Bug 94715

Summary:	dEQP-ES31 debug tests issue quality warnings due to shader-db messages
Product:	Mesa	Reporter:	Kenneth Graunke <kenneth>
Component:	Drivers/DRI/i965	Assignee:	Ian Romanick <idr>
Status:	RESOLVED NOTABUG	QA Contact:	Intel 3D Bugs Mailing List <intel-3d-bugs>
Severity:	normal
Priority:	medium	CC:	mattst88
Version:	git
Hardware:	Other
OS:	All
Whiteboard:
i915 platform:		i915 features:
Bug Depends on:
Bug Blocks:	94448

Description Kenneth Graunke 2016-03-26 09:33:31 UTC

Most of the dEQP-GLES31.functional.debug.error_filters.* and error_groups.* tests raise a "Quality Warning" error (resulting in a "warn" status when run via Piglit).  It appears that this is related to driver internal debug messages, such as the ones we print with shader compiler statistics for shader-db.  Commenting those out causes the tests to pass with no warnings.

This may or may not be a bug, but it would be worth figuring out why this happens, and if there's something we ought to be doing differently.

Note that these tests must be run with --deqp-gl-context-flags=debug.

Comment 1 Iago Toral 2016-10-21 13:11:01 UTC

I only see one such warning in the set mentioned groups of tests, this is running on HSW with Mesa master 17353ef043beede. The particular test case affected is:

dEQP-GLES31.functional.debug.error_filters.case_23

The problem is that the test runs twice, once without debug filters and then again with debug filters applied and checks that messages have been filtered properly. When we generate shader code, we emit debug messages with fragment and vertex shader code stats (see the call to compiler->shader_debug_log() in both vec4 and fs generators). These are logged as message type other when shaders are compiled on the first (unfiltered) run of the test, but on the second (filtered) run, they are not emitted because the shaders are already compiled and do not need to be compiled again. As a result, the debug messages with the copilation stats are not emitted and dEQP notices the change.

The reason why dEQP raises this as a warning is because with the filters used in the second run of the test, dEQP does not disable emission of debug messages of type OTHER, so even when it does not know why they were generated to begin with, it warns that they are going away with the filters on and they probably shouldn't (since it is not trying to filter them away).

In other words, dEQP does not know if the messages go away because of correct reasons or because we have a bug with the filters. In this case, it is for correct reasons and we don't need to fix anything.

I suggest that we close this as NOTABUG or WONTFIX at least if someone else can verify that this is indeed the only warning we see for the groups of tests mentioned in the bug report.

Comment 2 Mark Janes 2016-10-21 17:18:32 UTC

"Quality Warning" is a pass with respect to android requirements.

Comment 3 Kenneth Graunke 2016-10-25 06:16:41 UTC

Thanks for investigating, Iago!  It's nice to know why this is happening.  I agree, there's nothing to fix.

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.