[MITgcm-devel] omp and OMPFLAG

[Martin Losch]

Hi Jean-Michel,

OK, I’ll work in the build options file and thanks for your suggestion. I was trying something like this …

Martin

> On 8. Aug 2017, at 14:43, Jean-Michel Campin <jmc at mit.edu> wrote:
> 
> Hi Martin,
> 
> I have the impression that you are providing already the answer:
> (1) It make more sense to see $OMPFLAG as a compiler flag rather than OPTIM option; 
> (2) regarding changing genmake2, how can we be sure that an other compiler will not 
>  do the opposite (i.e., requiring OMPFLAG to come before OPTIM option) ?
> 
> I did not try, but may be changing the cray optfile by testing env. variable $OMP ?
> Just before
>  F90FLAGS=$FFLAGS
>  F90OPTIM=$FOPTIM
> adding something like:
> if test "x$OMP" = xtrue ; then
>   FOPTIM="$FOPTIM $OMPFLAG"
>   OMPFLAG=''
> fi
> 
> Cheers,
> Jean-Michel
> 
> On Tue, Aug 08, 2017 at 12:02:17PM +0200, Martin Losch wrote:
>> Hi,
>> 
>> I ran across a problem with omp and a cray compiler. Here???s the background:
>> (1) For a cray compiler, omp is defined by default, except for -O0 (no optimization, which is what we normally use when genmake2 is used with -ieee or when you run testreport (even without -devel).
>> (2) the order of flags matter, later flags override previous flags.
>> The consquence is, that if I want to run testreport I cannot use OpenMP, because the order of flags is
>> $FFLAGS (includes $OMPFLAG) $FOPTION (includes optimization flags): -homp ??? -O0
>> The latter will reset -homp to -hnoomp.
>> I could tweak the build_option file for a cray compiler, but wouldn???t it make more sense to add the OMPFLAG  to FOPTIM in genmake2 (even though it???s stricktly speaking not an optimization flag)?
>> 
>> Martin
>> 
>> _______________________________________________
>> MITgcm-devel mailing list
>> MITgcm-devel at mitgcm.org
>> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-devel
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-devel