2 * Copyright (c) 2011 Sveriges Television AB <info@casparcg.com>
\r
4 * This file is part of CasparCG (www.casparcg.com).
\r
6 * CasparCG is free software: you can redistribute it and/or modify
\r
7 * it under the terms of the GNU General Public License as published by
\r
8 * the Free Software Foundation, either version 3 of the License, or
\r
9 * (at your option) any later version.
\r
11 * CasparCG is distributed in the hope that it will be useful,
\r
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
\r
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
\r
14 * GNU General Public License for more details.
\r
16 * You should have received a copy of the GNU General Public License
\r
17 * along with CasparCG. If not, see <http://www.gnu.org/licenses/>.
\r
19 * Author: Robert Nagy, ronag89@gmail.com
\r
22 #include "../../stdafx.h"
\r
28 #include "../tbb_avcodec.h"
\r
29 #include "../../ffmpeg_error.h"
\r
31 #include <tbb/concurrent_unordered_map.h>
\r
32 #include <tbb/concurrent_queue.h>
\r
34 #include <core/producer/frame/frame_transform.h>
\r
35 #include <core/producer/frame/frame_factory.h>
\r
36 #include <core/producer/frame_producer.h>
\r
37 #include <core/mixer/write_frame.h>
\r
39 #include <common/exception/exceptions.h>
\r
41 #include <tbb/parallel_for.h>
\r
43 #include <common/assert.h>
\r
44 #include <boost/filesystem.hpp>
\r
45 #include <boost/lexical_cast.hpp>
\r
49 #if defined(_MSC_VER)
\r
50 #pragma warning (push)
\r
51 #pragma warning (disable : 4244)
\r
55 #include <libswscale/swscale.h>
\r
56 #include <libavcodec/avcodec.h>
\r
57 #include <libavformat/avformat.h>
\r
59 #if defined(_MSC_VER)
\r
60 #pragma warning (pop)
\r
63 namespace caspar { namespace ffmpeg {
\r
65 std::shared_ptr<core::audio_buffer> flush_audio()
\r
67 static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());
\r
71 std::shared_ptr<core::audio_buffer> empty_audio()
\r
73 static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());
\r
77 std::shared_ptr<AVFrame> flush_video()
\r
79 static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);
\r
83 std::shared_ptr<AVFrame> empty_video()
\r
85 static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);
\r
89 core::field_mode get_mode(const AVFrame& frame)
\r
91 if(!frame.interlaced_frame)
\r
92 return core::field_mode::progressive;
\r
94 return frame.top_field_first ? core::field_mode::upper : core::field_mode::lower;
\r
97 core::pixel_format get_pixel_format(PixelFormat pix_fmt)
\r
101 case CASPAR_PIX_FMT_LUMA: return core::pixel_format::luma;
\r
102 case PIX_FMT_GRAY8: return core::pixel_format::gray;
\r
103 case PIX_FMT_BGRA: return core::pixel_format::bgra;
\r
104 case PIX_FMT_ARGB: return core::pixel_format::argb;
\r
105 case PIX_FMT_RGBA: return core::pixel_format::rgba;
\r
106 case PIX_FMT_ABGR: return core::pixel_format::abgr;
\r
107 case PIX_FMT_YUV444P: return core::pixel_format::ycbcr;
\r
108 case PIX_FMT_YUV422P: return core::pixel_format::ycbcr;
\r
109 case PIX_FMT_YUV420P: return core::pixel_format::ycbcr;
\r
110 case PIX_FMT_YUV411P: return core::pixel_format::ycbcr;
\r
111 case PIX_FMT_YUV410P: return core::pixel_format::ycbcr;
\r
112 case PIX_FMT_YUVA420P: return core::pixel_format::ycbcra;
\r
113 default: return core::pixel_format::invalid;
\r
117 core::pixel_format_desc get_pixel_format_desc(PixelFormat pix_fmt, int width, int height)
\r
120 AVPicture dummy_pict;
\r
121 avpicture_fill(&dummy_pict, nullptr, pix_fmt == CASPAR_PIX_FMT_LUMA ? PIX_FMT_GRAY8 : pix_fmt, width, height);
\r
123 core::pixel_format_desc desc;
\r
124 desc.pix_fmt = get_pixel_format(pix_fmt);
\r
126 switch(desc.pix_fmt.value())
\r
128 case core::pixel_format::gray:
\r
129 case core::pixel_format::luma:
\r
131 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));
\r
134 case core::pixel_format::bgra:
\r
135 case core::pixel_format::argb:
\r
136 case core::pixel_format::rgba:
\r
137 case core::pixel_format::abgr:
\r
139 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0]/4, height, 4));
\r
142 case core::pixel_format::ycbcr:
\r
143 case core::pixel_format::ycbcra:
\r
145 // Find chroma height
\r
146 int size2 = static_cast<int>(dummy_pict.data[2] - dummy_pict.data[1]);
\r
147 int h2 = size2/dummy_pict.linesize[1];
