2004-03-28 00:57:13

by Glenn Johnson

[permalink] [raw]
Subject: 2.6.5-rc2-mm[34] causes drop in DRI FPS

The last two iterations of -mm kernels have caused my FPS count via
glxgears to drop from about 2000 fps to about 180 fps. I rebuilt X11 to
see if that would help but it did not.

This is with a Radeon 9100 AGP card and an Intel 865PE based
motherboard.



2004-03-28 02:09:35

by Andrew Morton

[permalink] [raw]
Subject: Re: 2.6.5-rc2-mm[34] causes drop in DRI FPS

Glenn Johnson <[email protected]> wrote:
>
> The last two iterations of -mm kernels have caused my FPS count via
> glxgears to drop from about 2000 fps to about 180 fps. I rebuilt X11 to
> see if that would help but it did not.
>
> This is with a Radeon 9100 AGP card and an Intel 865PE based
> motherboard.

It could be the AGP changes. Please do a `patch -p1 -R' of
bk-agpgart.patch and retest?

2004-03-28 03:18:24

by Glenn Johnson

[permalink] [raw]
Subject: Re: 2.6.5-rc2-mm[34] causes drop in DRI FPS

On Sat, 2004-03-27 at 20:09, Andrew Morton wrote:
> Glenn Johnson <[email protected]> wrote:
> >
> > The last two iterations of -mm kernels have caused my FPS count via
> > glxgears to drop from about 2000 fps to about 180 fps. I rebuilt X11 to
> > see if that would help but it did not.
> >
> > This is with a Radeon 9100 AGP card and an Intel 865PE based
> > motherboard.
>
> It could be the AGP changes. Please do a `patch -p1 -R' of
> bk-agpgart.patch and retest?

I reverted that patch and retested but it did not solve the problem. I
am still only seeing about 180 fps with glxgears.

2004-03-28 09:06:04

by Ingo Molnar

[permalink] [raw]
Subject: Re: 2.6.5-rc2-mm[34] causes drop in DRI FPS


* Glenn Johnson <[email protected]> wrote:

> > It could be the AGP changes. Please do a `patch -p1 -R' of
> > bk-agpgart.patch and retest?
>
> I reverted that patch and retested but it did not solve the problem. I
> am still only seeing about 180 fps with glxgears.

what does 'glxinfo | grep rendering' show?

Ingo

2004-03-28 16:19:34

by Glenn Johnson

[permalink] [raw]
Subject: Re: 2.6.5-rc2-mm[34] causes drop in DRI FPS

On Sun, 2004-03-28 at 03:06, Ingo Molnar wrote:

> * Glenn Johnson <[email protected]> wrote:
>
> > > It could be the AGP changes. Please do a `patch -p1 -R' of
> > > bk-agpgart.patch and retest?
> >
> > I reverted that patch and retested but it did not solve the problem. I
> > am still only seeing about 180 fps with glxgears.
>
> what does 'glxinfo | grep rendering' show?

It shows:

direct rendering: Yes

I had checked that first thing. I also verified that the AGPgart
detected the chipset, etc. Everything seems okay in the XFree86 log file
as well.

--
Glenn Johnson
[email protected]

2004-03-28 16:31:04

by Marc-Christian Petersen

[permalink] [raw]
Subject: Re: 2.6.5-rc2-mm[34] causes drop in DRI FPS

On Sunday 28 March 2004 18:17, Glenn Johnson wrote:

Hi Glenn,

> > > > It could be the AGP changes. Please do a `patch -p1 -R' of
> > > > bk-agpgart.patch and retest?
> > > I reverted that patch and retested but it did not solve the problem. I
> > > am still only seeing about 180 fps with glxgears.
> > what does 'glxinfo | grep rendering' show?

what about reverting "DRM-cvs-update.patch"?

ciao, Marc

2004-03-28 16:55:22

by Glenn Johnson

[permalink] [raw]
Subject: Re: 2.6.5-rc2-mm[34] causes drop in DRI FPS

On Sun, 2004-03-28 at 10:17, Glenn Johnson wrote:

> On Sun, 2004-03-28 at 03:06, Ingo Molnar wrote:
>
> > * Glenn Johnson <[email protected]> wrote:
> >
> > > > It could be the AGP changes. Please do a `patch -p1 -R' of
> > > > bk-agpgart.patch and retest?
> > >
> > > I reverted that patch and retested but it did not solve the problem. I
> > > am still only seeing about 180 fps with glxgears.
> >
> > what does 'glxinfo | grep rendering' show?
>
> It shows:
>
> direct rendering: Yes
>
> I had checked that first thing. I also verified that the AGPgart
> detected the chipset, etc. Everything seems okay in the XFree86 log file
> as well.

I just did a diff of my XFree86 log files with the system booted using
the 2.6.5-rc2-mm2 and 2.6.5-rc2-mm4 kernels, respectively. Here is the
diff, so maybe this will mean something.

