Created attachment 30239 [details] depth-tex-compare.png System environment: -- chipset: GM45 -- system architecture: i686 -- xserver-xorg-video-intel: 2:2.9.0-1ubuntu1 -- xserver-common: 2:1.6.3-1ubuntu7 -- mesa: db828ed7589d0a5687386c4b4268b4e7ff78c866 (master) -- libdrm: 2.4.15 -- kernel version: 2.6.31-13-generic -- Linux distribution: Ubuntu 9.10 -- Machine model: Dell Latitude E6400 Run piglit test depth-tex-compare. $ ./bin/depth-tex-compare Probe at (165,250) Expected: 1.000000 1.000000 1.000000 Observed: 1.000000 0.000000 1.000000 Left to Right: ALPHA, LUMINANCE, INTENSITY Top to Bottom: LESS, LEQUAL, GREATER, GEQUAL, ALWAYS, NEVER, NOTEQUAL, EQUAL
I confirm this failure in Q3 release.
I get the same behavior on 965 as on Nvidia hardware with their closed-source driver. I'm inclined to say that 965 is correct but the test and software Mesa are wrong. It seems odd that the GL_LESS and GL_LEQUAL rows are expected to be the same. Ditto for GL_GREATER and GL_GEQUAL. Opinions?
I have modified the test to expect the results produced on every single piece of hardware I could find. Software Mesa is the odd man out.
This is pretty hard/impossible to fix. As the R coordinate is interpolated across the triangles, the interpolation would have to _exactly_ produce R=0.5 for the middle pixel in order to satisfy the <= inequality test. That's not happening in Mesa. The closest we get is R=0.52. It looks the hardware you've tested does produce R=0.5 exactly. Perhaps Mesa's interpolation or point sampling is off a little bit somewhere, but I'm not inclined to spend a bunch of time on this or changing a bunch of code in order to fix this corner case. I'd rather fix the piglit test. We could add a "don't care" probe value to skip these corner cases, and then add a new test with r0=r1=0.5 which should guarantee hitting the R=0.5 case. I'll take a crack at that.
Created attachment 31233 [details] [review] proposed patch to depth-tex-compare.c Here's a patch for depth-tex-compare.c that adds don't-care values for the R=0.5 cases that are hard to exactly hit with Mesa swrast. It also adds additional tests for the GL_GEQUAL, GL_LEQUAL cases with R=0.5 to make sure we hit those cases in a more reliable manner.
(In reply to comment #5) > Created an attachment (id=31233) [details] > proposed patch to depth-tex-compare.c > > Here's a patch for depth-tex-compare.c that adds don't-care values for the > R=0.5 cases that are hard to exactly hit with Mesa swrast. It also adds > additional tests for the GL_GEQUAL, GL_LEQUAL cases with R=0.5 to make sure we > hit those cases in a more reliable manner. The patch no longer applies. Anyway I think it's useful to make the test less pedantic. A test that *cannot* pass on most drivers is useless.
-- GitLab Migration Automatic Message -- This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/293.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.