\r
149 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));
\r
150 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[1], h2, 1));
\r
151 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[2], h2, 1));
\r
153 if(desc.pix_fmt == core::pixel_format::ycbcra)
\r
154 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[3], height, 1));
\r
158 desc.pix_fmt = core::pixel_format::invalid;
\r
163 int make_alpha_format(int format)
\r
165 switch(get_pixel_format(static_cast<PixelFormat>(format)).value())
\r
167 case core::pixel_format::ycbcr:
\r
168 case core::pixel_format::ycbcra:
\r
169 return CASPAR_PIX_FMT_LUMA;
\r
175 safe_ptr<core::write_frame> make_write_frame(const void* tag, const safe_ptr<AVFrame>& decoded_frame, const safe_ptr<core::frame_factory>& frame_factory, int flags)
\r
177 static tbb::concurrent_unordered_map<int, tbb::concurrent_queue<std::shared_ptr<SwsContext>>> sws_contexts_;
\r
179 if(decoded_frame->width < 1 || decoded_frame->height < 1)
\r
180 return make_safe<core::write_frame>(tag);
\r
182 const auto width = decoded_frame->width;
\r
183 const auto height = decoded_frame->height;
\r
184 auto desc = get_pixel_format_desc(static_cast<PixelFormat>(decoded_frame->format), width, height);
\r
186 if(flags & core::frame_producer::flags::alpha_only)
\r
187 desc = get_pixel_format_desc(static_cast<PixelFormat>(make_alpha_format(decoded_frame->format)), width, height);
\r
189 std::shared_ptr<core::write_frame> write;
\r
191 if(desc.pix_fmt == core::pixel_format::invalid)
\r
193 auto pix_fmt = static_cast<PixelFormat>(decoded_frame->format);
\r
194 auto target_pix_fmt = PIX_FMT_BGRA;
\r
196 if(pix_fmt == PIX_FMT_UYVY422)
\r
197 target_pix_fmt = PIX_FMT_YUV422P;
\r
198 else if(pix_fmt == PIX_FMT_YUYV422)
\r
199 target_pix_fmt = PIX_FMT_YUV422P;
\r
200 else if(pix_fmt == PIX_FMT_UYYVYY411)
\r
201 target_pix_fmt = PIX_FMT_YUV411P;
\r
202 else if(pix_fmt == PIX_FMT_YUV420P10)
\r
203 target_pix_fmt = PIX_FMT_YUV420P;
\r
204 else if(pix_fmt == PIX_FMT_YUV422P10)
\r
205 target_pix_fmt = PIX_FMT_YUV422P;
\r
206 else if(pix_fmt == PIX_FMT_YUV444P10)
\r
207 target_pix_fmt = PIX_FMT_YUV444P;
\r
209 auto target_desc = get_pixel_format_desc(target_pix_fmt, width, height);
\r
211 write = frame_factory->create_frame(tag, target_desc, get_mode(*decoded_frame));
\r
213 std::shared_ptr<SwsContext> sws_context;
\r
215 //CASPAR_LOG(warning) << "Hardware accelerated color transform not supported.";
\r
217 int key = ((width << 22) & 0xFFC00000) | ((height << 6) & 0x003FC000) | ((pix_fmt << 7) & 0x00007F00) | ((target_pix_fmt << 0) & 0x0000007F);
\r
219 auto& pool = sws_contexts_[key];
\r
221 if(!pool.try_pop(sws_context))
\r
224 sws_context.reset(sws_getContext(width, height, pix_fmt, width, height, target_pix_fmt, SWS_BILINEAR, nullptr, nullptr, ¶m), sws_freeContext);
\r
229 BOOST_THROW_EXCEPTION(operation_failed() << msg_info("Could not create software scaling context.") <<
\r
230 boost::errinfo_api_function("sws_getContext"));
\r
233 safe_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);
\r
234 avcodec_get_frame_defaults(av_frame.get());
\r
235 if(target_pix_fmt == PIX_FMT_BGRA)
\r
237 auto size = avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), write->image_data().begin(), PIX_FMT_BGRA, width, height);
\r
238 CASPAR_VERIFY(size == write->image_data().size());
\r
242 av_frame->width = width;
\r
243 av_frame->height = height;
\r
244 for(int n = 0; n < target_desc.planes.size(); ++n)
\r
246 av_frame->data[n] = write->image_data(n).begin();
\r
247 av_frame->linesize[n] = target_desc.planes[n].linesize;
\r
251 sws_scale(sws_context.get(), decoded_frame->data, decoded_frame->linesize, 0, height, av_frame->data, av_frame->linesize);
\r
252 pool.push(sws_context);
\r
258 write = frame_factory->create_frame(tag, desc, get_mode(*decoded_frame));
\r
260 for(int n = 0; n < static_cast<int>(desc.planes.size()); ++n)
\r
262 auto plane = desc.planes[n];
\r
263 auto result = write->image_data(n).begin();
\r
264 auto decoded = decoded_frame->data[n];
\r
265 auto decoded_linesize = decoded_frame->linesize[n];
\r
267 CASPAR_ASSERT(decoded);
\r
268 CASPAR_ASSERT(write->image_data(n).begin());
\r
270 // Copy line by line since ffmpeg sometimes pads each line.
