catastrophic error with coarray and openmp

[Zuodong_Y_]

Compile following code with -qopenmp -coarray causes  010101_14921 catastrophic error on Version 18.0.1.163 Build 20171018. Any reason for it? Thanks!

module M_A
	implicit none
	integer :: ica
	type t_A
		integer :: map(10)
	contains
		procedure :: do_something
	end type
	complex(8) :: B(20,10)

contains
	subroutine do_something(self)
		class(t_A) :: self

		integer :: i,n
		self%map=[10:1:-1]
		do i=this_image(),10,num_images()
			!$OMP PARALLEL DO
			do n=1,20
				B(n,self%map(i))=1d0
			enddo
			!$OMP END PARALLEL DO
		enddo
	end subroutine
end module
program main
	use M_A
	implicit none
	type(t_A) :: tp

	call tp%do_something()
end program

[Steve_Lionel]

Internal compiler error is always a compiler bug. It still fails on 19.1. Please report it through the Intel Online Service Center.

[JohnNichols]

Is it just my imagination or are there quite a few compiler bug errors in the last few months, I do not remember seeing so many in my scant years on this site. 

[Steve_Lionel]

It's not unusual in my experience. They do come in waves - most often when a school semester starts. Another wave sometimes occurs when a new major compiler version is released and a lot of people jump in to try it, though this has decreased over the years.

[Zuodong_Y_]

I am not able to report the bug as I don't own license. Anyone help to report will be greatly appreciated. Thanks!

[Zuodong_Y_]

I further simplify the code a little bit. The problem seems to be referring member of coarray derived type inside of type-bound procedure in OMP scope. For now, a possible way to get rid of this is don't use type-bound procedure in coarray.

module debug
	implicit none
	type t_A
		integer :: a
	contains
		procedure :: set
	end type
contains
	subroutine set(self)
		class(t_A) :: self

		integer :: i
		!$OMP PARALLEL DO
		do i=1,10
			write(*,*)self%a
		enddo
		!$OMP END PARALLEL DO
	end subroutine
end module

[Barbara_P_Intel]

Thank you for simplifying the problem.  Let me check this out.

[Barbara_P_Intel]

The Intel Fortran compiler does not yet support polymorphic variables inside OpenMP parallel regions. That feature is part of OpenMP 5.0 and we haven't implemented that yet.

[Zuodong_Y_]

I didn't know that. I used a lot of polymorphic variables inside of openmp before and it seems to be fine. Thanks for the information!