[Bug middle-end/86437] New: runtime segfault on Fortran code with large array and -Ofast

[janus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86437

            Bug ID: 86437
           Summary: runtime segfault on Fortran code with large array and
                    -Ofast
           Product: gcc
           Version: 9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: janus at gcc dot gnu.org
  Target Milestone: ---

Reduced test case:


program ACT

   implicit none

   integer, parameter :: N = 53, M = 20000
   integer :: i
   integer, dimension(1:2*N) :: list
   real(8), dimension(:,:), allocatable :: Y1, Y2

   do i=1,2*N
      list(i) = min(i, N)
   end do

   call DoAllocate(Y1, 1,N,   1,M)
   call DoAllocate(Y2, 1,2*N, 1,2*M)

   call ArrayCopy(Y2(list(1:N),1:M), Y1)

contains

   subroutine DoAllocate(a, l1, u1, l2, u2)
      real(8), allocatable, intent(inout) :: a(:,:)
      integer,              intent(in)    :: l1, u1, l2, u2
      allocate(a(l1:u1,l2:u2), source=0.0_8)
   end subroutine

   subroutine ArrayCopy(arrFrom, arrTo)
      real(8), dimension(:,:), intent(in)    :: arrFrom
      real(8), dimension(:,:), intent(inout) :: arrTo
      arrTo(:,:) = arrFrom(:,:)
   end subroutine

end


When compiling this with any recent gfortran version (4.7.x up to 9-trunk)
using -Ofast, I get a segfault at runtime.

Things that make the segfault disappear:
* using gfortran 4.6.4
* using smaller M (e.g. M=2000)
* using -O1, -O2 or -O3 (with these I can even make M much larger, e.g.
M=200000)

Since the optimization level changes the behavior crucially, I assume this is
more of a middle-end than a front-end issue.