\r
271 tbb::affinity_partitioner ap;
\r
272 tbb::parallel_for(tbb::blocked_range<int>(0, desc.planes[n].height), [&](const tbb::blocked_range<int>& r)
\r
274 for(int y = r.begin(); y != r.end(); ++y)
\r
275 A_memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);
\r
282 if(decoded_frame->height == 480) // NTSC DV
\r
284 write->get_frame_transform().fill_translation[1] += 2.0/static_cast<double>(frame_factory->get_video_format_desc().height);
\r
285 write->get_frame_transform().fill_scale[1] = 1.0 - 6.0*1.0/static_cast<double>(frame_factory->get_video_format_desc().height);
\r
288 // Fix field-order if needed
\r
289 if(write->get_field_mode() == core::field_mode::lower && frame_factory->get_video_format_desc().field_mode == core::field_mode::upper)
\r
290 write->get_frame_transform().fill_translation[1] += 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);
\r
291 else if(write->get_field_mode() == core::field_mode::upper && frame_factory->get_video_format_desc().field_mode == core::field_mode::lower)
\r
292 write->get_frame_transform().fill_translation[1] -= 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);
\r
294 return make_safe_ptr(write);
\r
297 bool is_sane_fps(AVRational time_base)
\r
299 double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);
\r
300 return fps > 20.0 && fps < 65.0;
\r
303 AVRational fix_time_base(AVRational time_base)
\r
305 if(time_base.num == 1)
\r
306 time_base.num = static_cast<int>(std::pow(10.0, static_cast<int>(std::log10(static_cast<float>(time_base.den)))-1));
\r
308 if(!is_sane_fps(time_base))
\r
310 auto tmp = time_base;
\r
312 if(is_sane_fps(tmp))
\r
319 double read_fps(AVFormatContext& context, double fail_value)
\r
321 auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);
\r
322 auto audio_index = av_find_best_stream(&context, AVMEDIA_TYPE_AUDIO, -1, -1, 0, 0);
\r
324 if(video_index > -1)
\r
326 const auto video_context = context.streams[video_index]->codec;
\r
327 const auto video_stream = context.streams[video_index];
\r
329 AVRational time_base = video_context->time_base;
\r
331 if(boost::filesystem::path(context.filename).extension().string() == ".flv")
\r
335 auto meta = read_flv_meta_info(context.filename);
\r
336 return boost::lexical_cast<double>(meta["framerate"]);
\r
345 time_base.num *= video_context->ticks_per_frame;
\r
347 if(!is_sane_fps(time_base))
\r
349 time_base = fix_time_base(time_base);
\r
351 if(!is_sane_fps(time_base) && audio_index > -1)
\r
353 auto& audio_context = *context.streams[audio_index]->codec;
\r
354 auto& audio_stream = *context.streams[audio_index];
\r
356 double duration_sec = audio_stream.duration / static_cast<double>(audio_context.sample_rate);
\r
358 time_base.num = static_cast<int>(duration_sec*100000.0);
\r
359 time_base.den = static_cast<int>(video_stream->nb_frames*100000);
\r
364 double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);
\r
366 double closest_fps = 0.0;
\r
367 for(int n = 0; n < core::video_format::count; ++n)
\r
369 auto format = core::video_format_desc(core::video_format(n));
\r
371 double diff1 = std::abs(format.fps - fps);
\r
372 double diff2 = std::abs(closest_fps - fps);
\r
375 closest_fps = format.fps;
\r
378 return closest_fps;
\r
381 return fail_value;
\r
384 void fix_meta_data(AVFormatContext& context)
\r
386 auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);
\r
388 if(video_index > -1)
\r
390 auto video_stream = context.streams[video_index];
\r
391 auto video_context = context.streams[video_index]->codec;
\r
393 if(boost::filesystem::path(context.filename).extension().string() == ".flv")
\r
