git.sesse.net Git - casparcg/blob - modules/ffmpeg/producer/util/util.cpp

   1 /*
   2 * Copyright (c) 2011 Sveriges Television AB <info@casparcg.com>
   3 *
   4 * This file is part of CasparCG (www.casparcg.com).
   5 *
   6 * CasparCG is free software: you can redistribute it and/or modify
   7 * it under the terms of the GNU General Public License as published by
   8 * the Free Software Foundation, either version 3 of the License, or
   9 * (at your option) any later version.
  10 *
  11 * CasparCG is distributed in the hope that it will be useful,
  12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14 * GNU General Public License for more details.
  15 *
  16 * You should have received a copy of the GNU General Public License
  17 * along with CasparCG. If not, see <http://www.gnu.org/licenses/>.
  18 *
  19 * Author: Robert Nagy, ronag89@gmail.com
  20 */
  21
  22 #include "../../StdAfx.h"
  23
  24 #include "util.h"
  25
  26 #include "flv.h"
  27
  28 #include "../tbb_avcodec.h"
  29 #include "../../ffmpeg_error.h"
  30 #include "../../ffmpeg.h"
  31
  32 #include <tbb/concurrent_unordered_map.h>
  33 #include <tbb/concurrent_queue.h>
  34
  35 #include <core/frame/frame_transform.h>
  36 #include <core/frame/frame_factory.h>
  37 #include <core/frame/frame.h>
  38 #include <core/frame/audio_channel_layout.h>
  39 #include <core/producer/frame_producer.h>
  40
  41 #include <common/except.h>
  42 #include <common/array.h>
  43 #include <common/os/filesystem.h>
  44 #include <common/memcpy.h>
  45
  46 #include <tbb/parallel_for.h>
  47
  48 #include <common/assert.h>
  49 #include <boost/filesystem.hpp>
  50 #include <boost/lexical_cast.hpp>
  51 #include <boost/rational.hpp>
  52
  53 #include <fstream>
  54
  55 #include <asmlib.h>
  56
  57 #if defined(_MSC_VER)
  58 #pragma warning (push)
  59 #pragma warning (disable : 4244)
  60 #endif
  61 extern "C"
  62 {
  63         #include <libswscale/swscale.h>
  64         #include <libavcodec/avcodec.h>
  65         #include <libavformat/avformat.h>
  66 }
  67 #if defined(_MSC_VER)
  68 #pragma warning (pop)
  69 #endif
  70
  71 namespace caspar { namespace ffmpeg {
  72
  73 core::field_mode get_mode(const AVFrame& frame)
  74 {
  75         if(!frame.interlaced_frame)
  76                 return core::field_mode::progressive;
  77
  78         return frame.top_field_first ? core::field_mode::upper : core::field_mode::lower;
  79 }
  80
  81 core::pixel_format get_pixel_format(PixelFormat pix_fmt)
  82 {
  83         switch(pix_fmt)
  84         {
  85         case PIX_FMT_GRAY8:                     return core::pixel_format::gray;
  86         case PIX_FMT_RGB24:                     return core::pixel_format::rgb;
  87         case PIX_FMT_BGR24:                     return core::pixel_format::bgr;
  88         case PIX_FMT_BGRA:                      return core::pixel_format::bgra;
  89         case PIX_FMT_ARGB:                      return core::pixel_format::argb;
  90         case PIX_FMT_RGBA:                      return core::pixel_format::rgba;
  91         case PIX_FMT_ABGR:                      return core::pixel_format::abgr;
  92         case PIX_FMT_YUV444P:           return core::pixel_format::ycbcr;
  93         case PIX_FMT_YUV422P:           return core::pixel_format::ycbcr;
  94         case PIX_FMT_YUV420P:           return core::pixel_format::ycbcr;
  95         case PIX_FMT_YUV411P:           return core::pixel_format::ycbcr;
  96         case PIX_FMT_YUV410P:           return core::pixel_format::ycbcr;
  97         case PIX_FMT_YUVA420P:          return core::pixel_format::ycbcra;
  98         default:                                        return core::pixel_format::invalid;
  99         }
 100 }
 101
 102 core::pixel_format_desc pixel_format_desc(PixelFormat pix_fmt, int width, int height)
 103 {
 104         // Get linesizes
 105         AVPicture dummy_pict;
 106         avpicture_fill(&dummy_pict, nullptr, pix_fmt, width, height);
 107
 108         core::pixel_format_desc desc = get_pixel_format(pix_fmt);
 109
 110         switch(desc.format)
 111         {
 112         case core::pixel_format::gray:
 113         case core::pixel_format::luma:
 114                 {
 115                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));
 116                         return desc;
 117                 }
 118         case core::pixel_format::bgr:
 119         case core::pixel_format::rgb:
 120                 {
 121                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0]/3, height, 3));
 122                         return desc;
 123                 }
 124         case core::pixel_format::bgra:
 125         case core::pixel_format::argb:
 126         case core::pixel_format::rgba:
 127         case core::pixel_format::abgr:
 128                 {
 129                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0]/4, height, 4));
 130                         return desc;
 131                 }
 132         case core::pixel_format::ycbcr:
 133         case core::pixel_format::ycbcra:
 134                 {
 135                         // Find chroma height
 136                         int size2 = static_cast<int>(dummy_pict.data[2] - dummy_pict.data[1]);
 137                         int h2 = size2/dummy_pict.linesize[1];
 138
 139                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[0], height, 1));
 140                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[1], h2, 1));
 141                         desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[2], h2, 1));
 142
 143                         if(desc.format == core::pixel_format::ycbcra)
