Bug 62710

Summary:	[IVB bisected]x11perf performance reduced by 40%
Product:	DRI	Reporter:	meng <mengmeng.meng>
Component:	DRM/Intel	Assignee:	Intel GFX Bugs mailing list <intel-gfx-bugs>
Status:	CLOSED NOTABUG	QA Contact:	Intel GFX Bugs mailing list <intel-gfx-bugs>
Severity:	normal
Priority:	medium
Version:	unspecified
Hardware:	All
OS:	Linux (All)
Whiteboard:
i915 platform:		i915 features:

Description meng 2013-03-25 03:38:20 UTC

System Environment:       
--------------------------
Platform:IVB
Libdrm: (master)libdrm-2.4.42-7-g36a2daad2416ad55a859c483b0d7ed93a5eff6e0
Mesa:   (9.1)e6616948b74531eae3137585b3ae35a1fc0f8174
Xserver:(server-1.13-branch)xorg-server-1.13.3
Xf86_video_intel:(master)2.21.5-1-g9f8e86800345d19785eae2dd2099dd7d61a7a3b3
Kernel: (drm-intel-fixes) c12aba5aa0e60b7947bc8b6ea25ef55c4acf81a4

Bug detailed description:
-----------------------------
x11perf performance reduced by 40% on IVB. The problem exists on both xinit and gnome-session. It's kernel regression, by bisected show 3b4f819d5eac94ba8fe5 is the first bad commit.

commit 3b4f819d5eac94ba8fe5e8c061f6dabfe8d7b22c
Author: Takashi Iwai <tiwai@suse.de>
Date:   Mon Mar 11 18:40:16 2013 +0100
    Revert "drm/i915: try to train DP even harder"

Performance data
-------------------
x11perf -aa10text on IVB.
                   gnome-session         xinit               
good(3118a)        3620000.0/sec        4230000.0/sec       
bad(3b4f81)        2300000.0/sec        2480000.0/sec

Reproduce steps:
----------------------------
1, xinit&
2, vblank_mode=0 x11perf -aa10text

Comment 1 Chris Wilson 2013-03-25 09:39:10 UTC

Still in excess of 20MGlyphs/s here.

Comment 2 Jani Nikula 2013-03-25 13:10:07 UTC

(In reply to comment #0)
> x11perf performance reduced by 40% on IVB. The problem exists on both xinit
> and gnome-session. It's kernel regression, by bisected show
> 3b4f819d5eac94ba8fe5 is the first bad commit.
> 
> commit 3b4f819d5eac94ba8fe5e8c061f6dabfe8d7b22c
> Author: Takashi Iwai <tiwai@suse.de>
> Date:   Mon Mar 11 18:40:16 2013 +0100
>     Revert "drm/i915: try to train DP even harder"

That commit seems very unlikely to cause performance regressions.

Comment 3 meng 2013-03-26 06:39:11 UTC

The kernel configure changed cause the x11perf performance reduced, so close it.

Comment 4 Daniel Vetter 2013-03-26 12:32:38 UTC

Can you please attach the diff in .config which causes this? This isn't the first time where a random .config change caused a performance regression, we need to get this kinds of effects under control.

Since if we can't control for those random 40% spikes in different directions, performance testing isn't that useful.

Comment 5 meng 2013-03-27 01:44:28 UTC

It's caused by the kernel configure “CONFIG_PROVE_LOCKING” that is the same as bug 54795.

Comment 6 meng 2014-07-18 02:29:55 UTC

Closed it.

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.