[SciPy-dev] more 64 Bit testing

Arnd Baecker arnd.baecker at web.de
Thu Dec 8 03:59:35 CST 2005


On Thu, 8 Dec 2005, Travis Oliphant wrote:

> Arnd Baecker wrote:
>
> >Hi David,
> >
> >yes there is - from man icc
> >  -std=c99 Enable C99 support for C programs
> >
> >
> >
> http://cache-www.intel.com/cd/00/00/22/23/222301_222301.pdf
>
> This document gives some of the compatibility issues between the icc
> compiler and gcc.

Thanks - good to know about that!

> In particular, note that nested functions are not supported.   So, it
> looks like icc can't be used to build the fitpackmodule.c unless f2py is
> changed to not use nested functions.

It seems that nested functions  are not in the C99 standard
and some people have a strong opinion on them
http://lkml.org/lkml/2004/5/12/248
(no I did not google for negative statements, but just wanted
to learn about how widespread nested functions are ...;-).

I don't know the internals of f2py (way out of my league),
but at first glance it might be possible to
transfer those nested functions outside of the function
and call them with the corresponding parameters
- overall this is seems more complicated, already
on the C side, so I don't even want to speculate on
how difficult this would be on the f2py level...

> There are only a few incompatibilities in icc and gcc but that
> apparently is one of them.   I'm not sure how high on the priority list
> supporting the icc compiler is for most of us.

Fair enough - it *might* have some importance to us as
python itself has been reported to be 16% faster with icc
http://mail.python.org/pipermail/python-dev/2005-June/054233.html
OTOH, a lot of the time spent in our applications
is in LAPACK and FFT.
Still, when you have access to a machine with 192 CPUs
it would be good to get the maximal possible performance...

> But, we would gladly
> accept fixes to allow the icc compiler to be used....

Presently I see three problems
a) the NAN stuff for cephes
b) the dfitpackmodule problem above
c) the bunch of test failures

> Sorry about the trouble.

No problem at all! - I am only worried about c) as this might
indicate something more problematic. Unfortunately,
I did not have enough time to look into more details than I posted
so far.

Best, Arnd




More information about the Scipy-dev mailing list