 144                                 desc.planes.push_back(core::pixel_format_desc::plane(dummy_pict.linesize[3], height, 1));
 145                         return desc;
 146                 }
 147         default:
 148                 desc.format = core::pixel_format::invalid;
 149                 return desc;
 150         }
 151 }
 152
 153 core::mutable_frame make_frame(const void* tag, const spl::shared_ptr<AVFrame>& decoded_frame, core::frame_factory& frame_factory, const core::audio_channel_layout& channel_layout)
 154 {
 155         static tbb::concurrent_unordered_map<int64_t, tbb::concurrent_queue<std::shared_ptr<SwsContext>>> sws_contvalid_exts_;
 156
 157         if(decoded_frame->width < 1 || decoded_frame->height < 1)
 158                 return frame_factory.create_frame(tag, core::pixel_format_desc(core::pixel_format::invalid), core::audio_channel_layout::invalid());
 159
 160         const auto width  = decoded_frame->width;
 161         const auto height = decoded_frame->height;
 162         auto desc                 = pixel_format_desc(static_cast<PixelFormat>(decoded_frame->format), width, height);
 163
 164         if(desc.format == core::pixel_format::invalid)
 165         {
 166                 auto pix_fmt = static_cast<PixelFormat>(decoded_frame->format);
 167                 auto target_pix_fmt = PIX_FMT_BGRA;
 168
 169                 if(pix_fmt == PIX_FMT_UYVY422)
 170                         target_pix_fmt = PIX_FMT_YUV422P;
 171                 else if(pix_fmt == PIX_FMT_YUYV422)
 172                         target_pix_fmt = PIX_FMT_YUV422P;
 173                 else if(pix_fmt == PIX_FMT_UYYVYY411)
 174                         target_pix_fmt = PIX_FMT_YUV411P;
 175                 else if(pix_fmt == PIX_FMT_YUV420P10)
 176                         target_pix_fmt = PIX_FMT_YUV420P;
 177                 else if(pix_fmt == PIX_FMT_YUV422P10)
 178                         target_pix_fmt = PIX_FMT_YUV422P;
 179                 else if(pix_fmt == PIX_FMT_YUV444P10)
 180                         target_pix_fmt = PIX_FMT_YUV444P;
 181
 182                 auto target_desc = pixel_format_desc(target_pix_fmt, width, height);
 183
 184                 auto write = frame_factory.create_frame(tag, target_desc, channel_layout);
 185
 186                 std::shared_ptr<SwsContext> sws_context;
 187
 188                 //CASPAR_LOG(warning) << "Hardware accelerated color transform not supported.";
 189
 190                 int64_t key = ((static_cast<int64_t>(width)                      << 32) & 0xFFFF00000000) |
 191                                           ((static_cast<int64_t>(height)                 << 16) & 0xFFFF0000) |
 192                                           ((static_cast<int64_t>(pix_fmt)                <<  8) & 0xFF00) |
 193                                           ((static_cast<int64_t>(target_pix_fmt) <<  0) & 0xFF);
 194
 195                 auto& pool = sws_contvalid_exts_[key];
 196
 197                 if(!pool.try_pop(sws_context))
 198                 {
 199                         double param;
 200                         sws_context.reset(sws_getContext(width, height, pix_fmt, width, height, target_pix_fmt, SWS_BILINEAR, nullptr, nullptr, &param), sws_freeContext);
 201                 }
 202
 203                 if(!sws_context)
 204                 {
 205                         CASPAR_THROW_EXCEPTION(operation_failed() << msg_info("Could not create software scaling context.") <<
 206                                                                         boost::errinfo_api_function("sws_getContext"));
 207                 }
 208
 209                 auto av_frame = create_frame();
 210                 if(target_pix_fmt == PIX_FMT_BGRA)
 211                 {
 212                         auto size = avpicture_fill(reinterpret_cast<AVPicture*>(av_frame.get()), write.image_data(0).begin(), PIX_FMT_BGRA, width, height);
 213                         CASPAR_VERIFY(size == write.image_data(0).size());
 214                 }
 215                 else
 216                 {
 217                         av_frame->width  = width;
 218                         av_frame->height = height;
 219                         for(int n = 0; n < target_desc.planes.size(); ++n)
 220                         {
 221                                 av_frame->data[n]               = write.image_data(n).begin();
 222                                 av_frame->linesize[n]   = target_desc.planes[n].linesize;
 223                         }
 224                 }
 225
 226                 sws_scale(sws_context.get(), decoded_frame->data, decoded_frame->linesize, 0, height, av_frame->data, av_frame->linesize);
 227                 pool.push(sws_context);
 228
 229                 return std::move(write);
 230         }
 231         else
 232         {
 233                 auto write = frame_factory.create_frame(tag, desc, channel_layout);
 234
 235                 for(int n = 0; n < static_cast<int>(desc.planes.size()); ++n)
 236                 {
 237                         auto plane            = desc.planes[n];
 238                         auto result           = write.image_data(n).begin();
 239                         auto decoded          = decoded_frame->data[n];
 240                         auto decoded_linesize = decoded_frame->linesize[n];
 241
 242                         CASPAR_ASSERT(decoded);
 243                         CASPAR_ASSERT(write.image_data(n).begin());
 244
 245                         if (decoded_linesize != plane.linesize)
 246                         {
 247                                 // Copy line by line since ffmpeg sometimes pads each line.