--- XFree86-2.6.5-rc2-mm2.log 2004-03-28 10:32:00.070254034 -0600
+++ XFree86-2.6.5-rc2-mm4.log 2004-03-28 10:29:30.123163884 -0600
@@ -19 +19 @@
-(==) Log file: "/var/log/XFree86.0.log", Time: Sun Mar 28 10:31:22 2004
+(==) Log file: "/var/log/XFree86.0.log", Time: Sun Mar 28 10:28:32 2004
@@ -713,2 +713,2 @@
-(II) RADEON(0): [drm] added 8192 byte SAREA at 0xe188b000
-(II) RADEON(0): [drm] mapped SAREA 0xe188b000 to 0x48328000
+(II) RADEON(0): [drm] added 8192 byte SAREA at 0xe1889000
+(II) RADEON(0): [drm] mapped SAREA 0xe1889000 to 0x48328000
@@ -764 +764 @@
-(II) RADEON(0): [drm] dma control initialized, using IRQ 169
+(II) RADEON(0): [drm] dma control initialized, using IRQ 10



--
Glenn Johnson
[email protected]

2004-03-28 17:07:31

by Glenn Johnson

[permalink] [raw]
Subject: Re: 2.6.5-rc2-mm[34] causes drop in DRI FPS

On Sun, 2004-03-28 at 10:30, Marc-Christian Petersen wrote:
> On Sunday 28 March 2004 18:17, Glenn Johnson wrote:
>
> Hi Glenn,
>
> > > > > It could be the AGP changes. Please do a `patch -p1 -R' of
> > > > > bk-agpgart.patch and retest?
> > > > I reverted that patch and retested but it did not solve the problem. I
> > > > am still only seeing about 180 fps with glxgears.
> > > what does 'glxinfo | grep rendering' show?
>
> what about reverting "DRM-cvs-update.patch"?

I have not tried that but has that patch been in the tree for some time
now? My DRI works fine with 2.6.5-rc2-mm2, which has that patch.

--
Glenn Johnson
[email protected]

2004-03-28 19:35:36

by Zwane Mwaikambo

[permalink] [raw]
Subject: Re: 2.6.5-rc2-mm[34] causes drop in DRI FPS

On Sun, 28 Mar 2004, Glenn Johnson wrote:

> I just did a diff of my XFree86 log files with the system booted using
> the 2.6.5-rc2-mm2 and 2.6.5-rc2-mm4 kernels, respectively. Here is the
> diff, so maybe this will mean something.
>
> --- XFree86-2.6.5-rc2-mm2.log 2004-03-28 10:32:00.070254034 -0600
> +++ XFree86-2.6.5-rc2-mm4.log 2004-03-28 10:29:30.123163884 -0600
> @@ -19 +19 @@
> -(==) Log file: "/var/log/XFree86.0.log", Time: Sun Mar 28 10:31:22 2004
> +(==) Log file: "/var/log/XFree86.0.log", Time: Sun Mar 28 10:28:32 2004
> @@ -713,2 +713,2 @@
> -(II) RADEON(0): [drm] added 8192 byte SAREA at 0xe188b000
> -(II) RADEON(0): [drm] mapped SAREA 0xe188b000 to 0x48328000
> +(II) RADEON(0): [drm] added 8192 byte SAREA at 0xe1889000
> +(II) RADEON(0): [drm] mapped SAREA 0xe1889000 to 0x48328000

That's just a memory allocation, so that will never be consistent.

> @@ -764 +764 @@
> -(II) RADEON(0): [drm] dma control initialized, using IRQ 169
> +(II) RADEON(0): [drm] dma control initialized, using IRQ 10

You're not using CONFIG_PCI_USE_VECTOR in the newer kernel, but that also
wouldn't make a difference.

2004-03-28 22:50:56

by Glenn Johnson

[permalink] [raw]
Subject: Re: 2.6.5-rc2-mm[34] causes drop in DRI FPS

On Sun, 2004-03-28 at 13:35, Zwane Mwaikambo wrote:
> On Sun, 28 Mar 2004, Glenn Johnson wrote:
>
> > I just did a diff of my XFree86 log files with the system booted using
> > the 2.6.5-rc2-mm2 and 2.6.5-rc2-mm4 kernels, respectively. Here is the
> > diff, so maybe this will mean something.
> >
> > --- XFree86-2.6.5-rc2-mm2.log 2004-03-28 10:32:00.070254034 -0600
> > +++ XFree86-2.6.5-rc2-mm4.log 2004-03-28 10:29:30.123163884 -0600
> > @@ -19 +19 @@
> > -(==) Log file: "/var/log/XFree86.0.log", Time: Sun Mar 28 10:31:22 2004
> > +(==) Log file: "/var/log/XFree86.0.log", Time: Sun Mar 28 10:28:32 2004
> > @@ -713,2 +713,2 @@
> > -(II) RADEON(0): [drm] added 8192 byte SAREA at 0xe188b000
> > -(II) RADEON(0): [drm] mapped SAREA 0xe188b000 to 0x48328000
> > +(II) RADEON(0): [drm] added 8192 byte SAREA at 0xe1889000
> > +(II) RADEON(0): [drm] mapped SAREA 0xe1889000 to 0x48328000
>
> That's just a memory allocation, so that will never be consistent.
>
> > @@ -764 +764 @@
> > -(II) RADEON(0): [drm] dma control initialized, using IRQ 169
> > +(II) RADEON(0): [drm] dma control initialized, using IRQ 10
>
> You're not using CONFIG_PCI_USE_VECTOR in the newer kernel, but that also
> wouldn't make a difference.

Actually, I am using CONFIG_PCI_USE_VECTOR in both.