Problem with omp simd loop, ifx 2025.2

[puipp]

I experience the following problem with ifx (IFX) 2025.2.0 20250605

forrtl: severe (408): fort: (3): Subscript #2 of the array RES has value 0 which is less than the lower bound of 1

 which I could reproduce with the following test code

subroutine problem()

  integer, parameter :: np = 1000
  integer :: i, n, l, m, k, k_periodic, nk_stripped, j, lb, ub
  integer :: res(np, 1:4)
  nk_stripped = 4

  lb=-1
  ub=4
  !$omp parallel default(none) firstprivate(lb,ub) shared(nk_stripped, res) private(i, k, k_periodic)
  do n = 1,2
  do m = 1,10
  do l = 1,10
  do k = lb, ub
    !$omp do simd
    do i = 1, np
      k_periodic = modulo(k-1, nk_stripped) + 1
      if (k_periodic < 1 .or. k_periodic > nk_stripped) then
        print *, "ERROR: k_periodic out of bounds: ", k_periodic, " at k=", k, " i=", i
      end if
      res(i, k_periodic) = k
    end do
    !$omp end do simd
  end do
  end do
  end do
  end do
  !$omp end parallel

end subroutine

 produces

ERROR: k_periodic out of bounds:            0  at k=           0  i=         744
forrtl: severe (408): fort: (3): Subscript #2 of the array RES has value 0 which is less than the lower bound of 1

while I would expect k_periodic above to always be within 1 and 4.

I only encounter this with ifx in debug mode, not with gcc or ifort, and only when build in debug mode using the following flags

-g -fPIC -fpp -g -no-wrap-margin -traceback -fstack-security-check -fpe0 -O0 -warn all -check all,noarg_temp_created
 -debug inline-debug-info -ftrapv -Rno-debug-disables-optimization -fiopenmp

removing the simd from the code avoids the problem. Is this expected behaviour?

[jimdempseyatthecove]

    !$omp do simd
    do i = 1, np

Implies (requires?) that an array in the body of the do loop is referenced in a manner suitable for simd instructions

The body of the loop contains:

      k_periodic = modulo(k-1, nk_stripped) + 1

Which might confuse the code generation for

      res(i, k_periodic) = k

I suggest you relocate that (and the error test) code to the do k loop and see what happens:

  !$omp parallel default(none) firstprivate(lb,ub) shared(nk_stripped, res) private(i, k, k_periodic)
  do n = 1,2
  do m = 1,10
  do l = 1,10
  do k = lb, ub
    k_periodic = modulo(k-1, nk_stripped) + 1
    if (k_periodic < 1 .or. k_periodic > nk_stripped) then
      print *, "ERROR: k_periodic out of bounds: ", k_periodic, " at k=", k, " i=", i
    end if
    !$omp do simd
    do i = 1, np
      res(i, k_periodic) = k
    end do
    !$omp end do simd
  end do
  end do
  end do
  end do
  !$omp end parallel

BTW, I suggest you use:

k_periodic = modulo(k-lb, nk_stripped) + 1

as it is clearer as to the intent.

Jim Dempsey