Re: [PATCH] drm/fourcc: Add Arm 16x16 block modifier

From: Brian Starkey
Date: Mon Jun 24 2019 - 07:23:12 EST


On Mon, Jun 24, 2019 at 11:58:59AM +0200, Daniel Vetter wrote:
> On Mon, Jun 24, 2019 at 11:32 AM Brian Starkey <Brian.Starkey@xxxxxxx> wrote:
> >
> > Hi Daniel,
> >
> > On Fri, Jun 21, 2019 at 05:27:00PM +0200, Daniel Vetter wrote:
> > > On Fri, Jun 21, 2019 at 12:21 PM Raymond Smith <Raymond.Smith@xxxxxxx> wrote:
> > > >
> > > > Add the DRM_FORMAT_MOD_ARM_16X16_BLOCK_U_INTERLEAVED modifier to
> > > > denote the 16x16 block u-interleaved format used in Arm Utgard and
> > > > Midgard GPUs.
> > > >
> > > > Signed-off-by: Raymond Smith <raymond.smith@xxxxxxx>
> > > > ---
> > > > include/uapi/drm/drm_fourcc.h | 10 ++++++++++
> > > > 1 file changed, 10 insertions(+)
> > > >
> > > > diff --git a/include/uapi/drm/drm_fourcc.h b/include/uapi/drm/drm_fourcc.h
> > > > index 3feeaa3..8ed7ecf 100644
> > > > --- a/include/uapi/drm/drm_fourcc.h
> > > > +++ b/include/uapi/drm/drm_fourcc.h
> > > > @@ -743,6 +743,16 @@ extern "C" {
> > > > #define AFBC_FORMAT_MOD_BCH (1ULL << 11)
> > > >
> > > > /*
> > > > + * Arm 16x16 Block U-Interleaved modifier
> > > > + *
> > > > + * This is used by Arm Mali Utgard and Midgard GPUs. It divides the image
> > > > + * into 16x16 pixel blocks. Blocks are stored linearly in order, but pixels
> > > > + * in the block are reordered.
> > > > + */
> > > > +#define DRM_FORMAT_MOD_ARM_16X16_BLOCK_U_INTERLEAVED \
> > > > + fourcc_mod_code(ARM, ((1ULL << 55) | 1))
> > >
> > > This seems to be an extremely random pick for a new number. What's the
> > > thinking here? Aside from "doesnt match any of the afbc combos" ofc.
> > > If you're already up to having thrown away 55bits, then it's not going
> > > to last long really :-)
> > >
> > > I think a good idea would be to reserve a bunch of the high bits as
> > > some form of index (afbc would get index 0 for backwards compat). And
> > > then the lower bits would be for free use for a given index/mode. And
> > > the first mode is probably an enumeration, where possible modes simple
> > > get enumerated without further flags or anything.
> >
> > Yup, that's the plan:
> >
> > (0 << 55): AFBC
> > (1 << 55): This "non-category" for U-Interleaved
> > (1 << 54): Whatever the next category is
> > (3 << 54): Whatever comes after that
> > (1 << 53): Maybe we'll get here someday
>
> Uh, so the index would be encoded with least-significant bit first,
> starting from bit55 downwards?

Yeah.

> Clever idea, but I think this needs a
> macro (or at least a comment). Not sure there's a ready-made bitmask
> mirror function for this stuff, works case we can hand-code it and
> extend every time we need one more bit encoded. Something like:
>
> MIRROR_U32((u & (BIT(0)) << 31 | (u & BIT(1) << 30 | ...)
>

Is it really worth it? People can just use the definitions as written
in drm_fourcc.h. I agree that we should have the high bits described
in a comment though.

> And then shift that to the correct place. Probably want an
>
> ARM_MODIFIER_ENCODE(space_idx, flags) macro which assembles everything.
>
> > ...
> >
> > I didn't want to explicitly reserve some high bits, because we've no
> > idea how many to reserve. This way, we can assign exactly as many
> > high bits as we need, when we need them. If any of the "modes" start
> > encroaching towards the high bits, we'll have to make a decision at
> > that point.
> >
> > Also, this is the only U-Interleaved format (that I know of), so it's
> > not worth calling bit 55 "The U-Interleaved bit" because that would be
> > a waste of space. It's more like the "misc" bit, but that's not a
> > useful name to enshrine in UAPI.
>
> Yeah that's what I meant. Also better to explicitly reserve this, i.e.
>
> #define ARM_FBC_MODIFIER_SPACE 0
> #define ARM_MISC_MODIFIER_SPACE 1
>
> and then encode with the mirror trickery.
>

I don't really see the value in that either, it's just giving
userspace the opportunity to depend on more stuff: more future
headaches. So long as the 64-bit values are stable, that should be
enough.

> > Note that isn't the same as the "not-AFBC bit", because we may well
> > have something in the future which is neither AFBC nor "misc".
> >
> > We've been very careful in our code to enforce all
> > undefined/unrecognised bits to be zero, to ensure that this works.
> >
> > >
> > > The other bit: Would be real good to define the format a bit more
> > > precisely, including the layout within the tile.
> >
> > It's U-Interleaved, obviously ;-)
>
> :-) I mean full code exists in panfrost/lima, so this won't change
> anything really ...

Yeah, so for us to provide a more detailed description would require
another lengthy loop through our legal approval process, and I'm not
sure we can make a strong business case (which is what we need) for
why this is needed.

Of course, if someone happens to know the layout and wants to
contribute to this file... Then I don't know how ack/r-b would work in
that case, but I imagine the subsystem maintainer(s) might take issue
with us attempting to block that contribution.

Thanks,
-Brian

>
> Cheers, Daniel
>
> >
> > -Brian
> >
> > >
> > > Also ofc needs acks from lima/panfrost people since I assume they'll
> > > be using this, too.
> > >
> > > Thanks, Daniel
> > >
> > > > +
> > > > +/*
> > > > * Allwinner tiled modifier
> > > > *
> > > > * This tiling mode is implemented by the VPU found on all Allwinner platforms,
> > > > --
> > > > 2.7.4
> > > >
> > >
> > >
> > > --
> > > Daniel Vetter
> > > Software Engineer, Intel Corporation
> > > +41 (0) 79 365 57 48 - http://blog.ffwll.ch
>
>
>
> --
> Daniel Vetter
> Software Engineer, Intel Corporation
> +41 (0) 79 365 57 48 - http://blog.ffwll.ch