[Bug fortran/107679] New: [13 Regression] ICE in maybe_register_def, at tree-into-ssa.cc:1914

[Bug fortran/107679] New: [13 Regression] ICE in maybe_register_def, at tree-into-ssa.cc:1914

[gscfq@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107679

            Bug ID: 107679
           Summary: [13 Regression] ICE in maybe_register_def, at
                    tree-into-ssa.cc:1914
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gscfq@t-online.de
  Target Milestone: ---

Started between 20220918 and 20220925, at -O1+ :
(gcc configured with --enable-checking=yes)


$ cat z1.f90
subroutine s1(x)
   integer, intent(out) :: x
end
subroutine s2(z)
   integer, value :: z
   call s1(z)
end


$ cat z2.f90
subroutine s1(x)
   integer, intent(out) :: x
   x = 3
end
subroutine s2(z)
   integer, value :: z
   call s1(z)
end


$ gfortran-13-20221106 -c z1.f90 -O2
during GIMPLE pass: ccp
z1.f90:7:3:

    7 | end
      |   ^
internal compiler error: in maybe_register_def, at tree-into-ssa.cc:1914
0xffb188 maybe_register_def
        ../../gcc/tree-into-ssa.cc:1914
0xffb188 rewrite_update_stmt
        ../../gcc/tree-into-ssa.cc:2063
0xffb188 rewrite_update_dom_walker::before_dom_children(basic_block_def*)
        ../../gcc/tree-into-ssa.cc:2236
0x1cce2a7 dom_walker::walk(basic_block_def*)
        ../../gcc/domwalk.cc:311
0xff80e9 rewrite_blocks
        ../../gcc/tree-into-ssa.cc:2296
0xfff24e update_ssa(unsigned int)
        ../../gcc/tree-into-ssa.cc:3581
0x11f022e execute_update_addresses_taken()
        ../../gcc/tree-ssa.cc:2177
0xe47700 execute_function_todo
        ../../gcc/passes.cc:2066
0xe480b2 execute_todo
        ../../gcc/passes.cc:2145

[Bug fortran/107679] [13 Regression] ICE in maybe_register_def, at tree-into-ssa.cc:1914

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107679

anlauf at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |anlauf at gcc dot gnu.org
           Priority|P3                          |P4
   Last reconfirmed|                            |2022-11-14
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
   Target Milestone|---                         |13.0

--- Comment #1 from anlauf at gcc dot gnu.org ---
Confirmed.

[Bug fortran/107679] [13 Regression] ICE in maybe_register_def, at tree-into-ssa.cc:1914

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107679

--- Comment #2 from anlauf at gcc dot gnu.org ---
Interesting.  Adding a line

  print *, z

after the call

  call s1(z)

make the ICE go away.

Is it the CLOBBER that confuses us?

[Bug middle-end/107679] [13 Regression] ICE in maybe_register_def, at tree-into-ssa.cc:1914

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107679

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|fortran                     |middle-end
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
             Status|NEW                         |ASSIGNED
           Priority|P4                          |P3

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Kind-of, yes.  Somehow 'z' got marked for renaming, probably because we removed
a TREE_ADDRESABLE flag.  The SSA rewrite wants to replace

  <bb 2> [local count: 1073741824]:
  z ={v} {CLOBBER};
  return;

with an "uninitialized" def, but for PARM_DECLs the default def would be
the actual parameter value.  There's nothing wrong with clobbering 'z'
I think, just the into-ssa code isn't prepared for this - the fix would be
to create an anonymous default def in this case.

[Bug middle-end/107679] [13 Regression] ICE in maybe_register_def, at tree-into-ssa.cc:1914

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107679

--- Comment #4 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:8a5f3663f8de45b7030dfe7ce962783dc1a19885

commit r13-4101-g8a5f3663f8de45b7030dfe7ce962783dc1a19885
Author: Richard Biener <rguenther@suse.de>
Date:   Wed Nov 16 13:57:14 2022 +0100

    middle-end/107679 - fix SSA rewrite of clobber of parameter

    When the frontend clobbers a parameter and that parameter gets
    rewritten into SSA then we ICE because we didn't expect this.  Avoid
    using the parameter decl to create a SSA default def in this case.

            PR middle-end/107679
            * tree-into-ssa.cc (maybe_register_def): Use new temporary if
            we require an uninitialized value for a parameter decl.

            * gfortran.dg/pr107679.f90: New testcase.

[Bug middle-end/107679] [13 Regression] ICE in maybe_register_def, at tree-into-ssa.cc:1914

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107679

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.