OpenShot Library | libopenshot  0.3.0
FFmpegReader.cpp
Go to the documentation of this file.
1 
12 // Copyright (c) 2008-2019 OpenShot Studios, LLC, Fabrice Bellard
13 //
14 // SPDX-License-Identifier: LGPL-3.0-or-later
15 
16 #include <thread> // for std::this_thread::sleep_for
17 #include <chrono> // for std::chrono::milliseconds
18 #include <unistd.h>
19 
20 #include "FFmpegUtilities.h"
21 
22 #include "FFmpegReader.h"
23 #include "Exceptions.h"
24 #include "Timeline.h"
25 #include "ZmqLogger.h"
26 
27 #define ENABLE_VAAPI 0
28 
29 #if USE_HW_ACCEL
30 #define MAX_SUPPORTED_WIDTH 1950
31 #define MAX_SUPPORTED_HEIGHT 1100
32 
33 #if ENABLE_VAAPI
34 #include "libavutil/hwcontext_vaapi.h"
35 
36 typedef struct VAAPIDecodeContext {
37  VAProfile va_profile;
38  VAEntrypoint va_entrypoint;
39  VAConfigID va_config;
40  VAContextID va_context;
41 
42 #if FF_API_STRUCT_VAAPI_CONTEXT
43  // FF_DISABLE_DEPRECATION_WARNINGS
44  int have_old_context;
45  struct vaapi_context *old_context;
46  AVBufferRef *device_ref;
47  // FF_ENABLE_DEPRECATION_WARNINGS
48 #endif
49 
50  AVHWDeviceContext *device;
51  AVVAAPIDeviceContext *hwctx;
52 
53  AVHWFramesContext *frames;
54  AVVAAPIFramesContext *hwfc;
55 
56  enum AVPixelFormat surface_format;
57  int surface_count;
58  } VAAPIDecodeContext;
59 #endif // ENABLE_VAAPI
60 #endif // USE_HW_ACCEL
61 
62 
63 using namespace openshot;
64 
65 int hw_de_on = 0;
66 #if USE_HW_ACCEL
67  AVPixelFormat hw_de_av_pix_fmt_global = AV_PIX_FMT_NONE;
68  AVHWDeviceType hw_de_av_device_type_global = AV_HWDEVICE_TYPE_NONE;
69 #endif
70 
71 FFmpegReader::FFmpegReader(const std::string &path, bool inspect_reader)
72  : last_frame(0), is_seeking(0), seeking_pts(0), seeking_frame(0), seek_count(0), NO_PTS_OFFSET(-99999),
73  path(path), is_video_seek(true), check_interlace(false), check_fps(false), enable_seek(true), is_open(false),
74  seek_audio_frame_found(0), seek_video_frame_found(0),is_duration_known(false), largest_frame_processed(0),
75  current_video_frame(0), packet(NULL), max_concurrent_frames(OPEN_MP_NUM_PROCESSORS), audio_pts(0),
76  video_pts(0), pFormatCtx(NULL), videoStream(-1), audioStream(-1), pCodecCtx(NULL), aCodecCtx(NULL),
77  pStream(NULL), aStream(NULL), pFrame(NULL), previous_packet_location{-1,0},
78  hold_packet(false) {
79 
80  // Initialize FFMpeg, and register all formats and codecs
83 
84  // Init timestamp offsets
85  pts_offset_seconds = NO_PTS_OFFSET;
86  video_pts_seconds = NO_PTS_OFFSET;
87  audio_pts_seconds = NO_PTS_OFFSET;
88 
89  // Init cache
90  working_cache.SetMaxBytesFromInfo(max_concurrent_frames * info.fps.ToDouble() * 2, info.width, info.height, info.sample_rate, info.channels);
92 
93  // Open and Close the reader, to populate its attributes (such as height, width, etc...)
94  if (inspect_reader) {
95  Open();
96  Close();
97  }
98 }
99 
101  if (is_open)
102  // Auto close reader if not already done
103  Close();
104 }
105 
106 // This struct holds the associated video frame and starting sample # for an audio packet.
107 bool AudioLocation::is_near(AudioLocation location, int samples_per_frame, int64_t amount) {
108  // Is frame even close to this one?
109  if (abs(location.frame - frame) >= 2)
110  // This is too far away to be considered
111  return false;
112 
113  // Note that samples_per_frame can vary slightly frame to frame when the
114  // audio sampling rate is not an integer multiple of the video fps.
115  int64_t diff = samples_per_frame * (location.frame - frame) + location.sample_start - sample_start;
116  if (abs(diff) <= amount)
117  // close
118  return true;
119 
120  // not close
121  return false;
122 }
123 
124 #if USE_HW_ACCEL
125 
126 // Get hardware pix format
127 static enum AVPixelFormat get_hw_dec_format(AVCodecContext *ctx, const enum AVPixelFormat *pix_fmts)
128 {
129  const enum AVPixelFormat *p;
130 
131  for (p = pix_fmts; *p != AV_PIX_FMT_NONE; p++) {
132  switch (*p) {
133 #if defined(__linux__)
134  // Linux pix formats
135  case AV_PIX_FMT_VAAPI:
136  hw_de_av_pix_fmt_global = AV_PIX_FMT_VAAPI;
137  hw_de_av_device_type_global = AV_HWDEVICE_TYPE_VAAPI;
138  return *p;
139  break;
140  case AV_PIX_FMT_VDPAU:
141  hw_de_av_pix_fmt_global = AV_PIX_FMT_VDPAU;
142  hw_de_av_device_type_global = AV_HWDEVICE_TYPE_VDPAU;
143  return *p;
144  break;
145 #endif
146 #if defined(_WIN32)
147  // Windows pix formats
148  case AV_PIX_FMT_DXVA2_VLD:
149  hw_de_av_pix_fmt_global = AV_PIX_FMT_DXVA2_VLD;
150  hw_de_av_device_type_global = AV_HWDEVICE_TYPE_DXVA2;
151  return *p;
152  break;
153  case AV_PIX_FMT_D3D11:
154  hw_de_av_pix_fmt_global = AV_PIX_FMT_D3D11;
155  hw_de_av_device_type_global = AV_HWDEVICE_TYPE_D3D11VA;
156  return *p;
157  break;
158 #endif
159 #if defined(__APPLE__)
160  // Apple pix formats
161  case AV_PIX_FMT_VIDEOTOOLBOX:
162  hw_de_av_pix_fmt_global = AV_PIX_FMT_VIDEOTOOLBOX;
163  hw_de_av_device_type_global = AV_HWDEVICE_TYPE_VIDEOTOOLBOX;
164  return *p;
165  break;
166 #endif
167  // Cross-platform pix formats
168  case AV_PIX_FMT_CUDA:
169  hw_de_av_pix_fmt_global = AV_PIX_FMT_CUDA;
170  hw_de_av_device_type_global = AV_HWDEVICE_TYPE_CUDA;
171  return *p;
172  break;
173  case AV_PIX_FMT_QSV:
174  hw_de_av_pix_fmt_global = AV_PIX_FMT_QSV;
175  hw_de_av_device_type_global = AV_HWDEVICE_TYPE_QSV;
176  return *p;
177  break;
178  default:
179  // This is only here to silence unused-enum warnings
180  break;
181  }
182  }
183  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::get_hw_dec_format (Unable to decode this file using hardware decode)");
184  return AV_PIX_FMT_NONE;
185 }
186 
187 int FFmpegReader::IsHardwareDecodeSupported(int codecid)
188 {
189  int ret;
190  switch (codecid) {
191  case AV_CODEC_ID_H264:
192  case AV_CODEC_ID_MPEG2VIDEO:
193  case AV_CODEC_ID_VC1:
194  case AV_CODEC_ID_WMV1:
195  case AV_CODEC_ID_WMV2:
196  case AV_CODEC_ID_WMV3:
197  ret = 1;
198  break;
199  default :
200  ret = 0;
201  break;
202  }
203  return ret;
204 }
205 #endif // USE_HW_ACCEL
206 
208  // Open reader if not already open
209  if (!is_open) {
210  // Prevent async calls to the following code
211  const std::lock_guard<std::recursive_mutex> lock(getFrameMutex);
212 
213  // Initialize format context
214  pFormatCtx = NULL;
215  {
217  ZmqLogger::Instance()->AppendDebugMethod("Decode hardware acceleration settings", "hw_de_on", hw_de_on, "HARDWARE_DECODER", openshot::Settings::Instance()->HARDWARE_DECODER);
218  }
219 
220  // Open video file
221  if (avformat_open_input(&pFormatCtx, path.c_str(), NULL, NULL) != 0)
222  throw InvalidFile("File could not be opened.", path);
223 
224  // Retrieve stream information
225  if (avformat_find_stream_info(pFormatCtx, NULL) < 0)
226  throw NoStreamsFound("No streams found in file.", path);
227 
228  videoStream = -1;
229  audioStream = -1;
230 
231  // Init end-of-file detection variables
232  packet_status.reset(true);
233 
234  // Loop through each stream, and identify the video and audio stream index
235  for (unsigned int i = 0; i < pFormatCtx->nb_streams; i++) {
236  // Is this a video stream?
237  if (AV_GET_CODEC_TYPE(pFormatCtx->streams[i]) == AVMEDIA_TYPE_VIDEO && videoStream < 0) {
238  videoStream = i;
239  packet_status.video_eof = false;
240  packet_status.packets_eof = false;
241  packet_status.end_of_file = false;
242  }
243  // Is this an audio stream?
244  if (AV_GET_CODEC_TYPE(pFormatCtx->streams[i]) == AVMEDIA_TYPE_AUDIO && audioStream < 0) {
245  audioStream = i;
246  packet_status.audio_eof = false;
247  packet_status.packets_eof = false;
248  packet_status.end_of_file = false;
249  }
250  }
251  if (videoStream == -1 && audioStream == -1)
252  throw NoStreamsFound("No video or audio streams found in this file.", path);
253 
254  // Is there a video stream?
255  if (videoStream != -1) {
256  // Set the stream index
257  info.video_stream_index = videoStream;
258 
259  // Set the codec and codec context pointers
260  pStream = pFormatCtx->streams[videoStream];
261 
262  // Find the codec ID from stream
263  const AVCodecID codecId = AV_FIND_DECODER_CODEC_ID(pStream);
264 
265  // Get codec and codec context from stream
266  const AVCodec *pCodec = avcodec_find_decoder(codecId);
267  AVDictionary *opts = NULL;
268  int retry_decode_open = 2;
269  // If hw accel is selected but hardware cannot handle repeat with software decoding
270  do {
271  pCodecCtx = AV_GET_CODEC_CONTEXT(pStream, pCodec);
272 #if USE_HW_ACCEL
273  if (hw_de_on && (retry_decode_open==2)) {
274  // Up to here no decision is made if hardware or software decode
275  hw_de_supported = IsHardwareDecodeSupported(pCodecCtx->codec_id);
276  }
277 #endif
278  retry_decode_open = 0;
279 
280  // Set number of threads equal to number of processors (not to exceed 16)
281  pCodecCtx->thread_count = std::min(FF_NUM_PROCESSORS, 16);
282 
283  if (pCodec == NULL) {
284  throw InvalidCodec("A valid video codec could not be found for this file.", path);
285  }
286 
287  // Init options
288  av_dict_set(&opts, "strict", "experimental", 0);
289 #if USE_HW_ACCEL
290  if (hw_de_on && hw_de_supported) {
291  // Open Hardware Acceleration
292  int i_decoder_hw = 0;
293  char adapter[256];
294  char *adapter_ptr = NULL;
295  int adapter_num;
297  fprintf(stderr, "Hardware decoding device number: %d\n", adapter_num);
298 
299  // Set hardware pix format (callback)
300  pCodecCtx->get_format = get_hw_dec_format;
301 
302  if (adapter_num < 3 && adapter_num >=0) {
303 #if defined(__linux__)
304  snprintf(adapter,sizeof(adapter),"/dev/dri/renderD%d", adapter_num+128);
305  adapter_ptr = adapter;
307  switch (i_decoder_hw) {
308  case 1:
309  hw_de_av_device_type = AV_HWDEVICE_TYPE_VAAPI;
310  break;
311  case 2:
312  hw_de_av_device_type = AV_HWDEVICE_TYPE_CUDA;
313  break;
314  case 6:
315  hw_de_av_device_type = AV_HWDEVICE_TYPE_VDPAU;
316  break;
317  case 7:
318  hw_de_av_device_type = AV_HWDEVICE_TYPE_QSV;
319  break;
320  default:
321  hw_de_av_device_type = AV_HWDEVICE_TYPE_VAAPI;
322  break;
323  }
324 
325 #elif defined(_WIN32)
326  adapter_ptr = NULL;
328  switch (i_decoder_hw) {
329  case 2:
330  hw_de_av_device_type = AV_HWDEVICE_TYPE_CUDA;
331  break;
332  case 3:
333  hw_de_av_device_type = AV_HWDEVICE_TYPE_DXVA2;
334  break;
335  case 4:
336  hw_de_av_device_type = AV_HWDEVICE_TYPE_D3D11VA;
337  break;
338  case 7:
339  hw_de_av_device_type = AV_HWDEVICE_TYPE_QSV;
340  break;
341  default:
342  hw_de_av_device_type = AV_HWDEVICE_TYPE_DXVA2;
343  break;
344  }
345 #elif defined(__APPLE__)
346  adapter_ptr = NULL;
348  switch (i_decoder_hw) {
349  case 5:
350  hw_de_av_device_type = AV_HWDEVICE_TYPE_VIDEOTOOLBOX;
351  break;
352  case 7:
353  hw_de_av_device_type = AV_HWDEVICE_TYPE_QSV;
354  break;
355  default:
356  hw_de_av_device_type = AV_HWDEVICE_TYPE_VIDEOTOOLBOX;
357  break;
358  }
359 #endif
360 
361  } else {
362  adapter_ptr = NULL; // Just to be sure
363  }
364 
365  // Check if it is there and writable
366 #if defined(__linux__)
367  if( adapter_ptr != NULL && access( adapter_ptr, W_OK ) == 0 ) {
368 #elif defined(_WIN32)
369  if( adapter_ptr != NULL ) {
370 #elif defined(__APPLE__)
371  if( adapter_ptr != NULL ) {
372 #endif
373  ZmqLogger::Instance()->AppendDebugMethod("Decode Device present using device");
374  }
375  else {
376  adapter_ptr = NULL; // use default
377  ZmqLogger::Instance()->AppendDebugMethod("Decode Device not present using default");
378  }
379 
380  hw_device_ctx = NULL;
381  // Here the first hardware initialisations are made
382  if (av_hwdevice_ctx_create(&hw_device_ctx, hw_de_av_device_type, adapter_ptr, NULL, 0) >= 0) {
383  if (!(pCodecCtx->hw_device_ctx = av_buffer_ref(hw_device_ctx))) {
384  throw InvalidCodec("Hardware device reference create failed.", path);
385  }
386 
387  /*
388  av_buffer_unref(&ist->hw_frames_ctx);
389  ist->hw_frames_ctx = av_hwframe_ctx_alloc(hw_device_ctx);
390  if (!ist->hw_frames_ctx) {
391  av_log(avctx, AV_LOG_ERROR, "Error creating a CUDA frames context\n");
392  return AVERROR(ENOMEM);
393  }
394 
395  frames_ctx = (AVHWFramesContext*)ist->hw_frames_ctx->data;
396 
397  frames_ctx->format = AV_PIX_FMT_CUDA;
398  frames_ctx->sw_format = avctx->sw_pix_fmt;
399  frames_ctx->width = avctx->width;
400  frames_ctx->height = avctx->height;
401 
402  av_log(avctx, AV_LOG_DEBUG, "Initializing CUDA frames context: sw_format = %s, width = %d, height = %d\n",
403  av_get_pix_fmt_name(frames_ctx->sw_format), frames_ctx->width, frames_ctx->height);
404 
405 
406  ret = av_hwframe_ctx_init(pCodecCtx->hw_device_ctx);
407  ret = av_hwframe_ctx_init(ist->hw_frames_ctx);
408  if (ret < 0) {
409  av_log(avctx, AV_LOG_ERROR, "Error initializing a CUDA frame pool\n");
410  return ret;
411  }
412  */
413  }
414  else {
415  throw InvalidCodec("Hardware device create failed.", path);
416  }
417  }
418 #endif // USE_HW_ACCEL
419 
420  // Disable per-frame threading for album arts
421  // Using FF_THREAD_FRAME adds one frame decoding delay per thread,
422  // but there's only one frame in this case.
