[Libav-user] av_picture_copy to different linesize

[Vic Lee]

Hi FFmpeg developers,

I am a FreeRDP developer and I am currently working on using FFmpeg to 
decode video. The code I attached below is a simplified version of my 
real code. It basically just decode the video frame and store the image 
into my own buffer, which will be later sent to XVideo:

	/* Decode data into frame */
	AVPacket pkt;
	av_init_packet(&pkt);
	pkt.data = data;
	pkt.size = data_size;
	len = avcodec_decode_video2(context, frame, &decoded, &pkt);

	/* Copy the decoded frame to decoded_data */
	decoded_size = avpicture_get_size(context->pix_fmt,
		context->width, context->height);
	decoded_data = malloc(decoded_size);
	f = avcodec_alloc_frame();
	avpicture_fill((AVPicture *) f, decoded_data,
		context->pix_fmt,
		context->width, context->height);
	av_picture_copy((AVPicture *) f, (AVPicture *) frame,
		context->pix_fmt,
		context->width, context->height);
	av_free(f);

So far I only work with YUV420P. The above codes works fine if the 
decoded frame->linesize[0]==width; however in some cases 
frame->linesize[0] will be aligned to larger size than width, then color 
will be messed up in the target YUV data buffer (Y channel is ok but U 
and V is not).

To be more specific, the problematic test video is 852x480, the output 
frame has linesize 896/448/448, and the target frame has 852/426/426. I 
tried to read the av_picture_copy source code and dump the buffers, and 
seems everythig is alright: every first 426 bytes of the source lines 
are just copied to the target. But the color of the image does not look 
right.

Am I missing something here?

Vic