[Bug fortran/122388] New: [13/14/15/16 Regression] Fortran shadow_loopvar seems to lead to miscompilation

[tnfchris at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122388

            Bug ID: 122388
           Summary: [13/14/15/16 Regression] Fortran shadow_loopvar seems
                    to lead to miscompilation
           Product: gcc
           Version: 16.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tnfchris at gcc dot gnu.org
  Target Milestone: ---

The following example:

integer (8) i, u8, step8
  call test ((/ (i, i = u8, l4step8) /), u8,   0_8, step8)
contains
  subroutine test (a, l, u, step)
    integer (8), dimension (:) :: a
    integer (8)  l, u, step
    do i = u, step
      j = 1
    end do
    if (size (a) .ne. j - 1) STOP 2
  end
end

Compiled with

  gfortran array.f90 -O2 -o array.exe -fno-tree-loop-vectorize

fails but

  gfortran array.f90 -O2 -o array.exe -fno-tree-loop-vectorize -fno-tree-dce

or

  gfortran array.f90 -O1 -o array.exe -fno-tree-loop-vectorize

works.

The difference starts at the front-end where compared with -O1 at -O2 fortrans
trans_array transform seems to kick in.

I've been able to reproduce this on every GCC from 6 to 16 and can reproduce it
on latest HEAD on AArch64 and x86_64.