423  if (HasAlbumArt())
424  {
425  pCodecCtx->thread_type &= ~FF_THREAD_FRAME;
426  }
427 
428  // Open video codec
429  int avcodec_return = avcodec_open2(pCodecCtx, pCodec, &opts);
430  if (avcodec_return < 0) {
431  std::stringstream avcodec_error_msg;
432  avcodec_error_msg << "A video codec was found, but could not be opened. Error: " << av_err2string(avcodec_return);
433  throw InvalidCodec(avcodec_error_msg.str(), path);
434  }
435 
436 #if USE_HW_ACCEL
437  if (hw_de_on && hw_de_supported) {
438  AVHWFramesConstraints *constraints = NULL;
439  void *hwconfig = NULL;
440  hwconfig = av_hwdevice_hwconfig_alloc(hw_device_ctx);
441 
442 // TODO: needs va_config!
443 #if ENABLE_VAAPI
444  ((AVVAAPIHWConfig *)hwconfig)->config_id = ((VAAPIDecodeContext *)(pCodecCtx->priv_data))->va_config;
445  constraints = av_hwdevice_get_hwframe_constraints(hw_device_ctx,hwconfig);
446 #endif // ENABLE_VAAPI
447  if (constraints) {
448  if (pCodecCtx->coded_width < constraints->min_width ||
449  pCodecCtx->coded_height < constraints->min_height ||
450  pCodecCtx->coded_width > constraints->max_width ||
451  pCodecCtx->coded_height > constraints->max_height) {
452  ZmqLogger::Instance()->AppendDebugMethod("DIMENSIONS ARE TOO LARGE for hardware acceleration\n");
453  hw_de_supported = 0;
454  retry_decode_open = 1;
455  AV_FREE_CONTEXT(pCodecCtx);
456  if (hw_device_ctx) {
457  av_buffer_unref(&hw_device_ctx);
458  hw_device_ctx = NULL;
459  }
460  }
461  else {
462  // All is just peachy
463  ZmqLogger::Instance()->AppendDebugMethod("\nDecode hardware acceleration is used\n", "Min width :", constraints->min_width, "Min Height :", constraints->min_height, "MaxWidth :", constraints->max_width, "MaxHeight :", constraints->max_height, "Frame width :", pCodecCtx->coded_width, "Frame height :", pCodecCtx->coded_height);
464  retry_decode_open = 0;
465  }
466  av_hwframe_constraints_free(&constraints);
467  if (hwconfig) {
468  av_freep(&hwconfig);
469  }
470  }
471  else {
472  int max_h, max_w;
473  //max_h = ((getenv( "LIMIT_HEIGHT_MAX" )==NULL) ? MAX_SUPPORTED_HEIGHT : atoi(getenv( "LIMIT_HEIGHT_MAX" )));
475  //max_w = ((getenv( "LIMIT_WIDTH_MAX" )==NULL) ? MAX_SUPPORTED_WIDTH : atoi(getenv( "LIMIT_WIDTH_MAX" )));
477  ZmqLogger::Instance()->AppendDebugMethod("Constraints could not be found using default limit\n");
478  //cerr << "Constraints could not be found using default limit\n";
479  if (pCodecCtx->coded_width < 0 ||
480  pCodecCtx->coded_height < 0 ||
481  pCodecCtx->coded_width > max_w ||
482  pCodecCtx->coded_height > max_h ) {
483  ZmqLogger::Instance()->AppendDebugMethod("DIMENSIONS ARE TOO LARGE for hardware acceleration\n", "Max Width :", max_w, "Max Height :", max_h, "Frame width :", pCodecCtx->coded_width, "Frame height :", pCodecCtx->coded_height);
484  hw_de_supported = 0;
485  retry_decode_open = 1;
486  AV_FREE_CONTEXT(pCodecCtx);
487  if (hw_device_ctx) {
488  av_buffer_unref(&hw_device_ctx);
489  hw_device_ctx = NULL;
490  }
491  }
492  else {
493  ZmqLogger::Instance()->AppendDebugMethod("\nDecode hardware acceleration is used\n", "Max Width :", max_w, "Max Height :", max_h, "Frame width :", pCodecCtx->coded_width, "Frame height :", pCodecCtx->coded_height);
494  retry_decode_open = 0;
495  }
496  }
497  } // if hw_de_on && hw_de_supported
498  else {
499  ZmqLogger::Instance()->AppendDebugMethod("\nDecode in software is used\n");
500  }
501 #else
502  retry_decode_open = 0;
503 #endif // USE_HW_ACCEL
504  } while (retry_decode_open); // retry_decode_open
505  // Free options
506  av_dict_free(&opts);
507 
508  // Update the File Info struct with video details (if a video stream is found)
509  UpdateVideoInfo();
510  }
511 
512  // Is there an audio stream?
513  if (audioStream != -1) {
514  // Set the stream index
515  info.audio_stream_index = audioStream;
516 
517  // Get a pointer to the codec context for the audio stream
518  aStream = pFormatCtx->streams[audioStream];
519 
520  // Find the codec ID from stream
521  AVCodecID codecId = AV_FIND_DECODER_CODEC_ID(aStream);
522 
523  // Get codec and codec context from stream
524  const AVCodec *aCodec = avcodec_find_decoder(codecId);
525  aCodecCtx = AV_GET_CODEC_CONTEXT(aStream, aCodec);
526 
527  // Set number of threads equal to number of processors (not to exceed 16)
528  aCodecCtx->thread_count = std::min(FF_NUM_PROCESSORS, 16);
529 
530  if (aCodec == NULL) {
531  throw InvalidCodec("A valid audio codec could not be found for this file.", path);
532  }
533 
534  // Init options
535  AVDictionary *opts = NULL;
536  av_dict_set(&opts, "strict", "experimental", 0);
537 
538  // Open audio codec
539  if (avcodec_open2(aCodecCtx, aCodec, &opts) < 0)
540  throw InvalidCodec("An audio codec was found, but could not be opened.", path);
541 
542  // Free options
543  av_dict_free(&opts);
544 
545  // Update the File Info struct with audio details (if an audio stream is found)
546  UpdateAudioInfo();
547  }
548 
549  // Add format metadata (if any)
550  AVDictionaryEntry *tag = NULL;
551  while ((tag = av_dict_get(pFormatCtx->metadata, "", tag, AV_DICT_IGNORE_SUFFIX))) {
552  QString str_key = tag->key;
553  QString str_value = tag->value;
554  info.metadata[str_key.toStdString()] = str_value.trimmed().toStdString();
555  }
556 
557  // Init previous audio location to zero
558  previous_packet_location.frame = -1;
559  previous_packet_location.sample_start = 0;
560 
561  // Adjust cache size based on size of frame and audio
562  working_cache.SetMaxBytesFromInfo(max_concurrent_frames * info.fps.ToDouble() * 2, info.width, info.height, info.sample_rate, info.channels);
564 
565  // Scan PTS for any offsets (i.e. non-zero starting streams). At least 1 stream must start at zero timestamp.
566  // This method allows us to shift timestamps to ensure at least 1 stream is starting at zero.
567  UpdatePTSOffset();
568 
569  // Override an invalid framerate
570  if (info.fps.ToFloat() > 240.0f || (info.fps.num <= 0 || info.fps.den <= 0) || info.video_length <= 0) {
571  // Calculate FPS, duration, video bit rate, and video length manually
572  // by scanning through all the video stream packets
573  CheckFPS();
574  }
575 
576  // Mark as "open"
577  is_open = true;
578 
579  // Seek back to beginning of file (if not already seeking)
580  if (!is_seeking) {
581  Seek(1);
582  }
583  }
584 }
585 
587  // Close all objects, if reader is 'open'
588  if (is_open) {
589  // Prevent async calls to the following code
590  const std::lock_guard<std::recursive_mutex> lock(getFrameMutex);
591 
592  // Mark as "closed"
593  is_open = false;
594 
595  // Keep track of most recent packet
596  AVPacket *recent_packet = packet;
597 
598  // Drain any packets from the decoder
599  packet = NULL;
600  int attempts = 0;
601  int max_attempts = 128;
602  while (packet_status.packets_decoded() < packet_status.packets_read() && attempts < max_attempts) {
603  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::Close (Drain decoder loop)",
604  "packets_read", packet_status.packets_read(),
605  "packets_decoded", packet_status.packets_decoded(),
606  "attempts", attempts);
607  if (packet_status.video_decoded < packet_status.video_read) {
608  ProcessVideoPacket(info.video_length);
609  }
610  if (packet_status.audio_decoded < packet_status.audio_read) {
611  ProcessAudioPacket(info.video_length);
612  }
613  attempts++;
614  }
615 
616  // Remove packet
617  if (recent_packet) {
618  RemoveAVPacket(recent_packet);
619  }
620 
621  // Close the video codec
622  if (info.has_video) {
623  if(avcodec_is_open(pCodecCtx)) {
624  avcodec_flush_buffers(pCodecCtx);
625  }
626  AV_FREE_CONTEXT(pCodecCtx);
627 #if USE_HW_ACCEL
628  if (hw_de_on) {
629  if (hw_device_ctx) {
630  av_buffer_unref(&hw_device_ctx);
631  hw_device_ctx = NULL;
632  }
633  }
634 #endif // USE_HW_ACCEL
635  }
636 
637  // Close the audio codec
638  if (info.has_audio) {
639  if(avcodec_is_open(aCodecCtx)) {
640  avcodec_flush_buffers(aCodecCtx);
641  }
642  AV_FREE_CONTEXT(aCodecCtx);
643  }
644 
645  // Clear final cache
646  final_cache.Clear();
647  working_cache.Clear();
648 
649  // Close the video file
650  avformat_close_input(&pFormatCtx);
651  av_freep(&pFormatCtx);
652 
653  // Reset some variables
654  last_frame = 0;
655  hold_packet = false;
656  largest_frame_processed = 0;
657  seek_audio_frame_found = 0;
658  seek_video_frame_found = 0;
659  current_video_frame = 0;
660  last_video_frame.reset();
661  }
662 }
663 
664 bool FFmpegReader::HasAlbumArt() {
665  // Check if the video stream we use is an attached picture
666  // This won't return true if the file has a cover image as a secondary stream
667  // like an MKV file with an attached image file
668  return pFormatCtx && videoStream >= 0 && pFormatCtx->streams[videoStream]
669  && (pFormatCtx->streams[videoStream]->disposition & AV_DISPOSITION_ATTACHED_PIC);
670 }
671 
672 void FFmpegReader::UpdateAudioInfo() {
673  // Set values of FileInfo struct
674  info.has_audio = true;
675  info.file_size = pFormatCtx->pb ? avio_size(pFormatCtx->pb) : -1;
676  info.acodec = aCodecCtx->codec->name;
677  info.channels = AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channels;
678  if (AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channel_layout == 0)
679  AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channel_layout = av_get_default_channel_layout(AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channels);
680  info.channel_layout = (ChannelLayout) AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channel_layout;
681  info.sample_rate = AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->sample_rate;
682  info.audio_bit_rate = AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->bit_rate;
683  if (info.audio_bit_rate <= 0) {
684  // Get bitrate from format
685  info.audio_bit_rate = pFormatCtx->bit_rate;
686  }
687 
688  // Set audio timebase
689  info.audio_timebase.num = aStream->time_base.num;
690  info.audio_timebase.den = aStream->time_base.den;
691 
692  // Get timebase of audio stream (if valid) and greater than the current duration
693  if (aStream->duration > 0 && aStream->duration > info.duration) {
694  // Get duration from audio stream
695  info.duration = aStream->duration * info.audio_timebase.ToDouble();
696  } else if (pFormatCtx->duration > 0 && info.duration <= 0.0f) {
697  // Use the format's duration
698  info.duration = float(pFormatCtx->duration) / AV_TIME_BASE;
699  }
700 
701  // Calculate duration from filesize and bitrate (if any)
702  if (info.duration <= 0.0f && info.video_bit_rate > 0 && info.file_size > 0) {
703  // Estimate from bitrate, total bytes, and framerate
705  }
706 
707  // Check for an invalid video length
708  if (info.has_video && info.video_length <= 0) {
709  // Calculate the video length from the audio duration
711  }
712 
713  // Set video timebase (if no video stream was found)
714  if (!info.has_video) {
715  // Set a few important default video settings (so audio can be divided into frames)
716  info.fps.num = 24;
717  info.fps.den = 1;
718  info.video_timebase.num = 1;
719  info.video_timebase.den = 24;
721  info.width = 720;
722  info.height = 480;
723  }
724 
725  // Fix invalid video lengths for certain types of files (MP3 for example)
726  if (info.has_video && ((info.duration * info.fps.ToDouble()) - info.video_length > 60)) {
728  }
729 
730  // Add audio metadata (if any found)
731  AVDictionaryEntry *tag = NULL;
732  while ((tag = av_dict_get(aStream->metadata, "", tag, AV_DICT_IGNORE_SUFFIX))) {
733  QString str_key = tag->key;
734  QString str_value = tag->value;
735  info.metadata[str_key.toStdString()] = str_value.trimmed().toStdString();
736  }
737 }
738 
739 void FFmpegReader::UpdateVideoInfo() {
740  // Set values of FileInfo struct
741  info.has_video = true;
742  info.file_size = pFormatCtx->pb ? avio_size(pFormatCtx->pb) : -1;
743  info.height = AV_GET_CODEC_ATTRIBUTES(pStream, pCodecCtx)->height;