 248                                 tbb::affinity_partitioner ap;
 249                                 tbb::parallel_for(tbb::blocked_range<int>(0, desc.planes[n].height), [&](const tbb::blocked_range<int>& r)
 250                                 {
 251                                         for (int y = r.begin(); y != r.end(); ++y)
 252                                                 A_memcpy(result + y*plane.linesize, decoded + y*decoded_linesize, plane.linesize);
 253                                 }, ap);
 254                         }
 255                         else
 256                         {
 257                                 fast_memcpy(result, decoded, plane.size);
 258                         }
 259                 }
 260
 261                 return std::move(write);
 262         }
 263 }
 264
 265 spl::shared_ptr<AVFrame> make_av_frame(core::mutable_frame& frame)
 266 {
 267         std::array<uint8_t*, 4> data = {};
 268         for(int n = 0; n < frame.pixel_format_desc().planes.size(); ++n)
 269                 data[n] = frame.image_data(n).begin();
 270
 271         return make_av_frame(data, frame.pixel_format_desc());
 272 }
 273
 274 spl::shared_ptr<AVFrame> make_av_frame(std::array<uint8_t*, 4> data, const core::pixel_format_desc& pix_desc)
 275 {
 276         auto av_frame = create_frame();
 277
 278         auto planes              = pix_desc.planes;
 279         auto format              = pix_desc.format;
 280
 281         av_frame->width  = planes[0].width;
 282         av_frame->height = planes[0].height;
 283         for(int n = 0; n < planes.size(); ++n)
 284         {
 285                 av_frame->data[n]         = data[n];
 286                 av_frame->linesize[n] = planes[n].linesize;
 287         }
 288
 289         switch(format)
 290         {
 291         case core::pixel_format::rgb:
 292                 av_frame->format = PIX_FMT_RGB24;
 293                 break;
 294         case core::pixel_format::bgr:
 295                 av_frame->format = PIX_FMT_BGR24;
 296                 break;
 297         case core::pixel_format::rgba:
 298                 av_frame->format = PIX_FMT_RGBA;
 299                 break;
 300         case core::pixel_format::argb:
 301                 av_frame->format = PIX_FMT_ARGB;
 302                 break;
 303         case core::pixel_format::bgra:
 304                 av_frame->format = PIX_FMT_BGRA;
 305                 break;
 306         case core::pixel_format::abgr:
 307                 av_frame->format = PIX_FMT_ABGR;
 308                 break;
 309         case core::pixel_format::gray:
 310                 av_frame->format = PIX_FMT_GRAY8;
 311                 break;
 312         case core::pixel_format::ycbcr:
 313         {
 314                 int y_w = planes[0].width;
 315                 int y_h = planes[0].height;
 316                 int c_w = planes[1].width;
 317                 int c_h = planes[1].height;
 318
 319                 if(c_h == y_h && c_w == y_w)
 320                         av_frame->format = PIX_FMT_YUV444P;
 321                 else if(c_h == y_h && c_w*2 == y_w)
 322                         av_frame->format = PIX_FMT_YUV422P;
 323                 else if(c_h == y_h && c_w*4 == y_w)
 324                         av_frame->format = PIX_FMT_YUV411P;
 325                 else if(c_h*2 == y_h && c_w*2 == y_w)
 326                         av_frame->format = PIX_FMT_YUV420P;
 327                 else if(c_h*2 == y_h && c_w*4 == y_w)
 328                         av_frame->format = PIX_FMT_YUV410P;
 329
 330                 break;
 331         }
 332         case core::pixel_format::ycbcra:
 333                 av_frame->format = PIX_FMT_YUVA420P;
 334                 break;
 335         }
 336         return av_frame;
 337 }
 338
 339 bool is_sane_fps(AVRational time_base)
 340 {
 341         double fps = static_cast<double>(time_base.den) / static_cast<double>(time_base.num);
 342         return fps > 20.0 && fps < 65.0;
 343 }
 344
 345 AVRational fix_time_base(AVRational time_base)
 346 {
 347         if(time_base.num == 1)
 348                 time_base.num = static_cast<int>(std::pow(10.0, static_cast<int>(std::log10(static_cast<float>(time_base.den)))-1));
 349
 350         if(!is_sane_fps(time_base))
 351         {
 352                 auto tmp = time_base;
 353                 tmp.den /= 2;
 354                 if(is_sane_fps(tmp))
 355                         time_base = tmp;
 356         }
 357
 358         return time_base;
 359 }
 360
 361 double read_fps(AVFormatContext& context, double fail_value)
 362 {
 363         auto framerate = read_framerate(context, boost::rational<int>(static_cast<int>(fail_value * 1000000.0), 1000000));
 364
 365         return static_cast<double>(framerate.numerator()) / static_cast<double>(framerate.denominator());
