(), c->pix_fmt, format_desc_.width, format_desc_.height);\r
+ std::shared_ptr<AVFrame> in_frame(avcodec_alloc_frame(), av_free);\r
+ auto in_picture = reinterpret_cast<AVPicture*>(in_frame.get());\r
+\r
+ if (key_only_)\r
+ {\r
+ key_picture_buf_.resize(frame.image_data().size());\r
+ in_picture->linesize[0] = format_desc_.width * 4;\r
+ in_picture->data[0] = key_picture_buf_.data();\r
+\r
+ fast_memshfl(in_picture->data[0], frame.image_data().begin(), frame.image_data().size(), 0x0F0F0F0F, 0x0B0B0B0B, 0x07070707, 0x03030303);\r
+ }\r
+ else\r
+ {\r
+ avpicture_fill(in_picture, const_cast<uint8_t*>(frame.image_data().begin()), PIX_FMT_BGRA, format_desc_.width, format_desc_.height);\r
+ }\r
+\r
+ std::shared_ptr<AVFrame> out_frame(avcodec_alloc_frame(), av_free);\r
+ picture_buf_.resize(avpicture_get_size(c->pix_fmt, c->width, c->height));\r
+ avpicture_fill(reinterpret_cast<AVPicture*>(out_frame.get()), picture_buf_.data(), c->pix_fmt, c->width, c->height);\r
\r
- sws_scale(sws_.get(), av_frame->data, av_frame->linesize, 0, c->height, local_av_frame->data, local_av_frame->linesize);\r
+ sws_scale(sws_.get(), in_frame->data, in_frame->linesize, 0, format_desc_.height, out_frame->data, out_frame->linesize);\r
\r
- return local_av_frame;\r
+ return out_frame;\r
}\r
\r
- std::shared_ptr<AVPacket> encode_video_frame(const safe_ptr<core::read_frame>& frame)\r
+ void encode_video_frame(core::read_frame& frame)\r
{ \r
auto c = video_st_->codec;\r
+ \r
+ auto in_time = static_cast<double>(in_frame_number_) / format_desc_.fps;\r
+ auto out_time = static_cast<double>(out_frame_number_) / (static_cast<double>(c->time_base.den) / static_cast<double>(c->time_base.num));\r
+ \r
+ in_frame_number_++;\r
+\r
+ if(out_time - in_time > 0.01)\r
+ return;\r
\r
- auto av_frame = convert_video_frame(frame, c);\r
+ auto av_frame = convert_video(frame, c);\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
+ av_frame->pts = out_frame_number_++;\r
\r
int out_size = THROW_ON_ERROR2(avcodec_encode_video(c, video_outbuf_.data(), video_outbuf_.size(), av_frame.get()), "[ffmpeg_consumer]");\r
- if(out_size > 0)\r
+ if(out_size == 0)\r
+ return;\r
+ \r
+ safe_ptr<AVPacket> pkt(new AVPacket, [](AVPacket* p)\r
{\r
- safe_ptr<AVPacket> pkt(new AVPacket, [](AVPacket* p)\r
- {\r
- av_free_packet(p);\r
- delete p;\r
- });\r
- av_init_packet(pkt.get());\r
+ av_free_packet(p);\r
+ delete p;\r
+ });\r
+ av_init_packet(pkt.get());\r
\r
- if (c->coded_frame->pts != AV_NOPTS_VALUE)\r
- pkt->pts = av_rescale_q(c->coded_frame->pts, c->time_base, video_st_->time_base);\r
+ if (c->coded_frame->pts != AV_NOPTS_VALUE)\r
+ pkt->pts = av_rescale_q(c->coded_frame->pts, c->time_base, video_st_->time_base);\r
\r
- if(c->coded_frame->key_frame)\r
- pkt->flags |= AV_PKT_FLAG_KEY;\r
+ if(c->coded_frame->key_frame)\r
+ pkt->flags |= AV_PKT_FLAG_KEY;\r
\r
- pkt->stream_index = video_st_->index;\r
- pkt->data = video_outbuf_.data();\r
- pkt->size = out_size;\r
- \r
- av_dup_packet(pkt.get());\r
- return pkt;\r
- } \r
- return nullptr;\r
+ pkt->stream_index = video_st_->index;\r
+ pkt->data = video_outbuf_.data();\r
+ pkt->size = out_size;\r
+ \r
+ av_interleaved_write_frame(oc_.get(), pkt.get()); \r
}\r
\r
- std::shared_ptr<AVPacket> encode_audio_frame(const safe_ptr<core::read_frame>& frame)\r
+ byte_vector convert_audio(core::read_frame& frame, AVCodecContext* c)\r
+ {\r
+ if(!swr_) \r
+ swr_.reset(new audio_resampler(c->channels, format_desc_.audio_channels, \r
+ c->sample_rate, format_desc_.audio_sample_rate,\r
+ c->sample_fmt, AV_SAMPLE_FMT_S32));\r
+ \r
+\r
+ auto audio_data = frame.audio_data();\r
