xorg (much) slower than xfree ?

[Egbert Eich]

Stuart Kreitman writes:
 > Interesting. We should start discussing a reference platform(s) and 
 > stable benchmark method.
 > 
 > For starters, lets talk about more than 1 reference configuration, ie 
 > 32bit, 64bit processor, several
 > graphics engines, and the discontinuities introduced by post-core 
 > rendering models.
 > 

This problem only existed in the neomagic driver which is not a 
heavily used platform any more today. The problem was introduced
with the pseudocolor emulation. 
It is definitely true that we need to have reference platforms
and need to do build, spec conformity and performance regression
testing on a regular basis. However I doubt that we will be able 
to catch such corner cases as these platforms are not in widespread
use.
The dummy driver could serve as a reference platform as it should run
everywhere and it can be used to test a wide range of test cases
(apart from driver issues).
And if used as a test driver the reference implementation of certain
features could be done there.
We then need to look for people who are willing to donate CPU cycles
on a wide range of platforms.
On Linux we have:
x86
PPC
Sparc32
S390
x86-64
ia64
PPC64
Sparc64
S390x
Alpha
...

It would help to have a 32bit and 64bit system for each endianess.
A x86-64 system would cover both 32bit and 64bit for LE (it needs to be 
rebooted as it should support the apporpirate machine size natively).
I suppose you could provide both Sparc32 and Sparc64 systems for BE.

Cheers,
	Egbert.


Egbert.