[Mesa-dev] [PATCH] mesa: reference built-in uniforms into gl_uniform_storage

[Martin Peres]

On 04/06/15 14:02, Jose Fonseca wrote:
> On 04/06/15 11:53, Jose Fonseca wrote:
>> On 04/06/15 11:23, Martin Peres wrote:
>>> On 04/06/15 13:07, Jose Fonseca wrote:
>>>> On 04/06/15 10:56, Martin Peres wrote:
>>>>> On 04/06/15 12:53, Jose Fonseca wrote:
>>>>>> On 31/05/15 08:10, Tapani wrote:
>>>>>>>
>>>>>>> I've read this a couple of times now and cannot spot any users of
>>>>>>> storage that would be making a wrong assumption, you've fixed these
>>>>>>> and
>>>>>>> I trust Jenkins was OK for i915?
>>>>>>>
>>>>>>> Everything is ok if you remove 'I think' and 'Hopefully' from 
>>>>>>> commit
>>>>>>> message :)
>>>>>>
>>>>>> Martin's fears were legitimate, as this change just caused +2729 
>>>>>> tests
>>>>>> to fail with piglit. :)
>>>>>>
>>>>>> So there must be some code that still needs to be updated somewhere
>>>>>> (maybe in the gallium specific paths).  Question is then: Where's
>>>>>> Wally?
>>>>>>
>>>>>> Jose
>>>>>
>>>>> Thanks for the notification Jose! I will try running piglit on 
>>>>> llvmpipe
>>>>> and see if I can reproduce! I did not get any error on the intel
>>>>> driver.
>>>>> Will re-launch a test on our test platform, just in case!
>>>>
>>>> Thanks. No prob.
>>>>
>>>> I forgot to say -- this was with llvmpipe. Even something basic as
>>>>
>>>>   bin/shader_runner
>>>> tests/shaders/glsl-uniform-initializer-1.shader_test -auto
>>>>
>>>> regressed, so hopefully it should be easy to spot.
>>>
>>> Hopefully, indeed :) I just confirmed that I did not introduce any
>>> regression on the intel side. Will start debugging in the gallium 
>>> world!
>>>
>>> Martin
>>
>> BTW, isn't intel using NIR by default? I wonder if it can repro intel if
>> you disable it.
>>
>> Jose
>
> Yep, this is not gallium specific -- the problem can be reproed with 
> the classic swrast too:
>
> export 
> LD_LIBRARY_PATH=/home/jfonseca/work/vmware/opengl/mesa/build/linux-x86_64-debug/mesa/drivers/x11
>
> $ glxinfo | grep string
> server glx vendor string: Brian Paul
> server glx version string: 1.4 Mesa 10.7.0-devel
> client glx vendor string: Brian Paul
> client glx version string: 1.4 Mesa 10.7.0-devel
> OpenGL vendor string: Brian Paul
> OpenGL renderer string: Mesa X11
> OpenGL version string: 2.1 Mesa 10.7.0-devel (git-73ba4af)
> OpenGL shading language version string: 1.20
> $ tests/shaders/glsl-uniform-initializer-1.shader_test -auto
> Probe color at (62,125)
>   Expected: 0.000000 1.000000 0.000000
>   Observed: 0.000000 0.000000 0.000000
> Probe color at (187,125)
>   Expected: 0.000000 0.000000 1.000000
>   Observed: 0.000000 0.000000 0.000000
> PIGLIT: {"result": "fail" }
>
> Hopefully this would narrow it even further.

Found the bug, fixed it and tested for regressions. Now, I will hide 
under a very big brown bag for quite some time!

Will send the patch in reply to this email.