Hi Jan

On Wed, 24 Aug 2011, Jan Pohanka wrote:

Hello,

could please anyone explain me a bit situation about using memory buffers for
dma for video input devices? Unfortunately I don't understand it at all.
I want to capture images 1600x1200 from 2 mpix sensor on i.mx27 board.
I gave 8MB to mx2_camera device with dma_declare_coherent_memory.

Unfortunately it seems to be not enough. In UYVY format I need 1600x1200x2 for
one picture, it is cca 3.8MB.
After some digging I noticed, that dma_alloc_coherent() is called three times
and each time with the 3.8MB demand. Once it is directly from mx2_camera
driver and two times from videobuf_dma_contig. OK, that is more than 8MB
available, but why there are so big memory demands for one picture?

I'm using gstremer for capturing, so it probably requests several buffers with
VIDIOC_REQBUFS. Is this behaviour normal, even if there is explicitly noted
that I want only one buffer?

The mx2_camera driver is allocating one "discard" buffer of the same size, 
as regular buffers for cases, when the user is not fast enough to queue 
new buffers for the running capture. Arguably, this could be aliminated 
and the last submitted buffer could be re-used until either more buffers 
are available or the streaming is stopped. Otherwise, it could also be 
possible to stop capture until buffers are available again. In any case, 
this is the current driver implementation. As for 2 buffers instead of one 
for the actual capture, I think, gstreamer defines 2 as a minimum number 
of buffers, which is actually also required for any streaming chance.

Thanks
Guennadi

gst-launch \
 v4l2src num-buffers=1 device=/dev/video1 ! \
 video/x-raw-yuv,format=\(fourcc\)UYVY,width=$WIDTH,height=$HEIGHT ! \
 jpegenc ! \
 filesink location=col_image.jpg


with best regards
Jan

---
Guennadi Liakhovetski, Ph.D.
Freelance Open-Source Software Developer
http://www.open-technology.de/