Recherche avancée

Sur d’autres sites (10856)

• Trying to fix VPlayer's seeking ability, need some guidance [Android FFmpeg]

  1er juin 2016, par vxh.viet

  I’m trying to fix the currently broken seeking ability of VPlayer which is a FFmpeg player for Android. Being a Java developer, C code looks like alien language to me so can only fix it using common logic (which could make any C veteran have a good laugh).

  The relevant file is player.c and I’ll try my best to point out the relevant modification.

  So the basic idea is because FFmpeg’s av_seek_frame is very inaccurate even with AVSEEK_FLAG_ANY so I’m trying to follow this suggestion to seek backward to the nearest keyframe and then decode to the frame I want. One addition note is since I want to seek based on millisecond while the said solution show the way to seek by frame which is potentially a source of problem.

  In the Player I add the following fields :

  struct Player{
  
  ....
  
  AVFrame *frame;
  
  int64_t current_time_stamp;
  
  }; 

  In the player_read_from_stream I modify the seeking part as :

  void * player_read_from_stream(void *data) {
  
      ...
  
      struct DecoderData *decoder_data = data;
  
      int stream_no = decoder_data->stream_no;
  
      AVCodecContext * ctx = player->input_codec_ctxs[stream_no];
  
      ...
  
      // seeking, start my stuff
  
      if(av_seek_frame(player->input_format_ctx, seek_input_stream_number, seek_target, AVSEEK_FLAG_BACKWARD) >= 0){
  
          //seek to key frame success, now need to read every frame from the key frame to our target time stamp
  
  
  
  
  
          while(player->current_time_stamp < seek_target){
  
  
  
              int frame_done;
  
  
  
              while (av_read_frame(player->input_format_ctx, &packet) >= 0) {
  
                  if (packet.stream_index == seek_input_stream_number) {
  
  
  
                      avcodec_decode_video2(ctx, player->frame, &frame_done, &packet);
  
                      LOGI(1,"testing_stuff ctx %d", *ctx);
  
                      if (frame_done) {
  
  
  
                          player->current_time_stamp = packet.dts;
  
                          LOGI(1,"testing_stuff current_time_stamp: %"PRId64, player->current_time_stamp);
  
                          av_free_packet(&packet);
  
                          return;
  
                      }
  
                  }
  
                  av_free_packet(&packet);
  
              }
  
          }
  
  
  
  
  
      }
  
      //end my stuff
  
  
  
      LOGI(3, "player_read_from_stream seeking success");
  
  
  
      int64_t current_time = av_gettime();
  
      player->start_time = current_time - player->seek_position;
  
      player->pause_time = current_time;        
  
  }

  And in player_alloc_frames I allocate the memory for my frame as :

  int player_alloc_frames(struct Player *player) {
  
      int capture_streams_no = player->caputre_streams_no;
  
      int stream_no;
  
      for (stream_no = 0; stream_no < capture_streams_no; ++stream_no) {
  
          player->input_frames[stream_no] = av_frame_alloc();
  
  
  
          //todo: test my stuff
  
          player->frame = av_frame_alloc();
  
          //end test
  
  
  
          if (player->input_frames[stream_no] == NULL) {
  
              return -ERROR_COULD_NOT_ALLOC_FRAME;
  
          }
  
      }
  
      return 0;
  
  }

  Currently it just keep crashing and being a typical Android NDK’s "feature", it just provide a super helpful stack trace :

  libc: Fatal signal 11 (SIGSEGV), code 1, fault addr 0x40 in tid 2717 (FFmpegReadFromS)

  I very much appreciate if anyone could help me solve this problem. Thank you for your time.

• Revision 3d181a4516 : Adjust speed setting for temporal layers in 1 pass non-rd mode. For speed 7, re

  11 août 2015, par Marco

  Changed Paths :
   Modify /vp9/encoder/vp9_speed_features.c

  
  
  Adjust speed setting for temporal layers in 1 pass non-rd mode.

  For speed 7, real-time mode : Base layer frames are further apart
  (for #temporal layers = 3, this is every 4 frames) so worth keeping
  same motion search parameters (as in speed 6) on the base layer frames.

  Change-Id : Idebf49dda6ef4f3d9a55aee55129a68253f692fb

• Revision a46f5459c3 : improved speed of 4x4 sse2 fdct. * speed improvment of 30 percent achieved * mu

  3 mars 2014, par Andrew Russell

  Changed Paths :
   Modify /vp9/encoder/x86/vp9_dct_sse2.c

  
  
  improved speed of 4x4 sse2 fdct.

  * speed improvment of 30 percent achieved
  * multiplies and adds remain the same
  * non-arithmetic instructions minimized by hand, by :
  -expanding 2 pass loop
  -removing irrelivant "shuffles"
  -combining last two rounding steps
  * further improvments may be possible

  Change-Id : Idec2c3f52910c48e6a0e0f9aefed5cae31b0b8c0