744  info.width = AV_GET_CODEC_ATTRIBUTES(pStream, pCodecCtx)->width;
745  info.vcodec = pCodecCtx->codec->name;
746  info.video_bit_rate = (pFormatCtx->bit_rate / 8);
747 
748  // Frame rate from the container and codec
749  AVRational framerate = av_guess_frame_rate(pFormatCtx, pStream, NULL);
750  if (!check_fps) {
751  info.fps.num = framerate.num;
752  info.fps.den = framerate.den;
753  }
754 
755  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::UpdateVideoInfo", "info.fps.num", info.fps.num, "info.fps.den", info.fps.den);
756 
757  // TODO: remove excessive debug info in the next releases
758  // The debug info below is just for comparison and troubleshooting on users side during the transition period
759  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::UpdateVideoInfo (pStream->avg_frame_rate)", "num", pStream->avg_frame_rate.num, "den", pStream->avg_frame_rate.den);
760 
761  if (pStream->sample_aspect_ratio.num != 0) {
762  info.pixel_ratio.num = pStream->sample_aspect_ratio.num;
763  info.pixel_ratio.den = pStream->sample_aspect_ratio.den;
764  } else if (AV_GET_CODEC_ATTRIBUTES(pStream, pCodecCtx)->sample_aspect_ratio.num != 0) {
765  info.pixel_ratio.num = AV_GET_CODEC_ATTRIBUTES(pStream, pCodecCtx)->sample_aspect_ratio.num;
766  info.pixel_ratio.den = AV_GET_CODEC_ATTRIBUTES(pStream, pCodecCtx)->sample_aspect_ratio.den;
767  } else {
768  info.pixel_ratio.num = 1;
769  info.pixel_ratio.den = 1;
770  }
771  info.pixel_format = AV_GET_CODEC_PIXEL_FORMAT(pStream, pCodecCtx);
772 
773  // Calculate the DAR (display aspect ratio)
775 
776  // Reduce size fraction
777  size.Reduce();
778 
779  // Set the ratio based on the reduced fraction
780  info.display_ratio.num = size.num;
781  info.display_ratio.den = size.den;
782 
783  // Get scan type and order from codec context/params
784  if (!check_interlace) {
785  check_interlace = true;
786  AVFieldOrder field_order = AV_GET_CODEC_ATTRIBUTES(pStream, pCodecCtx)->field_order;
787  switch(field_order) {
788  case AV_FIELD_PROGRESSIVE:
789  info.interlaced_frame = false;
790  break;
791  case AV_FIELD_TT:
792  case AV_FIELD_TB:
793  info.interlaced_frame = true;
794  info.top_field_first = true;
795  break;
796  case AV_FIELD_BT:
797  case AV_FIELD_BB:
798  info.interlaced_frame = true;
799  info.top_field_first = false;
800  break;
801  case AV_FIELD_UNKNOWN:
802  // Check again later?
803  check_interlace = false;
804  break;
805  }
806  // check_interlace will prevent these checks being repeated,
807  // unless it was cleared because we got an AV_FIELD_UNKNOWN response.
808  }
809 
810  // Set the video timebase
811  info.video_timebase.num = pStream->time_base.num;
812  info.video_timebase.den = pStream->time_base.den;
813 
814  // Set the duration in seconds, and video length (# of frames)
815  info.duration = pStream->duration * info.video_timebase.ToDouble();
816 
817  // Check for valid duration (if found)
818  if (info.duration <= 0.0f && pFormatCtx->duration >= 0) {
819  // Use the format's duration
820  info.duration = float(pFormatCtx->duration) / AV_TIME_BASE;
821  }
822 
823  // Calculate duration from filesize and bitrate (if any)
824  if (info.duration <= 0.0f && info.video_bit_rate > 0 && info.file_size > 0) {
825  // Estimate from bitrate, total bytes, and framerate
827  }
828 
829  // Certain "image" formats do not have a valid duration
830  if (info.duration <= 0.0f && pStream->duration == AV_NOPTS_VALUE && pFormatCtx->duration == AV_NOPTS_VALUE) {
831  // Force an "image" duration
832  info.duration = 60 * 60 * 1; // 1 hour duration
833  info.video_length = 1;
834  info.has_single_image = true;
835  }
836 
837  // Get the # of video frames (if found in stream)
838  // Only set this 1 time (this method can be called multiple times)
839  if (pStream->nb_frames > 0 && info.video_length <= 0) {
840  info.video_length = pStream->nb_frames;
841  }
842 
843  // No duration found in stream of file
844  if (info.duration <= 0.0f) {
845  // No duration is found in the video stream
846  info.duration = -1;
847  info.video_length = -1;
848  is_duration_known = false;
849  } else {
850  // Yes, a duration was found
851  is_duration_known = true;
852 
853  // Calculate number of frames (if not already found in metadata)
854  // Only set this 1 time (this method can be called multiple times)
855  if (info.video_length <= 0) {
857  }
858  }
859 
860  // Add video metadata (if any)
861  AVDictionaryEntry *tag = NULL;
862  while ((tag = av_dict_get(pStream->metadata, "", tag, AV_DICT_IGNORE_SUFFIX))) {
863  QString str_key = tag->key;
864  QString str_value = tag->value;
865  info.metadata[str_key.toStdString()] = str_value.trimmed().toStdString();
866  }
867 }
868 
870  return this->is_duration_known;
871 }
872 
873 std::shared_ptr<Frame> FFmpegReader::GetFrame(int64_t requested_frame) {
874  // Check for open reader (or throw exception)
875  if (!is_open)
876  throw ReaderClosed("The FFmpegReader is closed. Call Open() before calling this method.", path);
877 
878  // Adjust for a requested frame that is too small or too large
879  if (requested_frame < 1)
880  requested_frame = 1;
881  if (requested_frame > info.video_length && is_duration_known)
882  requested_frame = info.video_length;
883  if (info.has_video && info.video_length == 0)
884  // Invalid duration of video file
885  throw InvalidFile("Could not detect the duration of the video or audio stream.", path);
886 
887  // Debug output
888  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetFrame", "requested_frame", requested_frame, "last_frame", last_frame);
889 
890  // Check the cache for this frame
891  std::shared_ptr<Frame> frame = final_cache.GetFrame(requested_frame);
892  if (frame) {
893  // Debug output
894  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetFrame", "returned cached frame", requested_frame);
895 
896  // Return the cached frame
897  return frame;
898  } else {
899 
900  // Prevent async calls to the remainder of this code
901  const std::lock_guard<std::recursive_mutex> lock(getFrameMutex);
902 
903  // Check the cache a 2nd time (due to the potential previous lock)
904  frame = final_cache.GetFrame(requested_frame);
905  if (frame) {
906  // Debug output
907  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetFrame", "returned cached frame on 2nd look", requested_frame);
908 
909  } else {
910  // Frame is not in cache
911  // Reset seek count
912  seek_count = 0;
913 
914  // Are we within X frames of the requested frame?
915  int64_t diff = requested_frame - last_frame;
916  if (diff >= 1 && diff <= 20) {
917  // Continue walking the stream
918  frame = ReadStream(requested_frame);
919  } else {
920  // Greater than 30 frames away, or backwards, we need to seek to the nearest key frame
921  if (enable_seek) {
922  // Only seek if enabled
923  Seek(requested_frame);
924 
925  } else if (!enable_seek && diff < 0) {
926  // Start over, since we can't seek, and the requested frame is smaller than our position
927  // Since we are seeking to frame 1, this actually just closes/re-opens the reader
928  Seek(1);
929  }
930 
931  // Then continue walking the stream
932  frame = ReadStream(requested_frame);
933  }
934  }
935  return frame;
936  }
937 }
938 
939 // Read the stream until we find the requested Frame
940 std::shared_ptr<Frame> FFmpegReader::ReadStream(int64_t requested_frame) {
941  // Allocate video frame
942  bool check_seek = false;
943  int packet_error = -1;
944 
945  // Debug output
946  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ReadStream", "requested_frame", requested_frame, "max_concurrent_frames", max_concurrent_frames);
947 
948  // Loop through the stream until the correct frame is found
949  while (true) {
950  // Check if working frames are 'finished'
951  if (!is_seeking) {
952  // Check for final frames
953  CheckWorkingFrames(requested_frame);
954  }
955 
956  // Check if requested 'final' frame is available (and break out of loop if found)
957  bool is_cache_found = (final_cache.GetFrame(requested_frame) != NULL);
958  if (is_cache_found) {
959  break;
960  }
961 
962  if (!hold_packet || !packet) {
963  // Get the next packet
964  packet_error = GetNextPacket();
965  if (packet_error < 0 && !packet) {
966  // No more packets to be found
967  packet_status.packets_eof = true;
968  }
969  }
970 
971  // Debug output
972  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ReadStream (GetNextPacket)", "requested_frame", requested_frame,"packets_read", packet_status.packets_read(), "packets_decoded", packet_status.packets_decoded(), "is_seeking", is_seeking);
973 
974  // Check the status of a seek (if any)
975  if (is_seeking) {
976  check_seek = CheckSeek(false);
977  } else {
978  check_seek = false;
979  }
980 
981  if (check_seek) {
982  // Packet may become NULL on Close inside Seek if CheckSeek returns false
983  // Jump to the next iteration of this loop
984  continue;
985  }
986 
987  // Video packet
988  if ((info.has_video && packet && packet->stream_index == videoStream) ||
989  (info.has_video && packet_status.video_decoded < packet_status.video_read) ||
990  (info.has_video && !packet && !packet_status.video_eof)) {
991  // Process Video Packet
992  ProcessVideoPacket(requested_frame);
993  }
994  // Audio packet
995  if ((info.has_audio && packet && packet->stream_index == audioStream) ||
996  (info.has_audio && !packet && packet_status.audio_decoded < packet_status.audio_read) ||
997  (info.has_audio && !packet && !packet_status.audio_eof)) {
998  // Process Audio Packet
999  ProcessAudioPacket(requested_frame);
1000  }
1001 
1002  // Remove unused packets (sometimes we purposely ignore video or audio packets,
1003  // if the has_video or has_audio properties are manually overridden)
1004  if ((!info.has_video && packet && packet->stream_index == videoStream) ||
1005  (!info.has_audio && packet && packet->stream_index == audioStream)) {
1006  // Keep track of deleted packet counts
1007  if (packet->stream_index == videoStream) {
1008  packet_status.video_decoded++;
1009  } else if (packet->stream_index == audioStream) {
1010  packet_status.audio_decoded++;
1011  }
1012 
1013  // Remove unused packets (sometimes we purposely ignore video or audio packets,
1014  // if the has_video or has_audio properties are manually overridden)
1015  RemoveAVPacket(packet);
1016  packet = NULL;
1017  }
1018 
1019  // Determine end-of-stream (waiting until final decoder threads finish)
1020  // Force end-of-stream in some situations
1021  packet_status.end_of_file = packet_status.packets_eof && packet_status.video_eof && packet_status.audio_eof;
1022  if ((packet_status.packets_eof && packet_status.packets_read() == packet_status.packets_decoded()) || packet_status.end_of_file) {
1023  // Force EOF (end of file) variables to true, if decoder does not support EOF detection.
1024  // If we have no more packets, and all known packets have been decoded
1025  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ReadStream (force EOF)", "packets_read", packet_status.packets_read(), "packets_decoded", packet_status.packets_decoded(), "packets_eof", packet_status.packets_eof, "video_eof", packet_status.video_eof, "audio_eof", packet_status.audio_eof, "end_of_file", packet_status.end_of_file);
1026  if (!packet_status.video_eof) {
1027  packet_status.video_eof = true;
1028  }
1029  if (!packet_status.audio_eof) {
1030  packet_status.audio_eof = true;
1031  }
1032  packet_status.end_of_file = true;
1033  break;
1034  }
1035  } // end while
1036 
1037  // Debug output
1038  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ReadStream (Completed)",
1039  "packets_read", packet_status.packets_read(),
1040  "packets_decoded", packet_status.packets_decoded(),
1041  "end_of_file", packet_status.end_of_file,
1042  "largest_frame_processed", largest_frame_processed,
1043  "Working Cache Count", working_cache.Count());
1044 
1045  // Have we reached end-of-stream (or the final frame)?
