[Piglit] [PATCH 1/3] arb_gpu_shader_fp64: add glsl algebraic tests

Tapani tapani.palli at intel.com
Thu Aug 21 05:55:22 PDT 2014


On 08/21/2014 03:50 PM, Ilia Mirkin wrote:
> On Thu, Aug 21, 2014 at 1:00 AM, Tapani <tapani.palli at intel.com> wrote:
>> On 08/21/2014 07:21 AM, Chris Forbes wrote:
>>> With that fixed, does it pass on fglrx?
>>
>> I can only try that tomorrow but I believe it might still fail (unless we
>> allow some fuzziness in results). All the double div calculations seemed to
>> have some issues, result for example for red channel 1.0 / 2.0 was not 0.5
>> but 0.500312 (don't remember the exact digits) so not quite the correct
>> answer but almost there. Same was for blue which resulted in
>> 0.250022<something>.
> That's fine -- the results are going into a texture that can't
> represent these things with infinite (or even double) precision.

Ah right, then it should be ok.

> FWIW all these tests pass except the div one on nvc0 as well, which
> seems to be due to a TGSI bug. (It's using 0.6 as the immediate for
> the 4th component instead of 1.0.)

This is good to know, I'll try to run these with nvidia as well when I 
get a chance.

>>
>>
>>> On Thu, Aug 21, 2014 at 4:16 PM, Tapani <tapani.palli at intel.com> wrote:
>>>> On 08/21/2014 05:26 AM, Chris Forbes wrote:
>>>>> On Thu, Aug 21, 2014 at 12:36 AM, Tapani Pälli <tapani.palli at intel.com>
>>>>> wrote:
>>>>>> +++
>>>>>>
>>>>>> b/tests/spec/arb_gpu_shader_fp64/execution/built-in-functions/glsl-algebraic-div-double.shader_test
>>>>> ...
>>>>>> +       gl_FragColor = vec4(dvec4(1.0lf, 0.6lf, 1.0lf, 1.0lf) /
>>>>>> dcolor);
>>>>> ...
>>>>>> +uniform vec4 color 2.0 3.0 4.0 1.0
>>>>>> +draw rect -1 -1 2 2
>>>>>> +probe all rgba 0.5 0.2 0.25 0.0
>>>>> Presumably you mean 1.0 in the alpha channel?
>>>>
>>>> Yes, thanks for catching this. The original test had probe for rgb so I
>>>> was
>>>> not careful when adding alpha.
>>>>
>>>> // Tapani
>>>>
>> _______________________________________________
>> Piglit mailing list
>> Piglit at lists.freedesktop.org
>> http://lists.freedesktop.org/mailman/listinfo/piglit



More information about the Piglit mailing list