CUDA_ERROR_ILLEGAL_ADDRESS in OpenACC Code with Pointer Alias – cuEventSynchronize Error

Accelerated Computing NGC GPU Cloud Container: HPC
1

I’m experiencing an issue with a very simple OpenACC program that uses a pointer alias. When I run my code, I get the following runtime error:

call to cuEventSynchronize returned error 700 (CUDA_ERROR_ILLEGAL_ADDRESS): Illegal address during kernel execution

program simple_acc_pointer
  implicit none
  integer, parameter :: n = 10
  integer :: i
  real, allocatable, target :: a(:) 
  real, pointer    :: p(:)     
  real :: local_val    
  

  allocate(a(n))
  p => a
  

  do i = 1, n
     a(i) = i * 1.0
  end do
  

  !$acc data copy(a(1:n))
  !$acc parallel loop deviceptr(p)
  do i = 1, n


     local_val = p(i) * 2.0
     p(i) = local_val

  end do
  !$acc end parallel loop
  !$acc end data
  
  deallocate(a)
  
end program simple_acc_pointer

Compiler: NVFORTRAN
Could the error be due to how local variables are handled within the OpenACC parallel region when using pointer aliasing?
Are there any known issues with this pattern (using a local variable together with a pointer alias in a collapsed data region)?