1046  if (!packet_status.end_of_file && requested_frame >= info.video_length) {
1047  // Force end-of-stream
1048  packet_status.end_of_file = true;
1049  }
1050  if (packet_status.end_of_file) {
1051  // Mark any other working frames as 'finished'
1052  CheckWorkingFrames(requested_frame);
1053  }
1054 
1055  // Return requested frame (if found)
1056  std::shared_ptr<Frame> frame = final_cache.GetFrame(requested_frame);
1057  if (frame)
1058  // Return prepared frame
1059  return frame;
1060  else {
1061 
1062  // Check if largest frame is still cached
1063  frame = final_cache.GetFrame(largest_frame_processed);
1064  int samples_in_frame = Frame::GetSamplesPerFrame(requested_frame, info.fps,
1066  if (frame) {
1067  // Copy and return the largest processed frame (assuming it was the last in the video file)
1068  std::shared_ptr<Frame> f = CreateFrame(largest_frame_processed);
1069 
1070  // Use solid color (if no image data found)
1071  if (!frame->has_image_data) {
1072  // Use solid black frame if no image data available
1073  f->AddColor(info.width, info.height, "#000");
1074  }
1075  // Silence audio data (if any), since we are repeating the last frame
1076  frame->AddAudioSilence(samples_in_frame);
1077 
1078  return frame;
1079  } else {
1080  // The largest processed frame is no longer in cache, return a blank frame
1081  std::shared_ptr<Frame> f = CreateFrame(largest_frame_processed);
1082  f->AddColor(info.width, info.height, "#000");
1083  f->AddAudioSilence(samples_in_frame);
1084  return f;
1085  }
1086  }
1087 
1088 }
1089 
1090 // Get the next packet (if any)
1091 int FFmpegReader::GetNextPacket() {
1092  int found_packet = 0;
1093  AVPacket *next_packet;
1094  next_packet = new AVPacket();
1095  found_packet = av_read_frame(pFormatCtx, next_packet);
1096 
1097  if (packet) {
1098  // Remove previous packet before getting next one
1099  RemoveAVPacket(packet);
1100  packet = NULL;
1101  }
1102  if (found_packet >= 0) {
1103  // Update current packet pointer
1104  packet = next_packet;
1105 
1106  // Keep track of packet stats
1107  if (packet->stream_index == videoStream) {
1108  packet_status.video_read++;
1109  } else if (packet->stream_index == audioStream) {
1110  packet_status.audio_read++;
1111  }
1112  } else {
1113  // No more packets found
1114  delete next_packet;
1115  packet = NULL;
1116  }
1117  // Return if packet was found (or error number)
1118  return found_packet;
1119 }
1120 
1121 // Get an AVFrame (if any)
1122 bool FFmpegReader::GetAVFrame() {
1123  int frameFinished = 0;
1124 
1125  // Decode video frame
1126  AVFrame *next_frame = AV_ALLOCATE_FRAME();
1127 
1128 #if IS_FFMPEG_3_2
1129  int send_packet_err = 0;
1130  int64_t send_packet_pts = 0;
1131  if ((packet && packet->stream_index == videoStream && !hold_packet) || !packet) {
1132  send_packet_err = avcodec_send_packet(pCodecCtx, packet);
1133 
1134  if (packet && send_packet_err >= 0) {
1135  send_packet_pts = GetPacketPTS();
1136  hold_packet = false;
1137  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (send packet succeeded)", "send_packet_err", send_packet_err, "send_packet_pts", send_packet_pts);
1138  }
1139  }
1140 
1141  #if USE_HW_ACCEL
1142  // Get the format from the variables set in get_hw_dec_format
1143  hw_de_av_pix_fmt = hw_de_av_pix_fmt_global;
1144  hw_de_av_device_type = hw_de_av_device_type_global;
1145  #endif // USE_HW_ACCEL
1146  if (send_packet_err < 0 && send_packet_err != AVERROR_EOF) {
1147  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (send packet: Not sent [" + av_err2string(send_packet_err) + "])", "send_packet_err", send_packet_err, "send_packet_pts", send_packet_pts);
1148  if (send_packet_err == AVERROR(EAGAIN)) {
1149  hold_packet = true;
1150  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (send packet: AVERROR(EAGAIN): user must read output with avcodec_receive_frame()", "send_packet_pts", send_packet_pts);
1151  }
1152  if (send_packet_err == AVERROR(EINVAL)) {
1153  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (send packet: AVERROR(EINVAL): codec not opened, it is an encoder, or requires flush", "send_packet_pts", send_packet_pts);
1154  }
1155  if (send_packet_err == AVERROR(ENOMEM)) {
1156  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (send packet: AVERROR(ENOMEM): failed to add packet to internal queue, or legitimate decoding errors", "send_packet_pts", send_packet_pts);
1157  }
1158  }
1159 
1160  // Always try and receive a packet, if not EOF.
1161  // Even if the above avcodec_send_packet failed to send,
1162  // we might still need to receive a packet.
1163  int receive_frame_err = 0;
1164  AVFrame *next_frame2;
1165 #if USE_HW_ACCEL
1166  if (hw_de_on && hw_de_supported) {
1167  next_frame2 = AV_ALLOCATE_FRAME();
1168  }
1169  else
1170 #endif // USE_HW_ACCEL
1171  {
1172  next_frame2 = next_frame;
1173  }
1174  pFrame = AV_ALLOCATE_FRAME();
1175  while (receive_frame_err >= 0) {
1176  receive_frame_err = avcodec_receive_frame(pCodecCtx, next_frame2);
1177 
1178  if (receive_frame_err != 0) {
1179  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (receive frame: frame not ready yet from decoder [\" + av_err2string(receive_frame_err) + \"])", "receive_frame_err", receive_frame_err, "send_packet_pts", send_packet_pts);
1180 
1181  if (receive_frame_err == AVERROR_EOF) {
1183  "FFmpegReader::GetAVFrame (receive frame: AVERROR_EOF: EOF detected from decoder, flushing buffers)", "send_packet_pts", send_packet_pts);
1184  avcodec_flush_buffers(pCodecCtx);
1185  packet_status.video_eof = true;
1186  }
1187  if (receive_frame_err == AVERROR(EINVAL)) {
1189  "FFmpegReader::GetAVFrame (receive frame: AVERROR(EINVAL): invalid frame received, flushing buffers)", "send_packet_pts", send_packet_pts);
1190  avcodec_flush_buffers(pCodecCtx);
1191  }
1192  if (receive_frame_err == AVERROR(EAGAIN)) {
1194  "FFmpegReader::GetAVFrame (receive frame: AVERROR(EAGAIN): output is not available in this state - user must try to send new input)", "send_packet_pts", send_packet_pts);
1195  }
1196  if (receive_frame_err == AVERROR_INPUT_CHANGED) {
1198  "FFmpegReader::GetAVFrame (receive frame: AVERROR_INPUT_CHANGED: current decoded frame has changed parameters with respect to first decoded frame)", "send_packet_pts", send_packet_pts);
1199  }
1200 
1201  // Break out of decoding loop
1202  // Nothing ready for decoding yet
1203  break;
1204  }
1205 
1206 #if USE_HW_ACCEL
1207  if (hw_de_on && hw_de_supported) {
1208  int err;
1209  if (next_frame2->format == hw_de_av_pix_fmt) {
1210  next_frame->format = AV_PIX_FMT_YUV420P;
1211  if ((err = av_hwframe_transfer_data(next_frame,next_frame2,0)) < 0) {
1212  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (Failed to transfer data to output frame)", "hw_de_on", hw_de_on);
1213  }
1214  if ((err = av_frame_copy_props(next_frame,next_frame2)) < 0) {
1215  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAVFrame (Failed to copy props to output frame)", "hw_de_on", hw_de_on);
1216  }
1217  }
1218  }
1219  else
1220 #endif // USE_HW_ACCEL
1221  { // No hardware acceleration used -> no copy from GPU memory needed
1222  next_frame = next_frame2;
1223  }
1224 
1225  // TODO also handle possible further frames
1226  // Use only the first frame like avcodec_decode_video2
1227  frameFinished = 1;
1228  packet_status.video_decoded++;
1229 
1230  av_image_alloc(pFrame->data, pFrame->linesize, info.width, info.height, (AVPixelFormat)(pStream->codecpar->format), 1);
1231  av_image_copy(pFrame->data, pFrame->linesize, (const uint8_t**)next_frame->data, next_frame->linesize,
1232  (AVPixelFormat)(pStream->codecpar->format), info.width, info.height);
1233 
1234  // Get display PTS from video frame, often different than packet->pts.
1235  // Sending packets to the decoder (i.e. packet->pts) is async,
1236  // and retrieving packets from the decoder (frame->pts) is async. In most decoders
1237  // sending and retrieving are separated by multiple calls to this method.
1238  if (next_frame->pts != AV_NOPTS_VALUE) {
1239  // This is the current decoded frame (and should be the pts used) for
1240  // processing this data
1241  video_pts = next_frame->pts;
1242  } else if (next_frame->pkt_dts != AV_NOPTS_VALUE) {
1243  // Some videos only set this timestamp (fallback)
1244  video_pts = next_frame->pkt_dts;
1245  }
1246 
1248  "FFmpegReader::GetAVFrame (Successful frame received)", "video_pts", video_pts, "send_packet_pts", send_packet_pts);
1249 
1250  // break out of loop after each successful image returned
1251  break;
1252  }
1253 #if USE_HW_ACCEL
1254  if (hw_de_on && hw_de_supported) {
1255  AV_FREE_FRAME(&next_frame2);
1256  }
1257  #endif // USE_HW_ACCEL
1258 #else
1259  avcodec_decode_video2(pCodecCtx, next_frame, &frameFinished, packet);
1260 
1261  // always allocate pFrame (because we do that in the ffmpeg >= 3.2 as well); it will always be freed later
1262  pFrame = AV_ALLOCATE_FRAME();
1263 
1264  // is frame finished
1265  if (frameFinished) {
1266  // AVFrames are clobbered on the each call to avcodec_decode_video, so we
1267  // must make a copy of the image data before this method is called again.
1268  avpicture_alloc((AVPicture *) pFrame, pCodecCtx->pix_fmt, info.width, info.height);
1269  av_picture_copy((AVPicture *) pFrame, (AVPicture *) next_frame, pCodecCtx->pix_fmt, info.width,
1270  info.height);
1271  }
1272 #endif // IS_FFMPEG_3_2
1273 
1274  // deallocate the frame
1275  AV_FREE_FRAME(&next_frame);
1276 
1277  // Did we get a video frame?
1278  return frameFinished;
1279 }
1280 
1281 // Check the current seek position and determine if we need to seek again
1282 bool FFmpegReader::CheckSeek(bool is_video) {
1283  // Are we seeking for a specific frame?
1284  if (is_seeking) {
1285  // Determine if both an audio and video packet have been decoded since the seek happened.
1286  // If not, allow the ReadStream method to keep looping
1287  if ((is_video_seek && !seek_video_frame_found) || (!is_video_seek && !seek_audio_frame_found))
1288  return false;
1289 
1290  // Check for both streams
1291  if ((info.has_video && !seek_video_frame_found) || (info.has_audio && !seek_audio_frame_found))
1292  return false;
1293 
1294  // Determine max seeked frame
1295  int64_t max_seeked_frame = std::max(seek_audio_frame_found, seek_video_frame_found);
1296 
1297  // determine if we are "before" the requested frame
1298  if (max_seeked_frame >= seeking_frame) {
1299  // SEEKED TOO FAR
1300  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::CheckSeek (Too far, seek again)",
1301  "is_video_seek", is_video_seek,
1302  "max_seeked_frame", max_seeked_frame,
1303  "seeking_frame", seeking_frame,
1304  "seeking_pts", seeking_pts,
1305  "seek_video_frame_found", seek_video_frame_found,
1306  "seek_audio_frame_found", seek_audio_frame_found);
1307 
1308  // Seek again... to the nearest Keyframe
1309  Seek(seeking_frame - (10 * seek_count * seek_count));
1310  } else {
1311  // SEEK WORKED
1312  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::CheckSeek (Successful)",
1313  "is_video_seek", is_video_seek,
1314  "packet->pts", GetPacketPTS(),
1315  "seeking_pts", seeking_pts,
1316  "seeking_frame", seeking_frame,
1317  "seek_video_frame_found", seek_video_frame_found,
1318  "seek_audio_frame_found", seek_audio_frame_found);
1319 
1320  // Seek worked, and we are "before" the requested frame
1321  is_seeking = false;
1322  seeking_frame = 0;
1323  seeking_pts = -1;
1324  }
1325  }
1326 
1327  // return the pts to seek to (if any)
1328  return is_seeking;
1329 }
1330 
1331 // Process a video packet
1332 void FFmpegReader::ProcessVideoPacket(int64_t requested_frame) {
1333  // Get the AVFrame from the current packet
1334  // This sets the video_pts to the correct timestamp
1335  int frame_finished = GetAVFrame();
1336 
1337  // Check if the AVFrame is finished and set it
1338  if (!frame_finished) {
1339  // No AVFrame decoded yet, bail out
1340  return;
1341  }
1342 
1343  // Calculate current frame #
1344  int64_t current_frame = ConvertVideoPTStoFrame(video_pts);
1345 
1346  // Track 1st video packet after a successful seek
1347  if (!seek_video_frame_found && is_seeking)
1348  seek_video_frame_found = current_frame;
1349 
1350  // Create or get the existing frame object. Requested frame needs to be created
1351  // in working_cache at least once. Seek can clear the working_cache, so we must
1352  // add the requested frame back to the working_cache here. If it already exists,
1353  // it will be moved to the top of the working_cache.
