Re: libm problems: log(9.8813129168249309e-324) = -infinity
On Fri, May 09, 2008 at 09:30:55PM +0000, Mark Dickinson wrote:
> I wish I knew! It's not my machine: it's a Python buildbot that I have
> no control over. The only debugging method I have available is to
> check in a change to the Python svn repository and then wait for
> the next run of the buildbot to see the results. The buildbot only
> runs once a day, which makes this whole method of debugging
> rather painful. I'm at least partially responsible for some of the
> code that's causing Python test failures, which is why I'm trying
> to track this down.
> I'll check in some autoconf tests to report the kernel version
> and the output from uname. Any idea what command I should
> use to figure out which CPU the machine has?
uname -a, cat /proc/cpuinfo.
There is always the option of asking the owner of the system.
> Interesting. Thanks for this---it gives me some new
> things to think about.
I believe any alpha 20164/166/064/066 needs math_emu modules loaded or
the feature built in (which it was on older debian kernels, but then
someone lost that setting for a while, and I am not sure that bug has
been fixed yet, since I just stuck 'math_emu' in /etc/modules on my
system and stopped worrying about it.) As of the 20264 the floating
point was IEEE compliant in hardware as far as I have understood things
and the 'math_emu' fixup routines are no longer needed (and do nothing
on the newer CPUs).