[virglrenderer-devel] [PATCH 0/2] more accurate format capabilities

[Erik Faye-Lund]

On 10. juli 2018 12:07, Erik Faye-Lund wrote:
>
> On 10. juli 2018 09:13, Erik Faye-Lund wrote:
>> On 10. juli 2018 06:35, Dave Airlie wrote:
>>> On 10 July 2018 at 01:03, Erik Faye-Lund 
>>> <erik.faye-lund at collabora.com> wrote:
>>>> This series enables the mesa-driver to more accurate (and
>>>> hence correct) format capablity checking.
>>>>
>>>> Together with the mesa-changes, this fixes a bunch of dEQP
>>>> failures on GLES 3.1. This was written on top of Dave
>>>> Airlie's gl-4.3 branch, but since these changes apply
>>>> cleanly on master, I decided to send them directly.
>>>>
>>>> Depends on changes to mesa, which can be found here:
>>>> https://gitlab.freedesktop.org/kusma/mesa/commits/virgl-format-cleanups 
>>>>
>>>>
>>> If these changes are applied and no mesa changes are applied does
>>> anything regress?
>>>
>>
>> No regressions in dEQP for me, at least. Sadly, I can't run full piglit
>> runs with any versions of this, as I get bad crashes in qemu with *any*
>> version of virgl it seems (so hard they even crash GDB if I try to
>> debug)...
>>
>> Any tips on how to run as much of piglit as possible would be very
>> welcome :)
>
> Ah, I think I misunderstood the question before. No, nothing should
> regress. Current mesa always checks *both* render caps and sampler
> caps for textures and render-targets, and does *not* check
> depthstencil caps at all.
>
> So I don't think there's any chance of change of behavior in that
> combination.

I just did a full piglit run, and nope, no regressions. (the trick was 
to disable the arb_shader_texture_image_samples test, which cause 
asserts in the intel graphics driver)