Re: AC_OPENMP broken for IBM xl Fortran compilers

[Christian Rössel]

Hi Ralf,

Am 2/20/2011 7:55 AM, schrieb Ralf Wildenhues:
> Hi Christian, Markus,
> 
> * Christian Rössel wrote on Fri, Feb 18, 2011 at 03:57:30PM CET:
>> AC_OPENMP returns "none needed" for the IBM xl Fortran compilers (xlf*,
>> bgxlf*, mpixlf*). This is wrong, you need to use -qsmp=omp.
>>
>> configure tries to compile following program:
>>
>> |       program main
>> |       call omp_get_num_threads
>> |       end
>>
>> Compilation succeeds without specifying any OpenMP flag. If you modify
>> the test program to
>>
>>       program main
>>       implicit none
>> !$    integer tid
>>       tid = 42
>>       call omp_set_num_threads(2)
>>       end
>>
>> the correct flags are detected.
> 
> Thanks for the bug report and patch.  I have one question (and I'd still
> like to do some testing before pushing):  Isn't the '!' line a Fortran
> 90 comment, and doesn't that have to be written differently for pure
> Fortran 77?

you are right, a '!' is usually a comment in Fortran. But in the context
of OpenMP and combined with the '$' it is interpreted as a conditional
compilation sentinel, i.e. if no OpenMP flags are passed to the
compiler, the line is interpreted as a comment, otherwise '!$' is
replaced by two spaces and the line gets compiled. See 2.2.1 and 2.2.2
in the OpenMP specification
(http://www.openmp.org/mp-documents/spec30.pdf). The patch should work
for Fortran 77 and 90.

Regards,
Christian

>> See the attached patch where I also
>> added OpenMP flags for the Cray and NEC compilers. If you are happy with
>> this patch, please add Markus Geimer to THANKS, too.
> 
> Yes, that part is quite obviously good (it has a near-zero chance of
> regressing anything!).
> 
> Thanks,
> Ralf