[Discuss] Test for positive gives wacko results for underflowed values for the gcc compiler suite (including the gcc compiler, and g77 and gfortran compilers)

[Alan W. Irwin]

On 2007-06-25 14:50-0700 pw wrote:

> Alan W. Irwin wrote:
>> 
>> 
>> Python gives the correct results both on my Ubuntu system
>> (Python 2.4.3) and my older Debian system (Python 2.3.5).
>
>
> What compile flags for python vs. gfortran?

Don't know.

> Is it compiled with gcc to use a floating point
> library rather than the floating point processor?

I am not quite sure what you are referring to here.  Is it the kernel option
for using software floating-point emulation versus using the floating-point
processor of the cpu?  My systems use just the default (which I am sure is
to do floating-point calculations in hardware since that should be much
faster than emulating them).

>
> The difference has to be in the compilation of
> the interpreter binary.

I assume both python and gcc use the C library and the math library so I
suspect the issue is not in those libraries and instead is caused by some
mishandling of underflows internally in gcc/g77/gfortran.

Test results for non-gcc compilers would help to sort out whether this
suspicion of a bug in gcc and friends is correct or not.

Alan
__________________________
Alan W. Irwin

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the FreeEOS equation-of-state implementation
for stellar interiors (freeeos.sf.net); PLplot scientific plotting software
package (plplot.org); the libLASi project (unifont.org/lasi); the Loads of
Linux Links project (loll.sf.net); and the Linux Brochure Project
(lbproject.sf.net).
__________________________

Linux-powered Science
__________________________