[Bug target/112359] New: [14 Regression] ICE: in expand_fn_using_insn, at internal-fn.cc:215 with -O -ftree-loop-if-convert -mavx512fp16

[Bug target/112359] New: [14 Regression] ICE: in expand_fn_using_insn, at internal-fn.cc:215 with -O -ftree-loop-if-convert -mavx512fp16

[zsojka at seznam dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112359

            Bug ID: 112359
           Summary: [14 Regression] ICE: in expand_fn_using_insn, at
                    internal-fn.cc:215 with -O -ftree-loop-if-convert
                    -mavx512fp16
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: zsojka at seznam dot cz
  Target Milestone: ---
              Host: x86_64-pc-linux-gnu
            Target: x86_64-pc-linux-gnu

Created attachment 56498
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56498&action=edit
reduced testcase

Compiler output:
$ x86_64-pc-linux-gnu-gcc -O -ftree-loop-if-convert -mavx512fp16 testcase.c
during RTL pass: expand
testcase.c: In function 'foo':
testcase.c:5:1: internal compiler error: in expand_fn_using_insn, at
internal-fn.cc:215
    5 | foo (void)
      | ^~~
0x788f3f expand_fn_using_insn
        /repo/gcc-trunk/gcc/internal-fn.cc:215
0xf4bbcf expand_call_stmt
        /repo/gcc-trunk/gcc/cfgexpand.cc:2737
0xf4bbcf expand_gimple_stmt_1
        /repo/gcc-trunk/gcc/cfgexpand.cc:3880
0xf4bbcf expand_gimple_stmt
        /repo/gcc-trunk/gcc/cfgexpand.cc:4044
0xf5218e expand_gimple_basic_block
        /repo/gcc-trunk/gcc/cfgexpand.cc:6100
0xf53e67 execute
        /repo/gcc-trunk/gcc/cfgexpand.cc:6835
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.

$ x86_64-pc-linux-gnu-gcc -v
Using built-in specs.
COLLECT_GCC=/repo/gcc-trunk/binary-latest-amd64/bin/x86_64-pc-linux-gnu-gcc
COLLECT_LTO_WRAPPER=/repo/gcc-trunk/binary-trunk-r14-5093-20231103113523-gc3847ca0571-checking-yes-rtl-df-extra-nobootstrap-amd64/bin/../libexec/gcc/x86_64-pc-linux-gnu/14.0.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: /repo/gcc-trunk//configure --enable-languages=c,c++
--enable-valgrind-annotations --disable-nls --enable-checking=yes,rtl,df,extra
--disable-bootstrap --with-cloog --with-ppl --with-isl
--build=x86_64-pc-linux-gnu --host=x86_64-pc-linux-gnu
--target=x86_64-pc-linux-gnu --with-ld=/usr/bin/x86_64-pc-linux-gnu-ld
--with-as=/usr/bin/x86_64-pc-linux-gnu-as --disable-libstdcxx-pch
--prefix=/repo/gcc-trunk//binary-trunk-r14-5093-20231103113523-gc3847ca0571-checking-yes-rtl-df-extra-nobootstrap-amd64
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 14.0.0 20231103 (experimental) (GCC)

[Bug target/112359] [14 Regression] ICE: in expand_fn_using_insn, at internal-fn.cc:215 with -O -ftree-loop-if-convert -mavx512fp16

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112359

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2023-11-03
             Status|UNCONFIRMED                 |NEW
                 CC|                            |rdapp at gcc dot gnu.org,
                   |                            |rguenth at gcc dot gnu.org
     Ever confirmed|0                           |1
           Priority|P3                          |P1
   Target Milestone|---                         |14.0

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  We get

_ifc__24 = .COND_IOR (_30, u_lsm.8_6, ivtmp.16_8, u_lsm.8_6);

we shouldn't have generated these outside of if (vectorized) versions or when
not vectorizing.  I see we discover this opportunities "late", after deciding
whether to version the loop or not.  We have to refrain from taking the
opportunity when we didn't version.

[Bug middle-end/112359] [14 Regression] ICE: in expand_fn_using_insn, at internal-fn.cc:215 with -O -ftree-loop-if-convert -mavx512fp16

[rdapp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112359

--- Comment #2 from Robin Dapp <rdapp at gcc dot gnu.org> ---
Would something like

+  bool allow_cond_op = flag_tree_loop_vectorize
+    && !gimple_bb (phi)->loop_father->dont_vectorize;

in convert_scalar_cond_reduction be sufficient or are the more conditions to
check still?  Testing with the above and the reduced testcase worked FWIW.

[Bug middle-end/112359] [14 Regression] ICE: in expand_fn_using_insn, at internal-fn.cc:215 with -O -ftree-loop-if-convert -mavx512fp16

[rguenther at suse dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112359

--- Comment #3 from rguenther at suse dot de <rguenther at suse dot de> ---
On Mon, 6 Nov 2023, rdapp at gcc dot gnu.org wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112359
> 
> --- Comment #2 from Robin Dapp <rdapp at gcc dot gnu.org> ---
> Would something like
> 
> +  bool allow_cond_op = flag_tree_loop_vectorize
> +    && !gimple_bb (phi)->loop_father->dont_vectorize;
> 
> in convert_scalar_cond_reduction be sufficient or are the more conditions to
> check still?  Testing with the above and the reduced testcase worked FWIW.

You want, in tree_if_conversion (), remember whether we versioned the
loop as part of

  if (need_to_lower_bitfields
      || need_to_predicate
      || any_complicated_phi
      || flag_tree_loop_if_convert != 1)
    {
...

and pass that as flag to combine_blocks and down to
predicate_all_scalar_phis (or add a new global flag).  Ideally we'd
do the convert_scalar_cond_reduction analysis beforehand and force
versioning for the case we'd generate a .COND_* instead of deciding
at code-generation only (when it's too late to do the required
versioning).

[Bug middle-end/112359] [14 Regression] ICE: in expand_fn_using_insn, at internal-fn.cc:215 with -O -ftree-loop-if-convert -mavx512fp16

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112359

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

https://gcc.gnu.org/g:fd940d248bfccb6994794152681dc4c693160919

commit r14-5231-gfd940d248bfccb6994794152681dc4c693160919
Author: Robin Dapp <rdapp@ventanamicro.com>
Date:   Mon Nov 6 11:24:37 2023 +0100

    vect/ifcvt: Add vec_cond fallback and check for vector versioning.

    This restricts tree-ifcvt to only create COND_OPs when we versioned the
    loop for vectorization.  Apart from that it re-creates a VEC_COND_EXPR
    in vect_expand_fold_left if we emitted a COND_OP.

    gcc/ChangeLog:

            PR tree-optimization/112361
            PR target/112359
            PR middle-end/112406

            * tree-if-conv.cc (convert_scalar_cond_reduction): Remember if
            loop was versioned and only then create COND_OPs.
            (predicate_scalar_phi): Do not create COND_OP when not
            vectorizing.
            * tree-vect-loop.cc (vect_expand_fold_left): Re-create
            VEC_COND_EXPR.
            (vectorize_fold_left_reduction): Pass mask to
            vect_expand_fold_left.

    gcc/testsuite/ChangeLog:

            * gcc.dg/pr112359.c: New test.

[Bug middle-end/112359] [14 Regression] ICE: in expand_fn_using_insn, at internal-fn.cc:215 with -O -ftree-loop-if-convert -mavx512fp16

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112359

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

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

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