Bug 91544

Summary:	[i965, regression, bisected] regression of several tests in 93977d3a151675946c03e
Product:	Mesa	Reporter:	Dylan Baker <baker.dylan.c>
Component:	glsl-compiler	Assignee:	Ian Romanick <idr>
Status:	RESOLVED FIXED	QA Contact:	Intel 3D Bugs Mailing List <intel-3d-bugs>
Severity:	normal
Priority:	medium	CC:	kenneth, mark.a.janes, t_arceri
Version:	git
Hardware:	x86-64 (AMD64)
OS:	Linux (All)
Whiteboard:
i915 platform:		i915 features:

Description Dylan Baker 2015-08-03 23:33:44 UTC

piglit.spec.glsl-1_50.execution.geometry.clip-distance-bulk-copy                                                                                                                                      
piglit.spec.glsl-1_50.execution.geometry.clip-distance-in-values                                                                                                                                    
piglit.spec.glsl-1_50.transform-feedback-type-and-size                                                                                                                                                         
piglit.spec.glsl-1_50.linker.intrastage-pervertex-in-redeclaration-unneeded                                                                                                               
piglit.spec.glsl-1_50.execution.geometry.clip-distance-in-explicitly-sized                                                                                                               
piglit.spec.arb_separate_shader_objects.validateprogrampipeline
piglit.spec.glsl-1_50.execution.geometry.clip-distance-in-bulk-read
piglit.spec.glsl-1_50.execution.gs-redeclares-pervertex-out-only

This seems like a red herring, but I can reproduce this on my HSW system personally. Note that some of the tests fail sporatically

Comment 1 Ian Romanick 2015-08-04 01:40:38 UTC

Ken mentioned similar failures that bisected to 7ac946e546bba440f87ce95ef022745201744f9c.

There's probably some field that was getting initialized to zero previously that's getting a non-zero value now.  I'll take a look at it tomorrow.

Comment 2 Ian Romanick 2015-08-04 01:55:34 UTC

Just skimming the code, I'll bet a quarter that the problem is stream being initialized to -1 instead of 0.  ir_variable has become such a complete disaster. :(

Comment 3 Ian Romanick 2015-08-05 15:55:51 UTC

*** Bug 91565 has been marked as a duplicate of this bug. ***

Comment 4 Ian Romanick 2015-08-05 15:56:14 UTC

There is a patch on the list for this:

http://lists.freedesktop.org/archives/mesa-dev/2015-August/090803.html

Comment 5 Michel Dänzer 2015-08-06 02:58:30 UTC

Module: Mesa
Branch: master
Commit: f7ac4ef4eeea737115d0b574fed7ecae46426072
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=f7ac4ef4eeea737115d0b574fed7ecae46426072

Author: Michel Dänzer <michel.daenzer@amd.com>
Date:   Wed Aug  5 18:17:14 2015 +0900

glsl: Initialize patch member of glsl_struct_field

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.