2 * Copyright 2013 Sveriges Television AB http://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
38 #include <core/mixer/audio/audio_util.h>
\r
40 #include <common/exception/exceptions.h>
\r
41 #include <common/utility/assert.h>
\r
42 #include <common/memory/memcpy.h>
\r
44 #include <tbb/parallel_for.h>
\r
46 #include <boost/filesystem.hpp>
\r
47 #include <boost/lexical_cast.hpp>
\r
48 #include <boost/algorithm/string.hpp>
\r
50 #if defined(_MSC_VER)
\r
51 #pragma warning (push)
\r
52 #pragma warning (disable : 4244)
\r
56 #include <libswscale/swscale.h>
\r
57 #include <libavcodec/avcodec.h>
\r
58 #include <libavformat/avformat.h>
\r
60 #if defined(_MSC_VER)
\r
61 #pragma warning (pop)
\r
64 namespace caspar { namespace ffmpeg {
\r
66 std::shared_ptr<core::audio_buffer> flush_audio()
\r
68 static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());
\r
72 std::shared_ptr<core::audio_buffer> empty_audio()
\r
74 static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());
\r
78 std::shared_ptr<AVFrame> flush_video()
\r
80 static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);
\r
84 std::shared_ptr<AVFrame> empty_video()
\r
86 static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);
\r
90 core::field_mode::type get_mode(const AVFrame& frame)
\r
92 if(!frame.interlaced_frame)
\r
93 return core::field_mode::progressive;
\r
95 return frame.top_field_first ? core::field_mode::upper : core::field_mode::lower;
\r
98 core::pixel_format::type get_pixel_format(PixelFormat pix_fmt)
\r
102 case CASPAR_PIX_FMT_LUMA: return core::pixel_format::luma;
\r
103 case PIX_FMT_GRAY8: return core::pixel_format::gray;
\r
104 case PIX_FMT_BGRA: return core::pixel_format::bgra;
\r
105 case PIX_FMT_ARGB: return core::pixel_format::argb;
\r
106 case PIX_FMT_RGBA: return core::pixel_format::rgba;
\r
107 case PIX_FMT_ABGR: return core::pixel_format::abgr;
\r
108 case PIX_FMT_YUV444P: return core::pixel_format::ycbcr;
\r
109 case PIX_FMT_YUV422P: return core::pixel_format::ycbcr;
\r
110 case PIX_FMT_YUV420P: return core::pixel_format::ycbcr;
\r
111 case PIX_FMT_YUV411P: return core::pixel_format::ycbcr;
\r
112 case PIX_FMT_YUV410P: return core::pixel_format::ycbcr;
\r
113 case PIX_FMT_YUVA420P: return core::pixel_format::ycbcra;
\r
114 default: return core::pixel_format::invalid;
\r
118 core::pixel_format_desc get_pixel_format_desc(PixelFormat pix_fmt, size_t width, size_t height)
\r
121 AVPicture dummy_pict;
\r
122 avpicture_fill(&dummy_pict, nullptr, pix_fmt == CASPAR_PIX_FMT_LUMA ? PIX_FMT_GRAY8 : pix_fmt, width, height);
\r
124 core::pixel_format_desc desc;
\r
125 desc.pix_fmt = get_pixel_format(pix_fmt);
\r
127 switch(desc.pix_fmt)
\r
129 case core::pixel_format::gray:
\r
130 case core::pixel_format::luma:
\r
132 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));
\r
135 case core::pixel_format::bgra:
\r
136 case core::pixel_format::argb:
\r
137 case core::pixel_format::rgba:
\r
138 case core::pixel_format::abgr:
\r
140 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0]/4, height, 4));
\r
143 case core::pixel_format::ycbcr:
\r
144 case core::pixel_format::ycbcra:
\r
146 // Find chroma height
\r
147 size_t size2 = dummy_pict.data[2] - dummy_pict.data[1];
\r
148 size_t h2 = size2/dummy_pict.linesize[1];
\r
150 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));
\r
151 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[1], h2, 1));
\r
152 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[2], h2, 1));
\r
154 if(desc.pix_fmt == core::pixel_format::ycbcra)
\r
155 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[3], height, 1));
\r
159 desc.pix_fmt = core::pixel_format::invalid;
\r
164 int make_alpha_format(int format)
\r
166 switch(get_pixel_format(static_cast<PixelFormat>(format)))
\r
168 case core::pixel_format::ycbcr:
\r
169 case core::pixel_format::ycbcra:
\r
