That's it.
Thank you Tom!





Il giorno 27 gennaio 2010 07.03, tom fogal <tfogal@alumni.unh.edu> ha scritto:
 writes:
> I've a simple FORTRAN code that does some operation (no more than
> addition and multiplication) with complex number.
>
> I've compiled my code with the -g option and with the -o3. And some
> of the resulting values are different.  So I started inserting some
> WRITE statements in the intermediate computation and some of the
> previous results that were different now are exactly the same!!
>
> I believe that this looks like a illegal value access problem, like
> reading uninitialized variables, reading over arrays boundary.

This is almost definitely the standard intel extra-precision issue and
not a real `bug' in your program.  I'm sure "Intel 80-bit" on google
will describe this in more detail.

If you're using the GNU toolchain, compile with -ffloat-store and you
should see identical results with and without the writes.

Cheers,

-tom