]> git.sesse.net Git - casparcg/blob - modules/ffmpeg/producer/util/util.cpp
2.1.0: Some mayor refactoring.
[casparcg] / modules / ffmpeg / producer / util / util.cpp
1 /*\r
2 * Copyright (c) 2011 Sveriges Television AB <info@casparcg.com>\r
3 *\r
4 * This file is part of CasparCG (www.casparcg.com).\r
5 *\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
10 *\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
15 *\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
18 *\r
19 * Author: Robert Nagy, ronag89@gmail.com\r
20 */\r
21 \r
22 #include "../../stdafx.h"\r
23 \r
24 #include "util.h"\r
25 \r
26 #include "flv.h"\r
27 \r
28 #include "../tbb_avcodec.h"\r
29 #include "../../ffmpeg_error.h"\r
30 \r
31 #include <tbb/concurrent_unordered_map.h>\r
32 #include <tbb/concurrent_queue.h>\r
33 \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 \r
39 #include <common/exception/exceptions.h>\r
40 \r
41 #include <tbb/parallel_for.h>\r
42 \r
43 #include <boost/assert.hpp>\r
44 #include <boost/filesystem.hpp>\r
45 #include <boost/lexical_cast.hpp>\r
46 \r
47 #include <asmlib.h>\r
48 \r
49 #if defined(_MSC_VER)\r
50 #pragma warning (push)\r
51 #pragma warning (disable : 4244)\r
52 #endif\r
53 extern "C" \r
54 {\r
55         #include <libswscale/swscale.h>\r
56         #include <libavcodec/avcodec.h>\r
57         #include <libavformat/avformat.h>\r
58 }\r
59 #if defined(_MSC_VER)\r
60 #pragma warning (pop)\r
61 #endif\r
62 \r
63 namespace caspar { namespace ffmpeg {\r
64                 \r
65 std::shared_ptr<core::audio_buffer> flush_audio()\r
66 {\r
67         static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());\r
68         return audio;\r
69 }\r
70 \r
71 std::shared_ptr<core::audio_buffer> empty_audio()\r
72 {\r
73         static std::shared_ptr<core::audio_buffer> audio(new core::audio_buffer());\r
74         return audio;\r
75 }\r
76 \r
77 std::shared_ptr<AVFrame>                        flush_video()\r
78 {\r
79         static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);\r
80         return video;\r
81 }\r
82 \r
83 std::shared_ptr<AVFrame>                        empty_video()\r
84 {\r
85         static std::shared_ptr<AVFrame> video(avcodec_alloc_frame(), av_free);\r
86         return video;\r
87 }\r
88 \r
89 core::field_mode get_mode(const AVFrame& frame)\r
90 {\r
91         if(!frame.interlaced_frame)\r
92                 return core::field_mode::progressive;\r
93 \r
94         return frame.top_field_first ? core::field_mode::upper : core::field_mode::lower;\r
95 }\r
96 \r
97 core::pixel_format get_pixel_format(PixelFormat pix_fmt)\r
98 {\r
99         switch(pix_fmt)\r
100         {\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
114         }\r
115 }\r
116 \r
117 core::pixel_format_desc get_pixel_format_desc(PixelFormat pix_fmt, int width, int height)\r
118 {\r
119         // Get linesizes\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
122 \r
123         core::pixel_format_desc desc;\r
124         desc.pix_fmt = get_pixel_format(pix_fmt);\r
125                 \r
126         switch(desc.pix_fmt.value())\r
127         {\r
128         case core::pixel_format::gray:\r
129         case core::pixel_format::luma:\r
130                 {\r
131                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));                                               \r
132                         return desc;\r
133                 }\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
138                 {\r
139                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0]/4, height, 4));                                             \r
140                         return desc;\r
141                 }\r
142         case core::pixel_format::ycbcr:\r
143         case core::pixel_format::ycbcra:\r
144                 {               \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
148 \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
152 \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
155                         return desc;\r
156                 }               \r
157         default:                \r
158                 desc.pix_fmt = core::pixel_format::invalid;\r
159                 return desc;\r
160         }\r
161 }\r
162 \r
163 int make_alpha_format(int format)\r
164 {\r
165         switch(get_pixel_format(static_cast<PixelFormat>(format)).value())\r
166         {\r
167         case core::pixel_format::ycbcr:\r
168         case core::pixel_format::ycbcra:\r
169                 return CASPAR_PIX_FMT_LUMA;\r
170         default:\r
171                 return format;\r
172         }\r
173 }\r
174 \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
176 {                       \r
177         static tbb::concurrent_unordered_map<int, tbb::concurrent_queue<std::shared_ptr<SwsContext>>> sws_contexts_;\r
178         \r
179         if(decoded_frame->width < 1 || decoded_frame->height < 1)\r
180                 return make_safe<core::write_frame>(tag);\r
181 \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