1354  working_cache.Add(CreateFrame(requested_frame));
1355 
1356  // Debug output
1357  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessVideoPacket (Before)", "requested_frame", requested_frame, "current_frame", current_frame);
1358 
1359  // Init some things local (for OpenMP)
1360  PixelFormat pix_fmt = AV_GET_CODEC_PIXEL_FORMAT(pStream, pCodecCtx);
1361  int height = info.height;
1362  int width = info.width;
1363  int64_t video_length = info.video_length;
1364  AVFrame *my_frame = pFrame;
1365  pFrame = NULL;
1366 
1367  // Create variables for a RGB Frame (since most videos are not in RGB, we must convert it)
1368  AVFrame *pFrameRGB = nullptr;
1369  uint8_t *buffer = nullptr;
1370 
1371  // Allocate an AVFrame structure
1372  pFrameRGB = AV_ALLOCATE_FRAME();
1373  if (pFrameRGB == nullptr)
1374  throw OutOfMemory("Failed to allocate frame buffer", path);
1375 
1376  // Determine the max size of this source image (based on the timeline's size, the scaling mode,
1377  // and the scaling keyframes). This is a performance improvement, to keep the images as small as possible,
1378  // without losing quality. NOTE: We cannot go smaller than the timeline itself, or the add_layer timeline
1379  // method will scale it back to timeline size before scaling it smaller again. This needs to be fixed in
1380  // the future.
1381  int max_width = info.width;
1382  int max_height = info.height;
1383 
1384  Clip *parent = (Clip *) ParentClip();
1385  if (parent) {
1386  if (parent->ParentTimeline()) {
1387  // Set max width/height based on parent clip's timeline (if attached to a timeline)
1388  max_width = parent->ParentTimeline()->preview_width;
1389  max_height = parent->ParentTimeline()->preview_height;
1390  }
1391  if (parent->scale == SCALE_FIT || parent->scale == SCALE_STRETCH) {
1392  // Best fit or Stretch scaling (based on max timeline size * scaling keyframes)
1393  float max_scale_x = parent->scale_x.GetMaxPoint().co.Y;
1394  float max_scale_y = parent->scale_y.GetMaxPoint().co.Y;
1395  max_width = std::max(float(max_width), max_width * max_scale_x);
1396  max_height = std::max(float(max_height), max_height * max_scale_y);
1397 
1398  } else if (parent->scale == SCALE_CROP) {
1399  // Cropping scale mode (based on max timeline size * cropped size * scaling keyframes)
1400  float max_scale_x = parent->scale_x.GetMaxPoint().co.Y;
1401  float max_scale_y = parent->scale_y.GetMaxPoint().co.Y;
1402  QSize width_size(max_width * max_scale_x,
1403  round(max_width / (float(info.width) / float(info.height))));
1404  QSize height_size(round(max_height / (float(info.height) / float(info.width))),
1405  max_height * max_scale_y);
1406  // respect aspect ratio
1407  if (width_size.width() >= max_width && width_size.height() >= max_height) {
1408  max_width = std::max(max_width, width_size.width());
1409  max_height = std::max(max_height, width_size.height());
1410  } else {
1411  max_width = std::max(max_width, height_size.width());
1412  max_height = std::max(max_height, height_size.height());
1413  }
1414 
1415  } else {
1416  // Scale video to equivalent unscaled size
1417  // Since the preview window can change sizes, we want to always
1418  // scale against the ratio of original video size to timeline size
1419  float preview_ratio = 1.0;
1420  if (parent->ParentTimeline()) {
1421  Timeline *t = (Timeline *) parent->ParentTimeline();
1422  preview_ratio = t->preview_width / float(t->info.width);
1423  }
1424  float max_scale_x = parent->scale_x.GetMaxPoint().co.Y;
1425  float max_scale_y = parent->scale_y.GetMaxPoint().co.Y;
1426  max_width = info.width * max_scale_x * preview_ratio;
1427  max_height = info.height * max_scale_y * preview_ratio;
1428  }
1429  }
1430 
1431  // Determine if image needs to be scaled (for performance reasons)
1432  int original_height = height;
1433  if (max_width != 0 && max_height != 0 && max_width < width && max_height < height) {
1434  // Override width and height (but maintain aspect ratio)
1435  float ratio = float(width) / float(height);
1436  int possible_width = round(max_height * ratio);
1437  int possible_height = round(max_width / ratio);
1438 
1439  if (possible_width <= max_width) {
1440  // use calculated width, and max_height
1441  width = possible_width;
1442  height = max_height;
1443  } else {
1444  // use max_width, and calculated height
1445  width = max_width;
1446  height = possible_height;
1447  }
1448  }
1449 
1450  // Determine required buffer size and allocate buffer
1451  const int bytes_per_pixel = 4;
1452  int buffer_size = (width * height * bytes_per_pixel) + 128;
1453  buffer = new unsigned char[buffer_size]();
1454 
1455  // Copy picture data from one AVFrame (or AVPicture) to another one.
1456  AV_COPY_PICTURE_DATA(pFrameRGB, buffer, PIX_FMT_RGBA, width, height);
1457 
1458  int scale_mode = SWS_FAST_BILINEAR;
1459  if (openshot::Settings::Instance()->HIGH_QUALITY_SCALING) {
1460  scale_mode = SWS_BICUBIC;
1461  }
1462  SwsContext *img_convert_ctx = sws_getContext(info.width, info.height, AV_GET_CODEC_PIXEL_FORMAT(pStream, pCodecCtx), width,
1463  height, PIX_FMT_RGBA, scale_mode, NULL, NULL, NULL);
1464 
1465  // Resize / Convert to RGB
1466  sws_scale(img_convert_ctx, my_frame->data, my_frame->linesize, 0,
1467  original_height, pFrameRGB->data, pFrameRGB->linesize);
1468 
1469  // Create or get the existing frame object
1470  std::shared_ptr<Frame> f = CreateFrame(current_frame);
1471 
1472  // Add Image data to frame
1473  if (!ffmpeg_has_alpha(AV_GET_CODEC_PIXEL_FORMAT(pStream, pCodecCtx))) {
1474  // Add image with no alpha channel, Speed optimization
1475  f->AddImage(width, height, bytes_per_pixel, QImage::Format_RGBA8888_Premultiplied, buffer);
1476  } else {
1477  // Add image with alpha channel (this will be converted to premultipled when needed, but is slower)
1478  f->AddImage(width, height, bytes_per_pixel, QImage::Format_RGBA8888, buffer);
1479  }
1480 
1481  // Update working cache
1482  working_cache.Add(f);
1483 
1484  // Keep track of last last_video_frame
1485  last_video_frame = f;
1486 
1487  // Free the RGB image
1488  AV_FREE_FRAME(&pFrameRGB);
1489 
1490  // Remove frame and packet
1491  RemoveAVFrame(my_frame);
1492  sws_freeContext(img_convert_ctx);
1493 
1494  // Get video PTS in seconds
1495  video_pts_seconds = (double(video_pts) * info.video_timebase.ToDouble()) + pts_offset_seconds;
1496 
1497  // Debug output
1498  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessVideoPacket (After)", "requested_frame", requested_frame, "current_frame", current_frame, "f->number", f->number, "video_pts_seconds", video_pts_seconds);
1499 }
1500 
1501 // Process an audio packet
1502 void FFmpegReader::ProcessAudioPacket(int64_t requested_frame) {
1503  AudioLocation location;
1504  // Calculate location of current audio packet
1505  if (packet && packet->pts != AV_NOPTS_VALUE) {
1506  // Determine related video frame and starting sample # from audio PTS
1507  location = GetAudioPTSLocation(packet->pts);
1508 
1509  // Track 1st audio packet after a successful seek
1510  if (!seek_audio_frame_found && is_seeking)
1511  seek_audio_frame_found = location.frame;
1512  }
1513 
1514  // Create or get the existing frame object. Requested frame needs to be created
1515  // in working_cache at least once. Seek can clear the working_cache, so we must
1516  // add the requested frame back to the working_cache here. If it already exists,
1517  // it will be moved to the top of the working_cache.
1518  working_cache.Add(CreateFrame(requested_frame));
1519 
1520  // Debug output
1521  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (Before)",
1522  "requested_frame", requested_frame,
1523  "target_frame", location.frame,
1524  "starting_sample", location.sample_start);
1525 
1526  // Init an AVFrame to hold the decoded audio samples
1527  int frame_finished = 0;
1528  AVFrame *audio_frame = AV_ALLOCATE_FRAME();
1529  AV_RESET_FRAME(audio_frame);
1530 
1531  int packet_samples = 0;
1532  int data_size = 0;
1533 
1534 #if IS_FFMPEG_3_2
1535  int send_packet_err = avcodec_send_packet(aCodecCtx, packet);
1536  if (send_packet_err < 0 && send_packet_err != AVERROR_EOF) {
1537  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (Packet not sent)");
1538  }
1539  else {
1540  int receive_frame_err = avcodec_receive_frame(aCodecCtx, audio_frame);
1541  if (receive_frame_err >= 0) {
1542  frame_finished = 1;
1543  }
1544  if (receive_frame_err == AVERROR_EOF) {
1545  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (EOF detected from decoder)");
1546  packet_status.audio_eof = true;
1547  }
1548  if (receive_frame_err == AVERROR(EINVAL) || receive_frame_err == AVERROR_EOF) {
1549  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (invalid frame received or EOF from decoder)");
1550  avcodec_flush_buffers(aCodecCtx);
1551  }
1552  if (receive_frame_err != 0) {
1553  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (frame not ready yet from decoder)");
1554  }
1555  }
1556 #else
1557  int used = avcodec_decode_audio4(aCodecCtx, audio_frame, &frame_finished, packet);
1558 #endif
1559 
1560  if (frame_finished) {
1561  packet_status.audio_decoded++;
1562 
1563  // This can be different than the current packet, so we need to look
1564  // at the current AVFrame from the audio decoder. This timestamp should
1565  // be used for the remainder of this function
1566  audio_pts = audio_frame->pts;
1567 
1568  // Determine related video frame and starting sample # from audio PTS
1569  location = GetAudioPTSLocation(audio_pts);
1570 
1571  // determine how many samples were decoded
1572  int plane_size = -1;
1573  data_size = av_samples_get_buffer_size(&plane_size,
1574  AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channels,
1575  audio_frame->nb_samples,
1576  (AVSampleFormat) (AV_GET_SAMPLE_FORMAT(aStream, aCodecCtx)), 1);
1577 
1578  // Calculate total number of samples
1579  packet_samples = audio_frame->nb_samples * AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channels;
1580  }
1581 
1582  // Estimate the # of samples and the end of this packet's location (to prevent GAPS for the next timestamp)
1583  int pts_remaining_samples = packet_samples / info.channels; // Adjust for zero based array
1584 
1585  // Bail if no samples found
1586  if (pts_remaining_samples == 0) {
1587  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (No samples, bailing)",
1588  "packet_samples", packet_samples,
1589  "info.channels", info.channels,
1590  "pts_remaining_samples", pts_remaining_samples);
1591  return;
1592  }
1593 
1594  while (pts_remaining_samples) {
1595  // Get Samples per frame (for this frame number)
1596  int samples_per_frame = Frame::GetSamplesPerFrame(previous_packet_location.frame, info.fps, info.sample_rate, info.channels);
1597 
1598  // Calculate # of samples to add to this frame
1599  int samples = samples_per_frame - previous_packet_location.sample_start;
1600  if (samples > pts_remaining_samples)
1601  samples = pts_remaining_samples;
1602 
1603  // Decrement remaining samples
1604  pts_remaining_samples -= samples;
1605 
1606  if (pts_remaining_samples > 0) {
1607  // next frame
1608  previous_packet_location.frame++;
1609  previous_packet_location.sample_start = 0;
1610  } else {
1611  // Increment sample start
1612  previous_packet_location.sample_start += samples;
1613  }
1614  }
1615 
1616  // Allocate audio buffer
1617  int16_t *audio_buf = new int16_t[AVCODEC_MAX_AUDIO_FRAME_SIZE + MY_INPUT_BUFFER_PADDING_SIZE];
1618 
1619  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (ReSample)",
1620  "packet_samples", packet_samples,
1621  "info.channels", info.channels,
1622  "info.sample_rate", info.sample_rate,
1623  "aCodecCtx->sample_fmt", AV_GET_SAMPLE_FORMAT(aStream, aCodecCtx),
1624  "AV_SAMPLE_FMT_S16", AV_SAMPLE_FMT_S16);
1625 
1626  // Create output frame
1627  AVFrame *audio_converted = AV_ALLOCATE_FRAME();
1628  AV_RESET_FRAME(audio_converted);
1629  audio_converted->nb_samples = audio_frame->nb_samples;
1630  av_samples_alloc(audio_converted->data, audio_converted->linesize, info.channels, audio_frame->nb_samples, AV_SAMPLE_FMT_S16, 0);
1631 
1632  SWRCONTEXT *avr = NULL;
1633  int nb_samples = 0;
1634 
1635  // setup resample context
1636  avr = SWR_ALLOC();
1637  av_opt_set_int(avr, "in_channel_layout", AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channel_layout, 0);
1638  av_opt_set_int(avr, "out_channel_layout", AV_GET_CODEC_ATTRIBUTES(aStream, aCodecCtx)->channel_layout, 0);
1639  av_opt_set_int(avr, "in_sample_fmt", AV_GET_SAMPLE_FORMAT(aStream, aCodecCtx), 0);
1640  av_opt_set_int(avr, "out_sample_fmt", AV_SAMPLE_FMT_S16, 0);
1641  av_opt_set_int(avr, "in_sample_rate", info.sample_rate, 0);
1642  av_opt_set_int(avr, "out_sample_rate", info.sample_rate, 0);
1643  av_opt_set_int(avr, "in_channels", info.channels, 0);
1644  av_opt_set_int(avr, "out_channels", info.channels, 0);
1645  SWR_INIT(avr);
1646 
1647  // Convert audio samples
1648  nb_samples = SWR_CONVERT(avr, // audio resample context
1649  audio_converted->data, // output data pointers
1650  audio_converted->linesize[0], // output plane size, in bytes. (0 if unknown)
1651  audio_converted->nb_samples, // maximum number of samples that the output buffer can hold
1652  audio_frame->data, // input data pointers
1653  audio_frame->linesize[0], // input plane size, in bytes (0 if unknown)
1654  audio_frame->nb_samples); // number of input samples to convert
1655 
1656  // Copy audio samples over original samples
1657  memcpy(audio_buf,
1658  audio_converted->data[0],
1659  static_cast<size_t>(audio_converted->nb_samples)
1660  * av_get_bytes_per_sample(AV_SAMPLE_FMT_S16)
1661  * info.channels);
1662 
1663  // Deallocate resample buffer
1664  SWR_CLOSE(avr);
1665  SWR_FREE(&avr);
1666  avr = NULL;
1667 
1668  // Free AVFrames
1669  av_free(audio_converted->data[0]);
1670  AV_FREE_FRAME(&audio_converted);
1671 
1672  int64_t starting_frame_number = -1;
1673  bool partial_frame = true;
1674  for (int channel_filter = 0; channel_filter < info.channels; channel_filter++) {
1675  // Array of floats (to hold samples for each channel)
1676  starting_frame_number = location.frame;
1677  int channel_buffer_size = packet_samples / info.channels;
1678  float *channel_buffer = new float[channel_buffer_size];
1679 
1680  // Init buffer array
1681  for (int z = 0; z < channel_buffer_size; z++)
1682  channel_buffer[z] = 0.0f;
1683 
1684  // Loop through all samples and add them to our Frame based on channel.