397 auto meta = read_flv_meta_info(context.filename);
\r
398 double fps = boost::lexical_cast<double>(meta["framerate"]);
\r
399 video_stream->nb_frames = static_cast<int64_t>(boost::lexical_cast<double>(meta["duration"])*fps);
\r
405 auto stream_time = video_stream->time_base;
\r
406 auto duration = video_stream->duration;
\r
407 auto codec_time = video_context->time_base;
\r
408 auto ticks = video_context->ticks_per_frame;
\r
410 if(video_stream->nb_frames == 0)
\r
411 video_stream->nb_frames = (duration*stream_time.num*codec_time.den)/(stream_time.den*codec_time.num*ticks);
\r
416 safe_ptr<AVPacket> create_packet()
\r
418 safe_ptr<AVPacket> packet(new AVPacket, [](AVPacket* p)
\r
424 av_init_packet(packet.get());
\r
428 safe_ptr<AVCodecContext> open_codec(AVFormatContext& context, enum AVMediaType type, int& index)
\r
431 index = THROW_ON_ERROR2(av_find_best_stream(&context, type, -1, -1, &decoder, 0), "");
\r
432 //if(strcmp(decoder->name, "prores") == 0 && decoder->next && strcmp(decoder->next->name, "prores_lgpl") == 0)
\r
433 // decoder = decoder->next;
\r
435 THROW_ON_ERROR2(tbb_avcodec_open(context.streams[index]->codec, decoder), "");
\r
436 return safe_ptr<AVCodecContext>(context.streams[index]->codec, tbb_avcodec_close);
\r
439 safe_ptr<AVFormatContext> open_input(const std::wstring& filename)
\r
441 AVFormatContext* weak_context = nullptr;
\r
442 THROW_ON_ERROR2(avformat_open_input(&weak_context, u8(filename).c_str(), nullptr, nullptr), filename);
\r
443 safe_ptr<AVFormatContext> context(weak_context, av_close_input_file);
\r
444 THROW_ON_ERROR2(avformat_find_stream_info(weak_context, nullptr), filename);
\r
445 fix_meta_data(*context);
\r
449 std::wstring print_mode(int width, int height, double fps, bool interlaced)
\r
451 std::wostringstream fps_ss;
\r
452 fps_ss << std::fixed << std::setprecision(2) << (!interlaced ? fps : 2.0 * fps);
\r
454 return boost::lexical_cast<std::wstring>(width) + L"x" + boost::lexical_cast<std::wstring>(height) + (!interlaced ? L"p" : L"i") + fps_ss.str();
\r
457 bool is_valid_file(const std::wstring filename)
\r
459 if(boost::filesystem::path(filename).extension() == ".m2t")
\r
462 std::ifstream file(filename);
\r
464 std::vector<unsigned char> buf;
\r
465 for(auto file_it = std::istreambuf_iterator<char>(file); file_it != std::istreambuf_iterator<char>() && buf.size() < 2048; ++file_it)
\r
466 buf.push_back(*file_it);
\r
472 pb.filename = u8(filename).c_str();
\r
473 pb.buf = buf.data();
\r
474 pb.buf_size = static_cast<int>(buf.size());
\r
477 return av_probe_input_format2(&pb, true, &score) != nullptr;
\r
480 std::wstring probe_stem(const std::wstring stem)
\r
482 auto stem2 = boost::filesystem::path(stem);
\r
483 auto dir = stem2.parent_path();
\r
484 for(auto it = boost::filesystem::directory_iterator(dir); it != boost::filesystem::directory_iterator(); ++it)
\r
486 if(boost::iequals(it->path().stem().wstring(), stem2.filename().wstring()) && is_valid_file(it->path().wstring()))
\r
487 return it->path().wstring();
\r
492 //void av_dup_frame(AVFrame* frame)
\r
494 // AVFrame* new_frame = avcodec_alloc_frame();
\r
497 // const uint8_t *src_data[4] = {0};
\r
498 // memcpy(const_cast<uint8_t**>(&src_data[0]), frame->data, 4);
\r
499 // const int src_linesizes[4] = {0};
\r
500 // memcpy(const_cast<int*>(&src_linesizes[0]), frame->linesize, 4);
\r
502 // av_image_alloc(new_frame->data, new_frame->linesize, new_frame->width, new_frame->height, frame->format, 16);
\r
504 // av_image_copy(new_frame->data, new_frame->linesize, src_data, src_linesizes, frame->format, new_frame->width, new_frame->height);
\r