[Bug c/97769] New: [11 Regression] vectorizer ICE when building perlbench in SPECCPU 2017

[Bug c/97769] New: [11 Regression] vectorizer ICE when building perlbench in SPECCPU 2017

[tnfchris at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97769

            Bug ID: 97769
           Summary: [11 Regression] vectorizer ICE when building perlbench
                    in SPECCPU 2017
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tnfchris at gcc dot gnu.org
  Target Milestone: ---

Created attachment 49528
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=49528&action=edit
reproducer for ice

With vectorizer enabled the perlbench benchmark ICEs if you bump up the inliner
limit:

during GIMPLE pass: vect
tmp.i: In function 'fn3':
tmp.i:25:1: internal compiler error: in vect_update_misalignment_for_peel, at
tree-vect-data-refs.c:1196
   25 | fn3() { fn2(&(&c)->otmp); }
      | ^~~
0x19f9b43 vect_update_misalignment_for_peel
        ../../gcc-fsf/gcc/tree-vect-data-refs.c:1193
0x19f9def vect_get_peeling_costs_all_drs
        ../../gcc-fsf/gcc/tree-vect-data-refs.c:1481
0x19fb83b vect_peeling_hash_get_lowest_cost(_vect_peel_info**,
_vect_peel_extended_info*)
        ../../gcc-fsf/gcc/tree-vect-data-refs.c:1508
0x1a068cb void hash_table<peel_info_hasher, false,
xcallocator>::traverse_noresize<_vect_peel_extended_info*,
&(vect_peeling_hash_get_lowest_cost(_vect_peel_info**,
_vect_peel_extended_info*))>(_vect_peel_extended_info*)
        ../../gcc-fsf/gcc/hash-table.h:1081
0x1a068cb void hash_table<peel_info_hasher, false,
xcallocator>::traverse<_vect_peel_extended_info*,
&(vect_peeling_hash_get_lowest_cost(_vect_peel_info**,
_vect_peel_extended_info*))>(_vect_peel_extended_info*)
        ../../gcc-fsf/gcc/hash-table.h:1102
0x1a068cb vect_peeling_hash_choose_best_peeling
        ../../gcc-fsf/gcc/tree-vect-data-refs.c:1557
0x1a068cb vect_enhance_data_refs_alignment(_loop_vec_info*)
        ../../gcc-fsf/gcc/tree-vect-data-refs.c:2058
0x10ef27f vect_analyze_loop_2
        ../../gcc-fsf/gcc/tree-vect-loop.c:2351
0x10f03b3 vect_analyze_loop(loop*, vec_info_tmpred*)
        ../../gcc-fsf/gcc/tree-vect-loop.c:2895
0x1118d83 try_vectorize_loop_1
        ../../gcc-fsf/gcc/tree-vectorizer.c:995
0x11194ff vectorize_loops()
        ../../gcc-fsf/gcc/tree-vectorizer.c:1230
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.

The attached testcase reproduces this at -O3 with no other options.

[Bug c/97769] [11 Regression] vectorizer ICE when building perlbench in SPECCPU 2017

[tnfchris at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97769

--- Comment #1 from Tamar Christina <tnfchris at gcc dot gnu.org> ---
This seems to go away if you initialize a in `int a`.

[Bug middle-end/97769] [11 Regression] vectorizer ICE when building perlbench in SPECCPU 2017

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97769

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2020-11-10
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
             Target|                            |aarch64
   Target Milestone|---                         |11.0
           Priority|P3                          |P1

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed on aarch64 with a cross, doesn't ICE on x86_64.  I can have a look.

[Bug middle-end/97769] [11 Regression] vectorizer ICE when building perlbench in SPECCPU 2017

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97769

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
OK, it's just an overzealeous assert that is too difficult to update for the
additional cases we now handle.  I'll remove it.

[Bug middle-end/97769] [11 Regression] vectorizer ICE when building perlbench in SPECCPU 2017

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97769

--- 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:1693746302e4306b43cb66a0afe589137069bd8e

commit r11-4877-g1693746302e4306b43cb66a0afe589137069bd8e
Author: Richard Biener <rguenther@suse.de>
Date:   Tue Nov 10 13:36:22 2020 +0100

    tree-optimization/97769 - fix assert in peeling for alignment

    The following removes an assert that can not easily be adjusted to
    cover the additional cases we now handle after the removal of
    the same-align DRs vector.

    2020-11-10  Richard Biener  <rguenther@suse.de>

            PR tree-optimization/97769
            * tree-vect-data-refs.c (vect_update_misalignment_for_peel):
            Remove assert.

            * gcc.dg/vect/pr97769.c: New testcase.

[Bug middle-end/97769] [11 Regression] vectorizer ICE when building perlbench in SPECCPU 2017

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97769

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

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

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

[Bug middle-end/97769] [11 Regression] vectorizer ICE when building perlbench in SPECCPU 2017

[tnfchris at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97769

--- Comment #6 from Tamar Christina <tnfchris at gcc dot gnu.org> ---
Thanks Richi!