[flang] Definitions of an array using subscript triplet of derived type is incorrect

[ohno-fj]

Version of flang-new : 17.0.0(633146093c4ea78dc8f56a27006fe4eed84be370)

Definitions of an array using subscript triplet of derived type are incorrect.

The following definitions are not correctly applied:

  v(1:2)%a(2,3)=2

There is no problem if you change the definition as follows:

  v(1)%a(2,3)=2
  v(2)%a(2,3)=2

The following are the test program, results of Flang-new, GFortran and ifort compilation and execution.

snggp155_4.f90:

subroutine sub
  type x
     integer, DIMENSION(2, 3) :: a
  end type x
  type(x),target::v(3)
  v(1)%a=3; v(2)%a=3; v(1:2)%a(2,3)=2

  write(6,*) "v(1)%a           = ", v(1)%a
  write(6,*) "v(2)%a           = ", v(2)%a
  write(6,*) "v(1:2)%a(2, 3)   = ", v(1:2)%a(2, 3)
  write(6,*) "v(1)%a(2, 3)     = ", v(1)%a(2, 3)
  write(6,*) "v(2)%a(2, 3)     = ", v(2)%a(2, 3)

end subroutine sub

program main
  call sub
end program main

$ flang-new -flang-experimental-exec snggp155_4.f90; ./a.out
 v(1)%a           =  3 3 3 3 2 3
 v(2)%a           =  3 3 3 3 2 3
 v(1:2)%a(2, 3)   =  2 2
 v(1)%a(2, 3)     =  3
 v(2)%a(2, 3)     =  3
$

$ gfortran snggp155_4.f90; ./a.out
 v(1)%a           =            3           3           3           3           3           2
 v(2)%a           =            3           3           3           3           3           2
 v(1:2)%a(2, 3)   =            2           2
 v(1)%a(2, 3)     =            2
 v(2)%a(2, 3)     =            2
$

$ ifort snggp155_4.f90; ./a.out
 v(1)%a           =            3           3           3           3           3
           2
 v(2)%a           =            3           3           3           3           3
           2
 v(1:2)%a(2, 3)   =            2           2
 v(1)%a(2, 3)     =            2
 v(2)%a(2, 3)     =            2
$

[llvmbot]

@llvm/issue-subscribers-bug

[luporl]

Issue also happens when built with flang-new -flang-experimental-hlfir.

[luporl]

https://reviews.llvm.org/D154214

[llvmbot]

@llvm/issue-subscribers-flang-ir