Bug 88912

Summary:	[IVB/BDW bisected] GLBenchmark manhattan performance reduce by 20% with gnome-session
Product:	Mesa	Reporter:	ye.tian <yex.tian>
Component:	Drivers/DRI/i965	Assignee:	Intel 3D Bugs Mailing List <intel-3d-bugs>
Status:	VERIFIED FIXED	QA Contact:	Intel 3D Bugs Mailing List <intel-3d-bugs>
Severity:	major
Priority:	high	CC:	eero.t.tamminen, valtteri.rantala
Version:	unspecified
Hardware:	All
OS:	Linux (All)
Whiteboard:
i915 platform:		i915 features:
Attachments:	Xorg.0.log info

Description ye.tian 2015-02-02 07:40:59 UTC

System Environment:       
Platform:  IVB
Kernel: (drm-intel-nightly)8b4216f91c7bf8d3459cadf9480116220bd6545e
Libdrm: (master)libdrm-2.4.59-6-g28ee135a37e10b9a6cd62d67df0332e38ee0b85c
Mesa:   (master)6094619c0297180db218e5b4e45fc07aff116413
Xserver: (master)xorg-server-1.16.99.902-4-g697b696e5e24d0679f133183a3bb0852025377c2
Xf86_video_intel: (master)2.99.917-75-g6d2754b1902e8bce37818c854fb890400b27343e
Cairo:  (master)cca8b1960b0091c53235d9dc4cde8a33010c5bbf
Libva:  (master)f9741725839ea144e9a6a1827f74503ee39946c3
Libva_intel_driver: (master)bf324e3440246a487997a1624ac862e3e4027f9e

Bug detailed description:
--------------------------------------------------
GLBenchmark manhattan performance reduce by 20% with gnome-session on IVB.
It’s Mesa regression. By bisected, show that 4c157d34 is the first bad commit.

commit 4c157d34c0163ab2389d74181121950aaf0ef121
Author:     Topi Pohjolainen <topi.pohjolainen@intel.com>
AuthorDate: Wed Jan 28 16:24:25 2015 +0200
Commit:     Topi Pohjolainen <topi.pohjolainen@intel.com>
CommitDate: Fri Jan 30 09:54:53 2015 +0200

    meta/blit: Add plumbing for shaders without depth

    Currently all blit programs are unconditionally compiled with
    gl_FragDepth.

    Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
    Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
    Reviewed-by: Matt Turner <mattst88@gmail.com>
---------------------------------------------------------------------
604ae33c is 4c157d34 parents commit.
              4c157d34  604ae33c  4c157d34 VS. 604ae33c
gnome-session  24.5           30.6       19.93%

Please see Xrog.

Reproduce steps:
----------------------------
1, xinit& 
2, gnome-session&
3, export MESA_GLSL_VERSION_OVERRIDE=400
  export MESA_GL_VERSION_OVERRIDE=4.1
4 ./mainapp -t gl_manhattan -fullscreen 1 -w 1920 -h 1080

Comment 1 ye.tian 2015-02-02 07:41:58 UTC

Created attachment 113038 [details]
Xorg.0.log info

Comment 2 ye.tian 2015-02-02 07:52:03 UTC

This bug also exists on BDW.

Comment 3 Tapani Pälli 2015-02-02 08:22:23 UTC

I'm not able to reproduce this regression, for me fps is stable both with Topi's comment and current Mesa master (6094619), around ~40 on my machine (I cannot run on full HD due to display being 1600x900) (fedora 21, 3.17.8 kernel, X.org 1.14).

Comment 4 Tapani Pälli 2015-02-02 08:25:33 UTC

Also, this particular commit cannot have any performance impact, it seems either bisect result is not correct or some other component is causing the regression.

Comment 5 ye.tian 2015-02-02 08:33:05 UTC

(In reply to Tapani Pälli from comment #4)
> Also, this particular commit cannot have any performance impact, it seems
> either bisect result is not correct or some other component is causing the
> regression.

Did you test the parents commit(604ae33c) of bisected bad commit(4c157d34)?

Comment 6 ye.tian 2015-02-02 09:35:32 UTC

Re-test results:
resolution               1600x1200(not fullscreen)   1920x1080(fullscreen)
parents commit(604ae33c)     31.5  fps                  30.6 fps

current Mesa master(6094619) 31.5  fps                  24.5 fps

Comment 7 Tapani Pälli 2015-02-02 10:10:07 UTC

Yep I've tested against parent and some older commits but I cannot seem to make the app fullscreen so that is quite a big difference here. I will try with some other machine.

Comment 8 Tapani Pälli 2015-02-03 08:06:55 UTC

Eero, could you try to reproduce?

Comment 9 Eero Tamminen 2015-02-03 13:25:38 UTC

I don't have any IVB machine.  Valtteri said that he can check whether he sees a BDW regression in Manhattan in the same time frame.

Comment 10 Kaveh 2015-02-05 00:13:37 UTC

Ken, could you please take a look at this?

Comment 11 Kenneth Graunke 2015-02-06 06:26:22 UTC

I can't reproduce this either.  I tried both current Mesa master (a338dc01866ce50bf7555ee8dc08491c7f63b585) and an older commit (fc6938d23e769497fd2acba7da0b59935ec9f1e4), and saw identical performance with each.

This was on IVB (Lenovo X230) running X server master.  I tried both xf86-video-modesetting+glamor+pageflipping and xf86-video-intel+sna.  No changes at all.

Comment 12 Kenneth Graunke 2015-02-06 06:27:15 UTC

(In reply to Kenneth Graunke from comment #11)
> This was on IVB (Lenovo X230) running X server master.  I tried both
> xf86-video-modesetting+glamor+pageflipping and xf86-video-intel+sna.  No
> changes at all.

Oh, and this was at 1920x1080, fullscreen.

Comment 13 ye.tian 2015-02-06 09:52:02 UTC

Verified this bug with latest Mesa(master)2335153f.

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.