170 return CASPAR_PIX_FMT_LUMA;
\r
176 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 hints, const core::channel_layout& audio_channel_layout)
\r
178 static tbb::concurrent_unordered_map<int64_t, tbb::concurrent_queue<std::shared_ptr<SwsContext>>> sws_contexts_;
\r
180 if(decoded_frame->width < 1 || decoded_frame->height < 1)
\r
181 return make_safe<core::write_frame>(tag, audio_channel_layout);
\r
183 const auto width = decoded_frame->width;
\r
184 const auto height = decoded_frame->height;
\r
185 auto desc = get_pixel_format_desc(static_cast<PixelFormat>(decoded_frame->format), width, height);
\r
187 if(hints & core::frame_producer::ALPHA_HINT)
\r
188 desc = get_pixel_format_desc(static_cast<PixelFormat>(make_alpha_format(decoded_frame->format)), width, height);
\r
190 std::shared_ptr<core::write_frame> write;
\r
192 if(desc.pix_fmt == core::pixel_format::invalid)
\r
194 auto pix_fmt = static_cast<PixelFormat>(decoded_frame->format);
\r
195 auto target_pix_fmt = PIX_FMT_BGRA;
\r
197 if(pix_fmt == PIX_FMT_UYVY422)
\r
198 target_pix_fmt = PIX_FMT_YUV422P;
\r
199 else if(pix_fmt == PIX_FMT_YUYV422)
\r
200 target_pix_fmt = PIX_FMT_YUV422P;
\r
201 else if(pix_fmt == PIX_FMT_UYYVYY411)
\r
202 target_pix_fmt = PIX_FMT_YUV411P;
\r
203 else if(pix_fmt == PIX_FMT_YUV420P10)
\r
204 target_pix_fmt = PIX_FMT_YUV420P;
\r
205 else if(pix_fmt == PIX_FMT_YUV422P10)
\r
206 target_pix_fmt = PIX_FMT_YUV422P;
\r
207 else if(pix_fmt == PIX_FMT_YUV444P10)
\r
208 target_pix_fmt = PIX_FMT_YUV444P;
\r
210 auto target_desc = get_pixel_format_desc(target_pix_fmt, width, height);
\r
212 write = frame_factory->create_frame(tag, target_desc, audio_channel_layout);
\r
213 write->set_type(get_mode(*decoded_frame));
\r
215 std::shared_ptr<SwsContext> sws_context;
\r
217 //CASPAR_LOG(warning) << "Hardware accelerated color transform not supported.";
\r
219 int64_t key = ((static_cast<int64_t>(width) << 32) & 0xFFFF00000000) |
\r
220 ((static_cast<int64_t>(height) << 16) & 0xFFFF0000) |
\r
221 ((static_cast<int64_t>(pix_fmt) << 8) & 0xFF00) |
\r
222 ((static_cast<int64_t>(target_pix_fmt) << 0) & 0xFF);
\r
224 auto& pool = sws_contexts_[key];
\r
226 if(!pool.try_pop(sws_context))
\r
229 sws_context.reset(sws_getContext(width, height, pix_fmt, width, height, target_pix_fmt, SWS_BILINEAR, nullptr, nullptr, ¶m), sws_freeContext);
\r
234 BOOST_THROW_EXCEPTION(operation_failed() << msg_info("Could not create software scaling context.") <<
\r
235 boost::errinfo_api_function("sws_getContext"));
\r
238 safe_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);
\r
239 avcodec_get_frame_defaults(av_frame.get());
\r
240 if(target_pix_fmt == PIX_FMT_BGRA)
\r
242 auto size = avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), write->image_data().begin(), PIX_FMT_BGRA, width, height);
\r
243 CASPAR_VERIFY(size == write->image_data().size());
\r
247 av_frame->width = width;
\r
248 av_frame->height = height;
\r
249 for(size_t n = 0; n < target_desc.planes.size(); ++n)
\r
251 av_frame->data[n] = write->image_data(n).begin();
\r
252 av_frame->linesize[n] = target_desc.planes[n].linesize;
\r
256 sws_scale(sws_context.get(), decoded_frame->data, decoded_frame->linesize, 0, height, av_frame->data, av_frame->linesize);
\r
257 pool.push(sws_context);
\r
263 write = frame_factory->create_frame(tag, desc, audio_channel_layout);
\r
264 write->set_type(get_mode(*decoded_frame));
\r
266 for(int n = 0; n < static_cast<int>(desc.planes.size()); ++n)
\r
268 auto plane = desc.planes[n];
\r
269 auto result = write->image_data(n).begin();
\r
270 auto decoded = decoded_frame->data[n];
\r
271 auto decoded_linesize = decoded_frame->linesize[n];
\r
273 CASPAR_ASSERT(decoded);
\r
274 CASPAR_ASSERT(write->image_data(n).begin());
\r
276 if(decoded_linesize != static_cast<int>(plane.width))
\r
278 // Copy line by line since ffmpeg sometimes pads each line.
