[Mesa-dev] i965: Transition the rest of surfaces (i.e., color) to isl

[Jason Ekstrand]

On Fri, Jul 28, 2017 at 8:17 PM, Tomasz Figa <tfiga at chromium.org> wrote:

> Hi Topi, Jason,
>
> On Sat, Jul 22, 2017 at 12:00 AM, Topi Pohjolainen
> <topi.pohjolainen at gmail.com> wrote:
> > First patch actually should have been included already when
> > gen6 stencil got transitioned - it has been giving warning ever
> > since.
> >
> > Most of the work actually got already done for depth surfaces (which
> > is y-tiled such as color surfaces). What is left are color surface
> > specifics, mostly preparing for corner cases.
> >
> > This is now all green in ci-system. For snb and older i965 wasn't
> > checking hardware incapabilities as hard as isl does. Certain
> > format/size/msaa combinations were allowed that shouldn't have.
> > Moving to isl exposed code paths that didn't report surface creation
> > failures resulting in asserts firing later on. Patches 10 and 11
> > now properly tell the client if the surface type can't be supported
> > allowing piglit tests to skip them.
>
> I think it might be related to previously merged patches, but the
> topic is still ISL, so let me ask my question here. Is there any
> possibility to add some diagnostic information to the validation code?
> We've been seeing EGL image import failures on new Mesa as a result of
> ISL catching issues in our allocator (cros_gralloc on top of ChromeOS
> minigbm), but it's close to impossible to identify the cause without
> manually inserting some printfs and recompiling the code. I think
> having some error messages printed in case of a buffer validation
> failure would be a great benefit.
>

What kind of prints are you looking for exactly?  I've run into some issues
myself but if you could provide a concrete example, that would help.

--Jason
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170728/d4994eb3/attachment.html>