[Flang][OpenMP] Incorrect execution result when reduction-identifier of reduction clause is specified as + and an array with a subscript of 0 is used in the reduction operation

[ohno-fj]

Version of flang : 21.0.0(3e6f618e86f5fbad2c2d5802416ec3d3366a2837)/AArch64

In parallel do reduction directive, when reduction-identifier of reduction clause is specified as + and an array with a subscript of 0 is used for the reduction operation, the result is incorrect.
The result of the reduction operation does not appear to be assigned.
The above program is test132.f90.

When an array with a subscript of 1 is used in the reduction operation, the result is correct.
The above program is test131.f90.

The following are the test program, Flang, Gfortran and ifx compilation/execution result.

test132.f90:

program main
  integer a(0:1)
  a=0
!$omp parallel do reduction(+:a)
  do i=1,10
     a(0)=a(0)+1
  enddo
!$omp end parallel do
  write(6,*) "a(0) = ", a(0)
 end program main

$ export OMP_NUM_THREADS=2; flang -fopenmp test132.f90; ./a.out
flang-20: warning: OpenMP support in flang is still experimental [-Wexperimental-option]
 a(0) =  0
$

$ export OMP_NUM_THREADS=2; gfortran -fopenmp test132.f90; ./a.out
 a(0) =           10
$

$ export OMP_NUM_THREADS=2; ifx -qopenmp test132.f90; ./a.out
 a(0) =           10
$

test131.f90:

program main
  integer a(0:1)
  a=0
!$omp parallel do reduction(+:a)
  do i=1,10
     a(1)=a(1)+1
  enddo
!$omp end parallel do
  write(6,*) "a(1) = ", a(1)
 end program main

$ export OMP_NUM_THREADS=2; flang -fopenmp test131.f90; ./a.out
flang-20: warning: OpenMP support in flang is still experimental [-Wexperimental-option]
 a(1) =  10
$

$ export OMP_NUM_THREADS=2; gfortran -fopenmp test131.f90; ./a.out
 a(1) =           10
$

$ export OMP_NUM_THREADS=2; ifx -qopenmp test131.f90; ./a.out
 a(1) =           10
$

[llvmbot]

@llvm/issue-subscribers-flang-ir

Author: None (ohno-fj)

``` Version of flang : 21.0.0(3e6f618)/AArch64 ```

In parallel do reduction directive, when reduction-identifier of reduction clause is specified as + and an array with a subscript of 0 is used for the reduction operation, the result is incorrect.
The result of the reduction operation does not appear to be assigned.
The above program is test132.f90.

When an array with a subscript of 1 is used in the reduction operation, the result is correct.
The above program is test131.f90.

The following are the test program, Flang, Gfortran and ifx compilation/execution result.

test132.f90:

program main
  integer a(0:1)
  a=0
!$omp parallel do reduction(+:a)
  do i=1,10
     a(0)=a(0)+1
  enddo
!$omp end parallel do
  write(6,*) "a(0) = ", a(0)
 end program main

$ export OMP_NUM_THREADS=2; flang -fopenmp test132.f90; ./a.out
flang-20: warning: OpenMP support in flang is still experimental [-Wexperimental-option]
 a(0) =  0
$

$ export OMP_NUM_THREADS=2; gfortran -fopenmp test132.f90; ./a.out
 a(0) =           10
$

$ export OMP_NUM_THREADS=2; ifx -qopenmp test132.f90; ./a.out
 a(0) =           10
$

test131.f90:

program main
  integer a(0:1)
  a=0
!$omp parallel do reduction(+:a)
  do i=1,10
     a(1)=a(1)+1
  enddo
!$omp end parallel do
  write(6,*) "a(1) = ", a(1)
 end program main

$ export OMP_NUM_THREADS=2; flang -fopenmp test131.f90; ./a.out
flang-20: warning: OpenMP support in flang is still experimental [-Wexperimental-option]
 a(1) =  10
$

$ export OMP_NUM_THREADS=2; gfortran -fopenmp test131.f90; ./a.out
 a(1) =           10
$

$ export OMP_NUM_THREADS=2; ifx -qopenmp test131.f90; ./a.out
 a(1) =           10
$