[Mesa-dev] [PATCH 00/64] i965: Start using ISL for filling out surface states

[Jason Ekstrand]

On Mon, Jun 13, 2016 at 2:24 AM, Emil Velikov <emil.l.velikov at gmail.com>
wrote:

> On 11 June 2016 at 17:02, Jason Ekstrand <jason at jlekstrand.net> wrote:
>
> > The next 20 patches or so are general ISL cleanups and fixes.  If no one
> is
> > too opposed, I'd like to back-port the whole pile to 12.0.  There are two
> > reasons for this: First, ISL is new and this is a substantial cleanup;
> > back-porting it will make back-porting will keep the initial release of
> ISL
> > cleaner and make back-porting other patches easier in the future.
> Second,
> > in the middle of the series are a couple of changes that fix some 850
> > Vulkan CTS tests on Haswell.
> >
> No serious objections on getting these in 12.0. A bunch of small
> suggestions though.
>
> >   i965/gen4: Subtract 1 from buffer sizes
> Please polish the commit message a bit.
>

I added "The PRM states that the values put in Width, Height, and Depth
should be various bits from the value size - 1.  We seem to have done this
wrong more-or-less from the start."


> >   isl/state: Put surface format setup at the top
> >   isl/state: Put all dimension setup together and towards the top
> >   isl/state: Put pitch calculations together
> Are these solely for cosmetic/ease of read reasons or there's
> something more to it ? Can you please mention in the commit message ?
>

How about: "This is purely cosmetic, but it makes things look a bit more
readable."
<https://www.khronos.org/registry/spir-v/specs/1.1/SPIRV.html#Integer>
<https://www.khronos.org/registry/spir-v/specs/1.1/SPIRV.html#Floating>


> >   genxml: Add enough XML for gens 4, 4.5, and 5 to get SURFACE_STATE
> Let me see if I can quickly cook something for the Android build.
>
>
> >   isl: Add support for filling out surface states all the way back to
> >     gen4
> Ditto.
>

If you want to give me something, I'm happy to squash it in.


>
> >   isl/formats: Mark RAW as having a block size of 1 byte
> Worth mentioning why we want that ?
>

That patch can actually be dropped.  The original idea was that, for buffer
surfaces, we needed to divide by the element size so the size should be
non-zero.  Turns out we need to divide by stride so it can be 0 and the
patch isn't needed.  I'll defer to Chad on whether it should be 0 or 1.


> Thanks
> Emil
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20160613/141a8568/attachment-0001.html>