[Mesa-dev] [PATCH] glsl: Remove field array_lvalue from ir_variable.

[Paul Berry]

On 16 September 2011 10:59, Paul Berry <stereotype441 at gmail.com> wrote:

> On 16 September 2011 09:01, Marek Olšák <maraeo at gmail.com> wrote:
>
>> Hi Paul,
>>
>> this commit breaks the piglit test
>> shaders/glsl-arb-fragment-coord-conventions on softpipe and r600g:
>>
>> bin/glsl-arb-fragment-coord-conventions -auto
>> Regular gl_FragCoord
>> Pixel center half integer
>> Pixel center integer
>> Probe at (0,0)
>>  Expected: 0.250000 0.250000 0.000000
>>  Observed: 0.749020 0.749020 0.000000
>> Probe at (99,99)
>>  Expected: 0.250000 0.250000 0.000000
>>  Observed: 0.749020 0.749020 0.000000
>> Pixel origin upper left
>> Probe at (0,0)
>>  Expected: 0.000000 1.000000 0.000000
>>  Observed: 0.003922 0.003922 0.000000
>> Probe at (99,99)
>>  Expected: 1.000000 0.000000 0.000000
>>  Observed: 0.996078 0.996078 0.000000
>> Pixel origin upper left and pixel center integer
>> PIGLIT: {'result': 'fail' }
>>
>> Reverting the commit fixes this. It looks like
>> layout(pixel_center_integer) is ignored, because "Pixel center half
>> integer" and "Pixel center integer" return the same color in the test.
>>
>> Do you have any idea what's wrong?
>>
>> Marek
>>
>>
> Wow, that's surprising, since the commit should only affect what is
> considered an error condition; it shouldn't affect the GLSL IR that gets
> generated.
>
> I'll do some investigation and let you know what I find.
>
> Paul
>

Marek,

I've reproduced the behavior you saw, and I believe it is due to a build
dependency problem of some sort, unrelated to the array_lvalue fix.

If I check out commit 373234ce17555376c5cb57a2cdce3038df3f51cd (the commit
before my array_lvalue fix) and do a clean build of Gallium,
glsl-arb-fragment-coord-conventions passes.
If I then check out commit 00792e3586746c833ffc9bb65712e38038916e06 (my
array_lvalue fix) and do an *incremental* build,
glsl-arb-fragment-coord-conventions fails.
If I then do a clean build of Gallium, glsl-arb-fragment-coord-conventions
passes again.

Unfortunately, I'm not familiar enough with the Gallium build system to
track this bug further.  But I'm confident that it's unrelated to this patch
since a clean build makes it go away.

Hope that helps,

Paul
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20110916/4da6aa44/attachment.htm>