[Dri-users] Re: ATI Mobility M6 artefacts/lockups (2nd try)

[Michel <mi...@da...>]

On Sam, 2002-09-07 at 18:13, Didier Moens wrote:
>=20
> Michel D=E4nzer wrote:
>=20
> >On Fre, 2002-09-06 at 02:10, Jos=E9 Fonseca wrote:=20
> > =20
> >>I'm not sure what's the DGA support status of the existing DRI drivers.
> >>Please check if there is a way to disable DGA support on VMWARE
> >>and try it to see if that problem is really DGA related. If not on
> >>VMWARE, it's possible to turn off DGA extension support on XF86Config-4=
=20
> >>(I don't know it by heart by I can find that out for you).
> >
> >Maybe DGA isn't broken but 2D acceleration with DRI enabled.
>=20
> I have no idea why the screen corruption seems to :
> - be mainly limited to the Windows mouse pointer vicinity ;
> - be induced when a specific action is performed (e.g. dragging a=20
> window, scrolling a window, ...) ;
> - only occur when DRI is enabled.

Please try if http://penguinppc.org/~daenzer/DRI/radeon-dga-sync.diff
helps; it could explain at least the last item.


> >>>* glxgears performance is limited to approx. 500 fps ; this seems rath=
er
> >>>below par to me, wrt. to benchmarks of equivalent Radeon desktop model=
s
> >>>(700-1000 fps).
> >
> >glxgears isn't really a meaningful benchmark, 4.2.0 was slower than the =
current
> >DRI drivers, and Mobility chips are slower than desktop chips.
>=20
> I do not see any improvement between 4.2 and current DRI : 500 fps (AGP 1=
x).

Hmm, the M6 doesn't benefit from the important new features like
hardware TCL and custom vertex formats. Page flipping (Option
"EnablePageFlip") should improve performance though.

> Switching from AGP 1x to AGP 2x yields an performance improvement of=20
> approx. 5%.

That's to be expected.

> Switching to AGP 4x (which, according to hardware and Windows specs,=20
> should be supported) locks the machine hard.

Higher AGP transfer rates tend to cause instabilites, for little gain as
you've found out.


> >>>/ Using Xree86 20020829 :
> >>>
> >>>* same screen corruption as with stock XFree 4.2 ;
> >
> >Can you try again with a 20020906 or later snapshot? I've cleaned up 2D
> >acceleration with DRI enabled and fixed some bugs.
>=20
> Tested with DRI 20020907 ; no improvement with DRI enabled.  Would you=20
> like a screenshot of the corruption ?

Might be interesting if the above patch doesn't help.


> >>>* with the DRI nightly binary driver packages, when running glxinfo,
> >>>glxgears, ... I get a message about TCL being disabled, after which a
> >>>segmentation fault occurs. All pointers to libGL.so seem correct ;
> >
> >Where did that libGL.so come from?
>=20
> DRI CVS 20020830.

I thought it worked with that?


> >>>* with a freshly compiled DRI CVS20020830, I can run glxinfo (please
> >>>find it below, LIBGL_DEBUG=3Dverbose), but glxgears opens a window wit=
h
> >>>some (perhaps not so) random garbage, and locks the computer hard (no
> >>>Alt-SysRq).
> >
> >That was a known problem with chips without a TCL unit. It's been fixed
> >for VEs at least, we're still looking for feedback with an M6. Again,
> >please try a current DRI snapshot and post your experience to the
> >dri-devel list.
>=20
> WFM with DRI 20020907, no more hard locks or segfaults

Great news! It seems the driver finally works with all r100 chips again.


> With DRI 20020907, when rebooting the machine, I get an OOPS in the=20
> shutdown sequence, I presume either in agpgart.o or radeon.o (sorry,=20
> couldn't capture it).

If it happens when the X server gets killed, I might have an idea what's
up.


--=20
Earthling Michel D=E4nzer (MrCooper)/ Debian GNU/Linux (powerpc) developer
XFree86 and DRI project member   /  CS student, Free Software enthusiast