Re: [Dri-devel] Matrox G450 _PCI_ and Alpha -- partial success

[Claus-Justus H. <he...@in...>]

Claus-Justus Heine <he...@in...> writes:

> Basically, it seems to work now -- in principle. Synchronization with
> the X-Server is flaky; maybe there is a mb() issue or something like
> that.

The deadlock was caused by debugging code inserted by myself (sending
output with printf() to an xterm while holding the lock, if I
understand it right).

So it seems to work now. Textures do not work (do they work at all
without HAL lib from Matrox? Found some comments in the mga driver
which suggest they are not (yet) supported).

According to the frame-rate reported by glxgears the performance burst
is not too amazing, about factor 4 for full-screen resolution
(compared to software shading).

Single-side filled triangle are deadly slow. Smooth shading is deadly
slow. Two-side filled triangle work quite nice.

When I find the time I compare the stuff with a G400 AGP card.

The changed files (kernel modules and XFree86 4.1.0 stuff) can be
found at (in case somebody is interested)

http://www.instmath.rwth-aachen.de:8000/linux/G450-PCI

or via anonymous cvs at

cvs -d :pserver:cv...@ww...:/usr/local/cvs/linux login
User: cvs
Password: ftape
cvs -d :pserver:cv...@ww...:/usr/local/cvs/linux co G450-PCI

If somebody is interested.

I'll post patches to the relevant maintainers when I find the time.

I was using the latest mga driver from the XFree86 cvs tree. The one
distributed with 4.1.0 does not restore the text-mode on the second head.


Maybe other PCI cards could work the same way. The case was
particularly easy with the G450 because the only difference between
APG/non-AGP mode is set/don't_set a particular bit in the DMA control
registers of the G450.

Greetings

Claus

-- 
  Claus-Justus Heine             
  he...@in...
  http://www.instmath.rwth-aachen.de/~heine/

  Ftape - the Linux Floppy Tape Project
  Home Page   : http://www.instmath.rwth-aachen.de/~heine/ftape/