Summary: | [HSW,BDW,SKL][GLES 3.1 CTS]ES31-CTS.shader_image_load_store.basic-allTargets-store-fs fail | ||
---|---|---|---|
Product: | Mesa | Reporter: | Marta Löfstedt <marta.lofstedt> |
Component: | Drivers/DRI/i965 | Assignee: | Francisco Jerez <currojerez> |
Status: | RESOLVED FIXED | QA Contact: | Intel 3D Bugs Mailing List <intel-3d-bugs> |
Severity: | normal | ||
Priority: | medium | CC: | currojerez |
Version: | unspecified | ||
Hardware: | Other | ||
OS: | All | ||
Whiteboard: | |||
i915 platform: | i915 features: | ||
Bug Depends on: | |||
Bug Blocks: | 92778 | ||
Attachments: | gen7_flush_caches_before_pipeline_select.patch |
Description
Marta Löfstedt
2015-12-10 12:40:29 UTC
imageCube appear to be broken, it is only the cases with layout and imageCube frag shader that fails. Very similar vs and cs test works. This issue is not reproducible on BDW and SKL with Jordans/Iagos patches: http://patchwork.freedesktop.org/patch/63033/ http://patchwork.freedesktop.org/patch/63034/ http://patchwork.freedesktop.org/patch/68471/ Issue on HSW remain. (In reply to Marta Löfstedt from comment #2) > This issue is not reproducible on BDW and SKL with Jordans/Iagos patches: > > http://patchwork.freedesktop.org/patch/63033/ > http://patchwork.freedesktop.org/patch/63034/ > http://patchwork.freedesktop.org/patch/68471/ > > Issue on HSW remain. I happen to have been looking into the HSW issue today. The problem seems to have nothing to do with cubemap images, the failure in the cubemap subtest seems to be fallout from incorrect switching between compute and render pipelines from the previous subtest: If you comment out all other subtests the failure disappears, and if you comment out the glDispatchCompute from the previous (passing) subtest the failure also disappears. What causes the failure from the glDispatchCompute call seems to be the switching to the GPGPU pipeline alone, the state emission and actual GPGPU walker commands seem to be irrelevant. The reason for the misrendering seems to be that we're missing the PIPE_CONTROL flushes that are necessary on Gen7 before switching pipelines (among a pile of other workarounds mentioned in the BSpec...), I'll attach a fix. Created attachment 120565 [details] [review] gen7_flush_caches_before_pipeline_select.patch Thanks Curro, Your patch fix the problem. Will you send it to the mesa list? (In reply to Marta Löfstedt from comment #5) > Thanks Curro, > > Your patch fix the problem. > > Will you send it to the mesa list? Sure, I'll send this to the mailing list as soon as I'm done with some other fixes I'm working on for the compute/render pipeline switching code. Fix for the Gen7 issue pushed as 18c76551ee425b981efefc61f663a7781df17882. |
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.