 366 }
 367
 368 boost::rational<int> read_framerate(AVFormatContext& context, const boost::rational<int>& fail_value)
 369 {
 370         auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);
 371         auto audio_index = av_find_best_stream(&context, AVMEDIA_TYPE_AUDIO, -1, -1, 0, 0);
 372
 373         if (video_index > -1)
 374         {
 375                 const auto video_context = context.streams[video_index]->codec;
 376                 const auto video_stream = context.streams[video_index];
 377
 378                 auto frame_rate_time_base = video_stream->avg_frame_rate;
 379                 std::swap(frame_rate_time_base.num, frame_rate_time_base.den);
 380
 381                 if (is_sane_fps(frame_rate_time_base))
 382                 {
 383                         return boost::rational<int>(frame_rate_time_base.den, frame_rate_time_base.num);
 384                 }
 385
 386                 AVRational time_base = video_context->time_base;
 387
 388                 if (boost::filesystem::path(context.filename).extension().string() == ".flv")
 389                 {
 390                         try
 391                         {
 392                                 auto meta = read_flv_meta_info(context.filename);
 393                                 return boost::rational<int>(static_cast<int>(boost::lexical_cast<double>(meta["framerate"]) * 1000000.0), 1000000);
 394                         }
 395                         catch (...)
 396                         {
 397                                 return fail_value;
 398                         }
 399                 }
 400                 else
 401                 {
 402                         time_base.num *= video_context->ticks_per_frame;
 403
 404                         if (!is_sane_fps(time_base))
 405                         {
 406                                 time_base = fix_time_base(time_base);
 407
 408                                 if (!is_sane_fps(time_base) && audio_index > -1)
 409                                 {
 410                                         auto& audio_context = *context.streams[audio_index]->codec;
 411                                         auto& audio_stream = *context.streams[audio_index];
 412
 413                                         double duration_sec = audio_stream.duration / static_cast<double>(audio_context.sample_rate);
 414
 415                                         time_base.num = static_cast<int>(duration_sec*100000.0);
 416                                         time_base.den = static_cast<int>(video_stream->nb_frames * 100000);
 417                                 }
 418                         }
 419                 }
 420
 421                 boost::rational<int> fps(time_base.den, time_base.num);
 422                 boost::rational<int> closest_fps(0);
 423
 424                 for (auto video_mode : enum_constants<core::video_format>())
 425                 {
 426                         auto format = core::video_format_desc(core::video_format(video_mode));
 427
 428                         auto diff1 = boost::abs(boost::rational<int>(format.time_scale, format.duration) - fps);
 429                         auto diff2 = boost::abs(closest_fps - fps);
 430
 431                         if (diff1 < diff2)
 432                                 closest_fps = boost::rational<int>(format.time_scale, format.duration);
 433                 }
 434
 435                 return closest_fps;
 436         }
 437
 438         return fail_value;
 439 }
 440
 441 void fix_meta_data(AVFormatContext& context)
 442 {
 443         auto video_index = av_find_best_stream(&context, AVMEDIA_TYPE_VIDEO, -1, -1, 0, 0);
 444
 445         if(video_index > -1)
 446         {
 447                 auto video_stream   = context.streams[video_index];
 448                 auto video_context  = context.streams[video_index]->codec;
 449
 450                 if(boost::filesystem::path(context.filename).extension().string() == ".flv")
 451                 {
 452                         try
 453                         {
 454                                 auto meta = read_flv_meta_info(context.filename);
 455                                 double fps = boost::lexical_cast<double>(meta["framerate"]);
 456                                 video_stream->nb_frames = static_cast<int64_t>(boost::lexical_cast<double>(meta["duration"])*fps);
 457                         }
 458                         catch(...){}
 459                 }
 460                 else
 461                 {
 462                         auto stream_time = video_stream->time_base;
 463                         auto duration    = video_stream->duration;
 464                         auto codec_time  = video_context->time_base;
 465                         auto ticks               = video_context->ticks_per_frame;
