[Piglit] [PATCH 1/3] fbo-blending-formats: Enable testing for channels with 0 bits of data.

[Marek Olšák]

On Wed, Jan 16, 2013 at 12:07 AM, Carl Worth <cworth at cworth.org> wrote:
> Marek Olšák <maraeo at gmail.com> writes:
>> The series is: Reviewed-by: Marek Olšák <maraeo at gmail.com>
>
> Thanks for the review, Marek.
>
>> I hope other tests are not broken by this change in the util code.
>
> That's a good point. I do expect my commits to cause some test results
> to change from PASS to FAIL, (since the whole point of this exercise was
> to improve the tests to catch some bugs I know exist in Mesa). But I
> need to be careful I'm not introducing any new bugs in tests.
>
> The original bug I was trying to exercise is the "DST_ALPHA bug" in i965
> as fixed in my recent commit 258453716f001eab1288d9976521 to mesa.
>
> After I changed the tests I also noticed some failures due to a "float
> ReadPixels bug". This is where glReadPixels give values other than 1.0
> for the alpha channel when reading from a floating-point format without
> an alpha channel. I haven't chased down the cause of this bug in mesa
> yet, but I did open a new bug report for it here:
>
>         glReadPixels doesn't smash alpha to 1.0 with RGB float formats
>         https://bugs.freedesktop.org/show_bug.cgi?id=59442
>
> From doing a run over quick.tests (on i965 SNB) now I see the following
> 8 new test failures after my patches:
>
> ARB_texture_rg/fbo-blending-formats
> ARB_texture_rg/fbo-blending-formats-float
>
>   These two look like they have failures only with DST_ALPHA. These are
>   apparently not fixed with my recent commit to mesa master. This could
>   require some additional fixing to the expected values in the tests
>   like I did in patch 3/3.
>
> ARB_texture_float/fbo-blending-formats
> ARB_texture_float/fbo-colormask-formats
> EXT_framebuffer_object/fbo-blending-formats
> EXT_texture_snorm/fbo-blending-formats
>
>   These four all look like the "float ReadPixels bug", (with a little
>   DST_ALPHA-specific bugginess thrown in as well). So if I can clear up
>   the DST_ALPHA stuff, these are expected failures.
>
> EXT_texture_snorm/fbo-colormask-formats
>
>   This one looks strictly like the "float ReadPixels" bug. So that's an
>   expected failure.
>
> EXT_texture_snorm/fbo-alphatest-formats
>
>   For this one, I'm not familiar with the test yet, and the output from
>   the failure doesn't look like anything I recognize:
>
>         Using test set: GL_EXT_texture_snorm
>           FAIL when testing window result, 1: 0.2 < 0.25.
>           FAIL when testing window result, 3: 0.6 > 0.55.
>           FAIL when testing window result, 4: 0.9 > 0.1.

The first FAIL report says that the input pixel value was 0.2, the
function was GL_LESS, and the alpha ref. value was 0.25, and the
driver/hardware did the opposite (discarded the pixel).

Mark