On Thu, 2 Jul 1998, Ignatios Souvatzis wrote:

> Basically, for 68040 and 68060, you should compile your X server (and
libraries
> used for it, and maybe all of X11) with the -m68040 (or -m68060) options.
> 
> Else gcc will create floating point instructions not there on these CPUs,
> making the process much slower than you'd think it should be. (And: the 
> kernel trapping will trash the instruction cache, so it gets even slower).
> 
> I think somebody wrote this on this mailing list a couple of months ago.
> Can somebody check the archive, please?

As far as I remember, it was assumed that recompiling with the -m68040 or
-m68060 option should improve the X11 performance on 68040 or 68060 machines
for the above stated reasons, but it was not mentioned if it really does.

Therefore, the following may perhaps be of general interest:

I have recently recompiled the complete X11 system for the 68060 and made a
performance comparison of the standard version and the recompiled version
with 'x11perf -all', which does about 330 different tests. My system is an
A4000 with CyberStorm MKII 060/50 and CyberVision 64. The tests were
performed on a 1024x768x16 screen. (I did not test AGA modes, because my
monitor cannot display them.)
Most of the operations were between 1% and 20% faster with the 68060
version, some were even more fast, but a few were slower than with the
standard version. So the overall improvement is not dramatic, but on the
other hand, the performance on the CyberVision 64 is already quite good
with the standard X11 distribution. Perhaps, the effect of the recompiled
version is more significant for other graphics devices depending on CPU
usage, blitter support etc..

Regards,

Stefan Hensen

----------------------------------------------------------
Stefan Hensen
e-mail: hensen@wpos4.physik.uni-wuppertal.de