1685  // Toggle through each channel number, since channel data is stored like (left right left right)
1686  int channel = 0;
1687  int position = 0;
1688  for (int sample = 0; sample < packet_samples; sample++) {
1689  // Only add samples for current channel
1690  if (channel_filter == channel) {
1691  // Add sample (convert from (-32768 to 32768) to (-1.0 to 1.0))
1692  channel_buffer[position] = audio_buf[sample] * (1.0f / (1 << 15));
1693 
1694  // Increment audio position
1695  position++;
1696  }
1697 
1698  // increment channel (if needed)
1699  if ((channel + 1) < info.channels)
1700  // move to next channel
1701  channel++;
1702  else
1703  // reset channel
1704  channel = 0;
1705  }
1706 
1707  // Loop through samples, and add them to the correct frames
1708  int start = location.sample_start;
1709  int remaining_samples = channel_buffer_size;
1710  float *iterate_channel_buffer = channel_buffer; // pointer to channel buffer
1711  while (remaining_samples > 0) {
1712  // Get Samples per frame (for this frame number)
1713  int samples_per_frame = Frame::GetSamplesPerFrame(starting_frame_number,
1715 
1716  // Calculate # of samples to add to this frame
1717  int samples = samples_per_frame - start;
1718  if (samples > remaining_samples)
1719  samples = remaining_samples;
1720 
1721  // Create or get the existing frame object
1722  std::shared_ptr<Frame> f = CreateFrame(starting_frame_number);
1723 
1724  // Determine if this frame was "partially" filled in
1725  if (samples_per_frame == start + samples)
1726  partial_frame = false;
1727  else
1728  partial_frame = true;
1729 
1730  // Add samples for current channel to the frame.
1731  f->AddAudio(true, channel_filter, start, iterate_channel_buffer,
1732  samples, 1.0f);
1733 
1734  // Debug output
1735  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (f->AddAudio)",
1736  "frame", starting_frame_number,
1737  "start", start,
1738  "samples", samples,
1739  "channel", channel_filter,
1740  "partial_frame", partial_frame,
1741  "samples_per_frame", samples_per_frame);
1742 
1743  // Add or update cache
1744  working_cache.Add(f);
1745 
1746  // Decrement remaining samples
1747  remaining_samples -= samples;
1748 
1749  // Increment buffer (to next set of samples)
1750  if (remaining_samples > 0)
1751  iterate_channel_buffer += samples;
1752 
1753  // Increment frame number
1754  starting_frame_number++;
1755 
1756  // Reset starting sample #
1757  start = 0;
1758  }
1759 
1760  // clear channel buffer
1761  delete[] channel_buffer;
1762  channel_buffer = NULL;
1763  iterate_channel_buffer = NULL;
1764  }
1765 
1766  // Clean up some arrays
1767  delete[] audio_buf;
1768  audio_buf = NULL;
1769 
1770  // Free audio frame
1771  AV_FREE_FRAME(&audio_frame);
1772 
1773  // Get audio PTS in seconds
1774  audio_pts_seconds = (double(audio_pts) * info.audio_timebase.ToDouble()) + pts_offset_seconds;
1775 
1776  // Debug output
1777  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::ProcessAudioPacket (After)",
1778  "requested_frame", requested_frame,
1779  "starting_frame", location.frame,
1780  "end_frame", starting_frame_number - 1,
1781  "audio_pts_seconds", audio_pts_seconds);
1782 
1783 }
1784 
1785 
1786 // Seek to a specific frame. This is not always frame accurate, it's more of an estimation on many codecs.
1787 void FFmpegReader::Seek(int64_t requested_frame) {
1788  // Adjust for a requested frame that is too small or too large
1789  if (requested_frame < 1)
1790  requested_frame = 1;
1791  if (requested_frame > info.video_length)
1792  requested_frame = info.video_length;
1793  if (requested_frame > largest_frame_processed && packet_status.end_of_file) {
1794  // Not possible to search past largest_frame once EOF is reached (no more packets)
1795  return;
1796  }
1797 
1798  // Debug output
1799  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::Seek",
1800  "requested_frame", requested_frame,
1801  "seek_count", seek_count,
1802  "last_frame", last_frame);
1803 
1804  // Clear working cache (since we are seeking to another location in the file)
1805  working_cache.Clear();
1806 
1807  // Reset the last frame variable
1808  video_pts = 0.0;
1809  video_pts_seconds = NO_PTS_OFFSET;
1810  audio_pts = 0.0;
1811  audio_pts_seconds = NO_PTS_OFFSET;
1812  hold_packet = false;
1813  last_frame = 0;
1814  current_video_frame = 0;
1815  largest_frame_processed = 0;
1816  bool has_audio_override = info.has_audio;
1817  bool has_video_override = info.has_video;
1818 
1819  // Init end-of-file detection variables
1820  packet_status.reset(false);
1821 
1822  // Increment seek count
1823  seek_count++;
1824 
1825  // If seeking near frame 1, we need to close and re-open the file (this is more reliable than seeking)
1826  int buffer_amount = std::max(max_concurrent_frames, 8);
1827  if (requested_frame - buffer_amount < 20) {
1828  // prevent Open() from seeking again
1829  is_seeking = true;
1830 
1831  // Close and re-open file (basically seeking to frame 1)
1832  Close();
1833  Open();
1834 
1835  // Update overrides (since closing and re-opening might update these)
1836  info.has_audio = has_audio_override;
1837  info.has_video = has_video_override;
1838 
1839  // Not actually seeking, so clear these flags
1840  is_seeking = false;
1841  if (seek_count == 1) {
1842  // Don't redefine this on multiple seek attempts for a specific frame
1843  seeking_frame = 1;
1844  seeking_pts = ConvertFrameToVideoPTS(1);
1845  }
1846  seek_audio_frame_found = 0; // used to detect which frames to throw away after a seek
1847  seek_video_frame_found = 0; // used to detect which frames to throw away after a seek
1848 
1849  } else {
1850  // Seek to nearest key-frame (aka, i-frame)
1851  bool seek_worked = false;
1852  int64_t seek_target = 0;
1853 
1854  // Seek video stream (if any), except album arts
1855  if (!seek_worked && info.has_video && !HasAlbumArt()) {
1856  seek_target = ConvertFrameToVideoPTS(requested_frame - buffer_amount);
1857  if (av_seek_frame(pFormatCtx, info.video_stream_index, seek_target, AVSEEK_FLAG_BACKWARD) < 0) {
1858  fprintf(stderr, "%s: error while seeking video stream\n", pFormatCtx->AV_FILENAME);
1859  } else {
1860  // VIDEO SEEK
1861  is_video_seek = true;
1862  seek_worked = true;
1863  }
1864  }
1865 
1866  // Seek audio stream (if not already seeked... and if an audio stream is found)
1867  if (!seek_worked && info.has_audio) {
1868  seek_target = ConvertFrameToAudioPTS(requested_frame - buffer_amount);
1869  if (av_seek_frame(pFormatCtx, info.audio_stream_index, seek_target, AVSEEK_FLAG_BACKWARD) < 0) {
1870  fprintf(stderr, "%s: error while seeking audio stream\n", pFormatCtx->AV_FILENAME);
1871  } else {
1872  // AUDIO SEEK
1873  is_video_seek = false;
1874  seek_worked = true;
1875  }
1876  }
1877 
1878  // Was the seek successful?
1879  if (seek_worked) {
1880  // Flush audio buffer
1881  if (info.has_audio)
1882  avcodec_flush_buffers(aCodecCtx);
1883 
1884  // Flush video buffer
1885  if (info.has_video)
1886  avcodec_flush_buffers(pCodecCtx);
1887 
1888  // Reset previous audio location to zero
1889  previous_packet_location.frame = -1;
1890  previous_packet_location.sample_start = 0;
1891 
1892  // init seek flags
1893  is_seeking = true;
1894  if (seek_count == 1) {
1895  // Don't redefine this on multiple seek attempts for a specific frame
1896  seeking_pts = seek_target;
1897  seeking_frame = requested_frame;
1898  }
1899  seek_audio_frame_found = 0; // used to detect which frames to throw away after a seek
1900  seek_video_frame_found = 0; // used to detect which frames to throw away after a seek
1901 
1902  } else {
1903  // seek failed
1904  seeking_pts = 0;
1905  seeking_frame = 0;
1906 
1907  // prevent Open() from seeking again
1908  is_seeking = true;
1909 
1910  // Close and re-open file (basically seeking to frame 1)
1911  Close();
1912  Open();
1913 
1914  // Not actually seeking, so clear these flags
1915  is_seeking = false;
1916 
1917  // disable seeking for this reader (since it failed)
1918  enable_seek = false;
1919 
1920  // Update overrides (since closing and re-opening might update these)
1921  info.has_audio = has_audio_override;
1922  info.has_video = has_video_override;
1923  }
1924  }
1925 }
1926 
1927 // Get the PTS for the current video packet
1928 int64_t FFmpegReader::GetPacketPTS() {
1929  if (packet) {
1930  int64_t current_pts = packet->pts;
1931  if (current_pts == AV_NOPTS_VALUE && packet->dts != AV_NOPTS_VALUE)
1932  current_pts = packet->dts;
1933 
1934  // Return adjusted PTS
1935  return current_pts;
1936  } else {
1937  // No packet, return NO PTS
1938  return AV_NOPTS_VALUE;
1939  }
1940 }
1941 
1942 // Update PTS Offset (if any)
1943 void FFmpegReader::UpdatePTSOffset() {
1944  if (pts_offset_seconds != NO_PTS_OFFSET) {
1945  // Skip this method if we have already set PTS offset
1946  return;
1947  }
1948  pts_offset_seconds = 0.0;
1949  double video_pts_offset_seconds = 0.0;
1950  double audio_pts_offset_seconds = 0.0;
1951 
1952  bool has_video_pts = false;
1953  if (!info.has_video) {
1954  // Mark as checked
1955  has_video_pts = true;
1956  }
1957  bool has_audio_pts = false;
1958  if (!info.has_audio) {
1959  // Mark as checked
1960  has_audio_pts = true;
1961  }
1962 
1963  // Loop through the stream (until a packet from all streams is found)
1964  while (!has_video_pts || !has_audio_pts) {
1965  // Get the next packet (if any)
1966  if (GetNextPacket() < 0)
1967  // Break loop when no more packets found
1968  break;
1969 
1970  // Get PTS of this packet
1971  int64_t pts = GetPacketPTS();
1972 
1973  // Video packet
1974  if (!has_video_pts && packet->stream_index == videoStream) {
1975  // Get the video packet start time (in seconds)
1976  video_pts_offset_seconds = 0.0 - (video_pts * info.video_timebase.ToDouble());
1977 
1978  // Is timestamp close to zero (within X seconds)
1979  // Ignore wildly invalid timestamps (i.e. -234923423423)
1980  if (std::abs(video_pts_offset_seconds) <= 10.0) {
1981  has_video_pts = true;
1982  }
1983  }
1984  else if (!has_audio_pts && packet->stream_index == audioStream) {
1985  // Get the audio packet start time (in seconds)
1986  audio_pts_offset_seconds = 0.0 - (pts * info.audio_timebase.ToDouble());
1987 
1988  // Is timestamp close to zero (within X seconds)
1989  // Ignore wildly invalid timestamps (i.e. -234923423423)
1990  if (std::abs(audio_pts_offset_seconds) <= 10.0) {
1991  has_audio_pts = true;
1992  }
1993  }
1994  }
1995 
1996  // Do we have all valid timestamps to determine PTS offset?
