wrong broadcast behavior in derived coarray object

[Zuodong_Y_]

Here is the code:

program main
	implicit none
	type gfs
		integer :: i
		integer :: Gc(1,1)
	end type
	integer :: ica
	type(gfs) :: self

	self%Gc=this_image()
	sync all
	if(this_image()==1) then
		do ica=2,num_images()
			self[ica]%Gc=self%Gc
		enddo
	endif
	sync all
	write(*,*)this_image(),"->Gc",self%Gc(1,1)
end program

compiled with Intel(R) 64, Version 18.0.1.163 Build 20171018 and run for example with four coarray images gives results like:

           1 ->Gc           1
           3 ->Gc           3
           4 ->Gc           4
           2 ->Gc           2

But what I expect is:

           1 ->Gc           1
           3 ->Gc           1
           4 ->Gc           1
           2 ->Gc           1

remove integer :: i in derived type or broadcast element by element (i.e. change to self[ica]%Gc(1,1)=self%Gc(1,1)) seems solve the problem. Any idea about this?

BTW, this there any performance cost to broadcast element by element for large array?

[FortranFan]

Looks like another bug in the compiler, a pernicious one since it involves erroneous run-time result.  Report this at Intel OSC if you can.

[FortranFan]

Zuodong Y. wrote:

.. BTW, this there any performance cost to broadcast element by element for large array?

I'm not sure how one can reliably measure performance under the circumstances.

If the whole array is to be broadcast, I think the intrinsic CO_BROADCAST will be as good as it gets for a given processor in terms of performance.  It will add to better readability of the code as well:

   integer, save :: a(3)

   integer :: i

   character(len=*), parameter :: fmtg = "(g0,t15,g0)"

   if ( this_image() == 1 ) then
      a = [ 1, 2, 3 ]
   else
      a = 0
   end if
   sync all

   call co_broadcast( a, source_image=1 )

   if ( this_image() == 1 ) then
      print fmtg, "Image #", "a(1)"
      do i = 1, num_images()
         print fmtg, i, a(1)
      end do
   end if

end

C:\Temp>ifort /standard-semantics /warn:all /stand:f18 /Qcoarray:shared /Qcoarray-num-images=8 p.f90
Intel(R) Visual Fortran Intel(R) 64 Compiler for applications running on Intel(R) 64, Version 19.1.1.216 Build 20200306
Copyright (C) 1985-2020 Intel Corporation.  All rights reserved.

Microsoft (R) Incremental Linker Version 14.25.28612.0
Copyright (C) Microsoft Corporation.  All rights reserved.

-out:p.exe
-subsystem:console
p.obj

C:\Temp>p.exe
Image #       a(1)
1             1
2             1
3             1
4             1
5             1
6             1
7             1
8             1

C:\Temp>

[Zuodong_Y_]

Thanks for comment. I believe CO_BROADCAST feature only available in ifort 19.1 and I remember have other issue to move to new version of ifort.

FortranFan wrote:

Quote:

Zuodong Y. wrote:

.. BTW, this there any performance cost to broadcast element by element for large array?

I'm not sure how one can reliably measure performance under the circumstances.

If the whole array is to be broadcast, I think the intrinsic CO_BROADCAST will be as good as it gets for a given processor in terms of performance.  It will add to better readability of the code as well:

   integer, save :: a(3)

   integer :: i

   character(len=*), parameter :: fmtg = "(g0,t15,g0)"

   if ( this_image() == 1 ) then
      a = [ 1, 2, 3 ]
   else
      a = 0
   end if
   sync all

   call co_broadcast( a, source_image=1 )

   if ( this_image() == 1 ) then
      print fmtg, "Image #", "a(1)"
      do i = 1, num_images()
         print fmtg, i, a(1)
      end do
   end if

end

C:\Temp>ifort /standard-semantics /warn:all /stand:f18 /Qcoarray:shared /Qcoarray-num-images=8 p.f90
Intel(R) Visual Fortran Intel(R) 64 Compiler for applications running on Intel(R) 64, Version 19.1.1.216 Build 20200306
Copyright (C) 1985-2020 Intel Corporation.  All rights reserved.

Microsoft (R) Incremental Linker Version 14.25.28612.0
Copyright (C) Microsoft Corporation.  All rights reserved.

-out:p.exe
-subsystem:console
p.obj

C:\Temp>p.exe
Image #       a(1)
1             1
2             1
3             1
4             1
5             1
6             1
7             1
8             1

C:\Temp>

[Barbara_P_Intel]

I filed a bug (moth) report for the original posting, CMPLRIL0-32769. 

[Zuodong_Y_]

Thanks for helping me file the bug (moth).

Barbara P (Intel) wrote:

I filed a bug (moth) report for the original posting, CMPLRIL0-32769. 

[Barbara_P_Intel]

This problem is fixed in PSXE 2020 update 2 (Fortran 19.1.2) that was released last week. Please check it out!