[Omp] PARALLEL DO REDUCTION of array-valued variables?

[Hoeflinger, Jay P]

This is not a valid Fortran program.  You have declared the function to
have an array-valued return value, but then use the function name (func)
as a scalar inside the function.  The function doesn't define an array
to return.  Also, no value is assigned to n, so you have no guarantee on
the number of iterations for the DO loop.

If you replace the 

    INTEGER, DIMENSION(n) :: func

with 

    INTEGER :: func

(and assign a value to n) the program runs correctly, with and without
OpenMP, with the Intel compiler.

Jay

-----Original Message-----
From: omp-bounces at openmp.org [mailto:omp-bounces at openmp.org] On Behalf
Of Daniel Franke
Sent: Saturday, October 28, 2006 4:21 PM
To: omp at openmp.org
Subject: [Omp] PARALLEL DO REDUCTION of array-valued variables?


Hi all.

The code below crashes GNU as well as INTEL fortran compilers. 

If either "func" is not array valued or not listed in the REDUCE-clause,
everything is fine (at least, from the compiler's point of view).

$> cat omp.f90
PROGRAM omp
  write (*,*) func(n)

  CONTAINS
   FUNCTION func(n)
    INTEGER, INTENT(in)   :: n
    INTEGER, DIMENSION(n) :: func
    INTEGER               :: k

    func = 0
    !$OMP PARALLEL DO PRIVATE(k), REDUCTION(+:func)
    DO k = 1, n
      func = func + 1
    END DO
    !$OMP END PARALLEL DO
   END FUNCTION
END PROGRAM


Could someone be so kind to comment on what is wrong?

Thanks
	Daniel


P.S. Detailed reports of compiler crahes:
gfortran: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29629
ifort: If someone from INTEL reads this, where is one supposed to report
ICEs?
_______________________________________________
Omp mailing list
Omp at openmp.org
http://openmp.org/mailman/listinfo/omp