[Bug c/109301] New: [13 Regression] ICE in format_helper, at real.h:233

[Bug c/109301] New: [13 Regression] ICE in format_helper, at real.h:233

[gscfq@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109301

            Bug ID: 109301
           Summary: [13 Regression] ICE in format_helper, at real.h:233
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gscfq@t-online.de
  Target Milestone: ---

Started between 20220717 and 20220724, at -Ofast :


$ cat z1.c
double x[256];
void f (void)
{
  int i, j;
  for (i=0; i<256; ++i)
    for (j=0; j<8; ++j)
      x[i] = __builtin_pow (x[i], 0.5);
}


$ gcc-13-20230326 -c z1.c -Ofast
during GIMPLE pass: cunroll
z1.c: In function 'f':
z1.c:2:6: internal compiler error: in format_helper, at real.h:233
    2 | void f (void)
      |      ^
0x94a671 format_helper::format_helper<machine_mode>(machine_mode const&)
        ../../gcc/real.h:233
0x94a671 const_binop
        ../../gcc/fold-const.cc:1340
0x94b027 const_binop(tree_code, tree_node*, tree_node*, tree_node*)
        ../../gcc/fold-const.cc:1761
0x10e23d8 gimple_resimplify2
        ../../gcc/gimple-match-head.cc:280
0x10e258a gimple_match_op::resimplify(gimple**, tree_node* (*)(tree_node*))
        ../../gcc/gimple-match-head.cc:516
0x1206eb8 gimple_simplify_CFN_POW
        .../gcc-13-20230326/zzz_rel/gcc/gimple-match.cc:188875
0x10e1112 gimple_simplify
        .../gcc-13-20230326/zzz_rel/gcc/gimple-match.cc:211319
0x10e22ca gimple_resimplify2
        ../../gcc/gimple-match-head.cc:323
0x10e258a gimple_match_op::resimplify(gimple**, tree_node* (*)(tree_node*))
        ../../gcc/gimple-match-head.cc:516
0x10fd1c0 gimple_simplify_CFN_SQRT
        .../gcc-13-20230326/zzz_rel/gcc/gimple-match.cc:87171
0x12a7aa7 gimple_simplify
        .../gcc-13-20230326/zzz_rel/gcc/gimple-match.cc:95836
0x12a8dc9 gimple_resimplify1
        ../../gcc/gimple-match-head.cc:246
0x10e259a gimple_match_op::resimplify(gimple**, tree_node* (*)(tree_node*))
        ../../gcc/gimple-match-head.cc:514
0x12a8f86 gimple_simplify(gimple*, gimple_match_op*, gimple**, tree_node*
(*)(tree_node*), tree_node* (*)(tree_node*))
        ../../gcc/gimple-match-head.cc:1115
0x99029d fold_stmt_1
        ../../gcc/gimple-fold.cc:6251
0xe234f8 eliminate_dom_walker::eliminate_stmt(basic_block_def*,
gimple_stmt_iterator*)
        ../../gcc/tree-ssa-sccvn.cc:7062
0xe298b7 process_bb
        ../../gcc/tree-ssa-sccvn.cc:8050
0xe2ad95 do_rpo_vn_1
        ../../gcc/tree-ssa-sccvn.cc:8501
0xe2bdc9 do_rpo_vn(function*, edge_def*, bitmap_head*, bool, bool,
vn_lookup_kind)
        ../../gcc/tree-ssa-sccvn.cc:8603
0xd9b03d tree_unroll_loops_completely
        ../../gcc/tree-ssa-loop-ivcanon.cc:1485

[Bug tree-optimization/109301] [13 Regression] ICE in format_helper, at real.h:233

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109301

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|c                           |tree-optimization
   Target Milestone|---                         |13.0

[Bug tree-optimization/109301] [13 Regression] ICE in format_helper, at real.h:233

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109301

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Started with r13-1763-g78d5e125c008d87cb2e1.

 (for sqrts (SQRT)
      cbrts (CBRT)
      pows (POW)
  /* sqrt(sqrt(x)) -> pow(x,1/4).  */
  (simplify
   (sqrts (sqrts @0))
   (pows @0 { build_real (type, dconst_quarter ()); }))
and various other simplifications after it aren't really suitable for floating
point vector types.
Well, we could do them as long as we'd verify that vector POW is actually
implemented and used build_uniform_cst (type, VECTOR_TYPE_P (type) ? TREE_TYPE
(type) : type, dconst_quarter ())); and similar stuff.
But given that no target currently implements pow_optab, I think that would be
a wasted effort.

[Bug tree-optimization/109301] [13 Regression] ICE in format_helper, at real.h:233 since r13-1763

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109301

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
                 CC|                            |jakub at gcc dot gnu.org
   Last reconfirmed|                            |2023-03-27
             Status|UNCONFIRMED                 |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |jakub at gcc dot gnu.org
           Priority|P3                          |P1
            Summary|[13 Regression] ICE in      |[13 Regression] ICE in
                   |format_helper, at           |format_helper, at
                   |real.h:233                  |real.h:233 since r13-1763

[Bug tree-optimization/109301] [13 Regression] ICE in format_helper, at real.h:233 since r13-1763

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109301

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 54774
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=54774&action=edit
gcc13-pr109301.patch

Actually, I think just that single case might be enough, all the others involve
cbrt or pow in the IL before.