\r
279 tbb::parallel_for<size_t>(0, desc.planes[n].height, [&](size_t y)
\r
281 fast_memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);
\r
286 fast_memcpy(result, decoded, plane.size);
\r
293 if(decoded_frame->height == 480) // NTSC DV
\r
295 write->get_frame_transform().fill_translation[1] += 2.0/static_cast<double>(frame_factory->get_video_format_desc().height);
\r
296 write->get_frame_transform().fill_scale[1] = 1.0 - 6.0*1.0/static_cast<double>(frame_factory->get_video_format_desc().height);
\r
299 // Fix field-order if needed
\r
300 if(write->get_type() == core::field_mode::lower && frame_factory->get_video_format_desc().field_mode == core::field_mode::upper)
\r
301 write->get_frame_transform().fill_translation[1] += 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);
\r
302 else if(write->get_type() == core::field_mode::upper && frame_factory->get_video_format_desc().field_mode == core::field_mode::lower)
\r
303 write->get_frame_transform().fill_translation[1] -= 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);
\r
305 return make_safe_ptr(write);
\r
308 bool is_sane_fps(AVRational time_base)
\r
310 double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);
\r
311 return fps > 20.0 && fps < 65.0;
\r
314 AVRational fix_time_base(AVRational time_base)
\r
316 if(time_base.num == 1)
\r
317 time_base.num = static_cast<int>(std::pow(10.0, static_cast<int>(std::log10(static_cast<float>(time_base.den)))-1));
\r
319 if(!is_sane_fps(time_base))
\r
321 auto tmp = time_base;
\r
323 if(is_sane_fps(tmp))
\r
330 double read_fps(AVFormatContext& context, double fail_value)
\r
332 auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);
\r
333 auto audio_index = av_find_best_stream(&context, AVMEDIA_TYPE_AUDIO, -1, -1, 0, 0);
\r
335 if(video_index > -1)
\r
337 const auto video_context = context.streams[video_index]->codec;
\r
338 const auto video_stream = context.streams[video_index];
\r
340 AVRational time_base = video_context->time_base;
\r
342 if(boost::filesystem2::path(context.filename).extension() == ".flv")
\r
346 auto meta = read_flv_meta_info(context.filename);
\r
347 return boost::lexical_cast<double>(meta["framerate"]);
\r
356 time_base.num *= video_context->ticks_per_frame;
\r
358 if(!is_sane_fps(time_base))
\r
360 time_base = fix_time_base(time_base);
\r
362 if(!is_sane_fps(time_base) && audio_index > -1)
\r
364 auto& audio_context = *context.streams[audio_index]->codec;
\r
365 auto& audio_stream = *context.streams[audio_index];
\r
367 double duration_sec = audio_stream.duration / static_cast<double>(audio_context.sample_rate);
\r
369 time_base.num = static_cast<int>(duration_sec*100000.0);
\r
370 time_base.den = static_cast<int>(video_stream->nb_frames*100000);
\r
375 double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);
\r
377 double closest_fps = 0.0;
\r
378 for(int n = 0; n < core::video_format::count; ++n)
\r
380 auto format = core::video_format_desc::get(static_cast<core::video_format::type>(n));
\r
382 double diff1 = std::abs(format.fps - fps);
\r
383 double diff2 = std::abs(closest_fps - fps);
\r
386 closest_fps = format.fps;
\r
389 return closest_fps;
\r
392 return fail_value;
\r
395 safe_ptr<AVPacket> create_packet()
\r
397 safe_ptr<AVPacket> packet(new AVPacket, [](AVPacket* p)
\r
403 av_init_packet(packet.get());
\r
407 safe_ptr<AVCodecContext> open_codec(AVFormatContext& context, enum AVMediaType type, int& index)
\r
410 index = THROW_ON_ERROR2(av_find_best_stream(&context, type, -1, -1, &decoder, 0), "");
\r
411 //if(strcmp(decoder->name, "prores") == 0 && decoder->next && strcmp(decoder->next->name, "prores_lgpl") == 0)
\r
412 // decoder = decoder->next;
\r
414 THROW_ON_ERROR2(tbb_avcodec_open(context.streams[index]->codec, decoder), "");
\r
415 return safe_ptr<AVCodecContext>(context.streams[index]->codec, tbb_avcodec_close);
\r
418 std::wstring print_mode(size_t width, size_t height, double fps, bool interlaced)
\r
420 std::wostringstream fps_ss;
\r
421 fps_ss << std::fixed << std::setprecision(2) << (!interlaced ? fps : 2.0 * fps);
\r
