[Piglit] [PATCH 0/8] piglits for arb_texture_multisample

Paul Berry stereotype441 at gmail.com
Mon Jan 7 14:40:47 PST 2013


On 5 January 2013 01:07, Chris Forbes <chrisf at ijw.co.nz> wrote:

> These cover most of the behavior specified by ARB_texture_multisample.
>
> _______________________________________________
> Piglit mailing list
> Piglit at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/piglit
>

This series looks like a great start.  I've sent out some comments on
individual patches, mostly concerning places where I think we need to
generalize the tests (e.g. to support sample counts other than 4).

In addition, I think we should add the following tests:

(1) A test to verify that the sample positions returned by SAMPLE_POSITION
actually match those used for rasterization.  For example, the test could
loop over all sample numbers, query SAMPLE_POSITION for that sample number,
then render a very tiny quad (say, 1/100 of a pixel in size) centered at
that SAMPLE_POSITION, and verify using texelFetch() that only the one
sample was affected.  I think it would be sufficient for this test to only
run when TEXTURE_FIXED_SAMPLE_LOCATIONS is true.

(2) A test to verify that rendering works properly for both very small (e.g
16x16) and large textures.  Rationale: on i965, we normally turn off tiling
for very small textures/renderbuffers.  However, for multisampled
textures/renderbuffers, tiling is required; we need to make sure the
implementation gets this right, otherwise the results of rendering to a
small multisampled buffer will be scrambled.

(3) A test to verify that the limits advertised by
MAX_COLOR_TEXTURE_SAMPLES, MAX_DEPTH_TEXTURE_SAMPLES, and
MAX_INTEGER_SAMPLES are actually respected by the hardware.  (For example,
query MAX_COLOR_TEXTURE_SAMPLES, then try to create a multisampled color
texture with that number of samples, and verify that it succeeds).

It would be ok with me if you wanted to do these additional tests as a
follow-up series, rather than insert them into the existing series.

Thanks again for working on this, Chris--it seems like we're getting close
to some code we can commit to master.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/piglit/attachments/20130107/0600d976/attachment.html>


More information about the Piglit mailing list