Le dimanche 12 février 2023 à 18:49 +0000, John Cox a écrit :
> Hi
>
> > Le jeudi 09 février 2023 à 19:06 +0000, John Cox a écrit :
> > > Hi
> > >
> > > > Le vendredi 27 janvier 2023 à 15:34 +0000, John Cox a écrit :
> > > > > Add fourccs for Broadcom 8 and 10-bit packed 128 byte column formats to
> > > > > videodev2.h
> > > > >
> > > > > Signed-off-by: John Cox <jc@xxxxxxxxxxxxx>
> > > > > ---
> > > > > include/uapi/linux/videodev2.h | 2 ++
> > > > > 1 file changed, 2 insertions(+)
> > > > >
> > > > > diff --git a/include/uapi/linux/videodev2.h b/include/uapi/linux/videodev2.h
> > > > > index 1befd181a4cc..a836322ae5d8 100644
> > > > > --- a/include/uapi/linux/videodev2.h
> > > > > +++ b/include/uapi/linux/videodev2.h
> > > > > @@ -656,6 +656,8 @@ struct v4l2_pix_format {
> > > > > #define V4L2_PIX_FMT_P010_4L4 v4l2_fourcc('T', '0', '1', '0') /* 12 Y/CbCr 4:2:0 10-bit 4x4 macroblocks */
> > > > > #define V4L2_PIX_FMT_NV12_8L128 v4l2_fourcc('A', 'T', '1', '2') /* Y/CbCr 4:2:0 8x128 tiles */
> > > > > #define V4L2_PIX_FMT_NV12_10BE_8L128 v4l2_fourcc_be('A', 'X', '1', '2') /* Y/CbCr 4:2:0 10-bit 8x128 tiles */
> > > > > +#define V4L2_PIX_FMT_NV12_C128 v4l2_fourcc('C', 'N', '1', '2') /* Y/CbCr 4:2:0 128 byte columns */
> > > > > +#define V4L2_PIX_FMT_P030_C128 v4l2_fourcc('C', 'N', '3', '0') /* Y/CbCr 4:2:0 10-bit packed 128 byte columns */
> > > > >
> > > > > /* Tiled YUV formats, non contiguous planes */
> > > > > #define V4L2_PIX_FMT_NV12MT v4l2_fourcc('T', 'M', '1', '2') /* 12 Y/CbCr 4:2:0 64x32 tiles */
> > > >
> > > > I would expect updates to v4l2-common.c and v4l2-ioctl.c to be in the same
> > > > patch. And then the driver should be using the helpers there whenever possible.
> > >
> > > Fair point - I'll fix that.
> > >
> > > What is the correct .bpp for 3 10-bit pixels packed into 4 bytes in the
> > > v4l2_format_info?
> >
> > Good question, maybe this can be done with the fractional bpp support. I must
> > admit, I didn't think about padded cases, I was handling 10bit fully packed over
> > 5 bytes.
> >
> > https://lore.kernel.org/linux-arm-kernel/20230103170058.810597-3-benjamin.gaignard@xxxxxxxxxxxxx/
> > My case ended working with:
> >
> > { .format = V4L2_PIX_FMT_NV12_10LE40_4L4, .pixel_enc = V4L2_PIXEL_ENC_YUV, .mem_planes = 1, .comp_planes = 2, .bpp = { 5, 10, 0, 0 }, .bpp_div = { 4, 4, 1, 1 }, .hdiv = 2, .vdiv = 2 },
> >
> > Question is what do we do about comp_planes, if we kind of fake it to be 2, then maybe this would work.
> >
> > { .format = V4L2_PIX_FMT_P030_C128, .pixel_enc = V4L2_PIXEL_ENC_YUV, .mem_planes = 1, .comp_planes = 2, .bpp = { 4, 8, 0, 0 }, .bpp_div = { 3, 3, 1, 1 }, .hdiv = 2, .vdiv = 2 },
> >
> > For weird format, this is a bit of hacky, all we want is to get the right
> > stride, and the offset part is not used for mem_planes = 1 formats.
>
> I'm happy with whatever the consensus says is "right".
>
> What tree/branch should I be patching against? Code to support the above
> doesn't seem to be in git://linuxtv.org/media_tree:master which is what
> I was using.
Not yet, this patch is part of "[PATCH v3 00/13] AV1 stateless decoder for
RK3588" serie from Benjamin, which now needs to be rebased on V5 of the API.
I see two options, you simply pick it as part of your patchset, worst case the
other serie get merge and we skip it or I split it out of AV1 so we can get it
merged sooner. I can probably make it on its own, as it is used to enable the
native (reference frame) format Hantro produces, including VP9, which is wrongly
mapped to P010_4L4 at the moment (not even sure this format exist).
Nicolas
>
> Thanks
>
> John Cox
>
> > let me know,
> > Nicolas
> >
> > >
> > > Regards
> > >
> > > John Cox