185         \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
188 \r
189         std::shared_ptr<core::write_frame> write;\r
190 \r
191         if(desc.pix_fmt == core::pixel_format::invalid)\r
192         {\r
193                 auto pix_fmt = static_cast<PixelFormat>(decoded_frame->format);\r
194                 auto target_pix_fmt = PIX_FMT_BGRA;\r
195 \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
208                 \r
209                 auto target_desc = get_pixel_format_desc(target_pix_fmt, width, height);\r
210 \r
211                 write = frame_factory->create_frame(tag, target_desc);\r
212                 write->set_type(get_mode(*decoded_frame));\r
213 \r
214                 std::shared_ptr<SwsContext> sws_context;\r
215 \r
216                 //CASPAR_LOG(warning) << "Hardware accelerated color transform not supported.";\r
217 \r
218                 int key = ((width << 22) & 0xFFC00000) | ((height << 6) & 0x003FC000) | ((pix_fmt << 7) & 0x00007F00) | ((target_pix_fmt << 0) & 0x0000007F);\r
219                         \r
220                 auto& pool = sws_contexts_[key];\r
221                                                 \r
222                 if(!pool.try_pop(sws_context))\r
223                 {\r
224                         double param;\r
225                         sws_context.reset(sws_getContext(width, height, pix_fmt, width, height, target_pix_fmt, SWS_BILINEAR, nullptr, nullptr, &param), sws_freeContext);\r
226                 }\r
227                         \r
228                 if(!sws_context)\r
229                 {\r
230                         BOOST_THROW_EXCEPTION(operation_failed() << msg_info("Could not create software scaling context.") << \r
231                                                                         boost::errinfo_api_function("sws_getContext"));\r
232                 }       \r
233                 \r
234                 safe_ptr<AVFrame> av_frame(avcodec_alloc_frame(), av_free);     \r
235                 avcodec_get_frame_defaults(av_frame.get());                     \r
236                 if(target_pix_fmt == PIX_FMT_BGRA)\r
237                 {\r
238                         auto size = avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), write->image_data().begin(), PIX_FMT_BGRA, width, height);\r
239                         BOOST_VERIFY(size == write->image_data().size()); \r
240                 }\r
241                 else\r
242                 {\r
243                         av_frame->width  = width;\r
244                         av_frame->height = height;\r
245                         for(int n = 0; n < target_desc.planes.size(); ++n)\r
246                         {\r
247                                 av_frame->data[n]               = write->image_data(n).begin();\r
248                                 av_frame->linesize[n]   = target_desc.planes[n].linesize;\r
249                         }\r
250                 }\r
251 \r
252                 sws_scale(sws_context.get(), decoded_frame->data, decoded_frame->linesize, 0, height, av_frame->data, av_frame->linesize);      \r
253                 pool.push(sws_context);\r
254 \r
255                 write->commit();                \r
256         }\r
257         else\r
258         {\r
259                 write = frame_factory->create_frame(tag, desc);\r
260                 write->set_type(get_mode(*decoded_frame));\r
261 \r
262                 for(int n = 0; n < static_cast<int>(desc.planes.size()); ++n)\r
263                 {\r
264                         auto plane            = desc.planes[n];\r
265                         auto result           = write->image_data(n).begin();\r
266                         auto decoded          = decoded_frame->data[n];\r
267                         auto decoded_linesize = decoded_frame->linesize[n];\r
268                         \r
269                         BOOST_ASSERT(decoded);\r
270                         BOOST_ASSERT(write->image_data(n).begin());\r
271 \r
272                         // Copy line by line since ffmpeg sometimes pads each line.\r
273                         tbb::affinity_partitioner ap;\r
274                         tbb::parallel_for(tbb::blocked_range<int>(0, desc.planes[n].height), [&](const tbb::blocked_range<int>& r)\r
275                         {\r
276                                 for(int y = r.begin(); y != r.end(); ++y)\r
277                                         A_memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);\r
278                         }, ap);\r
279 \r
280                         write->commit(n);\r
281                 }\r
282         }\r
283 \r
284         if(decoded_frame->height == 480) // NTSC DV\r
285         {\r
286                 write->get_frame_transform().fill_translation[1] += 2.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
287                 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         }\r
289         \r
290         // Fix field-order if needed\r