+\r
+ std::vector<int8_t, tbb::cache_aligned_allocator<int8_t>> audio_resample_buffer;\r
+ std::copy(reinterpret_cast<const uint8_t*>(audio_data.begin()), \r
+ reinterpret_cast<const uint8_t*>(audio_data.begin()) + audio_data.size()*4, \r
+ std::back_inserter(audio_resample_buffer));\r
+ \r
+ audio_resample_buffer = swr_->resample(std::move(audio_resample_buffer));\r
+ \r
+ return byte_vector(audio_resample_buffer.begin(), audio_resample_buffer.end());\r
+ }\r
+\r
+ void encode_audio_frame(core::read_frame& frame)\r
{ \r
auto c = audio_st_->codec;\r
\r
- auto audio_data = core::audio_32_to_16(frame->audio_data());\r
+ boost::range::push_back(audio_buf_, convert_audio(frame, c));\r
\r
- safe_ptr<AVPacket> pkt(new AVPacket, [](AVPacket* p)\r
+ std::size_t frame_size = c->frame_size;\r
+ auto input_audio_size = frame_size * av_get_bytes_per_sample(c->sample_fmt) * c->channels;\r
+ \r
+ while(audio_buf_.size() >= input_audio_size)\r
{\r
- av_free_packet(p);\r
- delete p;\r
- });\r
- av_init_packet(pkt.get());\r
+ safe_ptr<AVPacket> pkt(new AVPacket, [](AVPacket* p)\r
+ {\r
+ av_free_packet(p);\r
+ delete p;\r
+ });\r
+ av_init_packet(pkt.get());\r
+\r
+ if(frame_size > 1)\r
+ { \r
+ pkt->size = avcodec_encode_audio(c, audio_outbuf_.data(), audio_outbuf_.size(), reinterpret_cast<short*>(audio_buf_.data()));\r
+ audio_buf_.erase(audio_buf_.begin(), audio_buf_.begin() + input_audio_size);\r
+ }\r
+ else\r
+ {\r
+ audio_outbuf_ = std::move(audio_buf_); \r
+ audio_buf_.clear();\r
+ pkt->size = audio_outbuf_.size();\r
+ pkt->data = audio_outbuf_.data();\r
+ }\r
\r
- if (c->coded_frame && c->coded_frame->pts != AV_NOPTS_VALUE)\r
- pkt->pts = av_rescale_q(c->coded_frame->pts, c->time_base, audio_st_->time_base);\r
+ if(pkt->size == 0)\r
+ return;\r
\r
- pkt->flags |= AV_PKT_FLAG_KEY;\r
- pkt->stream_index = audio_st_->index;\r
- pkt->size = audio_data.size()*2;\r
- pkt->data = reinterpret_cast<uint8_t*>(audio_data.data());\r
+ if (c->coded_frame && c->coded_frame->pts != AV_NOPTS_VALUE)\r
+ pkt->pts = av_rescale_q(c->coded_frame->pts, c->time_base, audio_st_->time_base);\r
+\r
+ pkt->flags |= AV_PKT_FLAG_KEY;\r
+ pkt->stream_index = audio_st_->index;\r
+ pkt->data = reinterpret_cast<uint8_t*>(audio_outbuf_.data());\r
\r
- av_dup_packet(pkt.get());\r
- return pkt;\r
+ av_interleaved_write_frame(oc_.get(), pkt.get());\r
+ }\r
}\r
\r
void send(const safe_ptr<core::read_frame>& frame)\r
{\r
- executor_.begin_invoke([=]\r
+ encode_executor_.begin_invoke([=]\r
{ \r
- frame_timer_.restart();\r
+ boost::timer frame_timer;\r
\r
- auto video = encode_video_frame(frame);\r
- auto audio = encode_audio_frame(frame);\r
+ encode_video_frame(*frame);\r
\r
- graph_->set_value("frame-time", frame_timer_.elapsed()*format_desc_.fps*0.5);\r
- \r
- file_write_executor_.begin_invoke([=]\r
- {\r
- write_timer_.restart();\r
-\r
- if(video)\r
- av_write_frame(oc_.get(), video.get());\r
- if(audio)\r
- av_write_frame(oc_.get(), audio.get());\r
+ if (!key_only_)\r
+ encode_audio_frame(*frame);\r
\r
- graph_->set_value("write-time", write_timer_.elapsed()*format_desc_.fps*0.5);\r
- });\r
+ graph_->set_value("frame-time", frame_timer.elapsed()*format_desc_.fps*0.5); \r
});\r
}\r
+\r
+ bool ready_for_frame()\r
+ {\r
+ return encode_executor_.size() < encode_executor_.capacity();\r
+ }\r
+\r
+ void mark_dropped()\r
+ {\r
+ graph_->set_tag("dropped-frame");\r
+\r
+ // TODO: adjust PTS accordingly to make dropped frames contribute\r
+ // to the total playing time\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
+ const bool separate_key_;\r
\r
std::unique_ptr<ffmpeg_consumer> consumer_;\r
