[SciPy-dev] new core, Itanium/Intel compiler probs

[Arnd Baecker]

Hi Pearu,

On Mon, 5 Dec 2005, Pearu Peterson wrote:

> On Mon, 5 Dec 2005, Arnd Baecker wrote:
>
> > Hi,
> >
> > we just got access to a new 64 Bit toy with many CPUs - so scipy
> > has to be installed there ;-).
> >
> > The recommended compiler is the Intel one (icc,ifc).
> > Scipy core builds/installs alright, using
> >  python setup.py config --compiler=intel  build
>
> Could you check that the build was really done with intel compilers?

Pretty confident - the logfile only contains `icc` calls

> I would have suggested the following command
>
>    python setup.py config --fcompiler=intel install --prefix=$DESTDIR

To be sure I used that as well. icc is being used:

[...]
customize IntelFCompiler
customize IntelFCompiler using config
icc options: '-pthread -fno-strict-aliasing -OPT:Olimit=0 -DNDEBUG -g -O3
-Wall -Wstrict-prototypes -fPIC -fPIC'
compile options: '-I/home/baecker/python/include/python2.4
-Iscipy/base/src -I/home/baecker/python/include/python2.4 -c'
icc: _configtest.c
icc: Command line warning: ignoring option '-O'; no argument required
_configtest.c(50): warning #181: argument is incompatible with
corresponding format string conversion
            fprintf(fp,"#define SIZEOF_LONG_DOUBLE %d\n", sizeof(long
double));
[...]

(Note this is all with  `icc -v:  Version 9.0`)


> Also, could you try building scipy_core (and Python if necessary) with
> gcc? How the tests behave then?

With gcc version 3.3.3 (SuSE Linux)
everything compiles fine (BTW: much less warnings).

scipy.test(10,10): Ran 160 tests in 1.219s

So this looks fine!

I think that the problem is really related to icc.

Many thanks, Arnd