1997  if (has_video_pts && has_audio_pts) {
1998  // Set PTS Offset to the smallest offset
1999  // [ video timestamp ]
2000  // [ audio timestamp ]
2001  //
2002  // ** SHIFT TIMESTAMPS TO ZERO **
2003  //
2004  //[ video timestamp ]
2005  // [ audio timestamp ]
2006  //
2007  // Since all offsets are negative at this point, we want the max value, which
2008  // represents the closest to zero
2009  pts_offset_seconds = std::max(video_pts_offset_seconds, audio_pts_offset_seconds);
2010  }
2011 }
2012 
2013 // Convert PTS into Frame Number
2014 int64_t FFmpegReader::ConvertVideoPTStoFrame(int64_t pts) {
2015  // Apply PTS offset
2016  int64_t previous_video_frame = current_video_frame;
2017 
2018  // Get the video packet start time (in seconds)
2019  double video_seconds = (double(pts) * info.video_timebase.ToDouble()) + pts_offset_seconds;
2020 
2021  // Divide by the video timebase, to get the video frame number (frame # is decimal at this point)
2022  int64_t frame = round(video_seconds * info.fps.ToDouble()) + 1;
2023 
2024  // Keep track of the expected video frame #
2025  if (current_video_frame == 0)
2026  current_video_frame = frame;
2027  else {
2028 
2029  // Sometimes frames are duplicated due to identical (or similar) timestamps
2030  if (frame == previous_video_frame) {
2031  // return -1 frame number
2032  frame = -1;
2033  } else {
2034  // Increment expected frame
2035  current_video_frame++;
2036  }
2037  }
2038 
2039  // Return frame #
2040  return frame;
2041 }
2042 
2043 // Convert Frame Number into Video PTS
2044 int64_t FFmpegReader::ConvertFrameToVideoPTS(int64_t frame_number) {
2045  // Get timestamp of this frame (in seconds)
2046  double seconds = (double(frame_number - 1) / info.fps.ToDouble()) + pts_offset_seconds;
2047 
2048  // Calculate the # of video packets in this timestamp
2049  int64_t video_pts = round(seconds / info.video_timebase.ToDouble());
2050 
2051  // Apply PTS offset (opposite)
2052  return video_pts;
2053 }
2054 
2055 // Convert Frame Number into Video PTS
2056 int64_t FFmpegReader::ConvertFrameToAudioPTS(int64_t frame_number) {
2057  // Get timestamp of this frame (in seconds)
2058  double seconds = (double(frame_number - 1) / info.fps.ToDouble()) + pts_offset_seconds;
2059 
2060  // Calculate the # of audio packets in this timestamp
2061  int64_t audio_pts = round(seconds / info.audio_timebase.ToDouble());
2062 
2063  // Apply PTS offset (opposite)
2064  return audio_pts;
2065 }
2066 
2067 // Calculate Starting video frame and sample # for an audio PTS
2068 AudioLocation FFmpegReader::GetAudioPTSLocation(int64_t pts) {
2069  // Get the audio packet start time (in seconds)
2070  double audio_seconds = (double(pts) * info.audio_timebase.ToDouble()) + pts_offset_seconds;
2071 
2072  // Divide by the video timebase, to get the video frame number (frame # is decimal at this point)
2073  double frame = (audio_seconds * info.fps.ToDouble()) + 1;
2074 
2075  // Frame # as a whole number (no more decimals)
2076  int64_t whole_frame = int64_t(frame);
2077 
2078  // Remove the whole number, and only get the decimal of the frame
2079  double sample_start_percentage = frame - double(whole_frame);
2080 
2081  // Get Samples per frame
2082  int samples_per_frame = Frame::GetSamplesPerFrame(whole_frame, info.fps, info.sample_rate, info.channels);
2083 
2084  // Calculate the sample # to start on
2085  int sample_start = round(double(samples_per_frame) * sample_start_percentage);
2086 
2087  // Protect against broken (i.e. negative) timestamps
2088  if (whole_frame < 1)
2089  whole_frame = 1;
2090  if (sample_start < 0)
2091  sample_start = 0;
2092 
2093  // Prepare final audio packet location
2094  AudioLocation location = {whole_frame, sample_start};
2095 
2096  // Compare to previous audio packet (and fix small gaps due to varying PTS timestamps)
2097  if (previous_packet_location.frame != -1) {
2098  if (location.is_near(previous_packet_location, samples_per_frame, samples_per_frame)) {
2099  int64_t orig_frame = location.frame;
2100  int orig_start = location.sample_start;
2101 
2102  // Update sample start, to prevent gaps in audio
2103  location.sample_start = previous_packet_location.sample_start;
2104  location.frame = previous_packet_location.frame;
2105 
2106  // Debug output
2107  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAudioPTSLocation (Audio Gap Detected)", "Source Frame", orig_frame, "Source Audio Sample", orig_start, "Target Frame", location.frame, "Target Audio Sample", location.sample_start, "pts", pts);
2108 
2109  } else {
2110  // Debug output
2111  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::GetAudioPTSLocation (Audio Gap Ignored - too big)", "Previous location frame", previous_packet_location.frame, "Target Frame", location.frame, "Target Audio Sample", location.sample_start, "pts", pts);
2112  }
2113  }
2114 
2115  // Set previous location
2116  previous_packet_location = location;
2117 
2118  // Return the associated video frame and starting sample #
2119  return location;
2120 }
2121 
2122 // Create a new Frame (or return an existing one) and add it to the working queue.
2123 std::shared_ptr<Frame> FFmpegReader::CreateFrame(int64_t requested_frame) {
2124  // Check working cache
2125  std::shared_ptr<Frame> output = working_cache.GetFrame(requested_frame);
2126 
2127  if (!output) {
2128  // (re-)Check working cache
2129  output = working_cache.GetFrame(requested_frame);
2130  if(output) return output;
2131 
2132  // Create a new frame on the working cache
2133  output = std::make_shared<Frame>(requested_frame, info.width, info.height, "#000000", Frame::GetSamplesPerFrame(requested_frame, info.fps, info.sample_rate, info.channels), info.channels);
2134  output->SetPixelRatio(info.pixel_ratio.num, info.pixel_ratio.den); // update pixel ratio
2135  output->ChannelsLayout(info.channel_layout); // update audio channel layout from the parent reader
2136  output->SampleRate(info.sample_rate); // update the frame's sample rate of the parent reader
2137 
2138  working_cache.Add(output);
2139 
2140  // Set the largest processed frame (if this is larger)
2141  if (requested_frame > largest_frame_processed)
2142  largest_frame_processed = requested_frame;
2143  }
2144  // Return frame
2145  return output;
2146 }
2147 
2148 // Determine if frame is partial due to seek
2149 bool FFmpegReader::IsPartialFrame(int64_t requested_frame) {
2150 
2151  // Sometimes a seek gets partial frames, and we need to remove them
2152  bool seek_trash = false;
2153  int64_t max_seeked_frame = seek_audio_frame_found; // determine max seeked frame
2154  if (seek_video_frame_found > max_seeked_frame) {
2155  max_seeked_frame = seek_video_frame_found;
2156  }
2157  if ((info.has_audio && seek_audio_frame_found && max_seeked_frame >= requested_frame) ||
2158  (info.has_video && seek_video_frame_found && max_seeked_frame >= requested_frame)) {
2159  seek_trash = true;
2160  }
2161 
2162  return seek_trash;
2163 }
2164 
2165 // Check the working queue, and move finished frames to the finished queue
2166 void FFmpegReader::CheckWorkingFrames(int64_t requested_frame) {
2167 
2168  // Prevent async calls to the following code
2169  const std::lock_guard<std::recursive_mutex> lock(getFrameMutex);
2170 
2171  // Get a list of current working queue frames in the cache (in-progress frames)
2172  std::vector<std::shared_ptr<openshot::Frame>> working_frames = working_cache.GetFrames();
2173  std::vector<std::shared_ptr<openshot::Frame>>::iterator working_itr;
2174 
2175  // Loop through all working queue frames (sorted by frame #)
2176  for(working_itr = working_frames.begin(); working_itr != working_frames.end(); ++working_itr)
2177  {
2178  // Get working frame
2179  std::shared_ptr<Frame> f = *working_itr;
2180 
2181  // Was a frame found? Is frame requested yet?
2182  if (!f || f->number > requested_frame) {
2183  // If not, skip to next one
2184  continue;
2185  }
2186 
2187  // Calculate PTS in seconds (of working frame), and the most recent processed pts value
2188  double frame_pts_seconds = (double(f->number - 1) / info.fps.ToDouble()) + pts_offset_seconds;
2189  double recent_pts_seconds = std::max(video_pts_seconds, audio_pts_seconds);
2190 
2191  // Determine if video and audio are ready (based on timestamps)
2192  bool is_video_ready = false;
2193  bool is_audio_ready = false;
2194  double recent_pts_diff = recent_pts_seconds - frame_pts_seconds;
2195  if ((frame_pts_seconds <= video_pts_seconds)
2196  || (recent_pts_diff > 1.5)
2197  || packet_status.video_eof || packet_status.end_of_file) {
2198  // Video stream is past this frame (so it must be done)
2199  // OR video stream is too far behind, missing, or end-of-file
2200  is_video_ready = true;
2201  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::CheckWorkingFrames (video ready)",
2202  "frame_number", f->number,
2203  "frame_pts_seconds", frame_pts_seconds,
2204  "video_pts_seconds", video_pts_seconds,
2205  "recent_pts_diff", recent_pts_diff);
2206  if (info.has_video && !f->has_image_data) {
2207  // Frame has no image data (copy from previous frame)
2208  // Loop backwards through final frames (looking for the nearest, previous frame image)
2209  for (int64_t previous_frame = requested_frame - 1; previous_frame > 0; previous_frame--) {
2210  std::shared_ptr<Frame> previous_frame_instance = final_cache.GetFrame(previous_frame);
2211  if (previous_frame_instance && previous_frame_instance->has_image_data) {
2212  // Copy image from last decoded frame
2213  f->AddImage(std::make_shared<QImage>(previous_frame_instance->GetImage()->copy()));
2214  break;
2215  }
2216  }
2217 
2218  if (last_video_frame && !f->has_image_data) {
2219  // Copy image from last decoded frame
2220  f->AddImage(std::make_shared<QImage>(last_video_frame->GetImage()->copy()));
2221  } else if (!f->has_image_data) {
2222  f->AddColor("#000000");
2223  }
2224  }
2225  }
2226 
2227  double audio_pts_diff = audio_pts_seconds - frame_pts_seconds;
2228  if ((frame_pts_seconds < audio_pts_seconds && audio_pts_diff > 1.0)
2229  || (recent_pts_diff > 1.5)
2230  || packet_status.audio_eof || packet_status.end_of_file) {
2231  // Audio stream is past this frame (so it must be done)
2232  // OR audio stream is too far behind, missing, or end-of-file
2233  // Adding a bit of margin here, to allow for partial audio packets
2234  is_audio_ready = true;
2235  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::CheckWorkingFrames (audio ready)",
2236  "frame_number", f->number,
2237  "frame_pts_seconds", frame_pts_seconds,
2238  "audio_pts_seconds", audio_pts_seconds,
2239  "audio_pts_diff", audio_pts_diff,
2240  "recent_pts_diff", recent_pts_diff);
2241  }
2242  bool is_seek_trash = IsPartialFrame(f->number);
2243 
2244  // Adjust for available streams
2245  if (!info.has_video) is_video_ready = true;
2246  if (!info.has_audio) is_audio_ready = true;
2247 
2248  // Debug output
2249  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::CheckWorkingFrames",
2250  "frame_number", f->number,
2251  "is_video_ready", is_video_ready,
2252  "is_audio_ready", is_audio_ready,
2253  "video_eof", packet_status.video_eof,
2254  "audio_eof", packet_status.audio_eof,
2255  "end_of_file", packet_status.end_of_file);
2256 
2257  // Check if working frame is final
2258  if ((!packet_status.end_of_file && is_video_ready && is_audio_ready) || packet_status.end_of_file || is_seek_trash) {
2259  // Debug output
2260  ZmqLogger::Instance()->AppendDebugMethod("FFmpegReader::CheckWorkingFrames (mark frame as final)",
2261  "requested_frame", requested_frame,
2262  "f->number", f->number,
2263  "is_seek_trash", is_seek_trash,
2264  "Working Cache Count", working_cache.Count(),
2265  "Final Cache Count", final_cache.Count(),
2266  "end_of_file", packet_status.end_of_file);
2267 
2268  if (!is_seek_trash) {
2269  // Move frame to final cache
2270  final_cache.Add(f);
2271 
2272  // Remove frame from working cache
2273  working_cache.Remove(f->number);
2274 
2275  // Update last frame processed
2276  last_frame = f->number;
2277  } else {
2278  // Seek trash, so delete the frame from the working cache, and never add it to the final cache.
2279  working_cache.Remove(f->number);
2280  }
2281 
2282  }
2283  }
2284 
2285  // Clear vector of frames
2286  working_frames.clear();
2287  working_frames.shrink_to_fit();
2288 }
2289 
2290 // Check for the correct frames per second (FPS) value by scanning the 1st few seconds of video packets.
2291 void FFmpegReader::CheckFPS() {
2292  if (check_fps) {
2293  // Do not check FPS more than 1 time
2294  return;
2295  } else {
2296  check_fps = true;
2297  }
2298 
2299  int frames_per_second[3] = {0,0,0};
2300  int max_fps_index = sizeof(frames_per_second) / sizeof(frames_per_second[0]);
2301  int fps_index = 0;
2302 
2303  int all_frames_detected = 0;
2304  int starting_frames_detected = 0;
2305 
2306  // Loop through the stream
2307  while (true) {
2308  // Get the next packet (if any)
2309  if (GetNextPacket() < 0)
2310  // Break loop when no more packets found
2311  break;
2312 
2313  // Video packet
2314  if (packet->stream_index == videoStream) {
2315  // Get the video packet start time (in seconds)
2316  double video_seconds = (double(GetPacketPTS()) * info.video_timebase.ToDouble()) + pts_offset_seconds;
2317  fps_index = int(video_seconds); // truncate float timestamp to int (second 1, second 2, second 3)
2318 
2319  // Is this video packet from the first few seconds?
