Summary: | [Regression, bisected] spec.ext_transform_feedback.structs_gles3 tests fail | ||
---|---|---|---|
Product: | Mesa | Reporter: | Mark Janes <mark.a.janes> |
Component: | Drivers/DRI/i965 | Assignee: | Ian Romanick <idr> |
Status: | RESOLVED FIXED | QA Contact: | Intel 3D Bugs Mailing List <intel-3d-bugs> |
Severity: | normal | ||
Priority: | medium | CC: | mark.a.janes, t_arceri |
Version: | git | ||
Hardware: | x86-64 (AMD64) | ||
OS: | Linux (All) | ||
Whiteboard: | |||
i915 platform: | i915 features: |
Description
Mark Janes
2015-06-13 00:17:01 UTC
I've submitted a patch for piglit to stop running some of the test under gles here: http://lists.freedesktop.org/archives/piglit/2015-June/016235.html It seems desktop specs before 4.40 used the same line as the ES 3.0 spec, but the test don't fail to compile on closed source Nvidia drivers (although they do seem to fail execution). However arrays of structs or structs that contain arrays or structs are clearly allowed from 4.40 on. My head hurts, Ian please help. (In reply to Timothy Arceri from comment #2) > It seems desktop specs before 4.40 used the same line as the ES 3.0 spec, > but the test don't fail to compile on closed source Nvidia drivers (although > they do seem to fail execution). > > However arrays of structs or structs that contain arrays or structs are > clearly allowed from 4.40 on. > > My head hurts, Ian please help. Basically, the ES group interpreted the line differently than the desktop group. There was discussion about this while bringing arrays-of-arrays to ES, and that precipitated the clarification in the desktop spec. Frankly, unless there is a conformance test or a dEQP test that requires the restricted behavior, I'd rather not start enforcing it. Since the code is already landed, it's probably not worth the effort to undo it. My best guess is that we should just remove the gles3 tests that require out-of-spec behavior. Also... if a patch series is known to cause a big pile of tests to fail (you did run piglit before pushing, right?), that needs to be resolved first. (In reply to Ian Romanick from comment #3) > Basically, the ES group interpreted the line differently than the desktop > group. There was discussion about this while bringing arrays-of-arrays to > ES, and that precipitated the clarification in the desktop spec. Thanks for clearing this up Ian. > > Frankly, unless there is a conformance test or a dEQP test that requires the > restricted behavior, I'd rather not start enforcing it. Since the code is > already landed, it's probably not worth the effort to undo it. I added it because it seemed related to arrays of arrays, if you can't have an array of structs or nested structs then you can't have arrays of arrays of nested structs. I don't mind undoing it if you think its a good idea. > My best > guess is that we should just remove the gles3 tests that require out-of-spec > behavior. On the up side all the tests have a desktop version so not much is lost. > > Also... if a patch series is known to cause a big pile of tests to fail (you > did run piglit before pushing, right?), that needs to be resolved first. Yeah I didn't know I'd cased so many tests to fail until Mark pointed it out on IRC. Turns out I didn't have ES enabled in my piglit build. Piglit fixed with this commit f85bcce2be1f729d1d6b3f2b5738b17635a569c2 |
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.