 466
 467                         if(video_stream->nb_frames == 0)
 468                                 video_stream->nb_frames = (duration*stream_time.num*codec_time.den)/(stream_time.den*codec_time.num*ticks);
 469                 }
 470         }
 471 }
 472
 473 spl::shared_ptr<AVPacket> create_packet()
 474 {
 475         spl::shared_ptr<AVPacket> packet(new AVPacket(), [](AVPacket* p)
 476         {
 477                 av_free_packet(p);
 478                 delete p;
 479         });
 480
 481         av_init_packet(packet.get());
 482         return packet;
 483 }
 484
 485 spl::shared_ptr<AVFrame> create_frame()
 486 {
 487         spl::shared_ptr<AVFrame> frame(av_frame_alloc(), [](AVFrame* p)
 488         {
 489                 av_frame_free(&p);
 490         });
 491         return frame;
 492 }
 493
 494 std::shared_ptr<core::mutable_audio_buffer> flush_audio()
 495 {
 496         static std::shared_ptr<core::mutable_audio_buffer> audio(new core::mutable_audio_buffer());
 497         return audio;
 498 }
 499
 500 std::shared_ptr<core::mutable_audio_buffer> empty_audio()
 501 {
 502         static std::shared_ptr<core::mutable_audio_buffer> audio(new core::mutable_audio_buffer());
 503         return audio;
 504 }
 505
 506 std::shared_ptr<AVFrame> flush_video()
 507 {
 508         static auto video = create_frame();
 509         return video;
 510 }
 511
 512 std::shared_ptr<AVFrame> empty_video()
 513 {
 514         static auto video = create_frame();
 515         return video;
 516 }
 517
 518 spl::shared_ptr<AVCodecContext> open_codec(AVFormatContext& context, enum AVMediaType type, int& index, bool single_threaded)
 519 {
 520         AVCodec* decoder;
 521         index = THROW_ON_ERROR2(av_find_best_stream(&context, type, index, -1, &decoder, 0), "");
 522         //if(strcmp(decoder->name, "prores") == 0 && decoder->next && strcmp(decoder->next->name, "prores_lgpl") == 0)
 523         //      decoder = decoder->next;
 524
 525         THROW_ON_ERROR2(tbb_avcodec_open(context.streams[index]->codec, decoder, single_threaded), "");
 526         return spl::shared_ptr<AVCodecContext>(context.streams[index]->codec, tbb_avcodec_close);
 527 }
 528
 529 spl::shared_ptr<AVFormatContext> open_input(const std::wstring& filename)
 530 {
 531         AVFormatContext* weak_context = nullptr;
 532         THROW_ON_ERROR2(avformat_open_input(&weak_context, u8(filename).c_str(), nullptr, nullptr), filename);
 533         spl::shared_ptr<AVFormatContext> context(weak_context, [](AVFormatContext* p)
 534         {
 535                 avformat_close_input(&p);
 536         });
 537         THROW_ON_ERROR2(avformat_find_stream_info(weak_context, nullptr), filename);
 538         fix_meta_data(*context);
 539         return context;
 540 }
 541
 542 std::wstring print_mode(int width, int height, double fps, bool interlaced)
 543 {
 544         std::wostringstream fps_ss;
 545         fps_ss << std::fixed << std::setprecision(2) << (!interlaced ? fps : 2.0 * fps);
 546
 547         return boost::lexical_cast<std::wstring>(width) + L"x" + boost::lexical_cast<std::wstring>(height) + (!interlaced ? L"p" : L"i") + fps_ss.str();
 548 }
 549
 550 bool is_valid_file(const std::wstring& filename, bool only_video)
 551 {
 552         static const auto invalid_exts = {
 553                 L".png",
 554                 L".tga",
 555                 L".bmp",
 556                 L".jpg",
 557                 L".jpeg",
 558                 L".gif",
 559                 L".tiff",
 560                 L".tif",
 561                 L".jp2",
 562                 L".jpx",
 563                 L".j2k",
 564                 L".j2c",
 565                 L".swf",
 566                 L".ct"
 567         };
 568         static const auto only_audio = {
 569                 L".mp3",
 570                 L".wav",
 571                 L".wma"
 572         };
 573         static const auto valid_exts = {
 574                 L".m2t",
 575                 L".mov",
 576                 L".mp4",
 577                 L".dv",
 578                 L".flv",
 579                 L".mpg",
 580                 L".dnxhd",
 581                 L".h264",
 582                 L".prores"
 583         };
 584
 585         auto ext = boost::to_lower_copy(boost::filesystem::path(filename).extension().wstring());
 586
 587         if(std::find(valid_exts.begin(), valid_exts.end(), ext) != valid_exts.end())
 588                 return true;
 589
 590         if (!only_video && std::find(only_audio.begin(), only_audio.end(), ext) != only_audio.end())
 591                 return true;
 592
 593         if(std::find(invalid_exts.begin(), invalid_exts.end(), ext) != invalid_exts.end())
 594                 return false;
 595
 596         if (only_video && std::find(only_audio.begin(), only_audio.end(), ext) != only_audio.end())
