Bug 96252 - [SKL] GPU hang when replaying trace with SIMD32 + spilling + ?
Summary: [SKL] GPU hang when replaying trace with SIMD32 + spilling + ?
Status: RESOLVED FIXED
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/DRI/i965 (show other bugs)
Version: git
Hardware: Other All
: medium normal
Assignee: Jordan Justen
QA Contact: Intel 3D Bugs Mailing List
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2016-05-27 22:36 UTC by Ilia Mirkin
Modified: 2016-05-28 18:52 UTC (History)
0 users

See Also:
i915 platform:
i915 features:


Attachments

Description Ilia Mirkin 2016-05-27 22:36:05 UTC
Not sure what bits are required to mess things up, but with Curro's and Jordan's patches (jljusten's branch to enable GL 4.3), I get hangs on SKL:

[2782068.040596] [drm] stuck on render ring
[2782068.041487] [drm] GPU HANG: ecode 9:0:0x85ddfffb, in glretrace [2257], reason: Ring hung, action: reset
[2782068.961700] drm/i915: Resetting chip after gpu hang
[2782070.036607] [drm] RC6 on
[2782075.040687] [drm] stuck on render ring
[2782075.041542] [drm] GPU HANG: ecode 9:0:0x85ddfffb, in glretrace [2257], reason: Ring hung, action: reset
[2782075.043839] drm/i915: Resetting chip after gpu hang
[2782077.044990] [drm] RC6 on

This happens when replaying the trace from bug 94858 on my SKL GT2 (0x1912 pci id). Ken was able to reproduce the hangs, said something about barriers going crazy, but didn't make further progress. Note that the trace does contain some shaders that require SIMD32.
Comment 1 Kenneth Graunke 2016-05-27 22:51:53 UTC
The simulator gets very cranky about barriers - something about barrier ID being unset, and there being zero threads (so, one thread?)...but it still seems to hang even with barriers commented out.
Comment 2 Jordan Justen 2016-05-28 02:07:04 UTC
I found a bug in my merging of curro's SIMD32 and my
push constant/local IDs branches. Can you re-fetch and
retry my cs-simd32+ids branch?
Comment 3 Kenneth Graunke 2016-05-28 09:32:33 UTC
Works with Jordan's updated branch - renders correctly and no hangs.
Comment 4 Ilia Mirkin 2016-05-28 13:33:04 UTC
I won't be able to re-test until next week. Sorry for the false alarm! Feel free to close, or leave open until SIMD32 fully lands. Your call.


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.