I am trying to do the following with my IMX6Q board:
1. Capture frames from CMOS using the IPU CSI->PRP->MEM (involves converting YUYV->YUV420)
2. IPU resize using the /dev/mxc_ipu API
3. Some data processing on the down-scaled image
4. VPU Encoding of the pre-resized image
I cannot seem to get both IPU parts (1), (3) to work together.
If I do capture via "normal" v4l2 (usb-webcam, for example), all is fine.
If I do IPU capture and software resize it works fine.
I absolutely need BOTH sizes (one for encoding, one for processing).
My question -
Is there any limitation (HW?) on using both the CSI and the mxc_ipu api together?
How can I get this chain to work without CPU intensive calculations (csc / resize)?
Image size is VGA
Linux BSP 3.0.35
Freescale LTIB v4.0.0
I may not really get your point. /unit/mxc_v4l2_capture.out can resize the data from CSI, and CSC (kernel/drivers/media/video/mxc/capture/ipu_prp_enc.c, prp_enc_setup()) executes when CSI->PRP->MEM is adopted.
So why "2. IPU resize using the /dev/mxc_ipu API" is needed?