[Bug tree-optimization/96915] New: ICE in tree-switch-conversion

[Bug tree-optimization/96915] New: ICE in tree-switch-conversion

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96915

            Bug ID: 96915
           Summary: ICE in tree-switch-conversion
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ktkachov at gcc dot gnu.org
  Target Milestone: ---
            Target: aarch64*-*-*

The following SVE intrinsics code ICEs with -O3 -march=armv8.2-a+sve

#include <arm_sve.h>

void b (void);

void
a ()
{
  switch (svcntd())
  case 2:
  case 4:
    b();
}

during GIMPLE pass: switchconv
foo.c: In function 'a':
foo.c:12:1: internal compiler error: in expand, at tree-switch-conversion.c:988
   12 | }
      | ^
0xe475f1 tree_switch_conversion::switch_conversion::expand(gswitch*)
        $SRC/gcc/tree-switch-conversion.c:988
0xe478ae execute
        $SRC/gcc/tree-switch-conversion.c:2401
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.

[Bug tree-optimization/96915] ICE in tree-switch-conversion

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96915

ktkachov at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |10.1.1

--- Comment #1 from ktkachov at gcc dot gnu.org ---
Reproduced on the GCC 10 branch as well. Since SVE ACLE intrinsics were only
added in GCC 10 it's technically not a regression

[Bug tree-optimization/96915] ICE in tree-switch-conversion

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96915

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
  /* A switch on a constant should have been optimized in tree-cfg-cleanup.  */
  gcc_checking_assert (!TREE_CONSTANT (m_index_expr));

guess a POLY_INT_CST isn't really a constant so this check needs adjustment
(using TREE_CONSTANT is odd anyway).  The question is what goes wrong when
m_index_expr is constant?  I hope nothing.

[Bug tree-optimization/96915] ICE in tree-switch-conversion

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96915

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org,
                   |                            |marxin at gcc dot gnu.org

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
tree-cfg.c only handles INTEGER_CST, so either we should check for just that in
the assertion, or better just drop it.  Dependence on some other pass
performing some optimization is IMHO bad, what if that pass hasn't run for some
reason, due to debug counters or whatever else.
If it doesn't want to handle INTEGER_CSTs, then it can always punt on them with
a reason that some other pass should handle those cases.

[Bug tree-optimization/96915] ICE in tree-switch-conversion

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96915

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2020-09-21
                URL|                            |https://gcc.gnu.org/piperma
                   |                            |il/gcc-patches/2020-Septemb
                   |                            |er/554371.html
     Ever confirmed|0                           |1

--- Comment #4 from Martin Liška <marxin at gcc dot gnu.org> ---
I've just sent patch to ML.

[Bug tree-optimization/96915] ICE in tree-switch-conversion

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96915

--- Comment #5 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Martin Liska <marxin@gcc.gnu.org>:

https://gcc.gnu.org/g:d726ecd9554a805d4a5e044cb21ca23a7f7ca49f

commit r11-3320-gd726ecd9554a805d4a5e044cb21ca23a7f7ca49f
Author: Martin Liska <mliska@suse.cz>
Date:   Mon Sep 21 12:50:37 2020 +0200

    Fix ICE in tree-switch-conversion.

    With SVE we can end up with:
    switch (POLY_INT_CST [2, 2]) <default: <L2> [INV], case 2: <L3> [INV], case
4: <L3> [INV]>
    which is fine to expand and we can remove the assert.

    gcc/ChangeLog:

            PR tree-optimization/96915
            * tree-switch-conversion.c (switch_conversion::expand): Accept
            also integer constants.

    gcc/testsuite/ChangeLog:

            PR tree-optimization/96915
            * gcc.target/aarch64/sve/pr96915.c: New test.

[Bug tree-optimization/96915] ICE in tree-switch-conversion

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96915

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|11.0                        |
      Known to work|                            |11.0

--- Comment #6 from Martin Liška <marxin at gcc dot gnu.org> ---
Fixed on master.

[Bug tree-optimization/96915] ICE in tree-switch-conversion

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96915

Martin Liška <marxin at gcc dot gnu.org> changed:

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

--- Comment #7 from Martin Liška <marxin at gcc dot gnu.org> ---
Fixed.