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
40 #include <common/utility/assert.h>
\r
41 #include <common/memory/memcpy.h>
\r
43 #include <tbb/parallel_for.h>
\r
45 #include <boost/filesystem.hpp>
\r
46 #include <boost/lexical_cast.hpp>
\r
48 #if defined(_MSC_VER)
\r
49 #pragma warning (push)
\r
50 #pragma warning (disable : 4244)
\r
54 #include <libswscale/swscale.h>
\r
55 #include <libavcodec/avcodec.h>
\r
56 #include <libavformat/avformat.h>
\r
58 #if defined(_MSC_VER)
\r
59 #pragma warning (pop)
\r
62 namespace caspar { namespace ffmpeg {
\r
64 std::shared_ptr<core::audio_buffer> flush_audio()
\r
66 static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());
\r
70 std::shared_ptr<core::audio_buffer> empty_audio()
\r
72 static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());
\r
76 std::shared_ptr<AVFrame> flush_video()
\r
78 static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);
\r
82 std::shared_ptr<AVFrame> empty_video()
\r
84 static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);
\r
88 core::field_mode::type get_mode(const AVFrame& frame)
\r
90 if(!frame.interlaced_frame)
\r
91 return core::field_mode::progressive;
\r
93 return frame.top_field_first ? core::field_mode::upper : core::field_mode::lower;
\r
96 core::pixel_format::type get_pixel_format(PixelFormat pix_fmt)
\r
100 case CASPAR_PIX_FMT_LUMA: return core::pixel_format::luma;
\r
101 case PIX_FMT_GRAY8: return core::pixel_format::gray;
\r
102 case PIX_FMT_BGRA: return core::pixel_format::bgra;
\r
103 case PIX_FMT_ARGB: return core::pixel_format::argb;
\r
104 case PIX_FMT_RGBA: return core::pixel_format::rgba;
\r
105 case PIX_FMT_ABGR: return core::pixel_format::abgr;
\r
106 case PIX_FMT_YUV444P: return core::pixel_format::ycbcr;
\r
107 case PIX_FMT_YUV422P: return core::pixel_format::ycbcr;
\r
108 case PIX_FMT_YUV420P: return core::pixel_format::ycbcr;
\r
109 case PIX_FMT_YUV411P: return core::pixel_format::ycbcr;
\r
110 case PIX_FMT_YUV410P: return core::pixel_format::ycbcr;
\r
111 case PIX_FMT_YUVA420P: return core::pixel_format::ycbcra;
\r
112 default: return core::pixel_format::invalid;
\r
116 core::pixel_format_desc get_pixel_format_desc(PixelFormat pix_fmt, size_t width, size_t height)
\r
119 AVPicture dummy_pict;
\r
120 avpicture_fill(&dummy_pict, nullptr, pix_fmt == CASPAR_PIX_FMT_LUMA ? PIX_FMT_GRAY8 : pix_fmt, width, height);
\r
122 core::pixel_format_desc desc;
\r
123 desc.pix_fmt = get_pixel_format(pix_fmt);
\r
125 switch(desc.pix_fmt)
\r
127 case core::pixel_format::gray:
\r
128 case core::pixel_format::luma:
\r
130 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));
\r
133 case core::pixel_format::bgra:
\r
134 case core::pixel_format::argb:
\r
135 case core::pixel_format::rgba:
\r
136 case core::pixel_format::abgr:
\r
138 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0]/4, height, 4));
\r
141 case core::pixel_format::ycbcr:
\r
142 case core::pixel_format::ycbcra:
\r
144 // Find chroma height
\r
145 size_t size2 = dummy_pict.data[2] - dummy_pict.data[1];
\r
146 size_t h2 = size2/dummy_pict.linesize[1];
\r
148 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));
\r
149 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[1], h2, 1));
\r
150 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[2], h2, 1));
\r
152 if(desc.pix_fmt == core::pixel_format::ycbcra)
\r
153 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[3], height, 1));
\r
157 desc.pix_fmt = core::pixel_format::invalid;
\r
162 int make_alpha_format(int format)
\r
164 switch(get_pixel_format(static_cast<PixelFormat>(format)))
\r
166 case core::pixel_format::ycbcr:
\r
167 case core::pixel_format::ycbcra:
\r
168 return CASPAR_PIX_FMT_LUMA;
\r
174 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
176 static tbb::concurrent_unordered_map<size_t, tbb::concurrent_queue<std::shared_ptr<SwsContext>>> sws_contexts_;
\r
178 if(decoded_frame->width < 1 || decoded_frame->height < 1)
\r
179 return make_safe<core::write_frame>(tag);
\r
181 const auto width = decoded_frame->width;
\r
182 const auto height = decoded_frame->height;
\r
183 auto desc = get_pixel_format_desc(static_cast<PixelFormat>(decoded_frame->format), width, height);
\r
185 if(flags & core::frame_producer::ALPHA_ONLY_FLAG)
\r
186 desc = get_pixel_format_desc(static_cast<PixelFormat>(make_alpha_format(decoded_frame->format)), width, height);
