Need help dissecting Intel GM45 regression

bbarcher · 2014-04-15 09:35:33

Performance regression on Gen4 Intel GM45 GMA4500MHD
https://bugs.freedesktop.org/show_bug.cgi?id=77372

Suspected components:
graphics stack
Gnome shell

Note:
The maintainer could not reproduce the bug on his Intel GM45 system.

It's beyond my skills to handle this on my own. I'd be really grateful if someone could assist me. I know that in arch/pacman partial upgrades are not supported so I don't even know whether it is possible to conduct the troubleshooting on arch. Even if not, I would still seek help here, among archers.

To give you an idea, if I go for a 'we use fedora for this and first we upgrade just gnome shell' I wouldn't be sure what package names to give to yum to achieve this, e.g. should it be the entire gnome group? I would know how to blacklist some package to prevent it from upgrading though.

Last edited by bbarcher (2014-05-05 19:58:01)

drcouzelis · 2014-04-15 18:25:41

Is direct rendering enabled? What is the output of "glxinfo | grep render"? You focus a lot on GNOME3. Do other applications experience the poor performance?

The first step is to downgrade packages until the desired behaviour returns.

Does that bug report say that the time acceleration worked correctly it was in Feb 2013? Is that correct?

bbarcher · 2014-04-15 19:54:26

Thank you for your reply.

drcouzelis wrote:

Is direct rendering enabled? What is the output of "glxinfo | grep render"?

# glxinfo | grep render
direct rendering: Yes
    GLX_MESA_multithread_makecurrent, GLX_MESA_query_renderer, 
    GLX_MESA_multithread_makecurrent, GLX_MESA_query_renderer, 
OpenGL renderer string: Mesa DRI Mobile Intel® GM45 Express Chipset 
    GL_MESA_ycbcr_texture, GL_NV_blend_square, GL_NV_conditional_render,

drcouzelis wrote:

You focus a lot on GNOME3. Do other applications experience the poor performance?

Yes, e.g. scrolling in other apps is also affected. Haven't checked on windowed games yet.

drcouzelis wrote:

The first step is to downgrade packages until the desired behaviour returns.

This I know. What I don't know is the >list of packages< for each >investigated component<. Investigated components being:
1. Graphics stack.
(and its sub-components)
1a. Intel drivers.
1b. other components of graphics stack (?)
2. Kernel.
3. Gnome shell (I assume mutter, clutter etc. are integral parts of Gnome shell).

Another thing I don't know is the order in which I should downgrade the components. My noob guts tell me we should rule out gnome first.

Also, wouldn't it be easier to install clean Fedora 18, which ran smoothly, and then upgrade one component at a time instead of doing downgrades? It would require just 'yum upgrade <package name/group name>'. AFAIK downgrades are potentially more troublesome. Especially for a noob.

drcouzelis wrote:

Does that bug report say that the time acceleration worked correctly it was in Feb 2013? Is that correct?

If I remember correctly, yes, that's correct. Clean install of Fedora 18. Before Fedora 19 has been released some updates*** to Fedora 18 degraded the performance and it never got back to the old performance level on my system since then.

***read as: some linux components updated upstream at the time, i.e. it was not and it is not Fedora-specific)

Last edited by bbarcher (2014-04-15 19:54:57)

drcouzelis · 2014-04-15 20:28:18

bbarcher wrote:

drcouzelis wrote:
You focus a lot on GNOME3. Do other applications experience the poor performance?
Yes, e.g. scrolling in other apps is also affected. Haven't checked on windowed games yet.
Another thing I don't know is the order in which I should downgrade the components. My noob guts tell me we should rule out gnome first.

That's easy enough. Start the X server with a window manager other than GNOME (such as Openbox) and check the performance. It would be good if you could collect data to compare the two environments, such as FPS in a video game. Anything more than "It FEELS sluggish...".

Also, wouldn't it be easier to install clean Fedora 18, which ran smoothly, and then upgrade one component at a time instead of doing downgrades?

Yes. I've done stuff like that before. If you can find the package that causes the problem and then the actual version where the problem started, I can almost guarantee that the developers will fix the bug for you.

