Recherche avancée

Sur d’autres sites (8468)

• avformat/utils : Avoid copying packets unnecessarily

  20 septembre 2019, par Andreas Rheinhardt

  avformat/utils : Avoid copying packets unnecessarily
  Up until now, read_frame_internal in avformat/utils.c uses a spare
  
  packet on the stack that serves no real purpose : At no point in this
  
  function is there a need for another packet besides the packet destined
  
  for output :
  
  1. If the packet doesn't need a parser, but is output as is, the content
  
  of the spare packet (that at this point contains a freshly read packet)
  
  is simply copied into the output packet (via simple assignment, not
  
  av_packet_move_ref, thereby confusing ownership).
  
  2. If the packet needs parsing, the spare packet will be reset after
  
  parsing and any packets resulting from the packet read will be put into
  
  a packet list ; the output packet is not used here at all.
  
  3. If the stream should be discarded, the spare packet will be
  
  unreferenced ; the output packet is not used here at all either.
  Therefore the spare packet and the copies can be removed in principle.
  
  In practice, one more thing needs to be taken care of : If ff_read_packet
  
  failed, the output packet was not affected, now it is. But given that
  
  ff_read_packet returns a blank (as if reset via av_packet_unref) packet
  
  on failure, there is no problem from this side either.
  Signed-off-by : Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
  

  • [DH] libavformat/utils.c

• avfilter/vsrc_mandelbrot : change sin to sinf for color computation

  24 novembre 2015, par Ganesh Ajjanagadde

  avfilter/vsrc_mandelbrot : change sin to sinf for color computation
  lrintf is anyway used, suggesting we only care up to floating precision.
  
  Rurthermore, there is a compat hack in avutil/libm for this function,
  
  and it is used in avcodec/aacps_tablegen.h.
  This yields a non-negligible speedup. Sample benchmark :
  
  x86-64, Haswell, GNU/Linux :
  old (draw_mandelbrot) :
  
  274635709 decicycles in draw_mandelbrot,     256 runs,      0 skips
  
  300287046 decicycles in draw_mandelbrot,     512 runs,      0 skips
  
  371819935 decicycles in draw_mandelbrot,    1024 runs,      0 skips
  
  336663765 decicycles in draw_mandelbrot,    2048 runs,      0 skips
  
  581851016 decicycles in draw_mandelbrot,    4096 runs,      0 skips
  new (draw_mandelbrot) :
  
  269882717 decicycles in draw_mandelbrot,     256 runs,      0 skips
  
  296359285 decicycles in draw_mandelbrot,     512 runs,      0 skips
  
  370076599 decicycles in draw_mandelbrot,    1024 runs,      0 skips
  
  331478354 decicycles in draw_mandelbrot,    2048 runs,      0 skips
  
  571904318 decicycles in draw_mandelbrot,    4096 runs,      0 skips
  Reviewed-by : Michael Niedermayer <michael@niedermayer.cc>
  
  Signed-off-by : Ganesh Ajjanagadde <gajjanagadde@gmail.com>
  

  • [DH] libavfilter/vsrc_mandelbrot.c

• Direct ffmpeg -> OpenGL send without intermediary buffer copy

  3 juillet 2021, par Lucas Zanella

  To get a decoded frame from ffmpeg, there's the function avcodec_receive_frame :

  


  int avcodec_receive_frame   (   AVCodecContext *    avctx,
        AVFrame *   frame 
    )   


  


  Documentation : https://libav.org/documentation/doxygen/master/group__lavc__decoding.html#ga11e6542c4e66d3028668788a1a74217c

  


  The problem with this function is that I need to pass a frame to it like this :

  


  avcodec_receive_frame(&context, &frame)


  


  such that it will write the context into an AVFrame* just so I can pass it to OpenGL by doing another copy. That is, I make an unnecessary copy. For large video frames this could take time.

  


  There should be a function like this :

  


  int avcodec_receive_frame   (   AVCodecContext *    avctx,
        int (write_function*)(uint8_t data*, AVFrameParameters * parameters)
    )   


  


  this way I can pass a function pointer of a function that will take care of writing the video frame directly to OpenGL, without making an unnecessary copy. By the way, AVFrameParameters is a type I invented. It would have the parameters of the frame like color scheme, dimensions and stride, etc so I can upload properly to OpenGL.

  


  Is there a way to do something like this ? If not, why not ?

  


  PS : something even better would be for OpenGL to pass its GPU pointer to be so I can pass it to OpenGL, but I think this is too extreme