[Piglit] [PATCH] hiz, new depth test

[Chad Versace]

On 11/08/2012 01:52 PM, Eric Anholt wrote:
> Chad Versace <chad.versace at linux.intel.com> writes:
> 
>> On 11/05/2012 12:09 PM, Chad Versace wrote:
>>> On 11/05/2012 12:04 PM, Chad Versace wrote:
>>>> On 10/31/2012 06:38 AM, Tapani Pälli wrote:
>>>>> Hello,
>>>>>
>>>>> Here is a new test that tries to demonstrate failure with hiz code when
>>>>> frambuffer depth and color do not match. This failure happens for example
>>>>> with MESA_FORMAT_RGB565 color buffer and MESA_FORMAT_X8_Z24 depth buffer.
>>>>>
>>>>> This test seems to fail, however my suggested fix does not make it pass
>>>>> either so I am a bit puzzled if I am testing here the right thing. My
>>>>> suggested fix was to not use fast depth clear in this case which produces
>>>>> good visual results with test applications, however this piglit test does
>>>>> not pass.
>>>>>
>>>>> Any help appreciated;
>>>>
>>>> I suspect that the test failure may be unrelated to hiz. The failing probes are
>>>> *color* probes. I've ran the test in non-auto mode, and, at least visually,
>>>> the color at the failing probe points are correct.
>>>>
>>>> Maybe Piglit or Mesa is handling RGB565 incorrectly? I'm still looking into it.
>>>
>>> I found it. Adding `piglit_set_tolerance_for_bits(5, 6, 5, 0)` to the top of
>>> piglit_display() fixes the test.
>>>
>>> On what gen do you expect the test to fail? It passes for me on Sandybridge.
>>
>> Tapani,
>>
>> Do you have commit access to Piglit? I can commit this test for you, with the
>> tolerance fix, if you don't.
> 
> If the test is passing, then it's not testing the issue he's trying to
> get at, right?

IIRC, the bug occured on one gen but not the other, where the gens here are 6
and 7. Maybe I got lucky and ran the test on the nonproblematic gen. Or maybe
I memory fails me.