[Users] problem with ifort 10 compiler

[Christian Reisswig]

Hi,

but in my config-file I did not specify any processor-dependant optimization 
flags. I think this is set by Cactus during the configuration.
Thomas, you could try to take my config-file on 
damiana /home/reisswig/.cactus/ and test it whether it will set this flag or 
not.
(On my laptop by the way it is using the same flag and again I did not specify 
that flag explicitly (I did not even know about it beforehands)).

I think this is something that needs to be fixed in the Cactus 
configuration-scripts itself when no optimization flags are given at all and 
Cactus needs to guess them.

cheers,
Christian

On Friday 18 January 2008 18:18:50 Thomas Radke wrote:
> Erik Schnetter wrote:
> > On Jan 16, 2008, at 06:26:26, Christian Reisswig wrote:
> >> Hi,
> >>
> >> when I try to compile Cactus, I get problems compiling Fortran  code,
> >> since
> >> Cactus sets an optimization option "-xN".
> >> This option doesn't seem to be supported by the Intel Fortran 10
> >> compiler.
> >> So I guess the -xN option has to be replaced by -xW.
> >
> > I think the option should be omitted.  Which of these options should  be
> > used depends on the system; Cactus should either autoconfigure the
> > system and optimise for the system where it is compiled, or should  (by
> > default) produce binaries which run everywhere.  -xW seems to  break for
> > Pentium III systems -- I don't think they are in widespread  use, but
> > you never know.
> >
> > We could use -axP, which generates code that runs everywhere but is
> > optimised for modern machines.  For example, -fast also implies -xP.
>
> I suggest to omit any processor-dependent optimisation flags and just
> use '-O2 -ip' (same Cactus default optimisation flags as for C/C++ code).
>
> I just ran a BSSN_MoL benchmark comparing a '-O2 -ip' configuration
> against '-O2 -xN -ip'. It showed no difference in runtime on Damiana.



-- 
Christian Reisswig
Max-Planck-Institut fuer Gravitationsphysik
Albert-Einstein-Institut
Potsdam / Golm
Astrophysical Relativity (Numerical Relativity Group)
office: +49 331 567 7179