2320  if (fps_index >= 0 && fps_index < max_fps_index) {
2321  // Yes, keep track of how many frames per second (over the first few seconds)
2322  starting_frames_detected++;
2323  frames_per_second[fps_index]++;
2324  }
2325 
2326  // Track all video packets detected
2327  all_frames_detected++;
2328  }
2329  }
2330 
2331  // Calculate FPS (based on the first few seconds of video packets)
2332  float avg_fps = 30.0;
2333  if (starting_frames_detected > 0 && fps_index > 0) {
2334  avg_fps = float(starting_frames_detected) / std::min(fps_index, max_fps_index);
2335  }
2336 
2337  // Verify average FPS is a reasonable value
2338  if (avg_fps < 8.0) {
2339  // Invalid FPS assumed, so switching to a sane default FPS instead
2340  avg_fps = 30.0;
2341  }
2342 
2343  // Update FPS (truncate average FPS to Integer)
2344  info.fps = Fraction(int(avg_fps), 1);
2345 
2346  // Update Duration and Length
2347  if (all_frames_detected > 0) {
2348  // Use all video frames detected to calculate # of frames
2349  info.video_length = all_frames_detected;
2350  info.duration = all_frames_detected / avg_fps;
2351  } else {
2352  // Use previous duration to calculate # of frames
2353  info.video_length = info.duration * avg_fps;
2354  }
2355 
2356  // Update video bit rate
2358 }
2359 
2360 // Remove AVFrame from cache (and deallocate its memory)
2361 void FFmpegReader::RemoveAVFrame(AVFrame *remove_frame) {
2362  // Remove pFrame (if exists)
2363  if (remove_frame) {
2364  // Free memory
2365  av_freep(&remove_frame->data[0]);
2366 #ifndef WIN32
2367  AV_FREE_FRAME(&remove_frame);
2368 #endif
2369  }
2370 }
2371 
2372 // Remove AVPacket from cache (and deallocate its memory)
2373 void FFmpegReader::RemoveAVPacket(AVPacket *remove_packet) {
2374  // deallocate memory for packet
2375  AV_FREE_PACKET(remove_packet);
2376 
2377  // Delete the object
2378  delete remove_packet;
2379 }
2380 
2381 // Generate JSON string of this object
2382 std::string FFmpegReader::Json() const {
2383 
2384  // Return formatted string
2385  return JsonValue().toStyledString();
2386 }
2387 
2388 // Generate Json::Value for this object
2389 Json::Value FFmpegReader::JsonValue() const {
2390 
2391  // Create root json object
2392  Json::Value root = ReaderBase::JsonValue(); // get parent properties
2393  root["type"] = "FFmpegReader";
2394  root["path"] = path;
2395 
2396  // return JsonValue
2397  return root;
2398 }
2399 
2400 // Load JSON string into this object
2401 void FFmpegReader::SetJson(const std::string value) {
2402 
2403  // Parse JSON string into JSON objects
2404  try {
2405  const Json::Value root = openshot::stringToJson(value);
2406  // Set all values that match
2407  SetJsonValue(root);
2408  }
2409  catch (const std::exception& e) {
2410  // Error parsing JSON (or missing keys)
2411  throw InvalidJSON("JSON is invalid (missing keys or invalid data types)");
2412  }
2413 }
2414 
2415 // Load Json::Value into this object
2416 void FFmpegReader::SetJsonValue(const Json::Value root) {
2417 
2418  // Set parent data
2420 
2421  // Set data from Json (if key is found)
2422  if (!root["path"].isNull())
2423  path = root["path"].asString();
2424 
2425  // Re-Open path, and re-init everything (if needed)
2426  if (is_open) {
2427  Close();
2428  Open();
2429  }
2430 }
Header file for all Exception classes.
AVPixelFormat hw_de_av_pix_fmt_global
AVHWDeviceType hw_de_av_device_type_global
int hw_de_on
Header file for FFmpegReader class.
Header file for FFmpegUtilities.
#define AV_FREE_CONTEXT(av_context)
#define SWR_INIT(ctx)
#define AV_FREE_FRAME(av_frame)
#define SWR_CONVERT(ctx, out, linesize, out_count, in, linesize2, in_count)
#define SWR_ALLOC()
#define SWR_CLOSE(ctx)
#define AV_GET_CODEC_TYPE(av_stream)
#define PixelFormat
#define AV_GET_CODEC_PIXEL_FORMAT(av_stream, av_context)
#define AV_GET_CODEC_CONTEXT(av_stream, av_codec)
#define AV_FIND_DECODER_CODEC_ID(av_stream)
#define AV_ALLOCATE_FRAME()
#define AV_REGISTER_ALL
#define PIX_FMT_RGBA
#define SWR_FREE(ctx)
#define AV_COPY_PICTURE_DATA(av_frame, buffer, pix_fmt, width, height)
#define AV_FREE_PACKET(av_packet)
#define SWRCONTEXT
#define AVCODEC_REGISTER_ALL
#define AVCODEC_MAX_AUDIO_FRAME_SIZE
#define AV_GET_CODEC_ATTRIBUTES(av_stream, av_context)
#define MY_INPUT_BUFFER_PADDING_SIZE
#define AV_GET_SAMPLE_FORMAT(av_stream, av_context)
#define AV_RESET_FRAME(av_frame)
AVDictionary * opts
if(!codec) codec
#define FF_NUM_PROCESSORS
#define OPEN_MP_NUM_PROCESSORS
Header file for Timeline class.
Header file for ZeroMQ-based Logger class.
void SetMaxBytesFromInfo(int64_t number_of_frames, int width, int height, int sample_rate, int channels)
Set maximum bytes to a different amount based on a ReaderInfo struct.
Definition: CacheBase.cpp:30
int64_t Count()
Count the frames in the queue.
void Add(std::shared_ptr< openshot::Frame > frame)
Add a Frame to the cache.
Definition: CacheMemory.cpp:46
std::shared_ptr< openshot::Frame > GetFrame(int64_t frame_number)
Get a frame from the cache.
Definition: CacheMemory.cpp:80
std::vector< std::shared_ptr< openshot::Frame > > GetFrames()
Get an array of all Frames.
Definition: CacheMemory.cpp:96
void Remove(int64_t frame_number)
Remove a specific frame.
void Clear()
Clear the cache of all frames.
openshot::TimelineBase * ParentTimeline()
Get the associated Timeline pointer (if any)
Definition: ClipBase.h:91
This class represents a clip (used to arrange readers on the timeline)
Definition: Clip.h:90
openshot::Keyframe scale_x
Curve representing the horizontal scaling in percent (0 to 1)
Definition: Clip.h:286
openshot::Keyframe scale_y
Curve representing the vertical scaling in percent (0 to 1)
Definition: Clip.h:287
openshot::ScaleType scale
The scale determines how a clip should be resized to fit its parent.
Definition: Clip.h:156
double Y
The Y value of the coordinate (usually representing the value of the property being animated)
Definition: Coordinate.h:41
void Open() override
Open File - which is called by the constructor automatically.
FFmpegReader(const std::string &path, bool inspect_reader=true)
Constructor for FFmpegReader.
Json::Value JsonValue() const override
Generate Json::Value for this object.
bool GetIsDurationKnown()
Return true if frame can be read with GetFrame()
void SetJsonValue(const Json::Value root) override
Load Json::Value into this object.
CacheMemory final_cache
Final cache object used to hold final frames.
Definition: FFmpegReader.h:240
virtual ~FFmpegReader()
Destructor.
std::string Json() const override
Generate JSON string of this object.
std::shared_ptr< openshot::Frame > GetFrame(int64_t requested_frame) override
void Close() override
Close File.
void SetJson(const std::string value) override
Load JSON string into this object.
This class represents a fraction.
Definition: Fraction.h:30
int num
Numerator for the fraction.
Definition: Fraction.h:32
float ToFloat()
Return this fraction as a float (i.e. 1/2 = 0.5)
Definition: Fraction.cpp:35
double ToDouble() const
Return this fraction as a double (i.e. 1/2 = 0.5)
Definition: Fraction.cpp:40
int den
Denominator for the fraction.
Definition: Fraction.h:33
int GetSamplesPerFrame(openshot::Fraction fps, int sample_rate, int channels)
Calculate the # of samples per video frame (for the current frame number)
Definition: Frame.cpp:534
Exception when no valid codec is found for a file.
Definition: Exceptions.h:173
Exception for files that can not be found or opened.
Definition: Exceptions.h:188
Exception for invalid JSON.
Definition: Exceptions.h:218
Point GetMaxPoint() const
Get max point (by Y coordinate)
Definition: KeyFrame.cpp:245
Exception when no streams are found in the file.
Definition: Exceptions.h:286
Exception when memory could not be allocated.
Definition: Exceptions.h:349
Coordinate co
This is the primary coordinate.
Definition: Point.h:66
openshot::ReaderInfo info
Information about the current media file.
Definition: ReaderBase.h:88
virtual void SetJsonValue(const Json::Value root)=0
Load Json::Value into this object.
Definition: ReaderBase.cpp:162
virtual Json::Value JsonValue() const =0
Generate Json::Value for this object.
Definition: ReaderBase.cpp:107
std::recursive_mutex getFrameMutex
Mutex for multiple threads.
Definition: ReaderBase.h:79
openshot::ClipBase * ParentClip()
Parent clip object of this reader (which can be unparented and NULL)
Definition: ReaderBase.cpp:245
Exception when a reader is closed, and a frame is requested.
Definition: Exceptions.h:364
int DE_LIMIT_WIDTH_MAX
Maximum columns that hardware decode can handle.
Definition: Settings.h:77
int HW_DE_DEVICE_SET
Which GPU to use to decode (0 is the first)
Definition: Settings.h:80
int DE_LIMIT_HEIGHT_MAX
Maximum rows that hardware decode can handle.
Definition: Settings.h:74
static Settings * Instance()
Create or get an instance of this logger singleton (invoke the class with this method)
Definition: Settings.cpp:23
int HARDWARE_DECODER
Use video codec for faster video decoding (if supported)
Definition: Settings.h:62
int preview_height
Optional preview width of timeline image. If your preview window is smaller than the timeline,...
Definition: TimelineBase.h:44
int preview_width
Optional preview width of timeline image. If your preview window is smaller than the timeline,...
Definition: TimelineBase.h:43
This class represents a timeline.
Definition: Timeline.h:150
void AppendDebugMethod(std::string method_name, std::string arg1_name="", float arg1_value=-1.0, std::string arg2_name="", float arg2_value=-1.0, std::string arg3_name="", float arg3_value=-1.0, std::string arg4_name="", float arg4_value=-1.0, std::string arg5_name="", float arg5_value=-1.0, std::string arg6_name="", float arg6_value=-1.0)
Append debug information.
Definition: ZmqLogger.cpp:173
static ZmqLogger * Instance()
Create or get an instance of this logger singleton (invoke the class with this method)
Definition: ZmqLogger.cpp:35
This namespace is the default namespace for all code in the openshot library.
Definition: Compressor.h:29
@ SCALE_FIT
Scale the clip until either height or width fills the canvas (with no cropping)
Definition: Enums.h:38
@ SCALE_STRETCH
Scale the clip until both height and width fill the canvas (distort to fit)
Definition: Enums.h:39
@ SCALE_CROP
Scale the clip until both height and width fill the canvas (cropping the overlap)
Definition: Enums.h:37
ChannelLayout
This enumeration determines the audio channel layout (such as stereo, mono, 5 point surround,...
const Json::Value stringToJson(const std::string value)
Definition: Json.cpp:16
This struct holds the associated video frame and starting sample # for an audio packet.
Definition: FFmpegReader.h:42
bool is_near(AudioLocation location, int samples_per_frame, int64_t amount)
void reset(bool eof)
Definition: FFmpegReader.h:80
int audio_bit_rate
The bit rate of the audio stream (in bytes)
Definition: ReaderBase.h:59
int video_bit_rate
The bit rate of the video stream (in bytes)
Definition: ReaderBase.h:49
bool has_single_image
Determines if this file only contains a single image.
Definition: ReaderBase.h:42
float duration
Length of time (in seconds)
Definition: ReaderBase.h:43
openshot::Fraction audio_timebase
The audio timebase determines how long each audio packet should be played.
Definition: ReaderBase.h:64
int width
The width of the video (in pixesl)
Definition: ReaderBase.h:46
int channels
The number of audio channels used in the audio stream.
Definition: ReaderBase.h:61
openshot::Fraction fps
Frames per second, as a fraction (i.e. 24/1 = 24 fps)
Definition: ReaderBase.h:48
openshot::Fraction display_ratio
The ratio of width to height of the video stream (i.e. 640x480 has a ratio of 4/3)
Definition: ReaderBase.h:51
int height
The height of the video (in pixels)
Definition: ReaderBase.h:45
int pixel_format
The pixel format (i.e. YUV420P, RGB24, etc...)
Definition: ReaderBase.h:47
int64_t video_length
The number of frames in the video stream.
Definition: ReaderBase.h:53
std::string acodec
The name of the audio codec used to encode / decode the video stream.
Definition: ReaderBase.h:58
std::map< std::string, std::string > metadata
An optional map/dictionary of metadata for this reader.
Definition: ReaderBase.h:65
std::string vcodec
The name of the video codec used to encode / decode the video stream.
Definition: ReaderBase.h:52
openshot::Fraction pixel_ratio
The pixel ratio of the video stream as a fraction (i.e. some pixels are not square)
Definition: ReaderBase.h:50
openshot::ChannelLayout channel_layout
The channel layout (mono, stereo, 5 point surround, etc...)
Definition: ReaderBase.h:62
bool has_video
Determines if this file has a video stream.
Definition: ReaderBase.h:40
bool has_audio
Determines if this file has an audio stream.
Definition: ReaderBase.h:41
openshot::Fraction video_timebase
The video timebase determines how long each frame stays on the screen.
Definition: ReaderBase.h:55
int video_stream_index
The index of the video stream.
Definition: ReaderBase.h:54
int sample_rate
The number of audio samples per second (44100 is a common sample rate)
Definition: ReaderBase.h:60
int audio_stream_index
The index of the audio stream.
Definition: ReaderBase.h:63
int64_t file_size
Size of file (in bytes)
Definition: ReaderBase.h:44