291         if(write->get_type() == core::field_mode::lower && frame_factory->get_video_format_desc().field_mode == core::field_mode::upper)\r
292                 write->get_frame_transform().fill_translation[1] += 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
293         else if(write->get_type() == core::field_mode::upper && frame_factory->get_video_format_desc().field_mode == core::field_mode::lower)\r
294                 write->get_frame_transform().fill_translation[1] -= 1.0/static_cast<double>(frame_factory->get_video_format_desc().height);\r
295 \r
296         return make_safe_ptr(write);\r
297 }\r
298 \r
299 bool is_sane_fps(AVRational time_base)\r
300 {\r
301         double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);\r
302         return fps > 20.0 && fps < 65.0;\r
303 }\r
304 \r
305 AVRational fix_time_base(AVRational time_base)\r
306 {\r
307         if(time_base.num == 1)\r
308                 time_base.num = static_cast<int>(std::pow(10.0, static_cast<int>(std::log10(static_cast<float>(time_base.den)))-1));    \r
309                         \r
310         if(!is_sane_fps(time_base))\r
311         {\r
312                 auto tmp = time_base;\r
313                 tmp.den /= 2;\r
314                 if(is_sane_fps(tmp))\r
315                         time_base = tmp;\r
316         }\r
317 \r
318         return time_base;\r
319 }\r
320 \r
321 double read_fps(AVFormatContext& context, double fail_value)\r
322 {                                               \r
323         auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);\r
324         auto audio_index = av_find_best_stream(&context, AVMEDIA_TYPE_AUDIO, -1, -1, 0, 0);\r
325         \r
326         if(video_index > -1)\r
327         {\r
328                 const auto video_context = context.streams[video_index]->codec;\r
329                 const auto video_stream  = context.streams[video_index];\r
330                                                 \r
331                 AVRational time_base = video_context->time_base;\r
332 \r
333                 if(boost::filesystem::path(context.filename).extension().string() == ".flv")\r
334                 {\r
335                         try\r
336                         {\r
337                                 auto meta = read_flv_meta_info(context.filename);\r
338                                 return boost::lexical_cast<double>(meta["framerate"]);\r
339                         }\r
340                         catch(...)\r
341                         {\r
342                                 return 0.0;\r
343                         }\r
344                 }\r
345                 else\r
346                 {\r
347                         time_base.num *= video_context->ticks_per_frame;\r
348 \r
349                         if(!is_sane_fps(time_base))\r
350                         {                       \r
351                                 time_base = fix_time_base(time_base);\r
352 \r
353                                 if(!is_sane_fps(time_base) && audio_index > -1)\r
354                                 {\r
355                                         auto& audio_context = *context.streams[audio_index]->codec;\r
356                                         auto& audio_stream  = *context.streams[audio_index];\r
357 \r
358                                         double duration_sec = audio_stream.duration / static_cast<double>(audio_context.sample_rate);\r
359                                                                 \r
360                                         time_base.num = static_cast<int>(duration_sec*100000.0);\r
361                                         time_base.den = static_cast<int>(video_stream->nb_frames*100000);\r
362                                 }\r
363                         }\r
364                 }\r
365                 \r
366                 double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);\r
367 \r
368                 double closest_fps = 0.0;\r
369                 for(int n = 0; n < core::video_format::count; ++n)\r
370                 {\r
371                         auto format = core::video_format_desc(core::video_format(n));\r
372 \r
373                         double diff1 = std::abs(format.fps - fps);\r
374                         double diff2 = std::abs(closest_fps - fps);\r
375 \r
376                         if(diff1 < diff2)\r
377                                 closest_fps = format.fps;\r
378                 }\r
379         \r
380                 return closest_fps;\r
381         }\r
382 \r
383         return fail_value;      \r
384 }\r
385 \r
386 void fix_meta_data(AVFormatContext& context)\r
387 {\r
388         auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);\r
389 \r
390         if(video_index > -1)\r
391         {\r
392                 auto video_stream   = context.streams[video_index];\r
393                 auto video_context  = context.streams[video_index]->codec;\r
394                                                 \r
395                 if(boost::filesystem::path(context.filename).extension().string() == ".flv")\r
