[Libav-user] How to allocate memory to store the image data (when use av_frame_alloc) ?

[wm4]

On Thu, 5 Feb 2015 16:52:27 +0800
林其益 <antony.lin at topviewcorp.com> wrote:

> Dear Libav:
> 
>  
> 
>   Sorry for my poor english first.
> 
>  
> 
>   When using avcode_alloc_frame with avcodec_decode_video2 to decode h264
> stream, I found out that the data pointer of AVFrame will change everytime i
> decode a new frame.
> 
> So if i need to buffer the decoded frame, i must copy the image data to
> other place (which allocate to store the image data).
> 
>  
> 
> Could I not to allocate memory to store the image data (decode frame),
> because it will cost me 10 ms every frame ( 8 M resolution picture)?
> 
>  
> 
> If I just store the AVFrame 's data pointer for every frames, It only had
> three sets of data pointer, so frames will be overwrite if you decode more
> than three times before to use it (to display it on the screen).
> 
> Three sets of data pointer(3 frames buffer) is not enough for my implement
> of buffering decoded frame, I want to buffer 10 frames (decode 10 frames
> before to display it to the screen). Because I frame of H264 ( 8M
> resolution) will recevie very slow  (more than 100 ms) and decode very slow
> (more than 100 ms), so I need buffer 10 frames to let display the stream
> smoothly.
> 
>  
> 
> My question is : 
> 
>   1. Could ffmpeg decode h264 use 10 sets of data pointer (not just three
> sets of data pointer)?
> 
>   or 
> 
>   2. Could I allocate my own memory to the data[] array to AVFrame (not
> allocate by the ffmpeg library)?
> 
>  
> 
> ////////////////////////////////////////////////////////////////////////////
> ////////////////////////////////////////////////////////////////////////////
> ///
> 
> example code is describe as follow:
> 
> /* c code bellow */
> 
>   AVFrame * pFrame;
> 
>   pFrame = avcode_alloc_frame();
> 
>   while (still_has_h264_frame)
> 
>   {
> 
>     avcodec_decode_video2(avctx, pFrame,....);
> 
>     printf(" data pointer of pFrame: %d %d %d \n", pFrame->data[0],
> pFrame->data[1], pFrame->data[2]);
> 
>   }
> 
>  
> 
> /*  result print log */
> 
>  
> 
> data pointer of pFrame: 396361792 395445504 401670240
> 
> data pointer of pFrame: 405536832 402628512 402864768
> 
> data pointer of pFrame: 430506048 431489120 431725376
> 
> data pointer of pFrame: 396361792 395445504 401670240   <---  from here ,
> the data point is the same as 1'th frame
> 
> data pointer of pFrame: 405536832 402628512 402864768   <---  from here ,
> the data point is the same as 2'th frame
> 
> data pointer of pFrame: 430506048 431489120 431725376   <---  from here ,
> the data point is the same as 3'th frame
> 
> ......
> 
>  
> 
> Best Regards,
> 
> Antony Lin
> 
>  
> 
>  
> 

The decoder already uses a memory pool for AVFrame image data.