+ std::unique_ptr<ffmpeg_consumer> key_only_consumer_;\r
\r
public:\r
\r
- ffmpeg_consumer_proxy(const std::wstring& filename, const std::vector<option>& options)\r
+ ffmpeg_consumer_proxy(const std::wstring& filename, const std::vector<option>& options, bool separate_key_)\r
: filename_(filename)\r
, options_(options)\r
+ , separate_key_(separate_key_)\r
{\r
}\r
\r
virtual void initialize(const core::video_format_desc& format_desc, int)\r
{\r
consumer_.reset();\r
- consumer_.reset(new ffmpeg_consumer(narrow(filename_), format_desc, options_));\r
+ key_only_consumer_.reset();\r
+ consumer_.reset(new ffmpeg_consumer(narrow(filename_), format_desc, options_, false));\r
+\r
+ if (separate_key_)\r
+ {\r
+ boost::filesystem::wpath fill_file(filename_);\r
+ auto without_extension = fill_file.stem();\r
+ auto key_file = env::media_folder() + without_extension + L"_A" + fill_file.extension();\r
+ \r
+ key_only_consumer_.reset(new ffmpeg_consumer(narrow(key_file), format_desc, options_, true));\r
+ }\r
}\r
\r
- virtual bool send(const safe_ptr<core::read_frame>& frame) override\r
+ virtual boost::unique_future<bool> send(const safe_ptr<core::read_frame>& frame) override\r
{\r
- consumer_->send(frame);\r
- return true;\r
+ bool ready_for_frame = consumer_->ready_for_frame();\r
+\r
+ if (ready_for_frame && separate_key_)\r
+ ready_for_frame = ready_for_frame && key_only_consumer_->ready_for_frame();\r
+\r
+ if (ready_for_frame)\r
+ {\r
+ consumer_->send(frame);\r
+\r
+ if (separate_key_)\r
+ key_only_consumer_->send(frame);\r
+ }\r
+ else\r
+ {\r
+ consumer_->mark_dropped();\r
+\r
+ if (separate_key_)\r
+ key_only_consumer_->mark_dropped();\r
+ }\r
+\r
+ return caspar::wrap_as_future(true);\r
}\r
\r
virtual std::wstring print() const override\r
boost::property_tree::wptree info;\r
info.add(L"type", L"ffmpeg-consumer");\r
info.add(L"filename", filename_);\r
+ info.add(L"separate_key", separate_key_);\r
return info;\r
}\r
\r
{\r
if(params.size() < 1 || params[0] != L"FILE")\r
return core::frame_consumer::empty();\r
+\r
+ std::vector<std::wstring> params2 = params;\r
\r
- auto filename = (params.size() > 1 ? params[1] : L"");\r
+ auto filename = (params2.size() > 1 ? params2[1] : L"");\r
+ auto separate_key_it = std::find(params2.begin(), params2.end(), L"SEPARATE_KEY");\r
+ bool separate_key = false;\r
+\r
+ if (separate_key_it != params2.end())\r
+ {\r
+ separate_key = true;\r
+ params2.erase(separate_key_it);\r
+ }\r
\r
std::vector<option> options;\r
\r
- if(params.size() >= 3)\r
+ if(params2.size() >= 3)\r
{\r
- for(auto opt_it = params.begin()+2; opt_it != params.end();)\r
+ for(auto opt_it = params2.begin()+2; opt_it != params2.end();)\r
{\r
auto name = narrow(boost::trim_copy(boost::to_lower_copy(*opt_it++))).substr(1);\r
auto value = narrow(boost::trim_copy(boost::to_lower_copy(*opt_it++)));\r
}\r
}\r
\r
- return make_safe<ffmpeg_consumer_proxy>(env::media_folder() + filename, options);\r
+ return make_safe<ffmpeg_consumer_proxy>(env::media_folder() + filename, options, separate_key);\r
}\r
\r
safe_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
+ auto filename = ptree.get<std::wstring>(L"path");\r
+ auto codec = ptree.get(L"vcodec", L"libx264");\r
+ auto separate_key = ptree.get(L"separate-key", false);\r
\r
std::vector<option> options;\r
options.push_back(option("vcodec", narrow(codec)));\r
\r
- return make_safe<ffmpeg_consumer_proxy>(env::media_folder() + filename, options);\r
+ return make_safe<ffmpeg_consumer_proxy>(env::media_folder() + filename, options, separate_key);\r
}\r
\r
}}\r
projects / casparcg / blobdiff