Anyway, I agree that it would be difficult to do in Arch Linux. February 2013 is a LONG time ago in a rolling release distribution.

Pank · 2014-04-15 20:54:29

For what it is worth, I can "reproduce" in the sense that Gnome 3.12 also feels slower than previous versions on my PC with the same Intel graphics card. I can test whatever in i3 and Gnome 3.12 quite easily, though I'm not sure which tests to conduct.

bbarcher · 2014-04-16 17:55:32

Pank wrote:

For what it is worth, I can "reproduce" in the sense that Gnome 3.12 also feels slower than previous versions on my PC with the same Intel graphics card. I can test whatever in i3 and Gnome 3.12 quite easily, though I'm not sure which tests to conduct.

I see that performance got *even worse* on my system with Gnome 3.12 but please let's focus on the bug report linked in the OP, so pre-3.12.

Pank · 2014-04-16 19:03:29

Archer,
You are right. Turns out performance was even worse in i3 than in Gnome 3.12; I just didn't notice. Going from linux-ck-core2 v3.13.10-1 to linux-ck-core2 v3.12.13-1 restores great performance on my system (X200s). Will it help if I track down a Linux version where this occurs? If so I can look through the official Arch builds to track down where I start loosing performance.

Note, I'm not sure how to test the graphics stack in a solid manner. I use Gnome shell animation as well as playing a HD video. With V3.13.10-1 I'm not able to play the video in i3; it barely plays in gnome 3.12. In 3.12.13 I still use around 80% decoding the video but everything is smooth, and I can almost go back and forth from overview while the video is playing.

bbarcher · 2014-04-22 20:02:10

drcouzelis wrote:

That's easy enough. Start the X server with a window manager other than GNOME (such as Openbox) and check the performance. It would be good if you could collect data to compare the two environments, such as FPS in a video game. Anything more than "It FEELS sluggish...".

OK, got back from Easter holidays and I think I could give it a go. What reasonably-sized free linux game has a built-in benchamrking mode. I'd like to avoid giving bare FPS readout to devs as it may turn out not really more useful then 'sluggish/not sluggish.'

Also, is a windowed game really suited for troubleshooting desktop performance? Please keep in mind that the number one suspect is the partial screen redraws function of Gnome Shell (clutter?). I wouldn't want to measure with something that actually bypasses any of the investigated components (see post #3).

Last edited by bbarcher (2014-04-22 20:04:52)

bbarcher · 2014-05-04 19:39:57

I ran multiple quick benchmarks using Xonotic (full-screen versus high-res windowed) on Gnome and LXDE, Fedora 18, 19, recent Arch and all seem to indicate: no problems with content, problems with animations of actual desktop elements. Long story short I logged in to KDE session on Fedora 19 install affected by the performance regression. Result? Perfectly smooth animations! So it's the Gnome's fault.

Finally I have something conclusive I can show at the Fedora mailing lists. (haven't done it yet ; tomorrow)

Last edited by bbarcher (2014-05-07 06:21:44)

bbarcher · 2014-05-07 06:23:26

Update:
I have posted on Fedora devel mailing list. Here's the link:
https://lists.fedoraproject.org/piperma … 98981.html

Arch Linux

#1 2014-04-15 09:35:33

Need help dissecting Intel GM45 regression

#2 2014-04-15 18:25:41

Re: Need help dissecting Intel GM45 regression

#3 2014-04-15 19:54:26

Re: Need help dissecting Intel GM45 regression

#4 2014-04-15 20:28:18

Re: Need help dissecting Intel GM45 regression

#5 2014-04-15 20:54:29

Re: Need help dissecting Intel GM45 regression

#6 2014-04-16 17:55:32

Re: Need help dissecting Intel GM45 regression

#7 2014-04-16 19:03:29

Re: Need help dissecting Intel GM45 regression

#8 2014-04-22 20:02:10

Re: Need help dissecting Intel GM45 regression

#9 2014-05-04 19:39:57

Re: Need help dissecting Intel GM45 regression

#10 2014-05-07 06:23:26

Re: Need help dissecting Intel GM45 regression

Board footer