396                 {\r
397                         try\r
398                         {\r
399                                 auto meta = read_flv_meta_info(context.filename);\r
400                                 double fps = boost::lexical_cast<double>(meta["framerate"]);\r
401                                 video_stream->nb_frames = static_cast<int64_t>(boost::lexical_cast<double>(meta["duration"])*fps);\r
402                         }\r
403                         catch(...){}\r
404                 }\r
405                 else\r
406                 {\r
407                         auto stream_time = video_stream->time_base;\r
408                         auto duration    = video_stream->duration;\r
409                         auto codec_time  = video_context->time_base;\r
410                         auto ticks               = video_context->ticks_per_frame;\r
411 \r
412                         if(video_stream->nb_frames == 0)\r
413                                 video_stream->nb_frames = (duration*stream_time.num*codec_time.den)/(stream_time.den*codec_time.num*ticks);     \r
414                 }\r
415         }\r
416 }\r
417 \r
418 safe_ptr<AVPacket> create_packet()\r
419 {\r
420         safe_ptr<AVPacket> packet(new AVPacket, [](AVPacket* p)\r
421         {\r
422                 av_free_packet(p);\r
423                 delete p;\r
424         });\r
425         \r
426         av_init_packet(packet.get());\r
427         return packet;\r
428 }\r
429 \r
430 safe_ptr<AVCodecContext> open_codec(AVFormatContext& context, enum AVMediaType type, int& index)\r
431 {       \r
432         AVCodec* decoder;\r
433         index = THROW_ON_ERROR2(av_find_best_stream(&context, type, -1, -1, &decoder, 0), "");\r
434         //if(strcmp(decoder->name, "prores") == 0 && decoder->next && strcmp(decoder->next->name, "prores_lgpl") == 0)\r
435         //      decoder = decoder->next;\r
436 \r
437         THROW_ON_ERROR2(tbb_avcodec_open(context.streams[index]->codec, decoder), "");\r
438         return safe_ptr<AVCodecContext>(context.streams[index]->codec, tbb_avcodec_close);\r
439 }\r
440 \r
441 safe_ptr<AVFormatContext> open_input(const std::wstring& filename)\r
442 {\r
443         AVFormatContext* weak_context = nullptr;\r
444         THROW_ON_ERROR2(avformat_open_input(&weak_context, u8(filename).c_str(), nullptr, nullptr), filename);\r
445         safe_ptr<AVFormatContext> context(weak_context, av_close_input_file);                   \r
446         THROW_ON_ERROR2(avformat_find_stream_info(weak_context, nullptr), filename);\r
447         fix_meta_data(*context);\r
448         return context;\r
449 }\r
450 \r
451 std::wstring print_mode(int width, int height, double fps, bool interlaced)\r
452 {\r
453         std::wostringstream fps_ss;\r
454         fps_ss << std::fixed << std::setprecision(2) << (!interlaced ? fps : 2.0 * fps);\r
455 \r
456         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 }\r
458 \r
459 bool is_valid_file(const std::wstring filename)\r
460 {                       \r
461         std::ifstream file(filename);\r
462 \r
463         std::vector<unsigned char> buf;\r
464         for(auto file_it = std::istreambuf_iterator<char>(file); file_it != std::istreambuf_iterator<char>() && buf.size() < 2048; ++file_it)\r
465                 buf.push_back(*file_it);\r
466 \r
467         if(buf.empty())\r
468                 return nullptr;\r
469 \r
470         AVProbeData pb;\r
471         pb.filename = u8(filename).c_str();\r
472         pb.buf          = buf.data();\r
473         pb.buf_size = static_cast<int>(buf.size());\r
474 \r
475         int score = 0;\r
476         return av_probe_input_format2(&pb, true, &score) != nullptr;\r
477 }\r
478 \r
479 std::wstring probe_stem(const std::wstring stem)\r
480 {\r
481         auto stem2 = boost::filesystem::path(stem);\r
482         auto dir = stem2.parent_path();\r
483         for(auto it = boost::filesystem::directory_iterator(dir); it != boost::filesystem::directory_iterator(); ++it)\r
484         {\r
485                 if(boost::iequals(it->path().stem().wstring(), stem2.filename().wstring()) && is_valid_file(it->path().wstring()))\r
486                         return it->path().wstring();\r
487         }\r
488         return L"";\r
489 }\r
490 //\r
491 //void av_dup_frame(AVFrame* frame)\r
492 //{\r
493 //      AVFrame* new_frame = avcodec_alloc_frame();\r
494 //\r
495 //\r
496 //      const uint8_t *src_data[4] = {0};\r
497 //      memcpy(const_cast<uint8_t**>(&src_data[0]), frame->data, 4);\r
498 //      const int src_linesizes[4] = {0};\r
499 //      memcpy(const_cast<int*>(&src_linesizes[0]), frame->linesize, 4);\r
500 //\r
501 //      av_image_alloc(new_frame->data, new_frame->linesize, new_frame->width, new_frame->height, frame->format, 16);\r
502 //\r
503 //      av_image_copy(new_frame->data, new_frame->linesize, src_data, src_linesizes, frame->format, new_frame->width, new_frame->height);\r
504 //\r
505 //      frame =\r
506 //}\r
507 \r
508 }}