\r
188 std::shared_ptr<core::write_frame> write;
\r
190 if(desc.pix_fmt == core::pixel_format::invalid)
\r
192 auto pix_fmt = static_cast<PixelFormat>(decoded_frame->format);
\r
193 auto target_pix_fmt = PIX_FMT_BGRA;
\r
195 if(pix_fmt == PIX_FMT_UYVY422)
\r
196 target_pix_fmt = PIX_FMT_YUV422P;
\r
197 else if(pix_fmt == PIX_FMT_YUYV422)
\r
198 target_pix_fmt = PIX_FMT_YUV422P;
\r
199 else if(pix_fmt == PIX_FMT_UYYVYY411)
\r
200 target_pix_fmt = PIX_FMT_YUV411P;
\r
201 else if(pix_fmt == PIX_FMT_YUV420P10)
\r
202 target_pix_fmt = PIX_FMT_YUV420P;
\r
203 else if(pix_fmt == PIX_FMT_YUV422P10)
\r
204 target_pix_fmt = PIX_FMT_YUV422P;
\r
205 else if(pix_fmt == PIX_FMT_YUV444P10)
\r
206 target_pix_fmt = PIX_FMT_YUV444P;
\r
208 auto target_desc = get_pixel_format_desc(target_pix_fmt, width, height);
\r
210 write = frame_factory->create_frame(tag, target_desc);
\r
211 write->set_type(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 size_t 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(size_t 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);
\r
259 write->set_type(get_mode(*decoded_frame));
\r
261 for(int n = 0; n < static_cast<int>(desc.planes.size()); ++n)
\r
263 auto plane = desc.planes[n];
\r
264 auto result = write->image_data(n).begin();
\r
265 auto decoded = decoded_frame->data[n];
\r
266 auto decoded_linesize = decoded_frame->linesize[n];
\r
268 CASPAR_ASSERT(decoded);
\r
269 CASPAR_ASSERT(write->image_data(n).begin());
\r
271 if(decoded_linesize != static_cast<int>(plane.width))
\r
273 // Copy line by line since ffmpeg sometimes pads each line.
\r
274 tbb::parallel_for<size_t>(0, desc.planes[n].height, [&](size_t y)
\r
276 fast_memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);
\r
281 fast_memcpy(result, decoded, plane.size);
\r
288 if(decoded_frame->height == 480) // NTSC DV
\r
290 write->get_frame_transform().fill_translation[1] += 2.0/static_cast<double>(frame_factory->get_video_format_desc().height);
\r
291 write->get_frame_transform().fill_scale[1] = 1.0 - 6.0*1.0/static_cast<double>(frame_factory->get_video_format_desc().height);
\r
294 // Fix field-order if needed
\r
295 if(write->get_type() == core::field_mode::lower && frame_factory->get_video_format_desc().field_mode == core::field_mode::upper)
\r
296 write->get_frame_transform().fill_translation[1] += 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);
\r
297 else if(write->get_type() == core::field_mode::upper && frame_factory->get_video_format_desc().field_mode == core::field_mode::lower)
\r
298 write->get_frame_transform().fill_translation[1] -= 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);
\r
300 return make_safe_ptr(write);
\r
303 bool is_sane_fps(AVRational time_base)
\r
305 double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);
\r
306 return fps > 20.0 && fps < 65.0;
\r
309 AVRational fix_time_base(AVRational time_base)
\r
311 if(time_base.num == 1)
\r
312 time_base.num = static_cast<int>(std::pow(10.0, static_cast<int>(std::log10(static_cast<float>(time_base.den)))-1));
\r
314 if(!is_sane_fps(time_base))
\r
316 auto tmp = time_base;
\r
318 if(is_sane_fps(tmp))
\r
325 double read_fps(AVFormatContext& context, double fail_value)
\r
327 auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);
\r
328 auto audio_index = av_find_best_stream(&context, AVMEDIA_TYPE_AUDIO, -1, -1, 0, 0);
\r
330 if(video_index > -1)
\r
332 const auto video_context = context.streams[video_index]->codec;
\r
333 const auto video_stream = context.streams[video_index];
\r
335 AVRational time_base = video_context->time_base;
\r
337 if(boost::filesystem2::path(context.filename).extension() == ".flv")
\r
341 auto meta = read_flv_meta_info(context.filename);
\r
342 return boost::lexical_cast<double>(meta["framerate"]);
\r
351 time_base.num *= video_context->ticks_per_frame;
\r
353 if(!is_sane_fps(time_base))
\r
355 time_base = fix_time_base(time_base);
\r
357 if(!is_sane_fps(time_base) && audio_index > -1)
\r
359 auto& audio_context = *context.streams[audio_index]->codec;
\r
360 auto& audio_stream = *context.streams[audio_index];
\r
362 double duration_sec = audio_stream.duration / static_cast<double>(audio_context.sample_rate);
\r
364 time_base.num = static_cast<int>(duration_sec*100000.0);
\r
365 time_base.den = static_cast<int>(video_stream->nb_frames*100000);
