Hi Guennadi, Thanks for your reply. On Sat, Apr 7, 2012 at 4:21 PM, Guennadi Liakhovetski <g.liakhovetski@xxxxxx> wrote: > Hi Sergio > > On Sat, 7 Apr 2012, Aguirre, Sergio wrote: > >> Hi everyone, >> >> I'll like to request for your advice on adding NV12 support for my omap4iss >> camera driver, which is done after the resizer block in the OMAP4 ISS ISP >> (Imaging SubSystem Image Signal Processor). >> >> So, the problem with that, is that I don't see a match for V4L2_PIX_FMT_NV12 >> pixel format in "enum v4l2_mbus_pixelcode". >> >> Now, I wonder what's the best way to describe the format... Is this correct? >> >> V4L2_MBUS_FMT_NV12_1X12 >> >> Because every pixel is comprised of a 8-bit Y element, and it's UV components >> are grouped in pairs with the next horizontal pixel, whcih in combination >> are represented in 8 bits... So it's like that UV component per-pixel is 4-bits. >> Not exactly, but it's the best representation I could think of to >> simplify things. > > Do I understand it right, that your resizer is sending the data to the DMA > engine interleaved, not Y and UV planes separately, and it's only the DMA > engine, that is separating the planes, when writing to buffers? In such a > case I'd use a suitable YUV420 V4L2_MBUS_FMT_* format for that and have > the DMA engine convert it to NV12, similar to what sh_mobile_ceu_camera > does. No, it actually has 2 register sets for specifying the start address for each plane. So, I have one register that I program with "Y-start" address, and another register that I program with "UV-start" address. For both cases, you control the byte offset between every begin of each line. So, in theory, you could save it interleaved in memory if you want, or in 2 separate buffers depending on how you program the address/offset pair. Regards, Sergio > > Thanks > Guennadi > >> I mean, the HW itself writes in memory to 2 contiguous buffers, so there's 2 >> separate DMA writes. I have to program 2 starting addresses, which, in an >> internal non-v4l2-subdev implementation, I have been programming like this: >> >> paddr = start of 32-byte aligned physical address to store buffer >> x = width >> y = height >> >> Ysize = (x * y) >> UVsize = (x / 2) * y >> Total size = Ysize + UVsize >> >> Ystart = paddr >> UVstart = (paddr + Ysize) >> >> But, in the media controller framework, i have a single DMA output pad, that >> creates a v4l2 capture device node, and i'll be queueing a single buffer. >> >> Any advice on how to address this properly? Does anyone has/had a similar need? >> >> Regards, >> Sergio >> -- >> To unsubscribe from this list: send the line "unsubscribe linux-media" in >> the body of a message to majordomo@xxxxxxxxxxxxxxx >> More majordomo info at http://vger.kernel.org/majordomo-info.html >> > > --- > Guennadi Liakhovetski, Ph.D. > Freelance Open-Source Software Developer > http://www.open-technology.de/ -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