[Bug tree-optimization/109301] [13 Regression] ICE in format_helper, at real.h:233 since r13-1763

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109301

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Tried -Ofast -mavx512f
#include <math.h>

void
foo (double *a, double *b)
{
  for (int i = 0; i < 1024; i++)
    a[i] = pow (a[i], b[i]);
}

void
bar (double *a)
{
  for (int i = 0; i < 1024; i++)
    a[i] = cbrt (sqrt (a[i]));
}

void
baz (double *a)
{
  for (int i = 0; i < 1024; i++)
    a[i] = cbrt (cbrt (a[i]));
}
but that doesn't trigger, it uses simd clones rather than internal functions.
Strangely, bar isn't vectorized while foo and baz are.

[Bug tree-optimization/109301] [13 Regression] ICE in format_helper, at real.h:233 since r13-1763

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109301

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #3)
> Tried -Ofast -mavx512f
> #include <math.h>
> 
> void
> foo (double *a, double *b)
> {
>   for (int i = 0; i < 1024; i++)
>     a[i] = pow (a[i], b[i]);
> }
> 
> void
> bar (double *a)
> {
>   for (int i = 0; i < 1024; i++)
>     a[i] = cbrt (sqrt (a[i]));
> }
> 
> void
> baz (double *a)
> {
>   for (int i = 0; i < 1024; i++)
>     a[i] = cbrt (cbrt (a[i]));
> }
> but that doesn't trigger, it uses simd clones rather than internal functions.
> Strangely, bar isn't vectorized while foo and baz are.

I only see foo vectorized (with simdclones), neither bar nor baz are - probably
because cbrt isn't available in vector form and we don't resort to pow()
here for baz because

  /* cbrt(cbrt(x)) -> pow(x,1/9), iff x is nonnegative.  */
  (simplify
   (cbrts (cbrts tree_expr_nonnegative_p@0))
   (pows @0 { build_real_truncate (type, dconst_ninth ()); }))

doesn't trigger.  Later the 'powcabs' pass synthesizes cbrt(sqrt()) again
but at least for me no cbrt vectorized variant is available.

[Bug tree-optimization/109301] [13 Regression] ICE in format_helper, at real.h:233 since r13-1763

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109301

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

https://gcc.gnu.org/g:8dd57939c20b5e218404a838f514429f8e414dea

commit r13-6924-g8dd57939c20b5e218404a838f514429f8e414dea
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Wed Mar 29 08:33:30 2023 +0200

    tree-ssa-math-opts: Move PROP_gimple_opt_math from sincos pass to powcabs
[PR109301]

    The following testcase ICEs since the sincos and vect pass order has
    been swapped.  It is not valid to replace vector sqrt (sqrt (x)) with
    pow (x, 0.25) because build_real on vector type is invalid (could be
    handled by using build_uniform_cst and adjusting type passed to
    build_real) but more importantly because nothing checks if we can
    actually do vector pow.
    While we have pow_optab, apparently no target defines it, so it doesn't
    seem to be worth bothering with for now and the patch just punts on
    non-scalar sqrts.
    I think the other simplifications next to it are fine, as they mostly
    use CBRT which doesn't even have internal function (so is a builtin
    only and therefore always scalar), or have already pow in the IL (which
    doesn't have optab and shouldn't be thus vector either).
    It is true that with <bits/math-vector.h> we do vectorize some calls to
    pow or cbrt (but don't handle others strangely), but those aren't using
    internal functions but simd clones and so match.pd doesn't know anything
    about those (at least for now).

    The following patch fixes it by mostly restoring the state before
    r13-1763 where canonicalize_math_p () was true only until the end of the
    pass which transformed pow or pow-like calls before vectorization (formerly
    sincos pass, now it is powcabs pass).
    powcabs is a pass in the spot sincos was happening before, so the
    only change was defer the sin+cos simplification into cexpi to a later
    new pass (except for the name moving with it) and none of the
    canonicalize_math_p () guarded simplification in match.pd seem to rely
    on those sin+cos -> cexpi simplifications and canonicalize_math_p is
    the only user of this property.

    2023-03-29  Jakub Jelinek  <jakub@redhat.com>
                Richard Biener  <rguenther@suse.de>

            PR tree-optimization/109301
            * tree-ssa-math-opts.cc (pass_data_cse_sincos): Change
            properties_provided from PROP_gimple_opt_math to 0.
            (pass_data_expand_powcabs): Change properties_provided from 0 to
            PROP_gimple_opt_math.

            * gcc.dg/pr109301.c: New test.

[Bug tree-optimization/109301] [13 Regression] ICE in format_helper, at real.h:233 since r13-1763

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109301

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

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

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed.