On Fri, Feb 8, 2019 at 3:22 PM Steve Longerbeam <slongerbeam@xxxxxxxxx> wrote: > > > > On 2/8/19 1:23 PM, Tim Harvey wrote: > > On Thu, Feb 7, 2019 at 5:54 PM Steve Longerbeam <slongerbeam@xxxxxxxxx> wrote: > >> > > <snip> > >>> Ok there is definitely something wrong when using the IC with > >>> UYVY8_1X16 (passthrough) which works with UYVY8_2X8. It looks to me > >>> like the ipu1_ic_prp isn't negotiating its format properly. You can't > >>> re-create this because you don't have any UYVY8_1X16 (passthrough) > >>> sensors right? > >> Sorry, maybe I didn't mention this, but passthrough cannot go though the > >> IPU, you can only send passthrough pixels out the CSI directly to > >> /dev/videoN interface (the ipu_csi:2 pad). > >> > > crud... this has been my issue all along with that set of UYVY8_1X16 > > pipelines then. So this means the mem2mem driver also won't be able to > > handle 16bit pixel formats as well. > > Ugh, let me rephrase. UYVY8_1X16 incoming on a parallel bus (not MIPI > CSI-2) to the CSI cannot be sent through the IPU, those pixels must be > sent directly to the ipu_csi:2 pad to /dev/videoN. At least according > the the imx6 register manual. > > But that's not a limitation of the mem2mem driver because the pixels are > coming from a memory buffer and not a 16-bi parallel bus via the CSI > (and in any case mem2mem does not deal in media bus codes, it speaks > V4L2_PIX_FMT_* which contains no bus info). So yes, you can receive > UYVY8_1X16 on the CSI parallel bus, routed to ipu_csi:2 pad to > /dev/videoN, and then pipe that to mem2mem v4l2convertN element in a > gstreamer pipeline. > > > > > So while I can downscale this by > > multiples of 2 (independent width/height), CSC convert it from > > srgb/bt.601 to yuv/bt.709, and even pixel reorder it within YUV via > > the ipu_csi entitty I'll never be able to deinterlace, > > that's true, currently you can't use the VDIC to do motion compensated > de-interlacing, but you can still make use of the IDMAC interweave > feature to deinterlace without motion compensation. > > > scale with > > flexibility, flip/rotate or do a RGB->YUV CSC on it as those are all > > features of the IC path. > > mem2mem v4l2convertN element will do those for you (flexible up/down > scaling, flip/rotate, and CSC). > ah... ok right using mem2mem is precisely that 'mem to mem' and not direct CSI to IPU. Ok so i experimented with some pipelines and perhaps am setting them up wrong. I'll respond to the imx-media mem2mem patch with my observations and expiriments. Thanks, Tim