 597                 return false;
 598
 599         auto u8filename = u8(filename);
 600
 601         int score = 0;
 602         AVProbeData pb = {};
 603         pb.filename = u8filename.c_str();
 604
 605         if(av_probe_input_format2(&pb, false, &score) != nullptr)
 606                 return true;
 607
 608         std::ifstream file(u8filename);
 609
 610         std::vector<unsigned char> buf;
 611         for(auto file_it = std::istreambuf_iterator<char>(file); file_it != std::istreambuf_iterator<char>() && buf.size() < 1024; ++file_it)
 612                 buf.push_back(*file_it);
 613
 614         if(buf.empty())
 615                 return false;
 616
 617         pb.buf          = buf.data();
 618         pb.buf_size = static_cast<int>(buf.size());
 619
 620         return av_probe_input_format2(&pb, true, &score) != nullptr;
 621 }
 622
 623 bool try_get_duration(const std::wstring filename, std::int64_t& duration, boost::rational<std::int64_t>& time_base)
 624 {
 625         AVFormatContext* weak_context = nullptr;
 626         if (avformat_open_input(&weak_context, u8(filename).c_str(), nullptr, nullptr) < 0)
 627                 return false;
 628
 629         std::shared_ptr<AVFormatContext> context(weak_context, [](AVFormatContext* p)
 630         {
 631                 avformat_close_input(&p);
 632         });
 633
 634         context->probesize = context->probesize / 10;
 635         context->max_analyze_duration = context->probesize / 10;
 636
 637         if (avformat_find_stream_info(context.get(), nullptr) < 0)
 638                 return false;
 639
 640         const auto fps = read_fps(*context, 1.0);
 641
 642         const auto rational_fps = boost::rational<std::int64_t>(static_cast<int>(fps * AV_TIME_BASE), AV_TIME_BASE);
 643
 644         duration = boost::rational_cast<std::int64_t>(context->duration * rational_fps / AV_TIME_BASE);
 645
 646         if (rational_fps == 0)
 647                 return false;
 648
 649         time_base = 1 / rational_fps;
 650
 651         return true;
 652 }
 653
 654 std::wstring probe_stem(const std::wstring& stem, bool only_video)
 655 {
 656         auto stem2 = boost::filesystem::path(stem);
 657         auto parent = find_case_insensitive(stem2.parent_path().wstring());
 658
 659         if (!parent)
 660                 return L"";
 661
 662         auto dir = boost::filesystem::path(*parent);
 663
 664         for(auto it = boost::filesystem::directory_iterator(dir); it != boost::filesystem::directory_iterator(); ++it)
 665         {
 666                 if(boost::iequals(it->path().stem().wstring(), stem2.filename().wstring()) && is_valid_file(it->path().wstring(), only_video))
 667                         return it->path().wstring();
 668         }
 669         return L"";
 670 }
 671
 672 core::audio_channel_layout get_audio_channel_layout(int num_channels, std::uint64_t layout, const std::wstring& channel_layout_spec)
 673 {
 674         if (!channel_layout_spec.empty())
 675         {
 676                 if (boost::contains(channel_layout_spec, L":")) // Custom on the fly layout specified.
 677                 {
 678                         std::vector<std::wstring> type_and_channel_order;
 679                         boost::split(type_and_channel_order, channel_layout_spec, boost::is_any_of(L":"), boost::algorithm::token_compress_off);
 680                         auto& type                      = type_and_channel_order.at(0);
 681                         auto& order                     = type_and_channel_order.at(1);
 682
 683                         return core::audio_channel_layout(num_channels, std::move(type), order);
 684                 }
 685                 else // Preconfigured named channel layout selected.
 686                 {
 687                         auto channel_layout = core::audio_channel_layout_repository::get_default()->get_layout(channel_layout_spec);
 688
 689                         if (!channel_layout)
 690                                 CASPAR_THROW_EXCEPTION(user_error() << msg_info(L"No channel layout with name " + channel_layout_spec + L" registered"));
 691
 692                         channel_layout->num_channels = num_channels;
 693
 694                         return *channel_layout;
 695                 }
 696         }
 697
 698         if (!layout)
 699         {
 700                 if (num_channels == 1)
 701                         return core::audio_channel_layout(num_channels, L"mono", L"FC");
 702                 else if (num_channels == 2)
 703                         return core::audio_channel_layout(num_channels, L"stereo", L"FL FR");
 704                 else
 705                         return core::audio_channel_layout(num_channels, L"", L""); // Passthru without named channels as is.
