Michel Dänzer <mic...@daenzer.net> writes: > Are these tests unreliable themselves, or rather the driver paths > they're hitting? If it's the latter (a phenomenon which is well-known to > me with radeonsi :), neither approach seems appropriate.
Yes, that's a good question. For glsl-fs-user-varying-ff, it was probing pixels for particular values (or rather, probing for any value *other* than a particular value) after invoking undefined behavior. That was bogus and Eric patched it. For texture_integer_glsl130, the test actually computes a new random color on every run. That's a very un-useful thing for the test to do. (Though the failures at some random values might still be driver bugs, they also might have nothing to do with my patches---and that was what was annoying here.) I'll follow up on the piglit list with my ideas for addressing this bug in the test. And yes, fixing the tests is better than deleting them. My apologies for my impatience. -Carl -- carl.d.wo...@intel.com
pgpNQQhtbMJmB.pgp
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
