Still having incredibly frustrating problems on Athlon (MOLPRO bug #51)
Dave Moore
dtmoore at email.unc.edu
Sat May 26 17:15:11 BST 2001
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Hello,
Ok .. I posted a message about random crashes on my
Athlon-Linux-Redhat-7.0 platform a few weeks ago (see also MOLPRO
Bug-report #51), and several people wrote me back and said that it
was likely to be a hardware problem. So I began an *EXTENSIVE* test
of hardware components involving much swapping (and swearing), and
nothing seemed to help. So I began to doubt that it was a hardware
problem. As a final check, I set up a loop to compile the Linux
kernel over and over until a crash was detected ... I used "make -j
20" to force lots of virtual memory swapping and really give the
memory and hard disks a workout. This job ran flawlessly for 2 days
in a row .. my MOLPRO crashes usually happened after a few hours.
Therefore, I no longer believe that this is a hardware problem, and
I now think it is almost certainly a compiler bug or a MOLPRO bug. I
am now trying different g77 versions (I don't think this problem came
up until I started using gcc-2.59.3). I am also examining the
effects of reducing the optimization levels .. I noticed that the
default optimizations in the CONFIG file is now -O3 instead of -O2.
I have run through several combinations so far without success.
One other point that might be important ... this machine uses IBM
SCSI drives with an Adaptec Ultra 160 controller .. has anyone
successfully used these components with a Linux/Athlon setup before?
One last thought .. I noticed that MOLPRO (and g77) detect my
processor as an i686 rather than an Athlon or AMDK7 chip .. is it
possible that there are some i686-optimizations that might cause
problems with an Athlon? I have looked several places on the
internet for info about this and I haven't found anything yet ... I
noticed that there do seem to be some athlon specific flags in gcc,
but the documentation on them seems sketchy .. I haven't been able to
find any details yet.
Please, if anyone has any information .. or even wild crazy hunches
that might help me, let me know .. I am at my wits end with this!
Dave Moore
-----BEGIN PGP SIGNATURE-----
Version: PGPfreeware 7.0.3 for non-commercial use <http://www.pgp.com>
iQA/AwUBOw/Wj15cg+YcPBRAEQIfXACfWjHSHpzPYuoE68Kpou1fVPoNon4AoISJ
yS5dEws7DSjyCRBzkOtd0x2C
=4cfQ
-----END PGP SIGNATURE-----
More information about the Molpro-user
mailing list