 706         }
 707
 708         // What FFmpeg calls "channel layout" is only the "layout type" of a channel layout in
 709         // CasparCG where the channel layout supports different orders as well.
 710         // The user needs to provide additional mix-configs in casparcg.config to support more
 711         // than the most common (5.1, mono and stereo) types.
 712
 713         // Based on information in https://ffmpeg.org/ffmpeg-utils.html#Channel-Layout
 714         switch (layout)
 715         {
 716         case AV_CH_LAYOUT_MONO:
 717                 return core::audio_channel_layout(num_channels, L"mono",                        L"FC");
 718         case AV_CH_LAYOUT_STEREO:
 719                 return core::audio_channel_layout(num_channels, L"stereo",                      L"FL FR");
 720         case AV_CH_LAYOUT_2POINT1:
 721                 return core::audio_channel_layout(num_channels, L"2.1",                         L"FL FR LFE");
 722         case AV_CH_LAYOUT_SURROUND:
 723                 return core::audio_channel_layout(num_channels, L"3.0",                         L"FL FR FC");
 724         case AV_CH_LAYOUT_2_1:
 725                 return core::audio_channel_layout(num_channels, L"3.0(back)",           L"FL FR BC");
 726         case AV_CH_LAYOUT_4POINT0:
 727                 return core::audio_channel_layout(num_channels, L"4.0",                         L"FL FR FC BC");
 728         case AV_CH_LAYOUT_QUAD:
 729                 return core::audio_channel_layout(num_channels, L"quad",                        L"FL FR BL BR");
 730         case AV_CH_LAYOUT_2_2:
 731                 return core::audio_channel_layout(num_channels, L"quad(side)",          L"FL FR SL SR");
 732         case AV_CH_LAYOUT_3POINT1:
 733                 return core::audio_channel_layout(num_channels, L"3.1",                         L"FL FR FC LFE");
 734         case AV_CH_LAYOUT_5POINT0_BACK:
 735                 return core::audio_channel_layout(num_channels, L"5.0",                         L"FL FR FC BL BR");
 736         case AV_CH_LAYOUT_5POINT0:
 737                 return core::audio_channel_layout(num_channels, L"5.0(side)",           L"FL FR FC SL SR");
 738         case AV_CH_LAYOUT_4POINT1:
 739                 return core::audio_channel_layout(num_channels, L"4.1",                         L"FL FR FC LFE BC");
 740         case AV_CH_LAYOUT_5POINT1_BACK:
 741                 return core::audio_channel_layout(num_channels, L"5.1",                         L"FL FR FC LFE BL BR");
 742         case AV_CH_LAYOUT_5POINT1:
 743                 return core::audio_channel_layout(num_channels, L"5.1(side)",           L"FL FR FC LFE SL SR");
 744         case AV_CH_LAYOUT_6POINT0:
 745                 return core::audio_channel_layout(num_channels, L"6.0",                         L"FL FR FC BC SL SR");
 746         case AV_CH_LAYOUT_6POINT0_FRONT:
 747                 return core::audio_channel_layout(num_channels, L"6.0(front)",          L"FL FR FLC FRC SL SR");
 748         case AV_CH_LAYOUT_HEXAGONAL:
 749                 return core::audio_channel_layout(num_channels, L"hexagonal",           L"FL FR FC BL BR BC");
 750         case AV_CH_LAYOUT_6POINT1:
 751                 return core::audio_channel_layout(num_channels, L"6.1",                         L"FL FR FC LFE BC SL SR");
 752         case AV_CH_LAYOUT_6POINT1_BACK:
 753                 return core::audio_channel_layout(num_channels, L"6.1(back)",           L"FL FR FC LFE BL BR BC");
 754         case AV_CH_LAYOUT_6POINT1_FRONT:
 755                 return core::audio_channel_layout(num_channels, L"6.1(front)",          L"FL FR LFE FLC FRC SL SR");
 756         case AV_CH_LAYOUT_7POINT0:
 757                 return core::audio_channel_layout(num_channels, L"7.0",                         L"FL FR FC BL BR SL SR");
 758         case AV_CH_LAYOUT_7POINT0_FRONT:
 759                 return core::audio_channel_layout(num_channels, L"7.0(front)",          L"FL FR FC FLC FRC SL SR");
 760         case AV_CH_LAYOUT_7POINT1:
 761                 return core::audio_channel_layout(num_channels, L"7.1",                         L"FL FR FC LFE BL BR SL SR");
 762         case AV_CH_LAYOUT_7POINT1_WIDE_BACK:
 763                 return core::audio_channel_layout(num_channels, L"7.1(wide)",           L"FL FR FC LFE BL BR FLC FRC");
 764         case AV_CH_LAYOUT_7POINT1_WIDE:
 765                 return core::audio_channel_layout(num_channels, L"7.1(wide-side)",      L"FL FR FC LFE FLC FRC SL SR");
 766         case AV_CH_LAYOUT_STEREO_DOWNMIX:
 767                 return core::audio_channel_layout(num_channels, L"downmix",                     L"DL DR");
 768         default:
 769                 // Passthru
 770                 return core::audio_channel_layout(num_channels, L"", L"");
 771         }
 772 }
 773
 774 // av_get_default_channel_layout does not work for layouts not predefined in ffmpeg. This is needed to support > 8 channels.