423 return boost::lexical_cast<std::wstring>(width) + L"x" + boost::lexical_cast<std::wstring>(height) + (!interlaced ? L"p" : L"i") + fps_ss.str();
\r
426 bool is_valid_file(const std::wstring filename, const std::vector<std::wstring>& invalid_exts)
\r
428 static std::vector<std::wstring> valid_exts = boost::assign::list_of(L".m2t")(L".mov")(L".mp4")(L".dv")(L".flv")(L".mpg")(L".wav")(L".mp3")(L".dnxhd")(L".h264")(L".prores");
\r
430 auto ext = boost::to_lower_copy(boost::filesystem::wpath(filename).extension());
\r
432 if(std::find(invalid_exts.begin(), invalid_exts.end(), ext) != invalid_exts.end())
\r
435 if(std::find(valid_exts.begin(), valid_exts.end(), ext) != valid_exts.end())
\r
438 auto filename2 = narrow(filename);
\r
440 if(boost::filesystem::path(filename2).extension() == ".m2t")
\r
443 std::ifstream file(filename);
\r
445 std::vector<unsigned char> buf;
\r
446 for(auto file_it = std::istreambuf_iterator<char>(file); file_it != std::istreambuf_iterator<char>() && buf.size() < 2048; ++file_it)
\r
447 buf.push_back(*file_it);
\r
453 pb.filename = filename2.c_str();
\r
454 pb.buf = buf.data();
\r
455 pb.buf_size = buf.size();
\r
458 return av_probe_input_format2(&pb, true, &score) != nullptr;
\r
461 bool is_valid_file(const std::wstring filename)
\r
463 static const std::vector<std::wstring> invalid_exts = boost::assign::list_of(L".png")(L".tga")(L".bmp")(L".jpg")(L".jpeg")(L".gif")(L".tiff")(L".tif")(L".jp2")(L".jpx")(L".j2k")(L".j2c")(L".swf")(L".ct");
\r
465 return is_valid_file(filename, invalid_exts);
\r
468 std::wstring probe_stem(const std::wstring stem, const std::vector<std::wstring>& invalid_exts)
\r
470 auto stem2 = boost::filesystem2::wpath(stem);
\r
471 auto dir = stem2.parent_path();
\r
472 for(auto it = boost::filesystem2::wdirectory_iterator(dir); it != boost::filesystem2::wdirectory_iterator(); ++it)
\r
474 if(boost::iequals(it->path().stem(), stem2.filename()) && is_valid_file(it->path().file_string(), invalid_exts))
\r
475 return it->path().file_string();
\r
480 std::wstring probe_stem(const std::wstring stem)
\r
482 auto stem2 = boost::filesystem2::wpath(stem);
\r
483 auto dir = stem2.parent_path();
\r
484 for(auto it = boost::filesystem2::wdirectory_iterator(dir); it != boost::filesystem2::wdirectory_iterator(); ++it)
\r
486 if(boost::iequals(it->path().stem(), stem2.filename()) && is_valid_file(it->path().file_string()))
\r
487 return it->path().file_string();
\r
492 core::channel_layout get_audio_channel_layout(
\r
493 const AVCodecContext& context, const std::wstring& custom_channel_order)
\r
495 if (!custom_channel_order.empty())
\r
497 auto layout = core::create_custom_channel_layout(
\r
498 custom_channel_order,
\r
499 core::default_channel_layout_repository());
\r
501 layout.num_channels = context.channels;
\r
506 int64_t ch_layout = context.channel_layout;
\r
508 if (ch_layout == 0)
\r
509 ch_layout = av_get_default_channel_layout(context.channels);
\r
511 switch (ch_layout) // TODO: refine this auto-detection
\r
513 case AV_CH_LAYOUT_MONO:
\r
514 return core::default_channel_layout_repository().get_by_name(L"MONO");
\r
515 case AV_CH_LAYOUT_STEREO:
\r
516 return core::default_channel_layout_repository().get_by_name(L"STEREO");
\r
517 case AV_CH_LAYOUT_5POINT1:
\r
518 case AV_CH_LAYOUT_5POINT1_BACK:
\r
519 return core::default_channel_layout_repository().get_by_name(L"SMPTE");
\r
520 case AV_CH_LAYOUT_7POINT1:
\r
521 return core::default_channel_layout_repository().get_by_name(L"DOLBYE");
\r
524 return core::create_unspecified_layout(context.channels);
\r
528 //void av_dup_frame(AVFrame* frame)
\r
530 // AVFrame* new_frame = avcodec_alloc_frame();
\r
533 // const uint8_t *src_data[4] = {0};
\r
534 // memcpy(const_cast<uint8_t**>(&src_data[0]), frame->data, 4);
\r
535 // const int src_linesizes[4] = {0};
\r
536 // memcpy(const_cast<int*>(&src_linesizes[0]), frame->linesize, 4);
\r
538 // av_image_alloc(new_frame->data, new_frame->linesize, new_frame->width, new_frame->height, frame->format, 16);
\r
540 // av_image_copy(new_frame->data, new_frame->linesize, src_data, src_linesizes, frame->format, new_frame->width, new_frame->height);
\r