\r
370 double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);
\r
372 double closest_fps = 0.0;
\r
373 for(int n = 0; n < core::video_format::count; ++n)
\r
375 auto format = core::video_format_desc::get(static_cast<core::video_format::type>(n));
\r
377 double diff1 = std::abs(format.fps - fps);
\r
378 double diff2 = std::abs(closest_fps - fps);
\r
381 closest_fps = format.fps;
\r
384 return closest_fps;
\r
387 return fail_value;
\r
390 void fix_meta_data(AVFormatContext& context)
\r
392 auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);
\r
394 if(video_index > -1)
\r
396 auto video_stream = context.streams[video_index];
\r
397 auto video_context = context.streams[video_index]->codec;
\r
399 if(boost::filesystem2::path(context.filename).extension() == ".flv")
\r
403 auto meta = read_flv_meta_info(context.filename);
\r
404 double fps = boost::lexical_cast<double>(meta["framerate"]);
\r
405 video_stream->nb_frames = static_cast<int64_t>(boost::lexical_cast<double>(meta["duration"])*fps);
\r
411 auto stream_time = video_stream->time_base;
\r
412 auto duration = video_stream->duration;
\r
413 auto codec_time = video_context->time_base;
\r
414 auto ticks = video_context->ticks_per_frame;
\r
416 if(video_stream->nb_frames == 0)
\r
417 video_stream->nb_frames = (duration*stream_time.num*codec_time.den)/(stream_time.den*codec_time.num*ticks);
\r
422 safe_ptr<AVPacket> create_packet()
\r
424 safe_ptr<AVPacket> packet(new AVPacket, [](AVPacket* p)
\r
430 av_init_packet(packet.get());
\r
434 safe_ptr<AVCodecContext> open_codec(AVFormatContext& context, enum AVMediaType type, int& index)
\r
437 index = THROW_ON_ERROR2(av_find_best_stream(&context, type, -1, -1, &decoder, 0), "");
\r
438 //if(strcmp(decoder->name, "prores") == 0 && decoder->next && strcmp(decoder->next->name, "prores_lgpl") == 0)
\r
439 // decoder = decoder->next;
\r
441 THROW_ON_ERROR2(tbb_avcodec_open(context.streams[index]->codec, decoder), "");
\r
442 return safe_ptr<AVCodecContext>(context.streams[index]->codec, tbb_avcodec_close);
\r
445 safe_ptr<AVFormatContext> open_input(const std::wstring& filename)
\r
447 AVFormatContext* weak_context = nullptr;
\r
448 THROW_ON_ERROR2(avformat_open_input(&weak_context, u8(filename).c_str(), nullptr, nullptr), filename);
\r
449 safe_ptr<AVFormatContext> context(weak_context, av_close_input_file);
\r
450 THROW_ON_ERROR2(avformat_find_stream_info(weak_context, nullptr), filename);
\r
451 fix_meta_data(*context);
\r
455 std::wstring print_mode(size_t width, size_t height, double fps, bool interlaced)
\r
457 std::wostringstream fps_ss;
\r
458 fps_ss << std::fixed << std::setprecision(2) << (!interlaced ? fps : 2.0 * fps);
\r
460 return boost::lexical_cast<std::wstring>(width) + L"x" + boost::lexical_cast<std::wstring>(height) + (!interlaced ? L"p" : L"i") + fps_ss.str();
\r
463 bool is_valid_file(const std::wstring filename)
\r
465 std::ifstream file(filename);
\r
467 std::vector<unsigned char> buf;
\r
468 for(auto file_it = std::istreambuf_iterator<char>(file); file_it != std::istreambuf_iterator<char>() && buf.size() < 2048; ++file_it)
\r
469 buf.push_back(*file_it);
\r
475 pb.filename = u8(filename).c_str();
\r
476 pb.buf = buf.data();
\r
477 pb.buf_size = buf.size();
\r
480 return av_probe_input_format2(&pb, true, &score) != nullptr;
\r
483 std::wstring probe_stem(const std::wstring stem)
\r
485 auto stem2 = boost::filesystem2::wpath(stem);
\r
486 auto dir = stem2.parent_path();
\r
487 for(auto it = boost::filesystem2::wdirectory_iterator(dir); it != boost::filesystem2::wdirectory_iterator(); ++it)
\r
489 if(boost::iequals(it->path().stem(), stem2.filename()) && is_valid_file(it->path().file_string()))
\r
490 return it->path().file_string();
\r
495 //void av_dup_frame(AVFrame* frame)
\r
497 // AVFrame* new_frame = avcodec_alloc_frame();
\r
500 // const uint8_t *src_data[4] = {0};
\r
501 // memcpy(const_cast<uint8_t**>(&src_data[0]), frame->data, 4);
\r
502 // const int src_linesizes[4] = {0};
\r
503 // memcpy(const_cast<int*>(&src_linesizes[0]), frame->linesize, 4);
\r
505 // av_image_alloc(new_frame->data, new_frame->linesize, new_frame->width, new_frame->height, frame->format, 16);
\r
507 // av_image_copy(new_frame->data, new_frame->linesize, src_data, src_linesizes, frame->format, new_frame->width, new_frame->height);
\r