 775 std::uint64_t create_channel_layout_bitmask(int num_channels)
 776 {
 777         if (num_channels > 63)
 778                 CASPAR_THROW_EXCEPTION(invalid_argument() << msg_info(L"FFmpeg cannot handle more than 63 audio channels"));
 779
 780         const auto ALL_63_CHANNELS = 0x7FFFFFFFFFFFFFFFULL;
 781
 782         auto to_shift = 63 - num_channels;
 783         auto result = ALL_63_CHANNELS >> to_shift;
 784
 785         return static_cast<std::uint64_t>(result);
 786 }
 787
 788 std::string to_string(const boost::rational<int>& framerate)
 789 {
 790         return boost::lexical_cast<std::string>(framerate.numerator())
 791                 + "/" + boost::lexical_cast<std::string>(framerate.denominator()) + " (" + boost::lexical_cast<std::string>(static_cast<double>(framerate.numerator()) / static_cast<double>(framerate.denominator())) + ") fps";
 792 }
 793
 794 std::vector<int> find_audio_cadence(const boost::rational<int>& framerate)
 795 {
 796         static std::map<boost::rational<int>, std::vector<int>> CADENCES_BY_FRAMERATE = []
 797         {
 798                 std::map<boost::rational<int>, std::vector<int>> result;
 799
 800                 for (core::video_format format : enum_constants<core::video_format>())
 801                 {
 802                         core::video_format_desc desc(format);
 803                         boost::rational<int> format_rate(desc.time_scale, desc.duration);
 804
 805                         result.insert(std::make_pair(format_rate, desc.audio_cadence));
 806                 }
 807
 808                 return result;
 809         }();
 810
 811         auto exact_match = CADENCES_BY_FRAMERATE.find(framerate);
 812
 813         if (exact_match != CADENCES_BY_FRAMERATE.end())
 814                 return exact_match->second;
 815
 816         boost::rational<int> closest_framerate_diff = std::numeric_limits<int>::max();
 817         boost::rational<int> closest_framerate = 0;
 818
 819         for (auto format_framerate : CADENCES_BY_FRAMERATE | boost::adaptors::map_keys)
 820         {
 821                 auto diff = boost::abs(framerate - format_framerate);
 822
 823                 if (diff < closest_framerate_diff)
 824                 {
 825                         closest_framerate_diff = diff;
 826                         closest_framerate = format_framerate;
 827                 }
 828         }
 829
 830         if (is_logging_quiet_for_thread())
 831                 CASPAR_LOG(debug) << "No exact audio cadence match found for framerate " << to_string(framerate)
 832                 << "\nClosest match is " << to_string(closest_framerate)
 833                 << "\nwhich is a " << to_string(closest_framerate_diff) << " difference.";
 834         else
 835                 CASPAR_LOG(warning) << "No exact audio cadence match found for framerate " << to_string(framerate)
 836                 << "\nClosest match is " << to_string(closest_framerate)
 837                 << "\nwhich is a " << to_string(closest_framerate_diff) << " difference.";
 838
 839         return CADENCES_BY_FRAMERATE[closest_framerate];
 840 }
 841
 842 //
 843 //void av_dup_frame(AVFrame* frame)
 844 //{
 845 //      AVFrame* new_frame = avcodec_alloc_frame();
 846 //
 847 //
 848 //      const uint8_t *src_data[4] = {0};
 849 //      memcpy(const_cast<uint8_t**>(&src_data[0]), frame->data, 4);
 850 //      const int src_linesizes[4] = {0};
 851 //      memcpy(const_cast<int*>(&src_linesizes[0]), frame->linesize, 4);
 852 //
 853 //      av_image_alloc(new_frame->data, new_frame->linesize, new_frame->width, new_frame->height, frame->format, 16);
 854 //
 855 //      av_image_copy(new_frame->data, new_frame->linesize, src_data, src_linesizes, frame->format, new_frame->width, new_frame->height);
 856 //
 857 //      frame =
 858 //}
 859
 860 }}